Group Project 6 – Investigation of Response Features The third step in the SHM process is based on extracting damage-sensitive features from tested data, which will then be analyzed to gain a better understanding of the health of the structure1. This paper investigates the effectiveness and feasibility of six different response features. 1. Frequency Domain Curve Fitting Curve fitting tools can be used to extract features from the experimental data. The polyreference frequency and the polynomial rational fraction-z curve-fitting procedures perform well in identifying natural frequencies for light to heavily damped data and for systems with closely spaced modes. To perform such an extraction, the computer software PULSE Reflex Modal Analysis by Brüel & Kjær was implemented. A MatLAB script (as seen in the Appendix) was also developed for the polynomial rational fraction curve fitting. In this investigation, the polyreference frequency method failed to identify some natural frequencies, while the second method identified many spurious modes. Table 1 shows first 6 modes identified by each method. Polyreference and RFP are both sensitive to noise and represent the underlying dynamics of a structure. They can be computationally demanding and are not as feasible for real-time monitoring as other features. However, they can be extremely useful in detecting damage within a system. Table 1. Rational fraction polynomial and polyreference frequency curve-fitting results Rational Fraction Polynomial-Z Mode Damped Frequency (Hz) Damping (%) 1 49.31767 3.77062 2 83.69389 2.61411 3 93.92983 2.65468 4 294.2724 3.32843 5 307.47 3.73831 6 329.1031 2.83005 Polyreference Frequency Damped Frequency (Hz) 48.76165 81.66915 297.7363 333.0991 377.2522 543.3877 Damping (%) 3.80878 1.34111 4.44323 3.03485 2.84609 5.61648 2. Coherence Coherence helps to relate the output signals to the measured input signals. Values of a system’s coherence should be between 0 and 1, with 1 being the desired outcome. Any value less than 1 is an indication of error within the system, whether it is leakage, extraneous noise, or even non-linearities within the structure2. Figure 1 represents a coherence plot from a set of data from a hammer test. In the range of 25 to 65 Hz, the coherence value is 1 and is indicative that there was no extra instrumentation noise nor was the system behaving non-linearly. However, the low coherence values for the other ranges indicate that there were problems during the testing. This particular selected response feature is sensitive to noise, can convey the relevant underlying dynamics of a structure, is feasible for real-time monitoring since computer software is able to easily compute it, and is a quick and easy method to detecting whether damage is present in the system or not. 3. Frequency Response Function The frequency response function (FRF) of a system, which Avitabile considers the most important measurement needed for experimental modal analysis, is a ratio of the output response to the input 1 Farrar, C. R., & Worden, K. (2007). An introduction to structural health monitoring. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 365(1851), 303-315. 2 Allemang, R. J. (2001). Vibrations: Experimental Modal Analysis; Structural Dynamics Research Laboratory; University of Cincinnati. UC-SDRL-CN-20-263-663/664. Page | 1 response. Here, to quantitatively represent the characteristics of the FRF, the area under the curve of FRF will be employed, computed as: Area FRF( f )df . The effectiveness of this FRF measure can be easily illustrated using the following example, shown in Figure 2. Although there is no distinct variation of the resulting FRF between two tests, a considerable variation of that computed Area is noticed: the area under FRF of Test 2 is less than that of Test 1, and that difference, which can be approximated to 2.1%, can’t be neglected. The frequency response function is sensitive to noise and can reflect the underlying dynamics of a structure. The FRF is an efficient procedure and is able to be used during real time monitoring. Overall, the FRF is an important feature for experimental modal analysis and is helpful for detecting damage within a system. Figure 1: Coherence of hammer testing 90 80 Magnitude 70 60 50 40 30 20 10 0 0 50 100 Frequency (Hz) 150 200 Figure 2: Illustrative example to demonstrate the effectiveness using the area under the curve of FRF to feature the FRF 4. Modal Assurance Criterion (MAC) MACis