NEURONAL NETWORK USED FOR INVESTIGATION OF WATER IN POLYMER GELS *T. RUSU *, *C. IOJOIU and **V. BULACOVSCHI * P. Poni Institute of Macromolecular Chemistry, Ghica Voda Street, 41A, Iasi 6600 Romania, teia@tuiasi.ro ** Gh. Asachi Technical University, Iasi, 6600, vbulacov@ch.tuiasi.ro Abstract The paper deals with the synthesis and characterization of some copolymer networks build up from sequences of polydimethyilsiloxane (PDMS) and poly(methacrylic acid) (PMAA). The water diffusion properties of the networks are also investigated by using a neuronal network (NN) algorithm. The proposed algorithm is an empirical one, which simultaneously retrieves several physical parameters over the copolymer network properties as a relation between the molecular ratios of copolymer sequences. Comparison with other NN algorithms is presented. 1. Introduction Modern chemistry has made important efforts to develop and to test theoretical concepts for providing new macromolecular compounds with the structure related to imposed properties for new areas of applications. Hybrid materials with different molecular architectures have attracted an increased interest in recent years. This paper focuses on copolymers containing hydrophilic and hydrophobic sequences that confer them special transport properties. The molecular ratio between the two sequences has an important role for the properties of such copolymers like shape-selective separation and water (small size particles) delivery. It is known from literature that hydrogels reveal phase transitions making swelling-deflation changes in response to environmental modifications, e.g. solvent composition [1, 2], ionic strength [3], temperature [2-5], electronic field [6] and light [7]. Such stimuli-responsive polymers have been investigated with application to artificial muscles [8], immobilization of enzymes [9], concentration of dilute solutions [10, 11] and chemical valves [12]. Using stimuli-responsive amphoteric gels there have been attempts for temporal control of water release. Thus, a swollen gel in ethanol immersed in water undergoes immediately spontaneous translation and rotational motions with two-thirds of its volume immersed in water. In the course of its motion the gel gradually sinks and finally settles at the bottom of the vessel due to contraction and cease of motion [13]. The speed, duration and mode of gel motions are associated with its size, shape and chemical nature. Such polymer gels, exhibiting motion in water, have potential applications as soft-touch manipulators, target drug-delivery devices, micro-agitators, and micro generators. Their study is under way. Networks prepared by reacting functionalized polydimethylsiloxanes (PDMA) with different cross-linking agents are considered ideal models for studying network formation and properties [14]. PDMS is highly hydrophobic, very flexible and shows little variation of its properties with temperature. Since its mechanical properties are poor, for their improvement one should synthesize cross-linked copolymers based on PDMS [15]. This paper deals with the synthesis and characterization of amphoteric networks (gels) based on PDMS–co–PMAA sequences. The water-diffusion properties of the obtained gels are investigated by using some neuronal network algorithms. 2. Experimental The polymer gels were obtained by radical copolymerization of polydimethylsiloxane macro-initiators with methacrylic acid in presence of ethyleneglycol dimethacrylate as cross-linking agent [16]. The synthesis of the azoester macroinitiators (AzoPDMS), with different molecular weights of siloxane sequences and different contents of azo groups, was realized according to Scheme 1, and the experimental data are summarized in Table 1. The synthesis of PDMS-poly(methacrylic acid) (PMAA) hydrophobic– hydrophilic gels was realised according to Scheme 2 [17], their characteristics