TIME DOMAIN RESPONSES OF GLASSY POLYMERS by SHANKAR KOLLENGODU SUBRAMANIAN A DISSERTATION IN CHEMICAL ENGINEERING Submitted to the Graduate Faculty of Texas Tech University in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY Approved 111111111111111111111111111111 Gregory B. McKenna (Chairperson of the Committee) 111111111111111111111111111111 Sindee L. Simon 111111111111111111111111111111 Brandon Weeks 111111111111111111111111111111 Edward L. Quitevis 111111111111111111111111111111 Peggy Gordon Miller Dean of the Graduate School May, 2011 Copyright 2011 Shankar Kollengodu Subramanian Texas Tech University, Shankar Kollengodu Subramanian, May 2011 ACKNOWLEDGEMENT I would like to thank my advisor and chair of my committee Prof. Gregory B.McKenna for his role in mentoring, guiding and encouraging me. He provided a great environment to learn and ample space for me to think and grow. I would also like to extend my gratitude to Prof Sindee Simon for her positive words and valuable feedback on my dissertation. Further, I would like to extend my thanks to Prof Edward Quitevis and Brandon Weeks for agreeing to be part of my committee and for their insightful inputs. I would like to appreciate the encouragement and camaraderie of all my colleagues and coworkers at Texas Tech University. I would like to recognize Dr. Prashanth Bardinarayanan for his words of encouragement and Dr. Paul O Connell, Babji Srinivasan, Dr. Lameck Banda, Dr. Mataz Alcoutlabi, and Jing Zhao for helping me with my experiments. I would like to thank my mother, sisters, in-laws and my entire family for their unconditional love, belief and support. Finally I would like to acknowledge my wife, Harini for being my cheer leader. I would like to dedicate this thesis to my late uncle Dr Natrajan, who has been one of my inspirations for becoming a Chemical Engineer. ii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 TABLE OF CONTENTS ACKNOWLEDGEMENTS……………………………………………………….......ii ABSTRACT……………………………………………………………………...….viii LIST OF TABLES………………………………………………………………….....x LIST OF FIGURES…………………………………………………………………..xi CHAPTER 1 INTRODUCTION…………………………………………………………….1 1.1 Glassy Phenomena……………………………………………..……...1 1.2 Dielectric responses in glassy polymers……………………………....2 1.3 Structural recovery and aging in glassy polymers…………………….6 1.3.1 Intrinsic isopiestics…………………………………………....6 1.3.2 Memory effect ………………………………………………...7 1.3.3 Asymmetry of approach………………………………………7 1.3.4 τeff paradox……………………………………………………8 1.3.5 Physical aging………………………………………………...9 2 1.4 Effect of plasticizer in glassy behavior……………………………….9 1.5 References…………………………………………………………...12 EXPERIMENTAL SYSTEM ……………………………………………….29 2.1 Time domain dielectric spectrometer……………………………..…29 iii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 2.2 Experimental setup for studying the aging and structural recovery responses of glassy polymers subjected to plasticizer environment………..31 3 2.3 Linear Variable Differential Transformer calibration………………32 2.4 References…………………………………………………………..33 A DIELECTRIC STUDY OF POLY (VINYL ACETATE) USING A PULSEPROBE TECHNIQUE………………………………….…………………..39 3.1 Motivation…………………………………………………………..39 3.2 Introduction………………………………………………………....39 3.3 Sample preparation..………………………………………………...42 3.4 Method of analysis………………………..…………………………43 3.5 Results ………………………..……………………………………...45 3.5.1 Single step response………..………………………………..45 3.5.2 Time-frequency conversion……………...………………….48 3.5.3 Two step (pulse-probe) response………...………………….49 3.6 Discussion…………………………………………..……………….50 3.7 Conclusion…………...……………………………………………...52 3.8 Appendix……..……………………………………………………...53 3.8.1 Algorithm to obtain true compliance………………...……...53 iv Texas Tech University, Shankar Kollengodu Subramanian, May 2011 3.8.2 Boltzmann superposition algorithm to delineate linear and nonlinear behaviors………………….……………….……..57 3.9 4 References……………………………………………..…………...60 APPLICATION OF EMPIRICAL MODE DECOMPOSITION IN THE FIELD OF POLYMER PHYSICS................................................................78 4.1 Motivation..........................................................................................78 4.2 Introduction.......................................................................................79 4.3 Methodology......................................................................................86 4.3.1 EMD based FFT filerting algorithm..........................................87 4.4 Results and discussion........................................................................88 4.4.1 Simulation studies..................................................................88 4.4.2 Filtering of the simulated data...............................................90 4.4.3 Discussion of results obtained from filtering methods..........96 4.4.4 Experimental case study I......................................................97 4.4.5 Experimental study II............................................................99 4.4.6 Practical benefits due to use of the proposed EMD based FFT filtering algorithm............................................................................103 4.5 Conclusions.....................................................................................106 4.6 Appendix.........................................................................................107 v Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4.6.1 Mathematical details of various filtering techniques...........107 4.7 5 References........................................................................................110 AGING AND STRUCTURAL RECOVERY BEHAVIORS IN EPOXY FILMS SUBJECTED TO CARBON DIOXIDE PLASTICIZATION JUMPS: EVIDENCE FOR A NEW GLASSY STATE……………………………….……138 5.1 Motivation…………………………………………….…………...138 5.2 Introduction……………………………………..…………….…..138 5.3 Experimental……………….…………………………….………..140 5.4 Method of analysis………………………………..……………….141 5.5 Results and discussion……………………………………..……...142 5.5.1 Structural recovery experiments…………………..………142 5.5.2 Aging experiments………………………………………...143 5.6 Conclusions……………………………………………..…………146 5.7 References……………………………………..………………….147 6 SUMMARY AND CONCLUSIONS …………………………………….166 7 FUTURE WORK………………………………..………………………..170 7.1 Introduction ……………………………………………………..170 7.2 Time domain nonlinear dielectric……………………………......170 vi Texas Tech University, Shankar Kollengodu Subramanian, May 2011 7.3 Isobaric time domain dielectric measurements……………..……171 7.4 Physical aging and structural recovery of glassy polymers……...172 7.5 References……………………………………………..………...173 vii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 ABSTRACT The properties of glassy polymers are generally studied in the vicinity of the glass transition temperature using frequency and time domain responses. In this work, we focus on the dielectric, the structural recovery, and aging responses in the time domain. A proper understanding of these properties is essential for a better prediction of the performances of polymers. Time domain dielectric measurement provides a powerful means to study the glassy responses in short times (as low as micro seconds) to long times (>200 seconds) in a single measurement device. We built a Time Domain Dielectric Spectrometer (TDS) in our laboratory at Texas Tech to take advantage of this ability. Successful working of the spectrometer was demonstrated by studying the dielectric response of poly (vinyl acetate) in a pulse-probe experiment. We see memory effect and this was in quantitative agreement with linear Boltzmann superposition for small applied fields. However, evidence of breakdown of linearity was observed at larger applied fields. This is the first demonstration in time domain dielectric measurement of the ability to delineate between linear and nonlinear behaviors. Noise in the TDS set up was a hindrance for performing more sensitive experiments. The dielectric data obtained from TDS was non stationary in nature (with respect to time) and corrupted with nonlinear noise. Popular methods like fast Fourier transform or moving average are not entirely suitable to handle this type of noise. We introduce a filtration tool called Empirical Mode Decomposition (EMD) to improve the data analysis for these types of data corrupted with noise. An advantage of this method is viii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 that it aids in the improvement of the experimental setup by giving useful information on the noise. EMD based filtering was also applied to the data obtained from structural recovery experiments which are the other time domain response studied in this work. Structural recovery and aging experiments of glassy polymers are very well understood for temperature formed glasses compared to concentration formed glasses. Previous work from our group has shown that concentration formed glasses qualitatively mimic temperature formed glasses but were quantitatively different. Further, our preliminary work on the structural recovery of an epoxy film subjected to CO2 plasticizer jumps showed that the effective retardation time for concentration formed glass and temperature formed glass (subjected to same final condition) do not converge to the same point as equilibrium is approached. This result was unexpected; as we had hypothesized that both concentration and temperature formed glasses come to the same apparent equilibrium state. Hence, we further investigated this behavior by studying the aging and structural recovery of epoxy film subjected to CO2 plasticizer jumps. We observe evidence for the existence of a new metastable glassy state. ix Texas Tech University, Shankar Kollengodu Subramanian, May 2011 LIST OF TABLES 3.1 Fit parameters for equation 4 at 308 K. Errors represent standard error of estimate on the fit parameters……………………………………………………….. 64 3.2 A, B and T0 from VFT fits for the retardation time and the dc conductivity term vs. temperature obtained by fitting the data in Figure 2a to equation 4. Errors represent standard error of estimate on the VFT fit parameters……………...……... 64 4.1 Sum squared error between the MKWW model data and filtered data obtained from various filtering approaches…………………………………………...………116 5.1 KWW parameters for the longest aging time (230400s) of T and P jump experiments…………………………………………………………………….…...152 x Texas Tech University, Shankar Kollengodu Subramanian, May 2011 LIST OF FIGURES 1.1 Specific volume vs Temperature plot for different cooling rates (q). Here, q1> q2>q3 ………………………...………………………………………………………17 1.2 Classifications of dipoles in amorphous polymers………………...…………18 1.3 Schematic representation of two different sources of non-exponential correlation decays……………………………….…………………………………...19 1.4 Schematic of the protocol for dielectric hole burning experiment……...……20 1.5 Time dependent dielectric permittivity of propylene carbonate at 157.4K......21 1.6 Empirical mode decomposition schematic to filter noise in time domain responses………………………..……………………………………………………22 1.7 (a) Schematic of intrinsic isotherm experiment (b) Intrinsic isotherm experiment performed by Kovacs on PVAc ………………………………………...23 1.8 (a) Schematic of memory effect experiment (b) Memory effect experiment of PVAc performed by Kovacs………………………………..………………………..24 1.9 (a) Schematic for asymmetry of approach experiment (b) Asymmetry of approach experiment of PVAc performed by Kovacs…………………………...…..25 1.10 Schematic of Struik’s protocol……………………………..………………..26 xi Texas Tech University, Shankar Kollengodu Subramanian, May 2011 1.11 Creep compliance of poly(vinyl chloride) quenched from above to below Tg at different aging times after quenching…………...……………………………………27 1.12 Effective retardation time plot for temperature and concentration formed glasses to the same final condition….………………………………………………..28 2.1 Schematic of time domain dielectric spectrometer built at Texas Tech taken from reference ………………………………………………………….……………34 2.2 Sample setup schematic for the time domain dielectric spectrometer…..…...35 2.3 The creep apparatus that was built in our laboratory to perform experiments under different CO2-pressure and temperature conditions…………………………..36 2.4 Pressure vessel used to perform the experiments under CO2 …………...…..37 2.5 LVDT calibration plot where the displacement of the core is measured as a function of voltage……………………..……………………………………………38 3.1 (a) Isothermal measurement and (b) Master curve for PVAc with 308 K as reference temperature ………………………………………………………..……..65 3.2 Equation 3.4 parameters ε1 and ε2 as a function of temperature…...………..66 3.3 Dielectric retardation time τ and dc conductivity τc term plotted as a function of inverse temperature. The solid lines represent the VFT fits………………..……67 xii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 3.4 (a) Dielectric recoverable compliance and (b) master curve for PVAc with 308 K as reference temperature………..………………………………………….……68 3.5 Time-temperature shift factors as a function of temperature for PVAc. Comparison of current dielectric results with literature reports for dielectric and mechanical behaviors………………………………..…………………………….69 3.6 Dielectric storage compliance response for PVAc………………..………70 3.7 Dielectric loss compliance of PVAc ………………………..…………….71 3.8 Comparison of apparent dielectric compliance with true dielectric compliance for an applied electric field of 5.4*105Vm-1 ………………………………….…..72 3.9 Dielectric compliance response in two step pulse-probe experiments with varying time duration t1 of the first step ………………………………………….73 3.10 Dielectric compliance response in two step pulse-probe experiments with varying electric field E1 of the first step………………………………….……….74 3.11 Dielectric strain response for the 1second jump of Figure 8. The solid line represents the Boltzmann prediction……………………………………………...75 3.12 Single step responses for applied electric field of 44.6*105 Vm-1 and 89.2*105 Vm-1 …………………………………………………………………………...…76 xiii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 3.13 Two step dielectric compliance response (jump from 89.2*105 Vm-1 to 44.6*105 Vm-1) with linear Boltzmann prediction showing deviation from linear behavior…………………………………………………………………………77 4.1 Data generated using the MKWW model (without noise)……………..117 4.2 Simulated noise added to the data generated from the MKWW model (equation 4.4)…..………………………………………………………………118 4.3 Autocorrelation function of the white noise...........................................119 4.4 Simulated MKWW model data with noise n(t) added………………..120 4.5 MA filtered noisy model data – Window size 5................................121 4.6 MA filtered noisy model data – Window size 10..............................122 4.7 FFT filtered noisy model data...............................................................123 4.8 Schematic of DWT for the simulated data............................................124 4.9 Filtered data using Discrete Wavelet Transform – Five level decomposition...............................................................................................125 4.10 Filtered data using Discrete Wavelet Transform – Six level decomposition....................................................................................................126 4.11 IMFs and their corresponding magnitude at various frequencies obtained from FFT for the simulated data with noise.................................................127 4.12 Filtered data using EMD based FFT approach......................................128 xiv Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4.13 Squared error comparison plots for the filtered data obtained using FFT, DWT and EMD based FFT Approach…………………………………………….…129 4.14 (a) Zero voltage time domain data, (b) Magnitude spectrum of the data at various frequencies using FFT...........................................................................130 4.15 Experimental dielectric time domain data.........................................131 4.16 IMFs and their corresponding magnitude at various frequencies obtained for the experimental dielectric data………………………..……………………...132 4.17 Comparison of experimental dielectric data and filtered data from EMD based FFT approach (Inset shows a zoomed version)..................................................133 4.18 Comparison of volume recovery data after performing a down jump experiment from 85oC to 75oC and 72oC without filtering............................134 4.19 Comparison of volume recovery data after performing a down jump experiment from 85oC to 75oC and 72oC after filtering using EMD method.135 4.20 IMFs and their corresponding magnitude at various frequencies obtained for the experimental volume recovery after a down jump in temperature to 75o.....136 4.21 MKWW generated data after removing the noise at 14Hz........................137 5.1 Schematic representation of specific volume as a function of temperature or concentrations……………………..……………………………………………153 xv Texas Tech University, Shankar Kollengodu Subramanian, May 2011 5.2 A comparison of departure from equilibrium as a function of time for T jump and P jump experiment subjected to same final condition of 72oC and 0 MPa……152 5.3 Effective retardation time as a function of departure from equilibrium for T and P jump experiments of same final condition…………………………………..155 5.4 Creep compliance curves for different aging time plotted as a function of time for T jump experiment………………………………………………………….......156 5.5 Time-aging time superposition of creep curves of the T jump experiment…………………………………………………………………………..157 5.6 Creep compliance curves for different aging time plotted as a function of time for P jump experiment……………………………………………………………......................158 5.7 Time-aging time superposition of the creep curves of P jump experiment…159 5.8 Creep compliance curves for different aging time for T and P jump experiments subjected to same final condition 0MPa and 69.3oC…………...……..160 5.9 Time aging time superposition curves for P jump experiment superposed to the longest aging time of T jump experiment…………………………………….…161 5.10 Horizontal shift factor as a function of aging time. The concentration glasses are shifted with respect to the longest aging time of temperature formed glass……162 xvi Texas Tech University, Shankar Kollengodu Subramanian, May 2011 5.11 Retardation time obtained from KWW function for T and P jump creep curves plotted against the aging………………………………………………...…………..163 5.12 Retardation time as function of aging time for humidity and T jump experiments of same final condition taken from reference ………………………...164 5.13 Volume recovery of epoxy film showing the reversal of concentration formed glass to temperature formed glass upon heating above its Tg followed by cooling to room temperature and heating………………………………………….…………...165 7.1 Schematic of time domain dielectric spectrometer with modifications to perform isobaric measurements of dielectric compliance………….……………….174 xvii Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 1 INTRODUCTION 1.1 Glassy Phenomena Glassy behavior of amorphous polymers is one of the most widely studied topics in Applied Sciences and Engineering for the last half a century. When amorphous polymers are cooled from a higher temperature, the thermodynamic properties such as volume, enthalpy or entropy deviate from the equilibrium path. The point at which the thermodynamic property (volume, entropy or enthalpy) depart from equilibrium path is called the glass transition temperature (Tg) [1-3]. As shown in the Figure 1.1, the glass transition temperature is rate dependent and hence can be looked upon as a kinetic phenomenon [4, 5]. However, there are certain aspects of the glassy behavior which support the idea of thermodynamic origin [5-7]. The question of glassy behavior being a thermodynamic or kinetic phenomenon is not our area of interest in this work. For the purpose of clarity, as stated by McKenna, we can consider glassy behavior as a “kinetic phenomena with underlying thermodynamic transition” [2]. Understanding the properties of amorphous (glassy) polymers in the vicinity of, as well as below the glassy transition temperature is very important to predict their long term performance and stability [8-10]. The glassy behaviors are generally studied in both time and frequency domain. In this work, we primarily focus on 1 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 studying two time domain responses of glassy polymers, namely dielectric, and structural recovery and aging responses. 