an experimental modal analysis tool used to help identify damage within a system. MAC, which is a statistical indicator, is a leased squares based form of linear regression analysis3. The computer software PULSE Reflex Modal Analysis by Brüel & Kjær was used to perform this analysis. Figure 3 represents the MAC information that was obtained. MAC is not very sensitive to noise and can reflect the underlying dynamics of a structure. Although MAC requires numerous computational requirements and is not feasible for real time monitoring, it can be a very helpful tool in detecting damage within a system, especially when comparing a numerical model of a plate to measured data. 3 Pastor, M., Binda, M., & Harčarik, T. (2012). Modal assurance criterion. Procedia Engineering, 48, 543-548. Page | 2 543.4 0.00 0.01 0.00 0.01 0.03 1.00 Polyreference Frequency Modes 48.8 81.7 297.7 333.1 377.3 543.4 Polyreference Frequency 48.8 81.7 297.7 333.1 377.3 1.00 0.00 0.01 0.00 0.01 0.00 1.00 0.01 0.02 0.01 0.01 0.01 1.00 0.05 0.01 0.00 0.02 0.05 1.00 0.03 0.01 0.01 0.01 0.03 1.00 0.00 0.01 0.00 0.01 0.03 Rational Fraction Polynomial-Z Modes 49.5 62.5 87.8 94.7 291.3 295.5 48.8 1.00 0.29 0.01 0.05 0.02 0.01 81.7 0.00 0.54 0.99 0.87 0.59 0.28 297.7 0.01 0.01 0.00 0.02 0.14 0.21 333.1 0.00 0.18 0.02 0.10 0.18 0.43 377.3 0.01 0.12 0.01 0.08 0.01 0.06 543.4 0.00 0.02 0.01 0.02 0.08 0.08 Figure 3: MAC for Polyref curve fitting (left) and CrossMAC for comparing RFP and Polyref (right) 5. Log decrement Estimating the dampening coefficient of the time history data can be done using the log decrement method across the duration of the response. The extracted dampened value is that of an analogous single degree of freedom system. Figure 4 presents a log decrement procedure that has been performed. Log decrement is not sensitive to noise and cannot reflect the underlying dynamics of a structure. It does require computational requirements and is therefore not typically used during real time monitoring. Log decrement is an efficient tool in detecting damage within a structure. Figure 4: Log decrement results 6. Temporal moments Temporal moments, analogous to statistical moments, are numerical values that are descriptive of the overall nature of the output waveform. Each order represents a different temporal moment, with k = 0 – 4 being most common for Structural Health Monitoring applications. The temporal moments for these orders are, Energy, Central Time, RMS Duration, Skewness, and Kurtosis, respectively. The following table (Table 2) shows the temporal moments for the fifth test, accelerometers one and five. Temporal moments are not sensitive to noise and can reflect the underlying dynamics of the structure. There are some computational requirements making them not ideal for real-time monitoring. However, temporal moments can be extremely useful in detecting damage within a system. Table 2: Temporal Moments for the Fifth Test Temporal Moment Energy Central Time RMS Duration Skewness Kurtosis Accelerometer 1 0.0274 0.3741 7.2783 162.43 3.89E+03 Accelerometer 5 0.0336 0.5244 10.8442 252.7172 6.16E+03 Page | 3 Appendix MatLAB Rational Fraction Polynomial curve fitting for the frequency domain data function mainrfp rec=xlsread('Datatest.xls'); %get the test data, which contains the mangitude %of each node for each frequency domain omega=xlsread('Omega.xls'); %get the frequency domain [alpha,par]=rfp(rec(:,1),omega,15);%the number of 15 is selected that 15 %natural frequency will be generated; and this function is to fit the %tested frequency spectrum plot(omega,rec(:,1),'r'), hold on;% plot the figure of test data on node #1; %here,the number of 1 is selected to figure the frequency response spectrum %of node #1 xlabel('Frequency (rad/sec)'),ylabel('Magnitude (dB)'),%plot the x and y %lable of the figure plot(omega,20*log10(abs(alpha)),'b'), hold off,%plot the estiamted frequency %response spectrum of node #1 xlswrite('Node1.xlsx',par);%write the estimated natural frequency, damping %ration, and etc. in to a xlsx file end Shown in below is an illustrative application of the MatLab code in the Appendix: 140 80 Test data Fitted curve 60 Test data 120 Fitted curve 100 40 Magnitude (dB) Magnitude (dB) 80 20 0 60 40 20 -20 0 -40 -20 -40 -60 -80 0 -60 1000 200 400 600 Frequency (rad/sec) 800 1000 3 modes (frequency from 0 to 1000 Hz) 1200 1400 1600 1800 2000 2200 2400 Frequency (rad/sec) 2600 2800 3000 3200 8 modes (frequency from 1000 to 3200 Hz) Page | 4