being summarized in Table 2. CH3 CH3 CH3 Si O R O OC (CH 2)2 C N N C (CH 2)2 COO R Si O CH3 CN CN + + MAA EGDMA PDMS - co - PMAA HO - PDMS H2N - PDMS CH3 CH3 ClOC C2H4 C N=N C CN CH3 [O CH3 CH3 CH3 C3H6 ( Si O )mSi C 3H6 OOC C2H4 C N = N CH3 CH3 CN (AzoE-PDMS) CH3 [ HN C2H4 COCl CN C C2H4 CO] q CN CH3 CH3 CH3 C3H6 ( Si O ) Si C 3H6 NHOC n CH3 CH3 C2H4 C CN (AzoA-PDMS) N=N C C2H4 CO ] p CN Scheme 1 TABLE 1. Characteristic data for Azo-PDMS Sample Cod NH2 (mol/kg) Mn (GPC) Mw/Mn (GPC) Sample cod N2 (%) NH2 - DS NH2 – PDMS1 NH2 – PDMS1 8.06 0.92 1360 1.6 16.982 6.174 0.13 2000 1.4 Azo - DS Azo – PDMS1 Azo – PDMS2 Azo groups (mol/kg) 2.021 0.734 3.879 0.395 CH3 CH3 CH3 Si O R O OC (CH2)2 C N N C (CH2)2 COO R Si O CH3 CN CN + + MAA EGDMA PDMS - co - PMAA Scheme 2 TABLE 2. Characteristic data for PDMS–co-PMAA obtained by radical polymerization of MMA in presence of AzoPDMS Initial mixture* Sample Network MnPDMS SiO/MAA Yield, SiO/MAA** Aspect (from Molar % Molar ratio AEPS) ratio 5040 0.5 I.1 85.9 0.42 Gel 6620 0.5 I.2 87.6 0.44 Gel 14060 0.5 I.3 86.9 0.43 Gel 5040 2.0 II.1 89.4 1.63 Gel 6620 2.0 II.2 87.1 1.74 Dense gel 14060 2.0 II.3 90.4 1.81 Dense gel * Polymerization in sealed ampoules; 80C; 20 hours; solvent, toluene (total concentration 25%); EGDMA, 1% molar against MAA ** Determined from elemental analysis (Si content) 3. Characterization data 3.1. MACRO-INITIATOR CHARACTERIZATION The characterisation of macro-initiators was made by IR (SPECORD M80 IR) and NMR (Bruker AC 200). The IR spectra revealed the following absorptions: 1265 cm-1 - Si-CH3; 800, 1110, 1020 cm-1 - Si-O-Si; 1789 cm-1 -> 1738 cm-1 CACV inclusion in the PDMS sequence. In the 1NMR spectra (CDCl3) (Fig. 1) one finds: 0.01ppm (6H, CH3-Si, s), 0.43-0.47 ppm (2H, -CH2-Si, m; isomer), 0.81-0.88 ppm (3H, CH3-CH, d; isomer), 1.00-1.08 ppm (1H, CH3-CH, q; isomer), 1.40–1.47 ppm (2H, - CH2 -CH2 -CH2 -, m; isomer), 2.73-2.85 ppm (2H, CH2-NH2, m). For the AzoA-PDMS: 0.01ppm (6nH, CH3Si, s), 0.43-0.47 ppm (2H, -CH2Si, m; isomer), 0.80-0.88 ppm (3H, CH3-CH, d; isomer), 1.00-1.08 ppm (1H, CH3-CH, q; isomer), 1.20 ppm (3H, CH3-C(CN), s), 1.40–1.47 ppm (2H, CH2 -CH2 -CH2 -, m; isomer), 1.60-1.66 ppm (2H, CH2-C(CN), m), 2.262.43 ppm (2H, -CH2-CO, m), 3.11-3.19 ppm (2H, CH2-NHCO, m). Figure 1. 1H-RMN spectra for AzoA-PDMS 3.2. CHARACTERIZATION NETWORKS DATA FOR THE PDMS–co–PMAA The chemical structure and the thermal behavior of the cross-linked copolymers were established by IR spectroscopy and differential scanning calorimetry (DSC) (Perkin Elmer DSC-7 device), respectively (Fig. 2). Scanning Electron Microscopy (SEM) (Phillips XL 300 microscope) analysed the microstructure of the resulting networks (Fig. 3). Figure 2. Typical IR and DSC curves Figure 3. The SEM image of the network 4. Water diffusion tests The synthesized copolymers were submitted to the water delivery test. Equal amounts of samples from each of the six copolymers were llowed to achieve the swelling equilibrium by keeping them in an excess of water for three days. Then, the excess of water was removed by filtration and the swollen samples were submitted to control dryness thermosetting at 100C and weighed at every two hours [18]. The results are presented in Figure 4 and Table 2. The content of residual water was followed as a function of time and molar ratio of the two polymeric blocks. Residual water (%) Time (h) Figure 4. Residual water from the macromolecular network vs. time and molecular ratio. The water delivery tests were subsequently used in conceiving a Multilayer Perceptron Neuronal Network. A comparison of this algorithm with other global diffusion speed retrieval algorithms for amphoteric copolymers polymers is presented. We also examined different diffusion conditions related to different forms of polymer-soil moisture presenting various cross-sections of evaporation speed retrieval error analysis. 