1.2 Dielectric responses in glassy polymers Dielectric spectroscopy is a powerful method to study the glassy behavior of materials spanning 10 decades in both frequency and time response using a single measurement device, and more than 18 decades using multiple devices [11]. They work on the principle of measuring the polarization (dipole moment) of the material when subjected to an applied field (current or voltage) [11-13]. Dipole is a pair of electrical charges of “equal magnitude and opposite polarity” and dipole moment is defined as the vector quantity obtained by the “product of the magnitude of one of the poles and the distance separating the two poles” [14]. Polymers in general are weakly polar when compared to small molecule glass formers. However, based on the dipole’s arrangement they are classified as Type A, Type B and Type C, as shown in the Figure 2 [15]. In Type A polymers, relatively strong dipole moments typically occur along the backbone of the polymer chain. They are characterized by a slow relaxation mode with strong molecular weight dependence [15]. In Type B polymers, the dipole moments occur in the main chain, but perpendicular to the backbone chain. They are characterized by a fast segmental relaxation with no molecular weight dependence [15]. Type C polymers have the weakest dipole moments and they occur along the side chains of the polymers. They also show no molecular weight dependence like Type B polymers [15]. The material investigated in this work is poly (vinyl acetate) 2 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 (PVAc) which is a Type B dipole. PVAc is a model polymer used for dielectric studies of segmental relaxation. Most of the work on PVAc has been performed in frequency domain (linear and nonlinear studies) [16, 17] with limited work in time domain responses [18]. However, it may be noted that the dielectric responses can be transformed from time domain to frequency domain and vice versa using Laplace transformation in the linear regime [11, 19]. The relationship between dielectric time and frequency domain responses is given by equation1.1 and 1.2 [11, 18, 19]. = − " (1.1) = + " (1.2) Where ε(t) is the dielectric compliance, ε’(ω) is the dielectric storage compliance, ε”(ω) is the dielectric loss compliance, M(t) is the dielectric modulus, M’(ω) is dielectric storage modulus and M” (ω) is the dielectric loss modulus. Furthermore, the dielectric compliance and modulus can also be transformed from one form to the other using equations similar to that used for conversion between creep compliance and shear modulus [11, 20]. In general, most of the dielectric measurements reported in the literature are performed in frequency domain, despite time domain measurements being much faster compared to frequency sweeps because of the difficulty is filtering noises in time domain data [11, 16, 17, 21]. However, the potential of using time domain dielectric spectrometry (TDS) to study the glassy responses in polymers is tremendous. Probing dynamic heterogeneity is one of the potential areas of study which could be tapped using TDS. As shown in Figure 1.3, the overall macroscopic 3 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 picture of a homogeneous and heterogeneous material is same, although microscopically the each spatial domain in a heterogeneous material can be different [22]. This behavior can be well understood by subjecting these materials to a high sinusoidal frequency probe, wherein particular domain in the material relaxes slower or faster compared to the macroscopic response of the material. This is known as dynamic heterogeneity [22]. Hole burning experiment is an excellent technique to demonstrate the dynamic heterogeneity in materials [23, 24]. Figure 1.4 is the protocol for the hole burning experiment. It comprises of 4 experimental steps. In step 1, a positive sinusoidal probe of high frequency is applied, followed by a suitable waiting time and a small voltage step. In step 2, the same experiment is repeated with a negative pulse. Then, the signals from step 1 and step 2 are summed to give the modified response. Step 3 and step 4 are the positive and negative pulse respectively without the sinusoidal probe. Summation of the signals from step 3 and 4 is called the unmodified response. If there is no difference between the modified and unmodified response then the material is homogeneous; else it is the evidence for dynamic heterogeneity [23, 24]. Bohmer and coworkers, [23] and Richert and coworkers [24] have demonstrated the hole burning technique to probe dynamic heterogeneity in small molecule glass formers. Figure 1.5, is the result of a hole burning experiment of propylene carbonate showing horizontal and vertical holes from Bohmer’s group [23]. Shi and McKenna, [25] and Qin and McKenna [26] have also probed dynamic heterogeneity using mechanical hole burning experiments for small molecule glass formers and polymers. It may be noted that the small molecule glass formers used in 4 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 these dielectric studies have a larger signal to noise ratio compared to polymers. This is precisely the challenge we have currently in probing the dynamic heterogeneities in glassy polymers. We have demonstrated in our work on time domain dielectric responses (Chapter 3) that using the simple Boltzmann superposition principle, linear and nonlinear behaviors can be delineated [27]. It is a progress in the direction of tapping the potential in TDS experiments. However, noise in the data was proving detrimental for performing more sensitive experiments like the hole burning experiments with polymers as the signal to noise ratio is very low to sense holes as discussed above. Time domain dielectric data are time varying in nature and are corrupted with nonlinear noise. Widely popular filtration techniques like Fast Fourier Transform (FFT), and Moving Average (MA) filters are not appropriate for filtering these types of data [28]. We found out that the Empirical Mode Decomposition (EMD) with FFT based algorithm was an effective method to filter non stationary data corrupted with nonlinear noise. EMD filter works on the principle of splitting the given signal into various individual components in time domain called intrinsic mode functions (IMF), using in-situ generated cubic splines as shown in Figure 1.6 [ 28, 29]. Then the frequency information of each IMF component is obtained using FFT algorithm. Based on the prior knowledge of these experiments, we can then determine whether that IMF should be removed or retained. IMFs deemed as noise components are omitted and the signal is reconstructed using other IMFs. The biggest advantage of the EMD based approach is that we do not lose any information about the actual data 5 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 besides giving valuable information on the noise, which could be further used to improve the instrumentation wherever possible [28, 29]. The EMD based approach is covered more elaborately in chapter 4 of this thesis work. With improved data analysis using EMD based filtration; there is a potential to expand the horizon of time domain dielectric spectroscopy to probe the nonlinearity and dynamic heterogeneity in glassy polymers. Below is another study in time domain responses of glassy polymers, which has also been benefited from using EMD based filtration. 1.3 Structural recovery and aging in glassy polymers On cooling amorphous glassy polymers from above to below the glass transition point, the sample moves into a non equilibrium state and hence will tend to evolve back towards the equilibrium state. The study of thermodynamic properties (volume, enthalpy or entropy) as the material evolves towards equilibrium is called structural recovery [7] and the changes associated with the viscoelastic properties such as mechanical, optical or dielectric during this process is called physical aging [7]. Kovacs was the first to comprehensively demonstrate the structural recovery phenomenon using three classic experiments namely the intrinsic isotherm, the memory effect and the asymmetry of approach [7]. 1.3.1 Intrinsic isotherm Figure 1.7(a) is the schematic of the intrinsic isotherm experiment. When a glassy material is cooled from above the glass transition temperature (Tg) to a certain 6 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 temperature Ta below the glass transition temperature and held there isothermally, the thermodynamic property (volume, enthalpy or entropy) will slowly recover towards equilibrium. The further the departure from Tg , the longer it takes to reach the equilibrium. This could be well understood in terms of molecular mobility and free volume. As we go deeper into the glassy regime, the free volume, and hence the molecular mobility, decreases. Figure 1.7(b) is the intrinsic isotherm plot of PVAc, performed by Kovacs in his classic structural recovery experiment, after quenching from 40oC to different temperatures (Ta) [7]. 1.3.2 Memory effect Memory effect is a two step experiment as shown in Figure 1.8(a). Initially, the material is cooled from above the glass transition temperature to Point 1 far below Tg and partially aged for some time. After that, the material is then jumped to Point 2 (up jump), such that the departure to equilibrium is very close to zero and aged. Instead of directly returning to equilibrium, the sample goes through a maximum before reaching the equilibrium remembering the previous thermal history. It may be noted that the farther the first jump (point 2) is, the bigger is the memory. The memory effect is evidence that it needs more than one exponential function to capture the spectrum of data. Figure 1.8(b) is the memory experiment on PVAc performed by Kovacs [7]. 1.3.3 Asymmetry of approach Figure 1.9(a) is the schematic of the asymmetry of approach experiments. It comprises of two experiments, namely, up jump and down jump experiments. In 7 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 down jump experiment, a jump in temperature is made from Point A closer to glassy transition temperature, to Point B further away in the glassy regime. In up jump experiments, a jump is made from Point C (well below Point B) to point B. It may be noted that the magnitude of temperature jump for the up jump and down jump experiments should be the same. On measuring the departure from equilibrium for both cases, we can observe that the up jump and down jump responses are nonlinear. The down jump experiment behaves like an auto retardation experiment such that initially the recovery is faster. This is due to the higher mobility as a result of the departure from equilibrium (δ) being greater than zero. In time, the recovery slows down as the molecular mobility decreases. In the up jump experiments, the reverse happens and it behaves like an auto catalytic experiment. Initially, the recovery is slower as δ < 0. In time, the recovery accelerates due to the increase in molecular mobility as δ approaches zero. Figure 1.9 (b) is the asymmetry of approach experiment as performed by Kovacs on PVAc [7]. 1.3.4 τ eff paradox Kovacs’ initial work on the effective retardation time, which is the slope of departure from equilibrium (δ) as a function of time plot for different magnitudes of asymmetry of approach experiments, showed that the material doesn’t come to equilibrium at the same time as the departure from equilibrium approaches zero. This was considered a paradox because the material violates the fundamental law that the equilibrium is path independent [7]. McKenna and his coworkers later reanalyzed some of Kovacs’ original data, and concluded that the expansion gap and error in 8 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 measurements close to equilibrium could, to some extent, be the reason for the paradox [30]. Simon and coworkers have shown in their volume recovery work, that for a smaller δ, the paradox is resolved as the expansion gap vanishes [31]. Hence, the expansion gap occurs only for large T-jumps, i.e., when the response is nonlinear. 1.3.5 Physical Aging According to Struik’s protocol, the sample loading-unloading for the aging experiment is performed in such a way that the loading-unloading time is one tenth of the waiting time. This is done so that each loading-unloading step is independent of the previous event [8]. As seen in the Figure 1.10, the loading-unloading time is sequentially increased by a factor of 2. Figure 1.11 is the first demonstration of aging experiment performed by Struik [2] on poly (vinyl chloride). He showed that the creep curves at various aging times can be superimposed, similar to time temperature superposition and is called time-aging time superposition [2]. 1.4 Effect of plasticizers on glassy behavior Plasticizers are small molecules like moisture or carbon dioxide, which cause considerable changes in the properties of glassy polymers upon constant exposure [32-34]. Understanding these behaviors can add to exploring new areas of application as well as preventing unexpected material failures [35, 36]. Despite a considerable amount of work on the effect of plasticizers on polymers in general, there are very few studies in understanding the glassy behavior in terms of structural recovery and aging [37-41]. In our group, there has been tremendous impetus in this aspect. Previous work from our group has shown that concentration formed glasses using a 9 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 strongly polar plasticizer (H2O) and a weakly polar plasticizer (CO2) qualitatively mimic temperature formed glasses but were quantitatively different [37-41]. The concentration formed glasses had a larger departure from equilibrium than temperature formed glasses. Further, our preliminary work on the structural recovery of epoxy film subjected to CO2 plasticizer jumps, showed that the effective retardation time for concentration formed glasses and temperature formed glasses (subjected to same final condition) do not converge to the same point as equilibrium is approached [see Figure 1.12]. This is similar to the τeff paradox observed by Kovacs. This problem is discussed more elaborately in chapter 5. This dissertation is organized as follows: In chapter 1, the background of glassy phenomena, dielectric spectroscopy, its linear and nonlinear behavior in time domain responses, correlation between frequency and time domain responses, kinetic manifestation of glassy behaviors and impact of plasticizers in aging and recovery responses are discussed. In chapter 2, the experimental set up used for the time domain dielectric spectrometer and the pressure vessel set up for aging recovery experiments are discussed. Chapters 3, 4 and 5 are elaborated versions of publications in the peer reviewed journals. In chapter 3, we present an investigation of the dielectric behavior of poly(vinyl acetate) (PVAc) using a two step pulse-probe technique. Time domain dielectric experiments were performed in the vicinity of the glass transition temperature. After establishing the linear response function in single step 10 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 experiments, two types of pulse-probe experiments were performed. In one, the time duration t1 of the first step in the probe was varied. In the second case, the magnitude of the field E1 applied to the sample for the first step was varied. We observe the memory effect and the responses were analyzed in the context of a linear Boltzmann rule. Evidence of deviations from linear superposition at the highest electric fields are also presented. Noise in the dielectric experimental data limited the application of the time domain dilectric set up built at Texas Tech. In chapter 4, we propose an algorithm for effective filtering of noise using an EMD based FFT approach for applicatons in polymer physics. The advantages of the proposed approach are: (i) it uses the precise frequency information provided by the FFT and therefore efficiently filters a wide variety of noise and, (ii) the EMD approach can effectively obtain IMFs from both non-stationary as well as nonlinear experimental data. The utility of the proposed approach is illustrated using an analytical model and also through two typical laboratory experiments, namely, the dilectric experiments and structural recovery experiments in polymer physics, wherein the material response is nonstationary; standard filtering approaches are often inappropriate in such cases. By taking advantage of the EMD based filtering technique, in chapter 5, we investigate the structural recovery and physical aging of an epoxy film subjected to carbon dioxide pressure jumps and compare the results with temperature jump experiments, such that the final conditions are identical. This a continuation of work done previously in our group where we have shown using strong and weakly polar plasticizers, that they qualitatively mimic the behaviors of temperature jumps but 11 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 quantitatively they are different [8-10, 13] . In chapter 6, we summarize the conclusion of this work and the future scope is discussed in chapter 7. 1.5 References 1. Plazek DJ, Ngai KL. In: Mark JE, editor. Physical properties of polymers handbook. NY: American Institute of Physics; 1996 (chapter 12). 2. McKenna GB. In: Booth C, Price C, editors. Comprehensive Polymer Science, Polymer Properties, Vol.2, Oxford: Pergamon Press, 1989 (chapter 2). 3. McKenna GB, Simon SL. In: Cheng SZD, editor. Handbook of Thermal Analysis and Calorimetry, Applications to Polymers and Plastics, Vol.3, Elsevier Science, 2002. 4. Kovacs AJ. Transition vitreuse dans les polymères amorphes. Etude phénoménologique. Adv Polym Sci 1964; 3:394. 5. Kovacs AJ. La contraction isotherme du volume des polymères amorphes. J.Polym Sci. 1958; 30:131. 6. Chang SS. Thermodynamic properties and glass transition of polystyrene. J.Polym Sci, Poly. Sympo 1984; 71:59. 7. Kovacs AJ. Transition vitreuse dans les polymères amorphes. Etude phénoménologique. Fortschr. Hochpolym.-Forsch. 1963; 3:394. 8. Struik LCE. Physical aging in polymer and other amorphous materials. Elsevier: Amsterdam, 1978 9. McKenna GB. On the physics required for the prediction of long term performance of polymers and their composites. J. Res. NIST 1994; 99:169. 12 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 10. Alcoutlabi M, McKenna GB, Simon SL. Analysis of the development of isotropic residual stresses in a bismaleimide/sprio orthocarbonate thermosetting resin for composite materials. J Appl Polym Sci 2003; 88:227 11. Kremer F, Schonhals A. Broadband Dielectric Spectroscopy. 1st ed. New York: Springer-Verlog; 2003. 12. Mopsik FI. Precision Time-Domain Dielectric Spectrometer. Rev. Sci. Inst. 1984; 55: 79. 13. Smith JW. Electric dipole moments. London: Butterworths scientific; 1955. 14. www.awnsers.com 15. Watanabe H. Dielectric relaxation of type -A polymers in melts and solutions. Macromol. Rapid Commun. 2001; 22:127. 16. Mashimo S, Nozaki R, Yagihara S, Takeishi S. Dielectric relaxation of poly (vinyl acetate). Journal of Chemical Physics. 1982; 77:6259. 17. Rendell RW, Ngai KL, Mashimo S. Coupling model interpretation of dielectric relaxation of poly (vinyl acetate) near Tg. Journal of Chemical Physics. 1987; 87: 2359. 18. Richert R, Wagner H. The dielectric modulus: relaxation versus retardation. Solid State Ionics 1998; 105:167. 19. Mopsik FI. The transformation of time-domain relaxation data into the frequency domain. IEEE Trans. Elec. Insul 1985; 20:957. 20. Ferry JD. Viscoelastic Properties of Polymers. 3rd ed. New York: John Wiley and Sons; 1980. 13 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 21. Serghei A, Huth H, Schick C, Kremer F. Glass dynamics in thin polymer layers having a free upper interface. Macromolecules 2008; 41:3636. 22. Richert R. Homogeneous dispersion of dielectric responses in a simple glass. J Non Cryst Sol 1994; 209. 23. Bohmer R, Schiener B, Hemberger J, Chamberlin RV. Pulsed dielectric spectroscopy of supercooled liquids. Z.Phys. B. 1995; 99:91-99. 24. Duwuri K, Richert R. Dielectric hole burning in the high frequency wing of supercooled glycerol. J Chem. Phys 2003, 118: 1356. 25. Shi X, McKenna GB. Mechanical hole-burning spectroscopy. Demonstration of hole-burning in the terminal relaxation regime. Phys. Rev. B 2006, 73: 014203-1. 26. Qin Q, Shi X, McKenna GB. Mechanical holeburning spectroscopy in a SIS tri-block copolymer. J. Polym. Sci Part B Polym. Phys 2007; 46:3277. 27. Kollengodu-Subramanian S, McKenna GB. A dielectric study of poly (vinyl acetate) using a pulse probe technique. Journal of Thermal analysis and calorimetry 2010; 102:477. 28. Kollengodu-Subramanian S, Srinivasan B, Rengaswamy R, Zhao J, McKenna GB. Application of empirical mode decomposition in the field of polymer physics. J. Polym Sci. Part B Polym phys, 2011; 49:277. 29. Huang N.E.; et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proceedings of Royal society of London 1998; 454:903. 14 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 30. McKenna GB, Vangel MG, Rukhin AL, Leigh SD, Lotz B, Straupe C. The τEffective paradox revisited: An extended analysis of Kovacs volume recovery data on poly (vinyl acetate). Polymer 1999; 40:5183. 31. Kolla S, Simon SL. Tau-effective paradox: New measurements towards a resolution. Polymer 2005, 46:733. 32. Knauss WG, Kenner VH. On the hygrothermomechanical characterization of polyvinyl acetate. J. Appl. Phys. 1980; 51:5531. 33. Wang WCh, Kramer EJ, Sachse WH. Effect of high pressure CO2 on the glass transition temperature and mechanical properties of polystyrene. Journal of polymer science Part B: Polymer Physics 1982; 20:1371. 34. Chiou JS, Barlow JW, Paul DR. Plasticization of glassy polymers by CO2. J. Appl. Polym. Sci. 1985, 30, 2633-2642. 35. Van der Vegt NFA, Briels WJ, Wessling M, Strathman H. The sorption induced glass transition in amorphous glassy polymers. J Chem Phys 1999; 110:11061. 36. Cotugno S, Larobina D, Mensitieri G, Musto P, Ragotsa G. A novel spectroscopic approach to investigate transport process in polymers: The case of water-epoxy system. Polymer 2001; 42:6431. 37. Alcoutlabi M, Vangosa Briatico F, McKenna GB. Effect of chemical activity jumps on the viscoelastic behavior of an epoxy resin: physical aging response in carbon dioxide pressure jumps. Journal of polymer science Part B: Polymer Physics 2002, 40:2050. 15 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 38. Zheng Y, McKenna GB. Structural recovery in a model Epoxy: Comparison of responses after temperature and humidity jumps. Macromolecules 2003, 36:2387. 39. Zheng Y, Priestley RD, McKenna GB. Physical aging of an epoxy subsequent to relative humidity jumps through the glass concentration. Journal of polymer science Part B: Polymer Physics 2004; 42: 2107. 40. Alcoutlabi M, Banda L, McKenna, GB. A comparison of concentrationglasses and temperature-hyperquenched glasses: CO2 formed versus temperature formed glass. Polymers. 2004; 45:5629. 41. Alcoutlabi M, Banda L, Kollengodu-Subramanian S, Zhao J, McKenna GB. Environmental effects on the structural recovery responses of an epoxy resin after carbon dioxide pressure-jumps: Intrinsic isopiestics, asymmetry of approach and memory effect. Macromolecules 2010 (Under Review). 16 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.1: Specific volume vs Temperature plot for different cooling rates (q). Here, q1> q2>q3 17 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.2: Classifications of dipoles in amorphous polymers [15] 18 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.3: Schematic representation of two different sources of non-exponential correlation decays [24]. 19 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.4: Schematic of the protocol for dielectric hole burning experiment [23]. 20 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.5: Time dependent dielectric permittivity of propylene carbonate at 157.4K [23]. 21 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.6: Empirical mode decomposition schematic to filter noise in time domain responses [29] 22 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 δ x 1000 Figure 1.7 (a): Schematic of intrinsic isotherm experiment 5.0 o 4.5 T0 = 40 C 4.0 o 19.8 C 3.5 3.0 22.4 24.9 2.5 27.5 2.0 1.5 30 1.0 32.5 0.5 35 0.0 -3 -2 -1 0 1 10 10 10 10 10 2 10 t-ti (h) Figure 1.7(b): Intrinsic isotherm experiment performed by Kovacs on PVAc digitized by Zheng and McKenna [38] 23 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.8(a): Schematic of memory effect experiment 2.0 o T 0=40 C (1) δ x 1000 1.5 1.0 (2) (3) (4) 0.5 0.0 -2 10 -1 10 10 0 1 10 2 10 10 3 t-ti (h) Figure 1.8(b): Memory effect experiment of PVAc performed by Kovacs digitized by Zheng and McKenna [38]. 24 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.9(a): Schematic for asymmetry of approach experiment 1.5 o T0=40 C 1.0 δ x 1000 0.5 0.0 -0.5 o T a=35 C -1.0 -1.5 o T0=30 C -2.0 -2.5 -3 10 -2 10 -1 10 10 t-ti (h) 0 1 10 2 10 Figure 1.9(b): Asymmetry of approach experiment of PVAc performed by Kovacs digitized by Zheng and McKenna [38]. 25 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.10: Schematic of Struik’s protocol 26 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 1.11: Creep compliance of poly(vinyl chloride) quenched from above to below Tg at different aging times after quenching [8]. 27 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 -1 T-jump 85-62C at 0MPa PCO2-jump 3.9 to 0MPa at 62C -2 -log(τ)/s -3 -4 -5 -6 -1 0 1 2 3 4 5 6 δ*1000 Figure 1.12: Effective retardation time plot for temperature and concentration formed glasses to the same final condition [41]. 28 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 2 EXPERIMENTAL SYSTEM 2.1 Time domain dielectric spectrometer The dielectric responses of glassy polymers were studied using the time domain dielectric spectrometer built in the Polymer and Condensed Material lab at Texas Tech University. This setup was built as a part of this thesis work. The general idea for this experimental system was based on the earlier works pioneered by Mopsik [1] and Bohmer and his coworkers [2]. The working principle of the dielectric time domain spectrometer is based on the ability to measure the capacitance of the sample film as a function of time. It may be noted that when a polymer film is placed between the two well polished metal plates, it acts as a capacitor. Using the capacitance of the sample, we can calculate the dielectric compliance of the material as a function of time. The experimental set up comprises of a high voltage supply source, sample setup and electrometer interfaced with PC using a DAQ board and controlled using a LabView program [3]. The schematic of the system is shown in Figure 2.1 Trek model 610 E is used as the high voltage supply source. The Trek system can be used in two voltage ranges, namely, 0 to 1000V range and 0 to 10000V range. For the current work, we have used the low voltage range as it also gives a better resolution. Further, very high voltage on thin films also caused dielectric breakdown of the material. Keithley 6514 model is used as the electrometer to measure voltage or 29 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 current response of the material under investigation. In this work, the electrometer is used to measure the voltage response from the sample setup. The electrometer can measure a maximum output voltage of 200V. The electrometer operates in 3 ranges namely 0 to 2V, 0 to 20V and 0 to 200V range. The desired range is selected based on the response of the output voltage. This type of measurement leads to resolution problems which are addressed in chapter 4. A surge protector is used to prevent any damage to the electrometer if the voltage exceeds 200V range. The measured output voltage is then stored in the PC using an NI instrument DAQ board of 12 bit resolution. The schematic of the sample set is shown in Figure 2.2. It comprises of the sample and integrating capacitor. The sample capacitor (SC) comprises of two well polished flat stainless steel plates with thin polymer film placed between them. The thickness of the SS steel plate is about 0.5 mm. The diameter of the lower and upper plate is 3 cm and 2 cm respectively. The plates are held tight using a spring setup shown in Figure 2. Mylar capacitor of capacitance 2.2 nF purchased from Digi-Key was used as the integrating capacitor (IC). As shown in the schematic, the sample capacitor and integrating capacitor are connected in series, such that one end of the IC is connected to the SC and the other end is grounded. The high voltage is applied on the SC and the response is measured between the junction of SC-IC and ground using the electrometer. The entire sample set up is placed in an enclosed aluminum casing and the temperature is applied using a cone heater and controlled to a range of ±0.1oC. The total set up is placed in a temperature controlled box made from polycarbonate. 30 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 The time domain dielectric spectrometer built at Texas Tech University can operate in the temperature range of 25oC to 80oC. The capacitance of the sample C(t) is measured using the equation 2.1. × = (2.1) Where, Vin is the applied voltage from the high voltage supply, Vo is the output voltage measured by the electrometer, and Ci is the capacitance of the integrating capacitor. The dielectric compliance ε(t) of the sample is then measured using equation 2.2. = × × = ∆ × ∆ (2.2) Where, d is the thickness of the sample, A is the surface area of the plate, εo is the dielectric permittivity in vacuum, ∆P is the polarization and ∆E is the applied electric field. 2.2 Experimental setup for studying the aging and structural recovery responses of glassy polymers subjected to plasticizer environment The physical aging and structural recovery experiments after pressure jumps (P Jump) were performed using the experimental setup built at Texas Tech University shown in Figure 2.3 [4]. The set up comprises of a pressure vessel (Figure 2.4) with a capacity to handle up to 8 MPa of pressure, the sample set up comprising of the sample holder, 31 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 linear variable differential transformer (LVDT) to measure the change in length (HR 100, Lucas Schaevitz inc;) and a motor to apply and retract the load, pressure sensors (Omega Electrovalves, SV128) to control the pressurization and depressurization rates, an oil bath filled with silicone oil and a heating coil to heat the pressure vessel to the desired temperature with a stability of ±0.1oC. The entire system is controlled using a DAQ board interfaced with the computer using a LabView program. It may be noted that for the P jump experiments, the set up was first subjected to vacuum for 20 minutes, followed by pressurization to about 4MPa at a pressurization rate of 0.0016 MPa/s, maintained at 4 MPa for about an hour, and then depressurized to 0 MPa at the depressurization rate of 0.0016 MPa/s. The temperature jump experiments were performed in an oven using the same experimental setup instead of a pressure vessel. The reason for using the oven instead of a pressure vessel for temperature jump (T jump) experiments is because of the very low cooling rate in the pressure vessel. A complete detail on the experimental set up is given in the reference [4]. 2.3 Linear Variable Differential Transformer calibration The LVDT used for the aging and recovery experiments for the current work was performed using a higher resolution mode of the signal conditioner such that maximum length change measured is about 1 mm. Figure 2.5 is the calibration plot for the LVDT, where length change (the distance the core is moved inside the LVDT) is plotted against the voltage response. From the calibration plot it was that calculated that 0.0985mm length change corresponds to a 1V response. 32 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 2.4 References 1. Mopsik FI. Precision Time-Domain Dielectric Spectrometer. Rev. Sci. Inst. 1984; 55: 79. 2. Schiener B, Bohmer R, Loidl A, Chamberlin RV. Non resonant spectral hole burning in the slow dielectric response of super cooled liquids. Science. 1996; 274:752. 3. Kollengodu-Subramnain, S, McKenna, G.B. A dielectric study of poly(vinyl acetate) using pulse probe technique. Journal of Thermal analysis and calorimetry 2010, 102, 477. 4. Alcoutlabi, M.; Banda, L.; Kollengodu-Subramnain, S.; Zhao, J.; McKenna, G.B. Environmental effects on the structural recovery responses of an epoxy resin after carbon dioxide pressure-jumps: Intrinsic isopiestics, Asymmetry of approach and memory effect. Macromolecules (2010 : Under review) 33 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 2.1: Schematic of time domain dielectric spectrometer built at Texas Tech taken from reference 3 34 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 2.2: Sample setup schematic for the time domain dielectric spectrometer [3] 35 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 A D B1 E F1 G H Sample C F2 LVDT CO2 vent Weight B2 Lift Motor I Pressure vessel CO2 Supply Mixer Oil bath A Pressure Controller Temperature Controller Signal Conditioner Motor Controller PC A/D A/D Board Board Air Drive A) Regulator B1) Inlet automatic valve B2) Outlet automatic valve C) High pressure pump D) Filter E) Safety valve F1) Inlet needle valve F2) Outlet needle valve G) Pressure sensor H) One way valve I) Three-way valve Figure 2.3: The creep apparatus that was built in our laboratory to perform experiments under different CO2-pressure and temperature conditions [4]. 36 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 2.4: The pressure vessel used to perform the experiments under CO2 pressure [4]. 37 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 2.5: LVDT calibration plot where the displacement of the core is measured as a function of voltage 38 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 3 A DIELECTRIC STUDY OF POLY (VINYL ACETATE) USING A PULSEPROBE TECHNIQUE 3.1 Motivation There is considerable literature available that describes our understanding of the viscoelastic properties of polymers subjected to mechanical stresses or deformations. What we refer to here as a pulse-probe technique is one method that is commonly used to study the time dependent behavior of materials in histories, e.g., temperature-jump or step-deformations, that exhibit fading memory responses. In the linear case the behavior is well understood in the context of Boltzmann superposition ideas. However, there is only limited work available that investigates the dielectric response of materials within this same context 3.2 Introduction Dielectric spectroscopy is normally performed in the frequency domain and in the linear response regime [1-3]. It is used as a tool to characterize materials and often the results of dielectric response in the linear regime are compared with mechanical and rheological measurements. The pulse-probe technique is one method used to study time dependent responses in temperature jumps [4] and nonlinear mechanical or rheological measurements [5-8]. In the present work, we present results from time domain dielectric spectroscopy experiments in which we explore the limits of linearity in poly (vinyl acetate) (PVAc). In particular, we used a pulse-probe method. The way in which the experiments are carried out is similar to the single and 39 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 double step strain or stress experiments commonly used in nonlinear viscoelasticity investigations. [5- 8]. Dielectric properties such as dielectric compliance and modulus are analogous to mechanical properties such as shear compliance and modulus [9]. The equivalents to mechanical stress and strain in a dielectric measurement are the dielectric stress (applied electric field) and the dielectric strain (polarization). The equation relating the time dependent dielectric compliance (ε(t)) to the applied field (∆E) and the polarization (∆P) is [9] ε (t ) = DielectricStrain ∆P (t ) ⇔ ∆Eε 0 Dielectricstress (3.1) where ε0 is the dielectric permittivity of vacuum. Importantly, equation 3.1 is valid in ideal situations in which the field is applied instantaneously, i.e., a step-pulse measurement and does not decay with time. Here, we found that the time decay of ∆E is small enough that errors introduced by treating the data as ideal constant dielectric stress experiments are negligible. In the present work we use time domain dielectric spectroscopy, a method pioneered by Mopsik [10-12] to investigate the dielectric responses of small molecule liquids and polymers. The method has been extended by Richert and Wagner [13] through the development of time domain modulus spectroscopy to investigate dynamic heterogeneity in small molecule glass formers and polymers. These works were confined to the linear domain and the responses were well explained using classic descriptions such as the Kohlrausch-Williams-Watts (KWW) [14, 15] and 40 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 modified KWW [16] functions to describe the time-domain responses and the VogelFulcher-Tammann [17-19] expression to describe the temperature dependence of the relaxation or retardation times near to the glass transition temperature. As noted above, linear dielectric spectroscopy is generally used in the frequency domain rather than the time domain. Frequency domain measurements, however, have also been used to study the nonlinear dielectric response of materials. For example, Furukawa et al [20] studied the nonlinear dielectric response of PVAc by obtaining the first and third harmonics in the frequency domain in samples subjected to increasing electric field. This is similar to attempts by Davis and Macosko [21] to use modified Boltzmann superposition [22, 23] to study the nonlinear viscoelastic behavior of polymers subjected to large mechanical deformations. A similar body of work has recently appeared from Wilhelm’s group in which Fourier Transform Rheology is used to characterize the higher harmonics of the extremely nonlinear rheolgical response of polymers and other complex fluids [24, 25]. In addition, mixed mode experiments have been carried out in which large amplitude sine waves are followed by single step small probes or time domain measurements to examine the nonlinear response of glass-forming liquids and polymer melts and solutions. For example, Schiener et al [26] established the dielectric hole burning method using experiments on supercooled propylene carbonate, and Shi and McKenna [27] developed a mechanical hole burning experiments using a polyethylene melt and a polystyrene solution as example systems. Richert and coworkers [28, 29] have studied the nonlinear dielectric behavior of 41 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 supercooled liquids by subjecting them to several cycles of sinusoidal waves of high amplitude followed by a wait time and a small step (time domain) probe. Prior to the present work, there have been only limited reports in the literature of using the pulse-probe technique to study the dielectric behavior of materials. Bohmer et al [30] used the pulse-probe technique to study the dielectric responses of supercooled liquids. They observed the memory effect similar to that observed in thermal and mechanical measurements [4, 5]. An interesting early work from the 1890’s by Hopkinson reports a memory effect in simple glasses subjected to reversing polarity is cited by Whitehead in a 1927 treatise [31]. There, the Boltzmann superposition principle [22] was used to predict the same. In the present investigation, we test the limits of Boltzmann superposition [22] for the dielectric response of poly(vinyl acetate) (PVAc) in pulse-probe experiments. For this, we have performed a single step time domain response and two step (pulseprobe) measurements having different amplitudes and durations in the vicinity of the glass transition. This is the first of a series of work to be later extended to study the nonlinear dielectric time domain response using the modified Boltzmann superposition principle [21, 23] and the pulse-probe method as developed here. 3.3 Sample Preparation Dielectric experiments were performed using a time domain dielectric spectrometer built at Texas Tech University (Figure 2.1). The working of the instrument is well described in the chapter 2. 42 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Poly(vinyl acetate) of molecular weight 157,000 g mol-1 purchased from Scientific Polymer Products, Inc. was used for the experiments. The glass transition temperature Tg of this PVAc as previously measured by our group using DSC at a cooling rate of 10 K min-1 was reported to be 303.6 K [32]. The sample was made by placing the pellets between thin brass sheets placed between two thick brass plates and then pressed at 333 K in a platen press. After that, the sample is cut into a circular section to fit the electrode plate, then held tight using a spring support and annealed at 338 K (above the Tg) before performing the experiments. The loaded spring set up helps to establish and maintain good contact between electrodes and polymer. The figure for the sample support set up is given in Figure 2.2. PVAc films of thickness 185 ± 10 microns were used for the single step isothermal measurements used to examine the time-temperature superposition behavior of the PVAc. The same film thickness was used for two step pulse-probe experiments in which the first step duration t1 was varied. Films of 112 ± 8 micron thickness were used for the experiments in which the electric field E1 was varied. The sample was cut to size and placed into the dielectric cell for measurement. 3.4 Methods of analysis The dynamic response of many liquids can be described by the stretched exponential or so-called KWW function shown in equation 3.2 [14-16, 33]. However, because the shape of the dielectric compliance response shows a sigmoidal-like shape, it can be necessary to apply a modified KWW function [16] as shown in equation 3.3 to capture the full response. Furthermore, when there is a long time process such as 43 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 the viscosity in viscoelastic creep measurements [34], or the conductivity in dielectric creep it may be necessary to add an additional term as in equation 3.4. ε (t ) = ε 1 * e t τ β ε (t ) = ε 1 + ε 2 * (1 − e ε (t ) = ε 1 + ε 2 * (1 − e (3.2) t − τ β t − τ β ) (3.3) )+ t (3.4) τc Where ε1, ε2, τ, τc, and β are fit parameters. Here τ is the retardation time and τc is the conductivity term. We find that equation 3.3 describes the recoverable part of the dielectric compliance and the full response of ε(t) is well described by equation 3.4. This separation of equation 4 into a recoverable term and a conductivity term is used subsequently. To test the validity of Boltzmann superposition for the two step dielectric response, we used the Boltzmann equation as rearranged by Riande et al [35] and implemented numerically in our group [32]: ∆P (t ) = ε g E (t ) + ∫ E (t − t ' ) dε ( t ) dt ' dt ' (3.5) Hence, ∆P is the predicted dielectric strain, E(t) is the applied dielectric stress for the two step test in the experiment analyzed, εg is the dielectric compliance at zero time 44 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 and ε(t) is the modified KWW (equation 3) fit to the single step, recoverable, linear dielectric compliance response. 3.5.0 Results 3.5.1 Single step response Isothermal measurements of the dielectric response (compliance) for the PVAc were performed in a temperature range of 303 K to 333 K and the data are shown in Figure 3.1(a). The solid lines in Figure 3.1(a) are the fits to the data at each temperature using the modified KWW function with conductivity contribution given in equation 4. Similar studies on PVAc have been performed by various groups [1, 13, 36]. The reason for using equation 4 is to separate the effect of dc conductivity from the dielectric compliance response and also to examine the time-temperature superposability of the dielectric response separate from the dc conductivity. The fit parameters for equation 4 for the 308 K reference temperature are given in Table 3.1. The β parameter obtained at the reference temperature was kept fixed for fitting the data at the remaining temperatures. We observed an increase in the ε1 parameter and decrease in ε2 parameter with increasing temperature for the above used function as shown in Figure 3.2. Both the conductivity term and retardation time decreased with increasing temperature. Figure 3.1(a) shows that, with increasing temperature, the curves shift to shorter times. We attribute the steep rise after the secondary plateau to the dc conductivity [9, 13]. Richert and Wagner used two KWW functions to capture the entire spectrum of data in their work on dielectric modulus [13]. As noted above, we 45 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 have used a term similar to the viscosity term in linear viscoelasticity but for the conductivity added to a modified KWW function to fit our data (see equation 4). Figure 3.1(b) is the master curve for the data of Figure 3.1(a) with 308 K as the reference temperature. The data at the longer times do not superimpose and we hypothesize this to be due to the domination of dc conductivity over dielectric response. The retardation time τ and the conductivity term τc obtained from equation 4 for the isothermal measurements of PVAc shown in Figure 3.1(a) are plotted as a function of inverse temperature in Figure 3.3. The data are fitted using the VFT function given in equation 3.6 [37]. The fit parameters are given in Table 3.2. The fit parameters for the dielectric retardation time and the dc conductivity term are different which explains the spread of the dielectric response at longer times in Figure 3.2(b). log τ = − A + B (T − T o ) (3.6) Where A, B and To are fit parameters. To confirm the above hypothesis, the recoverable dielectric compliance (which is the difference between the dielectric compliance and the conductivity term) was estimated using equation 3.7. t ε (t ) − − ε 1 τc ε R (t ) = ε2 (3.7) 46 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Where εR(t) is the recoverable dielectric compliance For each fit of equation 3.4 to the data of Figure 3.1(a) εR(t) was determined and the recoverable dielectric compliance response functions are plotted in Figure 3.4(a). In fact, these are described by the modified KWW function presented in equation 3.3. Figure 3.4(b) gives the master curve for the recoverable compliance data with 308 K as the reference temperature. We also observed softening like behavior of the secondary plateau with increasing temperature. This can be explained by the increase in ε1 and decrease in ε2 parameters of equation 4 with temperature as shown in Figure 3.2. In equation 3.4 the parameter ε1 is related to the glassy (short time) response and ε2 is related to the transition towards the long time plateau response. This kind of softening behavior with increasing temperature is also observed for dielectric responses in the frequency domain in the literature [9]. By shifting the curves vertically for the higher temperature data in addition to the horizontal shift, we obtain a reasonable time temperature superposition. In Figure 3.5 we compare the horizontal shift factor data for the dielectric recoverable compliance with the Plazek’s recoverable creep compliance data [38] and Richert’s (digitized) dielectric modulus [13] data. The results are in good agreement with Richert’s dielectric data. Somewhat surprisingly, the data of Figure 3.5 show that the dielectric response seems to follow Plazek’s terminal dispersion data rather than the softening dispersion data. A similar sort of behavior has also been observed by Zorn et al. in their work on polybutadienes [12]. 47 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 3.5.2 Time-frequency conversion It is possible to calculate the frequency domain response namely the dielectric loss compliance (ε”() and the dielectric storage compliance (ε’() from time domain response and vice versa using Fast Fourier transformation. The relationship between the dielectric compliance and its corresponding frequency components is given by the equation 3.8. = ′ − " (3.8) Figure 3.6 is the dielectric storage compliance plot calculated from the MKWW fit (Equation 4) of isothermal measurements of dielectric compliance shown in Figure 3.2(a). Similar to what is observed in the compliance data, the secondary plateau shows softening like behavior with increasing temperature. Figure 3.7 is the dielectric loss compliance for the data given in Figure 3.1(a). Dielectric compliance in general is a sum of contribution from dielectric relaxation, electrode polarization and conductivity [9]. What is the contribution of electrode polarization and conductivity to memory effect in the dielectric response is a worthwhile question to ask here. In the dielectric loss compliance plot, the features of dielectric relaxation, electrode polarization and conductivity appear in that order as we move from a higher frequency to a lower frequency [9]. It can be clearly seen, for the above temperature data, the conductivity as well as electrode polarization effect in the given range is very small and we observe contribution mainly due to segmental relaxation. Also, it might be worthwhile to note that the range of temperature investigated is sufficient to the purpose of the study, which is to examine the 48 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Boltzmann superposition and the preliminary results of breakdown of Boltzmann superposition is going through the glass transition range. 3.5.3 Two step (pulse-probe) response Prior to performing a two step experiment, we checked the effects on the dielectric compliance due to the small drift in the dielectric stress (∆E in equation 1) as a function of time. We used the Boltzmann equation given in equation 5 to obtain the true compliance data for the single step as implemented by our group previously [32]. We observed that there is no significant difference between the apparent and true compliance for the above material indicating that the effect of the time varying dielectric stress is very weak. The comparison is shown in Figure 3.8. Hence, we have used the data as obtained for the experiments without the additional use of the full Boltzmann equation. The dielectric compliance response of PVAc in the variable duration first step, two step pulse-probe experiments is shown in Figure 3.9. The experiments were performed at 302.8 K. The electric field-jump is made from 10.8*105 Vm-1 to 5.4*105 Vm-1 at a varying jump times t1 of 0.25s, 0.5s and 1s. In all the cases, we see a typical non-monotonic memory response similar to what is observed in thermal [4] and mechanical responses [5] to step-wise histories. In Figure 3.10, two step pulse- probe responses after decreasing the field from E1 to 44.6*105 Vm-1 for different values of E1 are shown. This is a ‘down-jump’ experiment. As the down-jump magnitude increases the memory effect lasts for longer times. 49 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 When the response is linear, we expect Boltzmann [22] superposition to hold. To test this for the two step pulse-probe experiments, we applied equation 5, using the single step response function determined above (recoverable part) and the experimentally applied fields for each step, to calculate the expected second step response. Equation 5 was solved using a MATLAB program based on a numerical integration method used previously by our group [32]. Comparison of the dielectric strain response from the experiment for the 1s jump depicted in Figure 3.9 with the Boltzmann prediction is shown in Figure 3.11. The results are in good agreement confirming that the dielectric response of the two step pulse-probe is in the linear regime for the above experimental conditions. On the other hand, for the largest down-jump investigated, that from 89.2 *105 Vm-1 to 44.6 *105 Vm-1, the outcome is different although the single step response at this applied field is in good agreement with Boltzmann superposition (See Figure 3.12). As shown in Figure 3.13 the Boltzmann superposition prediction deviates from the observed second step response showing evidence of nonlinear behavior. For comparison, Richert and Weinstein observed nonlinear behaviors in small molecule liquids at a high applied electric field [27, 28] in frequency domain spectroscopy. 3.6 Discussion Time domain dielectric spectroscopy has been used to characterize the linear response of PVAc. We found that the response was well fitted using equation 4, which is basically the modified KWW function (equation 3) plus an additional term for conductivity. This is equivalent to the use of a viscosity tem in creep compliance 50 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 determinations in linear viscoelasticity [34]. The conductivity term and the retardation term show different temperature dependences which explains the spread of data at longer times. Richert and Wagner [13] also observed this kind of a behavior in their dielectric relaxation work on PVAc. By subtracting the conductivity term, we obtain the dielectric recoverable compliance of the material which is analogous to the recoverable creep compliance in linear viscoelasticity [34, 38]. The recoverable dielectric compliance plotted as a function of time in Figure 3.4a clearly shows a softening-like behavior with increasing temperature similar to that for frequency response data reported in the literature [9]. This can be attributed to the changing in ε1 and ε2 parameters which are related to glassy and long time plateau responses respectively. The recoverable dielectric compliance is found to follow timetemperature superposition with ε1 and ε2 changing with temperature. To the best of our knowledge, other workers have not explicitly written the dielectric compliance in an equivalent fashion (directly analogous to the creep compliance in mechanics; equation 4). The temperature shift factors for the dielectric recovery are found to be the same as those determined by Richert and Wagner [13] for dielectric modulus. Surprisingly, the dielectric shift factors seem to follow the terminal shift factors determined by Plazek [38] in shear experiments and differ from the mechanical segmental shift factors. This is a surprise because the type B dipole in PVAc should reflect molecular motions that are local rather than long chain motions related to the terminal relaxations. 51 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Two step pulse-probe experiments were performed to test the limits of linearity in PVAc. We observed the classical memory response similar to that observed by others for dielectric behavior [30, 31] and in volume recovery after temperature jumps [4] and in two step mechanical experiments [5]. The experiments demonstrate that memory depends on both the time duration of the first step and the magnitude of the field jump. We observed that for small jumps, Boltzmann superposition is valid. For larger jumps we find deviations from linearity as observed by the over prediction of the memory response by Boltzmann superposition in spite of the fact that the single step response at the same field magnitudes were in the linear regime as evidenced in Figure 3.13. This suggests that the two step pulse-probe method may be a sensitive approach in dielectric spectroscopy to delineate the linear to nonlinear transition in behavior. 3.7 Conclusion Time domain dielectric spectroscopic measurements were performed on a PVAc polymer near to its glass transition temperature. Time temperature superposition of the response was not strictly valid due to the existence of different shift factors for the recoverable portion of the dielectric compliance and for the conductivity contribution to the response. The shift factors agree with those for dielectric modulus measurements [13] and terminal relaxation response in mechanical measurements [38] reported in the literature. A modified KWW function with an additional term for the dc conductivity was able to capture the entire experimental 52 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 regime with the retardation time τ for the recoverable compliance term and the conductivity contribution τc showing different temperature dependence. A two step pulse-probe technique was used to study the dielectric behavior of PVAc in the context of the Boltzmann superposition. We observe a memory effect similar to those observed in mechanical [5] and thermal responses [4]. The memory effect observed was in quantitative agreement with linear Boltzmann superposition for small applied fields. Evidence of nonlinearity is observed when the polymer was subjected to higher electric field. However, the further probing of nonlinear behavior as well as dynamic heterogeneity was found to be difficult because of the noise in the data. Dielectric data are time varying in nature corrupted with nonlinear noise. This problem is addressed in the next chapter. 3.8 Appendix 3.8.1 Algorithm to obtain true compliance (Adapted from the thesis of Stephen Hutcheson, Texas Tech) % y=Strain, Z=time, n=stress % l=compliance obtained by experiment % SR=Compliance obtained by Boltzmann %n=zeros(1000); k=0; %y=zeros(1000);%y(strain) 53 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 %z=zeros(1000); %n=zeros(1000); for x=1:10000 z(x)=k+0.001; y(x)=dstrain(x);%MKWWFit n(x)=dstress(x);%Curve fit k=z(x); l(x)=y(x)/n(x); end semilogx(z,n) semilogx(z,y) semilogx(z,l) dTLab=0.001; dTLoop=0.001; phi_g=1.0; D1=zeros(1,10000); C1=zeros(1,10000); 54 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 numbers=zeros(1,10000); SR=zeros(1,10000); for x=1:9999 T_Lab=z(x); integral_tot=0; June=0; %Our D(z(x)) D=phi_g; x1=x; if (x1>=1) for x2=1:x1 T_loop=x2*dTLoop; T_loop_1=(x2-1)*dTLoop; mat_time=T_Lab-T_loop_1; c=int16(mat_time*1000); torque_local=n(c); integral_local=torque_local*numbers(x2); 55 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 integral_tot=integral_tot+integral_local; end Zero=5.0; while Zero>0.0001 torque_local=n(dTLab*1000); June=integral_tot+torque_local*D; fd=June-y(int16(T_Lab*1000)); Zero=abs(fd)*10^15; D1=D; D=D-fd/torque_local; end else D1=0; end numbers(x)=D1; if (x>1) SR(x)=numbers(x)+SR(x-1); 56 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 else SR(x)=numbers(x); end end semilogx(z,l,'+',z,SR) 3.8.2 Boltzmann superposition algorithm to delineate linear and nonlinear behaviors %MKWW Single step compliance response P1=2.8; P2=6.25; P3=81.64019; P4=0.38034; k=0; % Call the dielectric stress data of two step response % tress1, Stress 2 for x=1:50000 z(x)=k+0.001; 57 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 k=z(x); if (z(x) < 1.001) y(x) = stress1(:,1); J(x)= P1 + P2*(1-exp(-(z(x)/P3)^P4)); elseif (z(x) > 1.549) y(x) = Stress2(:,1); J(x)= P1 + P2*(1-exp(-(z(x)/P3)^P4)); end end semilogx(z,J,'+') for x=1:49999 m(x) = x/1000; DJ(x)=J(x+1)-J(x); end semilogx(m, DJ) q=0; 58 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 %torque=0; %torque=zeros(40000); for x=1:40000 q(x)=x/1000; torque(x)=0; for j=1:(x-1) if (x-j>0) torque(x)=torque(x) + y(x-j)*DJ(j); end end torque(x)=torque(x)+ 2.85*y(x); end for x=1: 40000 r(x)=J(x)*y(x) end semilogx(z,r,'o',q,torque) 59 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 3.9 References 1. Serghei A, Huth H, Schick C, Kremer F. Glass dynamics in thin polymer layers having a free upper interface. Macromolecules. 2008; 41:3636. 2. Kovacs AJ. Transition vitreuse dans les polymeres amorphes. etude phenomenoloqique. Fortschr. Hochpolym.-Forsch. 1963; 3:394. 3. McKenna GB, Zapas LJ. Non linear viscoelastic behavior of poly (methyl methacrylate) in torsion. J. Rheology. 1979; 23:151. 4. Zapas LJ, Craft T. Correlation of large longitudinal deformations with different strain histories. Res. Nat. Bur. Stand. 1965; 69A:541. 5. McKenna GB. Viscoelasticity. In: Encyclopedia of Polymer Science and Technology, John Wiley and Sons; 2002. 6. Schapery RA. On the characterization of non linear viscoelastic materials. Polym. Eng.Sci.1969; 9:295. 7. Kremer F, Schonhals A. Broadband Dielectric Spectroscopy. 1st ed. New York: Springer-Verlog; 2003. 8. Mopsik FI. Precision Time-Domain Dielectric Spectrometer. Rev. Sci. Inst. 1984; 55: 79. 9. Mopsik FI. The transformation of time-domain relaxation data into the frequency domain. IEEE Trans. Elec. Insul. 1985; 20:957. 10. Zorn R, Mopsik FI, McKenna GB, Willner L, Richter D. Dynamics of polybutadienes with different microstructure. 2. Dielectric response and comparisons with rheologigal behavior. Journal of Chemical Physics. 1997; 107 (9):3645. 60 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 11. Richert R, Wagner H. The dielectric modulus: relaxation versus retardation. Solid State Ionics. 1998; 105:167. 12. Kohlrausch F. Ueber die elastische Nachwirkung bei der Torsion. Pogg Ann Phys Chem. 1863; 119:337. 13. Williams G, Watts DC. Non-symmetrical dielectric relaxation behavior arising from a simple empirical decay function. Trans Faraday Soc. 1970; 66:80. 14. Alcoutlabi M, Francesco Briatico-Vangosa, McKenna GB. Effect of chemical activity jumps on the viscoelastic behavior of an epoxy resin: Physical aging response in carbon dioxide pressure jumps. Journal of Polymer Science Part B: Polymer Physics. 2002; 40:2050. 15. Vogel H. The law of relation between the viscosity of liquids and the temperature. Phys Z. 1921; 22:645. 16. Fulcher GS. Analysis of recent measurements of the viscosity of glasses. J Am Ceram Soc. 1923; 8:339-355. 17. Tammann G, Hesse W. The dependence of viscosity upon the temperature of supercooled liquids. Z Anorg Allg Chem. 1926; 156:245. 18. Furukawa T, Matsumoto K. Nonlinear dielectric relaxation spectra of polyvinylacetate. Japanese Journal of Applied Physics Part 1. 1992; 31, 840. 19. Davis WM, Macosko CW. Nonlinear dynamic mechanical moduli for polycarbonate and PMMA. Journal of Rheology. 1978; 22: 53.. 20. Boltzmann L. Zur theorie der elastischen nachwirkung. Akad. Wiss. Wien. Mathem.-Naturwiss. Kl. 1874; 70:275. 61 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 21. Findley WN, Lai JS, Onaran K. Creep and relaxation of nonlinear viscoelastic materials with an introduction to linear viscoelasticity. New York: North-Holland Publication; 1976. 22. Wilhelm M, Maring D, Spiess HW. Fourier transform rheology. Rheol. Acta. 1998; 37:399. 23. Wilhem M, Reinheimer P, Ortseifer M. High sensitivity Fourier transform rheology. Rheol. Acta. 1999; 38: 349. 24. Schiener B, Bohmer R, Loidl A, Chamberlin RV. Non resonant spectral hole burning in the slow dielectric response of super cooled liquids. Science. 1996; 274:752. 25. Shi X, Mckenna GB. Mechanical hole burning spectroscopy: Demonstration of hole burning in the terminal relaxation regime. Physical Review B. 2006; 73:0142303-1. 26. Richert R, Weinstein S. Nonlinear dielectric response and thermodynamic heterogeneity in liquids. Phys. Rev.Lett. 2006; 97:095703 -1. 27. Richert R, Weinstein S. Nonlinear features in the dielectric behavior of propylene glycol. Physical Review B. 2007; 75:064302-1. 28. Bohmer R, Schiener B, Hemberger J, Chamberlin RV. Pulsed dielectric spectroscopy of supercooled liquids. Z.Phys. B. 1995; 99:91. 29. Whitehead JB. Lectures on Dielectric Theory and Insulation. 1st ed. New York: McGraw Hill Book Company; 1927. 30. O'Connell PA, Hutcheson SA, McKenna GB. Creep behavior of ultra thin polymer films. J. Poly Sci Part B. Poly Phys. 2008; 46:1952. 62 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 31. Struik LCE. Physical aging in amorphous polymers and other materials. Amsterdam: Elsevier; 1978. 32. Ferry JD. Viscoelastic Properties of Polymers. 3rd ed. New York: John Wiley and Sons; 1980. 33. Riande E, Diaz-Calleja R , Prolongo M, Masegosa R. and Salom C . Polymer Viscoelasticity: Stress and Strain in Practice. New York: CRC Press, Marcel Dekker; 2000 34. Schlosser E, Schonhals A. Dielectric relaxation during physical aging. Polymer. 1991; 32:2135-2140. 35. Shelby JE. Introduction to glass science and technology. 2nd ed. Cambridge: The Royal Society of Chemistry; 2005. 36. Plazek DJ. The temperature dependence of the viscoelastic behavior of poly(vinyl acetate). Polymer Journal. 1980; 12: 43-53 63 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Table 3.1 Fit parameters for equation 4 at 308 K. Errors represent standard error of estimate on the fit parameters. ε1 ε2 Β τ/s τc /s 3.21± 0.0096 5.34± 0.017 0.5±0.005 1.43±0.03 164±5.25 Table 3.2 A, B and T0 from VFT fits for the retardation time and the dc conductivity term vs. temperature obtained by fitting the data in Figure 2a to equation 4. Errors represent standard error of estimate on the VFT fit parameters. Time constants A B /K T0/K log τ 13.5 ± 0.95 914 ±87 241± 4.8 log τc 6.54±0.85 437± 110 258± 9.8 64 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.1 (a): (a) Isothermal measurement for PVAc with 308 K as reference temperature Figure 3.1 (b): Master curve for PVAc with 308 K as reference temperature 65 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.2: Equation 3.4 parameters ε1 and ε2 as a function of temperature 66 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.3: Dielectric retardation time τ and dc conductivity τc term plotted as a function of inverse temperature. The solid lines represent the VFT fits 67 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.4 (a): Dielectric recoverable compliance for PVAc with 308 K as reference temperature Figure 3.4 (b): Master curve for PVAc with 308 K as reference temperature. 68 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.5: Time-temperature shift factors as a function of temperature for PVAc. Comparison of current dielectric results with literature reports for dielectric [13] and mechanical [37] behaviors. 69 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.6: Dielectric storage compliance response for PVAc 70 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.7: Dielectric loss compliance of PVAc 71 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.8: Comparison of apparent dielectric compliance with true dielectric compliance for an applied electric field of 5.4*105Vm-1 72 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.9: Dielectric compliance response in two step pulse-probe experiments with varying time duration t1 of the first step. 73 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.10: Dielectric compliance response in two step pulse-probe experiments with varying electric field E1 of the first step. 74 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.11: Dielectric strain response for the 1second jump of Figure 8. The solid line represents the Boltzmann prediction 75 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.12: Single step responses for applied electric field of 44.6*105Vm-1 and 89.2*105 Vm-1 76 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 3.13: Two step dielectric compliance response (jump from 89.2*105 Vm-1 to 44.6*105 Vm-1) with linear Boltzmann prediction showing deviation from linear behavior 77 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 4 APPLICATION OF EMPIRICAL MODE DECOMPOSITION IN THE FIELD OF POLYMER PHYSICS 4.1 Motivation Noisy data has always been a problem to the experimental community. Effective removal of noise from data is important for better understanding and interpretation of experimental results. Over the years, several methods have evolved for filtering the noise present in the data. Fast Fourier transform (FFT) based filters are widely used since they provide precise information about the frequency content of the experimental data which is used for filtering of noise. However, FFT assumes that the experimental data is stationary. This means that: (i) the deterministic part of the experimental data obtained from a system is at steady state without any transients and has frequency components which do not vary with respect to time and, (ii) noise corrupting the experimental data is wide sense stationary, i.e., mean and variance of the noise does not statistically vary with respect to time. Several approaches, e.g., Short Time Fourier Transform (STFT) and Wavelet Transform (WT) based filters, have been developed to handle transient data corrputed with nonstationary noise (mean and variance of noise varies with respect to time) data. Both these approaches provide time and frequency information about the data (time at which a particular frequency is present in the signal). However, these filtering approaches have the following drawbacks: (i) STFT requires identification of an optimal window length within 78 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 which the data is stationary, which is difficult and, (ii) there are theoretical limits on simultaneous time and frequency resolution. Hence, filtering of noise is compromised. Recently, Empirical Mode Decomposition (EMD) has been used in several applications to decompose a given nonstationary data segment into several characteristic oscillatory components called intrinsic mode functions (IMFs). Fourier transform of these IMFs identifies the frequency content in the signal, which can be used for removal of noisy IMFs and reconstruction of the filtered signal. In the present work, we propose an algorithm for effective filtering of noise using an EMD based FFT approach for applicatons in polymer physics. The advantages of the proposed approach are: (i) it uses the precise frequency information provided by the FFT and therefore efficiently filters a wide variety of noise and, (ii) the EMD approach can effectively obtain IMFs from both nonstationary as well as nonlinear experimental data. The utility of the proposed approach is illustrated using an analytical model and also through two typical laboratory experiments in polymer physics wherein the material response is nonstationary;standard filtering approaches are often inappropriate in such cases. 4.2 Introduction Experimental data are invariably corrupted with some form of noise [1]. Except for the few cases where noise is within the experimental uncertainties, it is necessary to eliminate this noise from the data [1,2]. Some common sources of noise are fluctuations in: (i) power supply (ii) instrument performance, especially when pushing the limits of a device, and (iii) ambient (temperature or other) 79 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 conditions involved in the experiments [1, 2]. Over the last half-century, much progress has been made in the area of denoising of signals, especially because of the advent of widespread digital data acquisition. In addition, experimentalists are invariably looking at ways to extend the boundaries of experimentation from macro to micro, micro to nano level [1-5], thereby pushing the limits of their instrumentation. Ways to reduce noise have historically involved physical methods. For example, shielded cables are used to eliminate or reduce the electrical noise present in the system [1, 6, 7]. However, since noise comes from various sources (corrupting the experimental data at various frequencies), it is impossible to completely physically eliminate all sources of noise. Recent advances in the field of digital signal processing (DSP) have addressed the denoising of signals by using various filtering algorithms [2] in addition to physical methods. Moving average and Fast Fourier transform (FFT) based filters are some of the most commonly applied DSP techniques in experimental studies [1, 2, 5-7]. Moving average (MA) filters are known to be optimal for removal of random noise present in the signal. However, if the data is corrupted with noise at specific frequencies, MA filters perform poorly by introducing biases [1]. Moving average filters act as low pass filters with poor ability to separate/filter noise at individual frequencies [8]. The drawbacks of using moving average filters are highlighted later in the article by applying this technique for data generated using a modified KohlrauschWilliams-Watts (MKWW) model corrupted with noise at specific frequencies. Fast Fourier transformation (FFT) is one of the most widely used methods to filter 80 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 the noise present in experimental data as it provides the best frequency information for a given time signal [9]. Unlike moving average filters, FFT based filters are unbiased, which is one of the main reasons for their use in multiple disciplines [1, 2]. In this approach, noise reduction is achieved by reducing the gain at user specified frequency regions. The fundamental assumptions involved in all FFT based filtering approaches are: (i) the noise is assumed to be predominantly present in the user specified frequency bands and, (ii) the experimental data is wide sense stationary (a weak form of stationarity condition) [10, 11]. A signal x(t) generated by the process is said to be wide sense stationary (WSS) if the following conditions are true: (i) mean of the signal x(t) generated from the process denoted as E(x) (E is the expectation operator) is constant and, (ii) the autocovariance function of the signal defined as E{x(t)x(t+τ)} = Rx(t,t+τ) is independent of time and does not vary with respect to time [8]. The data obtained from experiments is usually found to contain the following two components: (i) a deterministic component from the system under study and, (ii) a purely stochastic component arising due to the noise corrupting the process [12]. Fourier transformation (FT) assumes that the spectral content of the deterministic component of the signal given by = ∑ ! "#$% . Here Xk is the Fourier transform of the signal (which is complex) and ωk. Denotes the frequency of the signal for all k [12]. Therefore, nonstationary signals (amplitude and/or frequency characteristics changes with respect to time) cannot be analyzed or filtered using Fourier transformation (since it violates the assumptions made in FT) [13]. Regarding the purely stochastic component of the data, FT assumes that the noise 81 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 corrputing the data is wide sense stationary [14]. Fourier transform performs poorly if the WSS assumption of the data is violated [8]. A detailed explanation regarding the Fourier transform and its drawbacks for use in nonstationary signals is provided in reference 10. For nonlinear signals, FFT provides fundamental frequency along with harmonics (integer multiples of fundamental frequencies.) [15]. However, harmonics provided by FFT are defined based on the stationary nature of the signal. However, if the underlying signal is nonstationary, there is no certainty that the spectral components provided by FFT really exist in the data [16]. In short, traditional FFT based filtering approaches are not sutiable for filtering of experimental non-stationary data corrupted with nonlinear noise sources [10, 11]. Noise in non-stationary data can be handled using techniques like short time Fourier transform (STFT), Wavelet or EMD based filters [10, 11, 13]. STFT works on the principle of dividing the data into various stationary segments/windows (mean of the signal remains constant in this segment) followed by application of an FFT based filter for each individual segment [10]. In STFT, optimal window length selection is important to obtain the required frequency information from the data. Two major drawbacks associated with STFT are: (i) it is hard to obtain an optimal window (segment length) for a given system and (ii) a time-frequency resolution problem (for a detailed explanation refer to 9) occurs due to the finite size of the stationary segments [10, 11, 13]. In summary, when using STFT filters for experimental data, if the window size is small, it is not possible to separate narrow frequency bands. This in turn leads to difficulty in 82 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 filtering narrow band noise. On the other hand, it is also often not possible to find large stationary segments in the experimental data of interest [11, 13]. Discrete wavelet transform (DWT) filters are widely used to overcome the drawbacks associated with STFT filters [13] . While STFT uses a window function with infinitely long (in time) sinusoidal orthonormal basis functions, wavelet transform uses short lived (in time) mother wavelets as basis functions for signal processing[13, 17]. This is considered as one of the main advantages of DWT over STFT as this helps in better handling of both windowing and time-frequency resolution issues related to processing of non-stationary signals [13]. DWT works on the principal of splitting a signal into low and high frequency bands (levels) by passing it through a series of band pass filters obeying the Nyquist criterion [13]. Filtering is then performed by removing the noise components in the user specified frequency bands. The mathematical details involved in MA filters, FT based and DWT filters are briefly discussed in the appendix section of the article DWT based method is widely used for filtering wide band noise (has nonzero magnitude over a large band of frequencies) [13]. Notice that like STFT, DWT is also a time–frequency analysis algorithm for handling nonstationary signals. Also, in DSP, there always exists a trade-off between time and frequency resolution of the signal. This time-frequency tradeoff is referred to as the uncertainty principle in the DSP community since it is analogous to the Heisenberg uncertainty principle in quantum mechanics [10]. Therefore, DWT cannot be effectively used for filtering signals corrupted with narrow band and nonlinear noise sources.. Hence, there is a requirement for a filtering algorithm that can take advantage of 83 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 FFT as well as have the ability to handle non-stationary signals. Recently, Empirical Mode Decomposition (EMD), a time domain algorithm, has been developed for handling non-stationary and nonlinear signals [11, 18]. It works on the principal of dividing the experimental data into various intrinsic components called intrinsic mode functions (IMFs) which represent the characteristic features of the data at various time scales [11, 18]. Fourier transform of these characteristic IMFs are computed to obtain the frequency information from the experimental data over the various time scales. Unlike STFT which requires optimal window length selection for computing the frequency information from the data, EMD based filtering approach provides all the frequency information to the user without any user input. This frequency domain information about the IMFs is then utilized to identify the noisy IMFs which are removed to construct the filtered signal. The denoised signal is reconstructed by summing up of the remaining IMFs. This EMD based DSP filtering approach is more elaborately discussed in the Methodology section. Noise in experimental data is also a well recognized problem in the polymer field [6, 19-21]. A literature review over the last few decades clearly highlights the development of numerous DSP based filtering and model based approaches to treat experimental noise in this area [7, 22-25]. A discrete FFT method with oversampling has been developed to both filter noise and investigate nonlinear behavior in polymer rheology [7, 26-30]. Kremer and Schönhals [6] proposed a filtering approach based on the sinusoidal response of the system. In this approach, a sinusoidal input is provided to the system and the corresponding 84 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 output from the system is obtained. The output could be a phase shifted sinusoidal signal of the same frequency as the input along with harmonics (if the system is nonlinear). A correlation analysis in frequency domain (obtained using FFT) is performed between the input signal of known frequency and the output signal (with phase shifts and/or harmonics). The result of correlation analysis is a noise free sinusoidal output signal since it is well known that the correlation will be maximum between the input and output signals at the phase shifted value when there is no correlation between input and noise corrupting the system. This has been cited as one of the major advantages of the frequency domain approach over the time domain measurement in broad band dielectric spectroscopy (see page 39 of reference 6) as it is difficult to remove noise in time domain data. Kremer and Schönhals have also recognized the difficulty of handling noise and nonlinearity in time domain data. Further, it is to be noted that this frequency domain approach is time consuming for lower frequencies [6]. In the present article we propose a novel EMD based FFT filtering technique to remove the noise in time domain data and that offers improved measurement of time domain phenomena. Mopsik suggested the use of functional fitting techniques as an effective way to reduce noise in time-domain dielectric spectroscopy [31]. However, these are not filtering techniques and therefore can introduce artifacts in the data if the fitting function is improper. Moreover, it is difficult to arrive at the exact function which can capure all the features of experimental data and reduce the presence of noise. In the present work we demonstrate the application of the empirical mode decomposition (EMD) based FFT algorithm as an effective tool to filter the noise 85 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 from time-domain non-stationary data in the field of polymer physics. We illustrate the practical utility of the proposed methodology for filtering using the following case studies: (i) an analytical form, specifically the modified Kohlrausch-Williams-Watts (MKWW) [32, 33] model to generate data corrupted with harmonic and random noise, (ii) time domain dielectric compliance data for a poly(vinyl acetate polymer) in the glass transition regime, and (iii) volume recovery data after a temperature jump in an epoxy glass-former. 4.3 Methodology EMD is an adaptive data analysis technique used to decompose the given time domain signal into various time domain components called intrinsic mode functions (IMFs). The IMFs are representative of the characteristics of the signal at various time scales [11, 18]. In other words, the EMD algorithm obtains various IMFs by decomposing the given time domain signal into various shorter (high frequency) time scale & and longer time scale (low frequency) components '. Each of the IMFs obtained from the EMD algorithm satisfies the following constraints: (i) the number of extrema and zero crossings must be either equal or differ at most by one and, (ii) the mean value of the envelopes defined by the local maxima and minima is zero. The EMD algorithm used to obtain the IMFs is provided below as taken from references [34, 35]: 1. Identify all extrema of the given signal . 2. A cubic spline fit is used to obtain the envelope of minima (emin(t)) and maxima of the signal (emax(t)). 3. Mean of the envelopes m(t) = (emin(t)+emax(t))/2 is computed. 86 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4. Extract the component ( = – '. If ( satisfies the conditions of an IMF (mentioned earlier), then assign (=& , where & represents the kth short time scale IMF. Otherwise, assign = ( and iterate steps from 1 to 4. 5. Obtain the residual ) = x – & . If ) contains a minimum of two extrema, then set r(t) = x(t) and iterate steps 1 to 5. Otherwise set the trend (zero frequency) component ' * = ). Applying EMD algorithm to the signal gives = '* + ∑+,- & (4.1) where '* is the trend component, & is the kth intrinsic mode function with k varying from 1 to the number of IMFs, N. Once IMFs are obtained from the EMD algorithm, the next step is to identify and eliminate the IMFs corresponding to noise components. 4.3.1 EMD based FFT filtering algorithm The EMD based filtering procedure used in this work is outlined below: 1. Compute the magnitude at various frequencies using FFT of the IMFs obtained from the EMD algorithm. The magnitude of a particular IMF (x(t)) at various frequencies are computed using the following equation |X(ω)|= |! "# | 2. Identify the IMFs corresponding to the noisy signal using their magnitudes at the various frequencies obtained from step 1. Let the number of IMFs be 87 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 N. The frequency domain information provided by the FFT of the IMFs are used for selection of the noisy IMFs. Let the signal contain Y noisy IMFs with ki (i = 1 to Y) representing the index of the noisy IMFs. The magnitude of the IMFs are computed. Reconstruct the filtered signal / using the non noisy IMFs. The filtered signal is given by / = '* + ∑+0- & for all k ≠ ki (4.2) The proposed EMD based FFT filtering procedure is applied to various simulation and experimental data and the results obtained are discussed in the following section. We compare the FFT and DWT based filtering methods as well. 4.4 Results and Discussion 4.4.1 Simulation studies To illustrate the proposed method and to make comparisons with other filtering methods, we first use an analytical function to which noise is added: specifically, we use a modified form of the Kolrausch-Williams-Watt (MKWW) function [32, 33]. This model has been used to study the compliance and modulus behaviors in mechanical measurements of polymers. The model equation is given by: 1 = 2 + 31 − ! 6 9 7 5 8 (4.3) Here a, b, c and d are ‘material’ constants and the values used in the simulation are 7, 2.5, 10 and 0.5 respectively. Y is the response function. Often, one is 88 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 interested in determining the material parameters from noisy data. The data obtained using he above equation is shown in Figure 4.1. Noise n(t) is added to the MKWW function using the following expression : = 0.2 ∗ sin2π14t + sin2π60t + 0.2 ∗ sin2π180t + 0.1 ∗ sin2π300t + 0.1 ∗ randt (4.4) Here n(t) is the noise generated which includes sinusoidal noise at a fundamental frequency of 60Hz along with harmonic frequencies namely 180Hz and 300Hz. A low frequency noise at 14 Hz is introduced along with a random white noise (with zero mean and unit variance). The maximum variability in the signal due to noise is ± 5% of the data simulated using the MKWW function. The data is sampled at a sampling frequency of 1000 Hz which is the same as used in the experimental case studies. Therefore, according to the Nyquist criterion [34], the maximum frequency that can be present in the data is 500 Hz. The reasons for adding this kind of noise to the simulated data are as follows: (i) the dielctric spectroscopy data obtained from experiments (discussed in subsequently) is corrupted with a 60Hz sinusoidal power supply noise. However, due to truncation effects involved in the electrometer used for the time domain spectroscopy, this noise gets truncated such that there is a noise contribution to the signal with harmonic frequencies in addition to the fundamental at 60 Hz, and, (ii) the data obtained from volume recovery experiments (after performing a step change in temperature) is corrupted with ambient noise which can be considered to be white. Figure 4.2 shows the noise generated using equation 4.4. To ensure that the ranodm noise 89 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 added to the system is white, Autocorrelation function (Normalized autocovariance function) of the random noise is computed and is shown in Figure 4.3. The autocovariance function of a signal x(t) is defined as E{x(t)x(t+τ)} = Rx(t,t+τ), with τ denoting the lags which are used for computation of the correlation between the signal at the current instant and previous instant. In otherwords, autocovariance function provides the correlation between the signal at current instant and its values in the past. For a pure white noise, this should be maximum at at time t and zero at all instances. This is due to the fact that white noise is correlated with itself at time t and does not have any correlation with its past values. This can also be noticed from autocorrelation function shown in Figure 4.3. Further, the confidence interval (confidence interval, J = - √+ with N being the number of data points) for autocorrelation function is also provided in Figure 4.3. Figure 4.4 shows the simulated data corrupted with the noise, obtained by performing linear addition. 4.4.2 Filtering of the simulated data In the next paragraphs, we consider four filtering methods for the just described simulated data. We examine the results provided by MA filters (with two different window lengths), FFT based filtering method, the DWT method, and the proposed EMD based FFT filtering technique. Finally, we provide a comparison of the results obtained from all these approaches. Moving average filter Moving average filters compute the average of the given data over a 90 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 specified window length thereby reducing the noise present in the data. The results vary with different window size and selection of optimal window length is difficult. Further, as mentioned earlier, MA filters are known to perform poorly when the data is corrupted with noise at specific frequencies (colored noise). MA filter with window sizes 5 and 10 were used to filter the MKWW model generated signal with noise. The results obtained using this approach are shown in Figure 4.5 and Figure 4.6. From the figures, it can be clearly seen that the MA filters (with window sizes 5 and 10) perform poorly at lower time scales. This is due to: (i) presence of less data at short time scales compared to the large time scales and, (ii) the amount of frequency specific noise corrupting the system is high for the shorter time scales. This is an example of how MA filters are not optimal for filtering of nonstationary signals corrupted with colored noise [8]. FFT filter The MKWW function (used to study compliance of systems) used for simulation provides a transient signal. As mentioned in the introductory section of the paper, transient signals have amplitudes and freqencies varying with respect to time and hence they are nonstationary. Therefore, the current simulation data violates the stationarity assumption made in FFT. However, the noise corrupting this data set is stationary (mixture of sinusoidal signals and white noise with mean and variance constant with respect to time). An FFT based low pass filter with cut-off frequency as (60 Hz) is used to denoise the signal. This cut-off frequency is generally chosen to remove the 60Hz sinusoidal electrical noise. The result obtained after using the FFT filter is shown in Figure 4.7. From the Figure, it is 91 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 clear that FFT filter performs poorly when the data is nonstationary. Though the data looks stationary at the lower time scales, the number of data points is also less there and therefore, beginning from the sharp slope change, the FFT filter is unable to perform well. However, at longer time scales there is a large amount of data and there are not sharp slope changes (since the amount of data is large, differences between the previous and current data points in this region are less). Further, this is coupled with the fact that for the longer times the signal magnitude is higher while the noise magnitude remains the same. In other words, the signal to noise ratio is larger at longer time scales. Hence FFT is able to provide reasonable filtering at longer time scales. From, the figure, it can be seen clearly that the FFT filter performed well when the mean of the data is changing slowly, i.e. at long times. However, due to the non-stationarity of the short-time data, FFT filtering is inefficient at this time scale and this leads to oscillations in the filtered data. DWT filters A DWT based filter was applied to denoise the data. Discrete wavelet transform (DWT) is a time-frequency analysis tool which is widely used to analyze nonstationary data. Unlike Fourier transform which uses sine and cosine functions, DWT uses short lived orthogonal wave functions to analyze the data. The steps involved in DWT are as follows: (i) the signal x(t) = x(nT) (x(nT) is the experimental data sampled at discrete time intervals with a sampling time T and n = 1,2...N; N represents the maximum number of samples) is passed through a half band low pass filter h(n) (n is used to indicate that the filter is discrete) and a half 92 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 band high pass filter g(n) [13]. The half band low pass filter removes the frequencies that are above half of the maximum frequency content in the signal to give an approximation signal a(n) while the half band high pass filter removes the frequencies that are below half of the maximum frequency content in the signal, (ii) the signals obtained from these two filters are subsampled by a factor 2 (eliminating half the number of points based on the Nyquist criterion) [36]. These two steps correspond to single level decomposition of signal using DWT. Now the approximation signal s(n) is not decomposed any further, while the detail coefficient is further decomposed and these two steps are repeated until a desired level of decomposition is reached. After decomposition, detail components (corresponding to high frequency ranges) are removed or an amplitude based thresholding is performed (only frequencies at which amplitude is above a certain threshold magnitude are retained). Then, signal reconstruction is performed using the final approximation component and the retained detailed components. During reconstruction of the filtered signal, upsampling is performed at every step to obtain the final filtered signal [36]. A more detailed explanation on discrete wavelet filters is provided in reference 13. A five level decomposition of the data was performed using DWT. The schematic of the five level decomposition of the simulated signal is shown in Figure 4.8 (based on the Nyquist criterion, 500 Hz is the maximum frequency content in the signal since the sampling frequency is 1000 Hz) . The denoised signal is constructed using the coefficients a5 and d5 and the resulting denoised signal is shown in Figure 4.9. From Figure 4.9, it is clear that DWT filter 93 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 performs poor and has as still an oscillatory component present in the data. This suggests that the user has to perform more level of decomposition. However, there is no fixed specific decomposition level and results vary: (i) depending upon the number of decompositions performed and (ii) the number of approximation and detail components used for constructing the filtered signal. If the underlying original process data is unknown (usually this is the case in experiments), selection of decomposition levels and coefficients for construction of filtered data plays a key role in obtaining the filtered signal. Further, if the experimental data is corrupted with noise at specific frequencies, it is difficult to select these parameters. As an extreme case, one can include all the noise in the data (by including all detail and approximation coefficients, equivalent to no filtering ) or even remove the dynamics exhibited by the underlying process (excluding all the coefficients except for those closer to zero frequencies, thus filteriing the dynamic transient nature of the data produced by the process). In addition, this method does not provide precise frequency information about the noise corrupting the process output. Even if more levels of decomposition are performed, the user cannot identify which components have to be utilized for construction of the filtered data [13]. With the six level decomposition, as shown in Figure 4.10, the DWT filtered data matches well with that of the original signal, using DWT. As mentioned earlier, the results obtained from DWT vary depending upon the number of levels of decomposition performed and number of coefficients used for construction for filtered signal. From Table 4.1, it can be seen that the results 94 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 obtained from DWT vary depending upon the number of levels of decomposition. Further, they also vary with the number of components used for reconstruction of the data [6] . However, good filtering of the signal can be achieved if the number of coefficients and decomposition level are selected properly, which is generally difficult for narrow band noisy signals. In the following discussion, it will be shown that the proposed EMD based FFT filtering approach provides a time domain decomposition of the simulated signal which can be used to filter noise at specific frequencies along with the broad band noise. Proposed EMD based FFT filter approach The proposed EMD based FFT filtering approach was also used to denoise the MKWW function corrupted with noise The IMFs (components after time domain decomposition of the signal) and their corresponding FFTs obtained after the decomposition are shown in Figure 4.11. From the FFT of IMF 1, we can clearly see sharp spikes in the second and fourth harmonics of 60 Hz and at 14 Hz frequency. Thus IMF 1 provides information about the nonlinear noise corrupting the signal at specified frequencies. During the filtering procedure this IMF can be neglected since it represents the noise corrupting the signal. Similarly, FFT of the second IMF shows the presence of white noise with higher amplitudes between 0 and 50 Hz. Thus, FFT of the IMFs helps in identifying the nature of the noise corrupting the system. The data then obtained using the EMD based filtering approach is shown in Figure 4.12 matches well with the original data sans noise. The importance of this analysis is highlighted in the subsequently described experimental case studies. It is to be noted that unlike DWT based filters the 95 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 proposed approach provides: (i) precise frequency information about noise which can be used to improve the experimental setup (discussed as a separate subsection in the end of experimental studies) and (ii) filtered results do not depend on the number of decompositions, since the EMD algorithm itself has a termination criterion. 4.4.3 Discussion of results obtained from filtering methods In this section, we compare the goodness of the filtered data obtained from various filtering approaches. Total sum squared error (TSE) is a measure which indicates the effectivenss of the filtering technique to obtain better filtered data. The filtering approach which provides the minimum TSE compared to other filtering techniques performs the best filtering of the data. The total sum squared error (TSE) between the original non-noisy data and filtered data is defined as Q LMN = ∑+ ,-1 − 1/OP . The TSE values obtained from the various approaches are shown in Table 1. The plot of squared error (e2(t) =( y(t) – yfilt(t))2) from the various filtering approaches is shown in Figure 4.13. It can be clearly seen that the proposed approach provides minimum squared error at all times compared to other filtering approaches. Further, the TSE value for the proposed approach is also minimum compared to other techniques which explains the goodness of the proposed method over other methods. 96 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4.4.4 Experimental case study I Time-domain dilectric experiment Time-domain dielectric spectrometry is used to study the dielectric response of polymers and other materials of interest [6]. The data obtained from the dielectric response is time varying in nature and hence it is of interest to use such data to demonstate the applicability of EMD based filters. The data used for the present study was obtained from a single step response performed in the time domain dielectric spectrometer (TDS) built at Texas Tech. The TDS system consists of a high voltage supply using which a known voltage is applied to a sample capacitor. The voltage output is measured using an electrometer. The entire system is operated using a Labview program interfaced with the computer using a DAQ board. Figure 2.1 is the schematic of TDS system built at Texas Tech and a more detailed version of the experimental set up is given in reference 37. Poly vinyl (acetate) of molecular weight 198,000 g/mol and PDI of 2.73 with its glass transition temperature of 31oC obtained at the cooling rate of 10K/min is the material investigated in this experiment. The thickness of the film used is about 120 microns. The specific experiment considered here was performed at 29oC at a sampling frequency of 1000 Hz. The data obtained was found to be corrupted with experimental noise. To ascertain the noise in the system, a test was run on the same material at 0V (mean value =0V). The output data is shown in Figure 4.14 (a). It can be clearly observed that the data is corrupted with sinusoidal noise. Figure 4.14 (b) is 97 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 the FFT performed on the data shown in figure 4.14 (a). A distinct single peak at 60Hz can be observed in Figure 4.14 (b) indicating the presence of ac noise from the laboratory electrical power supply. However, there is also a small amplitude at a frequency of 180Hz due to minor distortions present in the observed sinusoidal signal. It is well known that this type of single frequency ac noise can be removed using the conventional FFT based filters [1]. However, the data collected as the single step response after a 200V applied voltage input shown in Figure 4.15 indicates the non stationary nature of the signal. Further, from the inset in Figure 4.15, the presence of irregular pulses can be noticed. This is because of the truncation effects arising due to the autoranging capacity of the electrometer system (Keithley 6514) at various voltage levels in the experimental set up. These truncations introduce nonlinear noise effects in the output voltage data. To summarize, filtering of the data shown in Figure 4.15, poses the following challenges: (i) nonstationary nature of the output signal of interest and (ii) nonlinear nature of the noise corrupting the system output. Therefore, as discussed in the simulation study using MKWW model, we use the EMD based FFT approaches to filter the experimental data. The advantages of the information provided by EMD based FFT filtering approach for filtering noisy data is highlighted. Proposed EMD based FFT filter approach We now apply the EMD algorithm to the experimental voltage data set shown in Figure 4.15. Following the algorithm presented earlier in section B, the data is decomposed into several IMFs. Consequently, the filtering approach (using 98 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 FFT of the IMFs) detailed in section B.1 is applied to obtain the filtered signal. During filtering, the noisy IMFs are identified using the frequency domain information provided by the FFT of the IMFs. The key noisy IMFs along with their FFTs are shown in Figure 4.16. Notice, the presence of magnitude values at frequencies of 60Hz (ω), 120Hz (2ω), 180 (3 ω) and so on from the FFT of the noisy IMFs shown in Figure 4.16. These IMFs provide information that the noise corrupting the experimental setup could be ac noise (60Hz signal) and their harmonics. Unlike DWT, we can justify the removal of these particular IMFs during reconstruction of filtered signal as detailed in section B.1. These keys IMFs along with other IMFs representing the truncated ac noise are removed to obtain the filtered dielectric time domain data (using equation 2). The ability of the proposed methodology to remove the noise can be readily seen using the comparison plot between the actual and filtered data shown in Figure 4.17 (magnified version is provided in the inset). 4.4.5 Experimental Study II Volume recovery after a temperature jump Structural recovery is the study of thermodynamic properties namely volume, enthalpy or entropy evolving towards equilibrium, when the glassforming material is quenched from above its glass transition temperature (Tg) to a known temperature below its Tg [38-40]. It is important to understand this behavior as it is known to have a strong impact on the long term durability of glassy materials, especially polymers [41, 42]. Structural recovery experiments 99 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 require very sensitive measurements that are often close to the limits of the transducing devices used to measure volume or length as a functions of time, and temperature fluctuations play an important role in determining how well one can estimate the equilibrium state [43, 44]. As a result, there is a need for a good filtering method to enhance the signal to noise ratio. Here, we analyze the results of a length-change dilatometric measurement from our own labs in which the volume recovery of an epoxy glass is followed subsequent to a tempereature jump from above to below, but still close to the glass transition temperature, i.e., the volume changes are small. The structural recovery test set up [45] shown in Figure 2.3 consists of a temperature controlled chamber, the sample set up comprising of a support holding the polymer film, and an LVDT system. The LVDT system comprises of the LVDT (HR100, Lucas Schaevitz, Inc.), and a signal conditioner (ATA/1000 Schaevitz) which is used for the length change measurement. The length change measurable using this set up is in the range of 1mm with the resolution of 1 micron. The data obtained from the LVDT system are collected using a DAQ board and stored in a PC. A more detailed version of the experimental set up can be found in references [39, 45]. The epoxy film used in the specific experiment analyzed was of thickness 60 microns and length of 40 mm. The Tg of the epoxy film measured using the DSC at the cooling rate of 10K/min is 79oC. The material is first heated to 85oC and maintained at that temperature for an hour. This treatment is performed to remove any previous thermal history effects. Then two sets of down-jump in 100 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 temperature experiments are performed from 85oC to 75oC and 72oC, respectively, with simultaneous measurement of the length change and the volume recovery is calculated using the following equation: R= S S P = % U − 1 PT . (4.5) Here, V(t) and V(∞), are the instantaneous and equilibrium volume respectively while l(t) and l(∞) denotes the instantaneous and equilibrium length respectively. The above experiments were performed using the sampling frequency of 1 Hz. It may be noted that each data point is obtained using the LabView program is an average of 150 points collected per second. This is due to the fact that the experiments are run for duration of 2 to 5 days leading to large volume of data. The length changes measured are in the range of 10 to 50 µm, which is at the low end of the measurement range of the LVDT. Figure 4.18 shows the time evolution of the volume recovery following the temperature jump from 85oC to 75oC and 72oC respectively. As seen, the data obtained from this experiment was also found to be corrupted with noise. Similar to case study 1, this signal is also nonstationary. Further, due to the non-linear nature of the noise corrupting the signal it is difficult to determine the approach towards equilibrium. The possible noise sources for this data are:(i) white noise due to thermal fluctuations in the experiment (ii) low frequency noise induced due to the use of the LVDT at its lower limits to measure the change in length, and (iii) unknown sources, e.g., due to vibrations in the lab. These types of noise can also be successfully removed using EMD based filtering approach as seen in Figure 4.19. The key IMFs and the corresponding FFT’s are shown in 101 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.20. From Figure 4.20, it is observed that the contributing noise is white noise and low frequency LVDT noise. Noises at lower frequencies might be due to the use of the LVDT at the extreme lower end of the measurement system or temperature variations. Due to the presence of a core, an inertial element, the LVDT acts as low pass filter [46] and can introduce noise at low frequencies due to smaller vibrations. We beleive that noise at low frequneices observed in the FFT of IMFs is contributed by the LVDT as well as temperature variations while the broad band noises are due to change in the ambient conditions. It may be noted that when performing the down jump experiments from above to below the glass transition, two factors contribute to volume/length change namely the temperature change and the structural recovery. The contribution of the changing temperature to the length change dominates for the first 1000 seconds as it takes about that time to reach the derised down jump temperature. After that, the length change observed is primarily due to structural recovery with some contribution due to thermal fluctuations. The results shown in Figure 4.18 were obtained after removing the data obtained during the first fifteen minutes which contributes to the length change mainly due to temperature change. Ideally, the length change goes from non equlibrium to equilibrium position leading to a zero volume change at a finite time. However, due to drift in the data, it is difficult to exactly zero in on the equilibrium position. Although, the DSP techniques helps in the reduction of noise, it cannot fix systematic biases like drifts occuring in the experiments. Hence, there remains a challenge in estimating the exact equilibrium position as observed in Figure 4.20. 102 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4.4.6 Practical benefits due to use of the proposed EMD based FFT filtering algorithm The proposed EMD based approach provides a time domain decomposition of the given signal (IMFs) along with the precise frequency information present in each of the individual IMF. This can be used to identify the noise corrupting the experimental setup and improve upon the same. In simulation studies, the proposed approach provides correct information about the simulated noise added to the MKWW model generated data. However, as seen in Figure 4.12, at the shortest time scales there is a deviation of the filtered data from the original data leading to a relatively higher squared error at this time scale compared to other time scales. This can also be seen in Figure 4.13. Now, the question of interest is, can we use the information about the noise provided by the proposed approach to improve the experimental setup and obtain better filtered data?. Let us assume that we have improved the experimental setup to eliminate the source of the 14 Hz noise. Therefore, the modified KWW data contains noises at 60Hz, 180 Hz and 300 Hz sinusoidal noise signals. The results obtained from the EMD filtering approach for this dataset are shown in Figure 4.21. The sum squared error value for this case is 0.0047 which is twice less than the earlier simulation study. This improvment by a factor of 2 is achieived by removing a single frequency noise component in the data. That is, by improving a single sensor, the noise present in the output is reduced by a factor of 2. The information provided by EMD based approach along with process knowledge can be used to improve upon several sensors which could result in drastic reduction in the amount 103 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 of noise in the system. Further, for the particular case study, the improvement factor turns out to be 2. However, this factor could be more in specific applications. Thus, the information about the noise content from proposed EMD based approach can be used to improve the experimental setup to obtain better experimental data. In experimental case study I, frequency information obtained from the IMFs of the time domain dielectric response data revealed the presence of truncated ac noise (can be seen clearely from Figure 4.16). The truncation is due to the use of an autoranging electrometer for different voltage ranges. Therefore, to avoid the truncation effects and the ac noise the experimental would be improved by: (i) using a high voltage power supply with inbuilt sophisticated 60 Hz hardware filters and (ii) electrometers with higher resolution (to avoid the problem with autoranging). Thus, the proposed approach helps in identification of noise sources which can be used to improve the experimental setup. Further, if these improvements are economically not attractive, as discussed in the simulation studies using the MKWW model, it is better to use the EMD based FFT filtering approach which practically requires zero additional input from the user except the experimental data to do the filtering. In the experimental case study II, the FFT of the individual IMFs revealed the presence of low frequency noise. This is mainly due to the use of the LVDT at its specifed lower limits. One possible improvement is to use a more sensitive LVDT with much lower limits along with a better temperature controller unit. The proposed filtering approach can also be used in several areas such as (i) Time 104 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 domain dilatometry techniques (ii) time domain diffusive wave spectroscopy [47] and (iii) all time varying signals corrutped with noises, obtained from various experiments conducted in the field of polymer physics. As discussed earlier in the introduction section, Kremer and Schönhals [6] have argued that it is difficult to perform time domain dielectric spectroscopy analysis due to the isssues invovled in filtering. Now, using the proposed filtering approach filtering can be performed effectively so that time domain dielectric spectroscopy experiments can be conducted with improved short time capability. Here, by improved short time capability we refer to the following alternative approach that instead of using sinusoidal waves as input to the system, arbitrary waveforms containing multiple frequencies can be used as input. Thus, with the design of single arbitray wave like Pseudo random binary signals (PRBS, which contains all frequencies) as inputs, the corresponding noisy outputs have be obtained. These outputs can be filtered effectively using the proposed method effectively. Thus information at all frequencies can be obtained effectively with use of signals apart from sinusoidal waves. Further, they claim that material behaviors at low frequencies can be studied more readily in time domain as opposed to frequency domain conditions since experiments with low frequency sinusoidal signals take longer time In fact, design of input signals to obtain maximum information from the object of interest is a growing area of interest in the fields of process control and electrical engineering [48] and also in rheology [49, 50] and thermal analysis [51]. The proposed EMD based filtering approach provides an opportunity to broaden the benefits of time domain experiments by 105 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 improving the short time data reliability by filtering. It may also be noted that each frequency measurement requires multiple sine waves to capture the necessary information whereas we could acheive much better results in a much less time applying the EMD approach to the time domain data. 4.5 Conclusions Application of Empirical Mode Decomposition (EMD) as an effective tool to filter noise in time domain data in the field of polymer physics has been proposed. The proposed EMD filtering algorithm provides: (i) information which can be used to identify the noise corrupting the process and, (ii) precise frequency content of the signal using FFT of the IMFs (Intrinsic Mode Functions). The utility of this method as a tool to filter non-stationary data corrupted with nonlinear noise is demonstarted successfully using the following case studies: (i) modified Kohlrausch Williams Watts (MKWW) model generated data corrupted with harmonic and random noise, (ii) time domain dielectric compliance data and, (iii) volume recovery data after temperature jumps in polymer materials. In the case study using the MKWW model, we have also illustrated the effectiveness of the EMD based filtering approach by performing a comparison study with MA filters, FFT and DWT based filtering methods. Further, the ability of EMD based algorithm to separate data from noise, for dielertic time domain data, without losing any features of the actual data, can increase the horizon of the current experimental setup, to further probe into the nonlinear behavior and dynamic heterogenity of glassy polymers using the sensitive holeburning 106 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 technique. We have demonstated that EMD based algoritm is also sensitive to filter noise in the structural recovery experiments after temperature jumps, where the signal to noise ratio is relatively low since measurements are made in the lower limits of the LVDT. This method is applied in the next chapter to filter such noises from the structural recovery experiments of temprature formed and concentration formed glasses which further aid in confoming some suprising results. 4.6 Appendix 4.6.1 Mathematical Details of various filtering techniques Moving Average (MA) filters or Boxcar averages Moving average filters are the most commonly used filter in the field of Digital Signal Processing (DSP) [8]. This is mainly due to the fact that it is easy to use and can significantly reduce the random noises present in the data. MA filter averages a number of points in the input signal (signal to be filtered) to produce a single point corresponding to the filtered signal (output of the MA filter). In otherwords, for a given noisy signal x, the output of the MA filter y is obtained using the followoing formula Z- 1 X V + YW 1 V W = ",[ In the above equation, M is the number of points over which average is computed. For example, in a 3 point MA filter, M is 3 MA filter can also be thought of as performing a convolution of the input noisy signal with a simple rectangular 107 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 kernel of size M to obtain the filtered signal. This moving average technique with rectangular kernels are also known as Boxcar averages. Several variants of this filter with different kernels are also widely used for filtering of noisy signals. Moving average filters have little ability to separate one band of frequencies from another [8]. Since MA filters operate by performing averages over a particular rectangular window, it can introduce biases when the signal is nonstationary. Fourier Transform (FT) based filters Fourier transform is applied to implement a well known class of filters called as Finite impulse response (FIR) filters. The equation governing the FIR filters is given by [52] 1: = X \V]W ∗ : − ] In the above equation, y(n) is the filtered signal and x(n) is the input signal. At any instant, the output of the filter is depends on the past input signal. The corresponding frequency response of the system is given by ^_ = X \ V]W ∗ ! 5 Q`" 8/a In the above equation, f denotes the frequency and j is the complex number. This frequency response of the FIR filter is the same as the of the Fourier transform of the filter coefficients. Therefore, the value of the filter coefficients of FIR filter is obtained by taking inverse Fourier Transform of the desired frequency response.. For instance, if it is required to filter the 60 Hz frequency signal from the data, the deisred frequency response corresponding to this requirement is constructed. Then, 108 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 an inverse FT of the desired response is computed to obtain the filter coefficients. Once the filter coefficients are obtained, the data is passed through the FIR filter to obtain the filtered signal. Discrete Wavelet Transform (DWT) based filters DWT analyzes the given signal by passing it through a series of low and high pass filters. In otherwords, the signal is decomposed into coarse and detail components using the low pass and high pass filters respectively. Once the given signal is decomposed at first stage as coarse and detailed components, the amount of samples can be halved (which drastically reduces the computation burden) due the Nyquist criterion after which next level of decomposition is performed. The following equations describe the operation involved in DWT at every level [53] 1:bOcb = X V]W ∗ dV2: − ]W 1:Pef = X V]W ∗ ℎV2: − ]W In the above equations, g(n) and h(n) denote the high and low pass frequency filters respectively. The factor of 2 is present to indicate that subsampling is performed in computation of DWT. Once the signal is decomposed to obtain various frequencies along with the time information, the desired coarse and details components are used for obtaining the filtered signal. Thus DWT can be thought of as a bank of low and high pass filters providing both time and frequency information to the user for filtering of the noisy signal. 109 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 4.7 References 1. Henry O. Noise reduction techniques in electronic systems; Wiley, Johns & Sons, 1988. 2. Ingad U. Noise reduction analysis; Jones and Barlett publishers: LLC, 2009. 3. Vincent TC, Dhruti T, Matthew W, Nahid NJ. Noise analysis and reduction in solid state nanopores. Nanotechnology 2007; 18: 305505 1. 4. Ogawa T, Kurachi S, Kageshima M, Naitoh Y, Li JY, Sugawara Y. Step response measurement of AFM cantilever for analysis of frequency-resolved viscoelasticity. Ultramicroscopy 2010; 110: 612. 5. Feuerstein D, Parker HK, Boutelle. Practical methods for noise removal: Applications to spikes, nonstationary Quasi- periodic noise, and baseline shift. G.M. Anal. Chem. 2009; 81:4987. 6. Kremer F, Schönhals A. Broadband Dielectric Spectroscopy. Springer-Verlag: Berlin, Heidelberg, 2003. 7. Wilhelm M, Maringl D, Spiess, HW. Fourier transform rheology. Rheol. Acta 1998, 37:399. 8. Smith SW. The Scientists and Engineers guide to Digital Signal Processing.: California Technical Publishing: California, 1996. 9. Boaz Porat. A course in digital signal processing. John Wiley, 1997 10. Cohen.L. Time frequency analysis: theory and application. Prentice Hall: NJ, 1995. 110 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 11. Huang NE.; et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proceedings of Royal society of London 1998; 454:903. 12. Mecklenbrauker TCM, Classen WFG. On stationary linear time-varying systems. IEEE transactions on circuits and systems 1982; 29: 169. 13. Mallat S. Wavelet tour of signal processing. Academic Press, 2nd edition, 1999. 14. Piersol JS, Bendat, AG. Random data: Analysis and measurement procedures. John Wiley& sons: NJ, 3rd edition, 2000. 15. Brigitte Forster, Peter Massopust. Four short courses on harmonic analysis. Birkhauser: Boston, 2000. 16. Chun Li. Inter Harmonics: Basic concepts and techniques for their detection and measurement. Electrical Power Systems research 2003; 68:39. 17. Barclay VJ, Bonner RF, Hamilton PI. Application of wavelet transforms to experimental spectra: smoothing, denoising and data set compression. Anal. Chem. 1997; 69:78. 18. Huang NE, Shen SP. Hilbert Huang transform in Engineering. CRC Press, 2006. 19. Ferry JD. Viscoelastic properties of polymers. John Wiley and Sons: New York , 3rd edition, 1980. 20. Pizzolato N, Fiasconaro A, Spagnolo B. Noise driven translocation of short polymers in crowded solutions. Journal of stastical mechanics: Theory and experiments..2009; 1: P01011. 111 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 21. Balik CM. On the extraction of diffusion coefficients from gravimetric data for sorption of small molecules by polymer thin films. Macromolecules 1996; 29:3025. 22. Ma CB, Xu D, Ren Q, Lv ZH, Yang L. Simple transmission technique for measuring the elctro poled polymer optic coefficients of films. Journal of materials science letters 2003; 22:49. 23. Kranthi KM, Sanjay G, Mohammad RKM. On the cytoskeleton and soft glass rheology. Journal of Biomechanics 2008; 41:1467. 24. Ghosh A, Badiger MV, Tapadia PS, Ravi Kumar V, Kulkarni BD. Characterization of chaotic dynamics I:dynamical invariants of sheared polymer solutions. Chemical Engineering Science 2001; 56: 5635. 25. Isambert H, Maggs AC. Dynamics and rheology of actin solutions. Macromolecules 1996; 29:1036. 26. Wilhelm M, Reinheimer P, Ortseifer M. High sensitivity fourier-transform rheology. Rheol.Acta 1999; 38:349. 27. Wilhelm, M. Fourier transform rheology.Macromolecular materials and engineering 2002; 287:83. 28. Carotenuto C, Grosso M, Maffettone PL. Fourier transform rheology of dilute immisible polymer blends: A novel procedure to probe blend rheology. Macromolecules 2008; 41:4492. 29. Clemeur N, Rutgers RPG, Debbaut B. On the evaluation of some differential formulations for the pom-pom constitutive model. Rheol. Acta 2003; 42: 217. 112 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 30. Hyun K, Wilhelm M. Establishing a new mechanical nonlinear coefficient Q from FT-Rheology: First investigation of entagled linear and comb polymer systems. Macromolecules 2009; 42: 411-422. 31. Mopsik, FI. Relaxation: The quantitative application of time domain techniques to dielectrics. IEEE Transactions on Dielectrics and Electrical Insulation 2002; 9:829-837. 32. Kohlrausch F. Ueber die elastische Nachwirkung bei der Torsion. Pogg Ann Phys Chem. 1863; 119:337. 33. Williams G, Watts DC. Non-symmetrical dielectric relaxation behavior arising from a simple empirical decay. Trans Faraday Soc. 1970; 66:80.. 34. Srinivasan B, Gorai P, Tangirala AK. Detection and Quantification. Advances in adaptive data analysis 2009; 2: 309. 35. Ranganathan S, Rengaswamy R, Miller R. A Modified Empirical Mode Decomposition (EMD) Process for Oscillation Characterization in Control Loops. Contrtol Engineering Practice 2007; 59:1135. 36. Addison PS. The illustrated wavelet transform handbook. Institute of Physis: London, 2002. 37. Kollengodu-Subramanian S, McKenna GB. A dielectric study of poly(vinyl acetate) using pulse probe technique. Journal of Thermal analysis and calorimetry 2010; 102:477. 38. Kovacs AJ. Transition vitreuse dans les polymeres amorphes. etude phenomenoloqique. Fortschritte der. Hochpolymeren-Forschung 1963; 3:394. 113 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 39. Zheng Y, McKenna GB. Structural recovery in a model Epoxy: Comparison of responses after temperature and humidity jumps. Macromolecules 2003; 36:2387-2396. 40. Simon SL, Soieseki W, Plazek DJ. Volume and enthalpy recovery of polystyrene. Polymer 2001; 42:2555. 41. Alcoutlabi M, McKenna GB, Simon SL. Analysis of the development of isotropic residual stresses in a bismaleimide/sprio Orthocarbonate thermosetting resin for composite materials. Journal of Applied Polymer science 2003; 88: 227. 42. McKenna GB. On the physics required for the prediction of long term performance of polymers and their composites. J. Res. NIST 1994; 99:169. 43. McKenna GB, Vangel MG, Rukhin AL, Leigh SD, Lotz B, Straupe C. The tau effective paradox revisited: an extended analysis of Kovacs volume recovery data on poly(vinyl acetate). Polymer 1999; 40: 5183. 44. McKenna GB, Simon SL. The glass transition: its measurement and underlying physics. Handbook of thermal analysis and calorimetry, 3, Chapter 2, 2002. 45. Alcoutlabi M, Vangosa FB, McKenna GB . Effect of chemical activity jumps on the viscoelastic behavior of an epoxy resin: physical aging response in carbon dioxide pressure jumps. Journal of polymer science Part B: Polymer Physics 2002; 40: 2050. 46. Northrop RB. Introduction to instrumentation and measurements. Taylor and Francis group: LLC, 2nd edition, 2005. 114 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 47. Viasnoff V, Lequeux F, Pine DJ. Multispeckle diffusive wave spectroscopy: A tool to study slow relaxations and time dependent dynamics. Review of scientific instruments 2002; 73: 2336. 48. Parker RS, Heemstra FJ, Pearson RK, Ogunnaike BA. The idendification of nonlinear models for the process control using tailored”plant friendly” input squences. J. Process control 2001; 11: 237. 49. Klein C, Venema P, Sagis L, Van der Linden E. Rheological discriminization and characterization of carrageenans and starches by fourier- transform rheology in the non-linear viscous regime. J. non newtonian fluid mech. 2008;151:145. 50. Liu TY, Mead D W, Soong DS, Williams MC. A parallel plate rheometer for the measurement of steady state and transient rheological properties. Rheologica acta 1983; 22: 81. 51. Wunderlich B, Androsch R, Pyda M, Kwon YK. Heat capacity by multi frequencies saw tooth modulations. Thermochimica acta 2000; 348:181. 52. Baher H. Analog & digital signal processing. John Wiley & Sons: West Sussex, 2nd Edition, 2002. 53. Robi Polikar. Wavelet tutorial http://www.cs.ucf.edu/courses/cap5015/WTpart4.pdf 115 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Table 4.1: Sum squared error between the MKWW model data and filtered data obtained from various filtering approaches Filtered data from various approaches MA filtered data Window size 5 MA filtered data Window size 10 FFT filtered data DWT filtered data – 5 level decomposition(with a5 ) DWT filtered data – 6 level decomposition(with a5 ) EMD based FFT approach Mean squared error (e2(t) = y(t) – yfilt(t))2) 9.3259 2.2614 3.2240 0.3476 0.0172 0.0151 116 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.1: Data generated using the MKWW model (without noise). 117 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.2: Simulated noise added to the data generated from the MKWW model (equation 4.4). 118 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.3: Autocorrelation function of the white noise 119 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.4: Simulated MKWW model data with noise n(t) added. 120 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.5: MA filtered noisy model data – Window size 5 121 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.6: MA filtered noisy model data – Window size 10. 122 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.7: FFT filtered noisy model data. 123 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.8: Schematic of DWT for the simulated data. 124 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.9: Filtered data using Discrete Wavelet Transform – Five level decomposition 125 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.10: Filtered data using Discrete Wavelet Transform – Six level decomposition 126 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.11: IMFs and their corresponding magnitude at various frequencies obtained from FFT for the simulated data with noise 127 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.12: Filtered data using EMD based FFT approach 128 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.13: Squared error comparison plots for the filtered data obtained using FFT, DWT and EMD based FFT Approach 129 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.14: (a) Zero voltage time domain data, (b) Magnitude spectrum of the data at various frequencies using FFT 130 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.15: Experimental dielectric time domain data. 131 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.16: IMFs and their corresponding magnitude at various frequencies obtained for the experimental dielectric data 132 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.17: Comparison of experimental dielectric data and filtered data from EMD based FFT approach (Inset shows a zoomed version) 133 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.18: Comparison of volume recovery data after performing a down jump experiment from 85oC to 75oC and 72oC without filtering 134 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.19: Comparison of volume recovery data after performing a down jump experiment from 85oC to 75oC and 72oC after filtering using EMD method 135 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.20: IMFs and their corresponding magnitude at various frequencies obtained for the experimental volume recovery after a down jump in temperature to 75oC 136 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 4.21: MKWW generated data after removing the noise at 14Hz 137 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 5 AGING AND STRUCTURAL RECOVERY BEHAVIORS IN EPOXY FILMS SUBJECTED TO CARBON DIOXIDE PLASTICIZATION JUMPS: EVIDENCE FOR A NEW GLASSY STATE 5.1 Motivation Structural recovery and physical aging of glassy polymers after temperature jumps have been very well studied in literature [1-7]. On the contrary, there is only limited work available on the aging and recovery behaviors of glassy polymers subjected to plasticizer jumps [8-13]. Plasticizers are known for causing a reduction in the glass transition temperature, which in turn alters the mechanical properties of the glassy polymers. We have shown in our previous works, using strong and weakly polar plasticizers, that they qualitatively mimic the behaviors of temperature jumps but quantitatively are different [8-10, 13]. In this work, we further investigate this anomalous behavior by studying the structural recovery and physical aging of an epoxy film subjected to carbon dioxide pressure jumps and compare the results with temperature jump experiments such that the final conditions are identical. 5.2 Introduction Glassy polymers are used extensively in very many industrial applications. Most of the applications use polymers in the vicinity or below its glass transition temperature or concentration (Tg or Φg). As shown in Figure 5.1, when a material is cooled from a higher temperature, the point at which the thermodynamic property such as volume, 138 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 enthalpy or entropy deviates from the equilibrium path is called the glass transition temperature (Tg) [1] and such a material is called temperature-formed glasses. The point of deviation upon depressurization instead of cooling when the material is subjected to a plasticizer environment is called the glass transition concentration (Φg) and the material formed during this process is called concentration-formed glasses. Below the glass transition point, the material is in non equilibrium state and hence will tend to evolve towards the equilibrium state. The study of thermodynamic properties as the material evolves towards equilibrium is called structural recovery [1] and the changes associated in the viscoelastic properties such as mechanical, optical, or dielectric during this process is called physical aging [1]. A proper understanding of aging and structural recovery behavior is essential to predict the long term stability of glassy polymers [2, 14, 15]. Aging and structural recovery of temperature-formed glasses have been widely studied and very well documented in literature [1-7]. However, there is only limited work available in understanding these behaviors of concentration-formed glasses [8-13, 16]. It is well documented that the plasticizers are known to depress the glass transition temperature of glassy polymers which can also create drastic impact on the properties of glassy polymers [16-28]. Therefore, it is important to understand these behaviors of concentration-formed glasses and their impact on material applications. Concentration glasses formed using supercritical CO2 have been known to show retrograde vitrification phenomenon [20, 25, 28]. Flemings and Koros [16] also observed hysteresis effect while studying the CO2 sorption-desorption and volume dilation work on polymers. The structural recovery work after humidity [9] and carbon dioxide jumps [13] 139 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 from our group were found to mimic the behaviors of temperature jump experiments qualitatively. However, quantitatively they were different. The concentration-formed glasses showed a larger departure from equilibrium compared to temperature formed glasses despite the excess volume [9, 13]. This anomalous behavior was also confirmed with physical aging experiments for both humidity and CO2 jumps [8, 10]. In our prior work on structural recovery of epoxy film subjected to CO2 plasticizer jumps, we surprisingly observed that at 10oC below the glass transition temperature, the effective retardation time for concentration-formed glasses and temperature-formed glasses (Subjected to same final condition) do not converge to the same point as equilibrium is approached [13]. This was kind of surprising. As a continuation of that work, we report the aging and structural recovery results of an epoxy film subjected to carbon dioxide plasticizer jumps (P-CO2) and compare the results with temperature jump experiments to same final conditions. 5.3 Experimental The material investigated in the current work is a thin epoxy film obtained by curing diglycidyl ether of bisphenol A (DGEBA, DER332) with amine terminated poly (propylene oxide) (T403, Huntsman). Epoxy resin DER 332 was first preheated at 55oC for duration of one hour to ensure that the resin is free of any crystal. A stoichiometric ratio of resin and curing agent is then mixed well in a container using a magnetic stirrer for a period of one hour under vacuum at room temperature. The degassed mixture is then neatly spread between two smooth brass plates and clamped. This is then placed 140 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 inside an oven and cured at 2 atm pressure and 100oC for 1 day. The sample preparation methodology is well described in references 9 and 13. Two different samples were used to study the aging behavior and structural recovery of glassy polymers respectively. The glass transition temperature of epoxy film, used for structural recovery and aging studies measured at the cooling rate of 10o/min using DSC, are 79oC and 75oC respectively. The dimension of the film used for structural recovery is 40mm in length, 60 micron thickness and 10 mm in width and that for physical experiments were 30 mm in length, 60 micron in thickness and 4mm in width. The stress applied for the aging experiments performed was 0.9 MPa. Both the physical aging and structural recovery experiments after pressure jump (P Jump) experiments were performed using the experimental setup built at Texas Tech University shown in Figure 2.3 [13]. A detailed explanation on the working of the experimental set up is given in chapter 2. 5.4 Method of analysis The departure from equilibrium (δ) for the structural recovery experiments were calculated using the equation 5.1 [1]. Here, V(t) is the instantaneous volume, and V(∞) is assumed to be the plateau volume. R= S S (5.1) The effective retardation time (τ-1eff ) is obtained using the equation 5.2 [1]. - j hi// = − j (5.2) 141 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 The creep compliance (D(t)) curves for aging experiments is calculated using equation 5.3 [8, 10]. k = (5.3) l Here, ε(t) is the instantaneous strain and σo is the applied stress. The creep curves obtained for T jump and P jump experiments were captured well using the Kohlrauch William Watt function (KWW) given in equation 5.4 [29, 30]. % n k = k- ! 5m8 (5.4) Here, D1, τ (retardation time) and β (shape factor) are fit parameters. 5.5 Results and discussion 5.5.1 Structural recovery experiments The comparison of the structural recovery experiments performed to same final condition of 0 MPa and 72oC after T jump and P-CO2 jump respectively is shown in Figure 5.2. The results reported in Figure 5.2, are the data obtained after performing EMD based filtering to remove the noise in the raw data [31]. We observe that the anomalous behavior, which is the departure from equilibrium, is much higher for P jump experiments compared to T jump experiments. Further, the P jump experiments take a longer time to reach the equilibrium compared to the T jump experiments. To further understand this result, we studied the comparison in terms of effective retardation plot (τ-1eff ) as shown in Figure 5.3. It may be noted that data were smoothened using the origin software before calculating the effective retardation time for concentration and 142 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 temperature formed glasses. Similar to results obtained previously in our group [13], we observed that the results of temperature and concentration formed glasses do not converge to the same value as equilibrium is approached. This result indicates that the kinetics of the concentration formed glasses is different from temperature formed glasses and the equilibrium is path dependant. To investigate this behavior further, we performed the physical aging experiments on the epoxy film subjected to temperature and pressure jumps. 5.5.2 Aging experiments Aging results reported in this work, were performed following the Struik protocol shown in Figure 1.9 [2]. According to Struik’s protocol, the sample loading-unloading for the aging experiment is performed in such a way that the loading-unloading time is one tenth of the waiting time. This is done so that each loading unloading step is independent of the previous event [2]. As seen in the Figure 1.9, the loading-unloading time is sequentially increased by a factor of 2. The creep compliance after a T- jump from 84oC to 69.3oC for different aging times is shown in Figure 5.4. The time-aging time superposition for the creep curves plotted in Figure 5.4 is shown in Figure 5.5. The solid line in Figure 5.5 is the KWW fit for the creep curves of longest aging time. Figure 5.6 shows the creep curves for different aging times after a PCO2- jump from 4.2 to 0 MPa and 69.3oC, and Figure 5.7 is time-aging time superposition curves for the P- jump experiments. Similar to the T- jump experiments, the KWW model captures the entire spectrum of creep curves reasonably well. The parameters for the KWW model obtained for both T- jump and PCO2 -jump experiments for the longest aging time of 143 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 230400s is given in Table 1. It may be noted that the aging experiments for T- jumps and PCO2 -jumps were performed to the same final condition of 0 MPa and 69.3oC. Figure 5.8 shows the creep response from the T jump and PCO2 -jump experiments of Figure 5.4 and 5.6 plotted together. Similar to the results in structural recovery experiments, we observe that the concentration-formed glasses take longer time to relax compared to the temperature-formed glasses, although the final conditions are same for both these experiments. The creep curves of PCO2 -jump experiments collapsed reasonably well to the longest aging time of T- jump experiments as shown in Figure 5.9, although the shape parameters (β) are different as seen in Table 5.1. Figure 5.10 is the horizontal shift factor plot for the time-aging time superposition curves of concentration and temperature formed glasses. Similar to the departure from equilibrium experiments, the horizontal shift factor of temperature and concentration formed glasses do not come to equilibrium at the same aging time. To further investigate the aging behavior of T- jump and PCO2- jump experiments, the retardation time obtained using KWW function for the creep curves of both these experiments were plotted as a function of aging time in Figure 5.11. The relaxation time for different aging times was estimated using the relationship between the shift factor and relaxation time given in equation 5.5. 2i = o o pqr (5.5) Where, ate is the time-aging time shift factor, τref is retardation time of the reference creep curve and τ is the retardation time of required creep curve. 144 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 It can be clearly observed that the retardation time as a function of aging time for T -jump and PCO2 -jump experiments do not converge to the same point as the equilibrium is approached. Moreover, there is close to a decade difference between the retardation times of these two glasses as equilibrium is reached. A similar result was reported in the earlier work of Alcoutlabi, Briatico-Vangossa, and McKenna [8], except that the aging experiments were not performed long enough for the experiments to attain equilibrium. Further, this result is surprising, as the retardation time results from Zheng and McKenna [10], on humidity and T- jump experiments converge to the same point as the equilibrium is reached as shown in Figure 5.12. It may be noted that Berens and Hodge’s work [32-34] on the DSC study of enthalpy recovery of concentration-formed glasses and hyper-quenched (temperature-formed) observed a sub Tg endotherm for both these glasses. Based on this result, they concluded that this is a common trait of aging phenomenon irrespective of type of glasses. In our prior work, we have shown that their finding is only partially correct since it is not a direct measurement method like volume recovery experiments [11]. The current results of structural recovery and aging experiments confirm that the equilibrium reached by temperature- and concentration-formed glasses due to carbon dioxide jumps is not the same equilibrium state. Further, the kinetics of temperature- and concentration-formed glasses are different. A question which could be asked at this point is whether the properties altered during the formation of concentration-formed glasses are reversible. Alcoutlabi, Banda and McKenna [11], have shown that by heating the concentration-formed glasses after complete depressurization ( partial equilibration for 24 hours), to above its glass 145 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 transition temperature, followed by cooling to room temperature of 25oC, and heating again above the glass transition temperature, the excess volume is shed and the material returns to its normal state. This can be seen in Figure 5.13. This reversing of the properties and kinetics of concentration formed glasses serves as evidence that the equilibrium state attained during the PCO2 -jump experiments is not a true equilibrium but a meta stable state. 5.6 Conclusion Structural recovery experiments after T- jump and P- jump experiments to the same final conditions were performed. We observe anomalous behavior similar to that previously observed in our group for humidity and carbon dioxide jump experiments [810, 13]. In addition, we also observe that volume recovery of temperature- and concentration-formed glasses do not reach the equilibrium at the same time. Further, the fact that the effective retardation times as a function of departure from equilibrium do not converge to same point as the departure from equilibrium approaches zero indicates two things (1) The kinetics of concentration-formed glasses are different from that of the temperature-formed glasses (2) The equilibrium state reached during the PCO2 -jump experiments is not the same as the T- jump experiments. To confirm this behavior, physical aging experiments were performed to the same final conditions for T-jump and PCO2 -jump experiments. They also confirm the anomalous behavior. Further, the retardation time plotted as a function of aging time shows clearly that the equilibrium state obtained for concentration formed glasses using carbon dioxide is different from the temperature formed glasses. The above result further 146 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 enhances the finding of structural recovery experiments about the existence of different equilibrium state for temperature- and concentration-formed glasses. Alcoutlabi, Banda, and McKenna [11], have shown that the concentration-formed glasses return to normal state upon heating above the glass transition temperature followed by cooling to room temperature and heating again. This confirms that the equilibrium obtained during the PCO2 -jump is an evidence for the existence of a new meta stable glassy state. This is interesting because such a behavior was not observed for PRH -jump experiments. There is no concrete explanation at this point for this underlying behavior and it needs to be further investigated. 5.7 References 1. Kovacs AJ. Transition vitreuse dans les polymeres amorphes. etude phenomenoloqique. Fortschritte der. Hochpolymeren-Forschung 1963; 3: 394. 2. Struik LCE. Physical aging in polymer and other amorphous materials. Elsevier: Amsterdam, 1978 3. McKenna, GB.“Glass formation and glassy behavior,” in Comprehensive Polymer Science. Vol.2, Polymer properties, ed. By C. Booth and C. Price, Pergamon, Oxford 1989; 311-363. 4. Hutchinson JM. Physical aging of polymers. Progress in Polymer Science 1995; 20:703. 5. Hodge IM. Enthalpy relaxation and recovery in amorphous materials. J. NonCryst. Solids 1994; 169:211. 6. Simon SL, Soieseki JW, Plazek DJ. Volume and enthalpy recovery of polystyrene. Polymer 2001; 42:2555. 147 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 7. Bernazzani P, Simon SL. Volume Recovery of Polystyrene: Evolution of the Characteristic Relaxation Time. Journal of Non-Crystalline Solids 2002; 307310C:470. 8. Alcoutlabi M, Vangosa Briatico F, McKenna GB. Effect of chemical activity jumps on the viscoelastic behavior of an epoxy resin: physical aging response in carbon dioxide pressure jumps. Journal of polymer science Part B: Polymer Physics 2002; 40:2050. 9. Zheng Y, McKenna GB. Structural recovery in a model Epoxy: Comparison of responses after temperature and humidity jumps. Macromolecules 2003; 36: 2387. 10. Zheng Y, Priestley RD, McKenna GB. Physical aging of an epoxy subsequent to relative humidity jumps through the glass concentration. Journal of polymer science Part B: Polymer Physics 2004; 42:2107. 11. Alcoutlabi M, Banda L, McKenna GB. A Comparison of Concentration-Glasses and Temperature-Hyperquenched Glasses: CO2-Formed Glass vs. TemperatureFormed Glass. Polymer 2004; 45:5629. 12. McKenna GB. Glassy States: Concentration Glasses and Temperature Glasses Compared. J. Non-Crystalline Solids 2007; 353:3820. 13. Alcoutlabi M, Banda L, Kollengodu-Subramanian S, Zhao J, McKenna GB. Environmental effects on the structural recovery responses of an epoxy resin after carbon dioxide pressure-Jumps: Intrinsic isopiestics, asymmetry of approach and memory effect. Macromolecules 2010(Under Review). 14. McKenna GB. On the physics required for the prediction of long term performance of polymers and their composites. J. Res. NIST 1994; 99:169. 148 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 15. Alcoutlabi M, McKenna GB, Simon SL. Analysis of the development of isotropic residual stresses in a bismaleimide/sprio Orthocarbonate thermosetting resin for composite materials. Journal of Applied Polymer science 2003; 88: 227. 16. Fleming GK, Koros WJ. Dilation of polymers by sorption of carbon dioxide at elevated pressures. Macromolecules 1986; 19: 2285. 17. Chan AH, Paul DR. Influence of history on gas sorption, thermal, mechanical properties of glassy polycarbonate. J. Appl. Polym. Sci.1979; 24: 1539. 18. Koros WJ, Paul DR. Sorption and transport of CO2 above and below the glass transition of poly(ethylene terephthalate). Polym. Eng. Sci. 1980; 20:14. 19. Knauss WG, Kenner VH. On the hygrothermomechanical characterization of polyvinyl acetate. J. Appl. Phys. 1980; 51: 5531-5536. 20. Wang WCh, Kramer EJ, Sachse WH. Effect of high pressure CO2 on the glass transition temperature and mechanical properties of polystyrene. Journal of polymer science Part B: Polymer Physics 1982; 20: 1371. 21. Chiou JS, Barlow JW, Paul DR. Plasticization of glassy polymers by CO2. J. Appl. Polym. Sci. 1985; 30:2633. 22. Chiou JS, Barlow JW, Paul DR. Polymer crystallization induced by sorption of CO2 gas. J. Appl. Polym. Sci. 1985; 30:3911. 23. Wissinger RG, Paulaitis ME. Swelling and sorption in polymer-CO2 mixtures at elevated pressures. Journal of polymer science Part B: Polymer Physics 1987; 25: 2497. 149 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 24. Wissinger RG, Paulaitis ME. Glass transition in polymer/CO2 mixtures at elevated pressures. Journal of polymer science Part B: Polymer Physics 1991; 29: 631. 25. Condo PD, Johnston KP. Retrograde virtrification of polymers with compressed fluid diluents: Experimental confirmation. Macromolecules 1992; 25:6730. 26. Condo PD, Johnston KP. In situ measurement of glass transition temperature of polymers with compressed fluid diluents. Journal of polymer science Part B: Polymer Physics 1994; 32; 523. 27. Zhang C, Cappleman BP, Defibaugh-Chavez M, Weinkauf DH. Glassy polymersorption phenomena measured with quartz crystal microbalance technique. Journal of polymer science Part B: Polymer Physics 2003; 41:2109. 28. Handa YP, Zhang Z. A new technique for measuring retrograde vitrification in polymer-gas systems and for making ultramicrocellular foams from retrograde phase. Journal of polymer science Part B: Polymer Physics 2000; 38:716. 29. Kohlrausch F. Ueber die elastische Nachwirkung bei der Torsion. Pogg Ann Phys Chem. 1863; 119:337. 30. Williams G, Watts DC. Non-symmetrical dielectric relaxation behavior arising from a simple empirical decay. Trans Faraday Soc. 1970; 66: 80. 31. Kollengodu-Subramanian S, Babji S, Zhao J, Rengaswamy R, McKenna G B. Application of empirical mode decomposition in polymer physics. Journal of polymer science. Part B. Polymer Physics 2011; 49:277. 150 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 32. Berens AR, Hodge IM. Effects of annealing and prior history on the enthalpy relaxation in glassy polymers. I Experimental study of polyvinyl chloride. Macromolecules 1982; 756. 33. Hodge IM, Berens AR. Effects of annealing and prior history on the enthalpy relaxation in glassy polymers. 2. Mathematical modeling. Macromolecules 1982; 762. 34. Hodge IM. Effects of annealing and prior history on the enthalpy relaxation in glassy polymers. 4. Comparison of 5 polymers. Macromolecules 1983, 16:898. 151 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Table 5.1: KWW parameters for the longest aging time (230400s) of T and P jump experiments Parameter T jump P jump D1/Pa 1.2 E-9 1.43E-9 β 0.28 ± 0.2 0.46 ± 0.2 τ/s 1820 ± 80 60534 ± 500 152 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.1: Schematic representation of specific volume as a function of temperature or concentrations 153 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.2: A comparison of departure from equilibrium as a function of time for T jump and P jump experiment subjected to same final condition of 72oC and 0 MPa. 154 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.3: Effective retardation time as a function of departure from equilibrium for T and P jump experiments of same final condition 155 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.4: Creep compliance curves for different aging time plotted as a function of time for T jump experiment 156 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.5: Time-aging time superposition of creep curves of the T jump experiment 157 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.6: Creep compliance curves for different aging time plotted as a function of time for P jump experiment 158 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.7: Time-aging time superposition of the creep curves of P jump experiment 159 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.8: Creep compliance curves for different aging time for T and P jump experiments subjected to same final condition 0MPa and 69.3oC 160 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.9: Time aging time superposition curves for P jump experiment superposed to the longest aging time of T jump experiment 161 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.10: Horizontal shift factor as a function of aging time. The concentration glasses are shifted with respect to the longest aging time of temperature formed glass 162 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.11: Retardation time obtained from KWW function for T and P jump creep curves plotted against the aging time 163 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.12: Retardation time as function of aging time for humidity and T jump experiments of same final condition taken from reference [10]. 164 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 5.13. Volume recovery of epoxy film showing the reversal of concentration formed glass to temperature formed glass upon heating above its Tg followed by cooling to room temperature and heating [11]. 165 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 6 SUMMARY AND CONCLUSIONS In this work, we have investigated the time domain responses of glassy polymers in the vicinity of its glass transition temperature. As a part of this thesis work, Time Domain Dielectric Spectrometer (TDS) was built in the Polymer and Condensed matter group at Texas Tech. Its successful working is demonstrated by studying the dielectric response of poly(vinyl acetate) subjected to pulse probe technique. After establishing the linear response function in single step experiments, two types of pulse-probe experiments were performed. In one, the time duration t1 of the first step in the probe was varied. In the second case, the magnitude of the field E1 applied to the sample for the first step was varied. The memory effect was observed similar to what is observed in mechanical or thermal responses. This was in a quantitative agreement with linear Boltzmann superposition for small applied fields. However, evidence of breakdown of linearity was observed at a relatively larger applied field. This is an interesting result, as we were able to demonstrate the ability to delineate between linear and nonlinear behaviors in time domain, instead of the traditionally used high amplitude sinusoidal pulses in frequency domain. Further, by using a modified KWW function with an additional term to capture the dc conductivity, we successfully separated the conductivity effect from the dielectric relaxation in the time domain itself. Dielectric time domain data obtained from the TDS were nonstationary in nature and were found to be corrupted with nonlinear noise. This hindered the application of the current TDS setup for more sensitive dielectric applications, like probing the dynamic 166 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 heterogeneity. To circumvent the noise problems in instrumentation, and in general to address the filtration issues associated with nonstationary data corrupted with nonlinear noise in the field of polymer physics, we used an Empirical Mode Decomposition (EMD) algorithm to filter the noise. EMD is an effective tool developed by the signal processing community to filter non stationary data (time varying signal with no constant mean) corrupted with nonlinear noises. They work on the principle of breaking the time domain data into various components called intrinsic mode functions (IMF) such that the actual signal is the sum of all IMFs. The frequency information of each IMF is then obtained using a fast Fourier transform (FFT). Then based on the information obtained, the IMF’s related to the noises are omitted and the remaining IMF’s are added to construct the filtered data. The effectiveness of EMD filter for data analysis over the traditionally used MA, FFT and DWT filters was demonstrated using a comparative study of modified KWW generated model data corrupted with nonlinear and random noise. Further, dielectric data obtained from TDS was also successfully filtered using the EMD based filtering approach. Besides filtering the data, the EMD method also gives insightful information about the noise, which could be used to improve the experimental set up. Structural recovery and aging experiments of glassy polymers are also time domain responses essential to predict the long term behavior of glassy polymers. These behaviors are very well understood for temperature-formed glasses. On the contrary, there is only limited work done in understanding these behaviors in the context of concentration (Plasticizers)-formed glasses. Previous work from McKenna’s group has 167 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 shown that concentration-formed glasses (P-RH, P-CO2) qualitatively mimic the temperature-formed glasses but quantitatively they were different. The initial departure from equilibrium for concentration-formed glasses after both humidity (strongly polar) and carbon dioxide (weakly polar) jumps, were higher than the temperature-formed glasses. Further, our preliminary work on the structural recovery of epoxy film subjected to CO2 plasticizer jumps showed that the effective retardation time for concentrationformed glasses and temperature-formed glasses (subjected to same final condition) do not converge to the same point as equilibrium is approached. Hence, we further investigated this behavior by studying the aging and structural recovery of epoxy film subjected to carbon dioxide plasticizer jumps and compared them with the temperature jump experiments of same final condition. It may be noted that the data obtained from the structural recovery experiments after temperature and plasticizer (P-CO2) jumps were also found to be time varying in nature corrupted with nonlinear noise. Moreover, this is a very sensitive experiment with measurements made in the lower limits of LVDT which lead to relatively lower signal to noise ratio. EMD based algorithm was successfully used to filter these noise in these experiments. The data obtained after filtering the raw structural recovery data from the temperature and pressure jump experiments to the same final condition validated the anomalous observations made previously in our group. Further, due to better filtering, we were able to clearly observe that the volume recovery for T jumps and P-CO2 jumps were not coming to equilibrium at the same time. In addition, the effective retardation plot showed that the concentration- and temperature-formed glasses do not converge to the same value as the departure from equilibrium approaches zero, in agreement with 168 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 preliminary results of our previous work. This behavior was further confirmed with a comparative study of the aging experiments of temperature- and concentration-formed glasses to the same final condition. It was found that the concentration-formed glasses relax slower than the temperature-formed glasses. Further, the time-aging time superposition curves of concentration-formed glasses to longest aging time curve of temperature-formed glasses was good. The retardation time of aging creep curves of temperature- and concentrationformed glasses was estimated using the KWW function. It was found that the retardation time as equilibrium was approached, differed by almost a decade between T jump and P jump experiments. Similar to the volume recovery results, the retardation time for PCO2 jump experiments were longer than T-jump experiments. This result indicates that the equilibrium attained by the concentration- and temperature-formed glasses were not the same. The existence of two equilibriums is a surprise, as our previous work with humidity jumps did not show such a behavior. However, Alcoutlabi, Banda, and McKenna have also shown that the equilibrium attained by the concentration formed glasses is reversible. This result further suggests that this equilibrium obtained with concentration-formed glasses is only a metastable. 169 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 CHAPTER 7 FUTURE WORK 7.1 Introduction Time domain responses of glassy polymers in the vicinity of glass transition have been studied in this work using dielectric and mechanical measurements. The understanding of these properties is essential to predict and modify the long term stability and performance of glassy polymers. There is a tremendous potential in this area of research. 7.2 Time domain nonlinear dielectric In the time domain dielectric study of glassy polymer, we have demonstrated that it is possible to delineate between linear and nonlinear behavior in glassy polymers using linear Boltzmann superposition. This is a good start in this area which needs to be further explored. The objective of future work would be to explore the possibility of using nonlinear Boltzmann model to capture and explain this nonlinear behavior. The modified Boltzmann equation similar to the one used in studying nonlinear mechanical responses, given in equation 1 [1, 2, 3] can be applied to verify the same. v = ℎe s te s + ue ∆t − h vo ℎ- VshW&h (7.1) Where Jo is the glassy creep compliance, σ(τ) is the instantaneous stress and ε (t) is the instantaneous stain. The modified Boltzmann for nonlinear behavior is based on that assumption that when you double your stress, the strain is doubled as a function of stress instead of just 170 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 the stress as in linear case [1, 2, 3]. The effective working of this model for a single step response in mechanical measurement of polycarbonate is well demonstrated by Luo and his coworkers [1]. Experimental methodology The nonlinear dielectric experiments can be performed using the time domain dielectric spectrometer built at Texas Tech. The necessary protocols are discussed in the experimental section in chapter 3. It may be noted that to study the nonlinear behaviors, the experiments have to be performed at a higher applied electric field. This could be achieved by reducing the thickness of the film, increasing the applied voltage or by using the combination of both. 7.3 Isobaric time domain dielectric measurements We have successfully demonstrated the working of time domain dielectric spectrometer by performing isothermal measurements of PVAc in chapter 3. As per our knowledge, there is no work done on understanding the dielectric behaviors of polymers when they are subjected to plasticizer environment. Since we have the facility to perform such experiments, it would be of fundamental interest to perform the same. This work will be based on the hypothesis that, similar to mechanical measurements in polymers, the dielectric measurements are sensitive to plasticizer concentration. Experimental methodology The schematic of the experimental set up for studying the dielectric responses of glassy polymers subjected to plasticizer environment (Carbon dioxide) is shown in Figure 171 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 7.1. The set up comprises of a pressure vessel with a maximum capacity of 6MPa inside which the sample set up (See chapter 3, Figure) is placed. The working of the pressure vessel is well explained in the experimental section in chapter 2. Dielectric compliance can then be measured as a function of time at various pressures (Plasticizer concentration) to verify the hypothesis. Further, there is also a possibility to probe dynamic heterogeneity, by replacing carbon dioxide with a more polar plasticizer, which then can be used to further perform hole burning experiments which are discussed in Chapter 1. 7.4 Physical aging and structural recovery of glassy polymers The other time domain study, namely, the aging and structural recovery of glassy polymers subjected to carbon dioxide plasticizer jump, has indicated two things. One is that similar to humidity jump experiments, P-CO2 jump experiments show anomalous behavior compared to the temperature jump experiments of the same final condition. The second thing is that we have observed the evidence for a new glassy state in P-CO2 jump experiments. This is very interesting; as such a behavior was not observed with the concentration-formed glasses subjected to humidity jumps and needs further investigation. An enthalpy recovery study after plasticizer and temperature jump would be another method which could complement the current work. This experiment will not only help in confirming the anomalous behavior irrespective of the measurement technique but can also validate evidence of the new glassy state observed with the carbon dioxide concentration glasses. 172 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 7.5 References 1. Findley WN, Lai JS, Onaran K. Creep and relaxation of nonlinear viscoelastic materials with an introduction to linear viscoelasticity. New York: North-Holland Publication; 1976. 2. Schapery RA. On the characterization of nonlinear viscoelastic material. Polym. Eng.Sci.1969; 9:295. 173 Texas Tech University, Shankar Kollengodu Subramanian, May 2011 Figure 7.1: Schematic of time domain dielectric spectrometer with modifications to perform isobaric measurements of dielectric compliance 174