5. Neural network method A generic neuron Multilayer Perceptron NN employing feed forward, fully connected topology Neural networks (NNs) are well suited for a very broad class of non-linear approximations and mappings. Neural networks consist of layers of uniform processing elements, nodes, units, or neurons. The neurons and layers are connected according to a specific architecture or topology. The number of input neurons, n, in the input layer is equal to the dimension of the input vector X. The number of output neurons, m, in the output layer is equal to the dimension of the output vector Y. A Multilayer Perception NN always has at least one hidden layer with k generic neurons. The neuron is a non-linear element because its output zj is a non-linear function of its inputs X. So, problem which can be mathematically reduced to a non-linear mapping can be solved using the NN represented. NNs are robust with respect to random noise and sensitive to systematic, regular signals. From a mathematical point of view, the multi-parameter retrieval algorithm corresponds to a continuous mapping [19]. It maps a vector of ECs, T, onto a vector of retrieved physical parameters, g. NNs are well suited for performing a wide variety of continuous mappings. The architecture of the NN that we used is the OMBNN3 algorithm (Scheme 3). The NN which represents this algorithm has n = 4 inputs, m = 3 outputs, and one hidden layer with k = 12 neurons. This NN can be also written explicitly as: k n j 1 j 1 g q bq aq tanh{ wqj [tanh( jiTi B j )] q },.....q 1,..., m (1) where the matrix ji and the vector j represent weights and biases in the neurons of the hidden layer; qj and the q represent weights and biases in the neurons of the output layer; the aq and bq are scaling parameters. Scheme 3 However, as shown in Scheme 3, even such an approximate separation of signals allows the reducing of the random and systematic errors in diffusion speed and to reduce the dependence of the diffusion speed bias on V and L. For comparison were selected three alghoritms: - the original global operational (cal/val) algorithm developed by Goodberlet [20] (GSW); - the current operational algorithm (GSWP) which is the same as the GSW algorithm but corrected for water vapor by Petty [21]; - a physically-based (PB) retrieval algorithm developed by Wentz [22]. Table 3 shows that the NN algorithm provides significant improvement in retrieval accuracy at high diffusion speeds. Error budgets (m/s) for different diffusion speed algorithms and separately for higher diffusion speeds are also considered. TABLE 3. Algorithm Bias GSW GSWP PB OMBNN3 -0.2 (-0.5) -0.1 (-0.3) 0.1 (-0.1) -0.1 (-0.2) Algorithm RMSE 1.4 (1.8) 1.3 (1.6) 1.3 (1.8) 1.0 (1.3) Total RMSE 1.8 (2.1) 1.7 (1.9) 1.7 (2.1) 1.5 (1.7) W>15 m/s RMSE (2.7) (2.6) (2.6) (2.3) 6. Results and Discussion The OMBNN3 algorithm has the ability to retrieve not only diffusion speed but also three other parameters: columnar water vapor V and columnar liquid water L as shown in Figure 5. Figure 6 presents diffusion speed related to molecular ratio of PDMS/PMAA and the number of matches for the OMBNN3 algorithm. Figure 5. Water diffusion speed retrieved by four different algorithms as functions of columnar water vapor (Solid line - OMBNN3, dotted line - GSWP, dashed line - GSW, and dash-dotted line the Wentz algorithm). Figure 6. Water diffusion speed related to molecular ratio of PDMS/PMAA 7. Conclusion. One of the challenges of modern chemistry deals with the design of macromolecules with imposed properties. With this idea in mind we have used/tested an empirical neural network (NN) algorithm, which simultaneously retrieves several physical parameters over some polymer network properties from evaporation conditions (EC), primary emphasizing on water diffusion speed in a copolymer gel. This NN-based algorithm, which retrieves diffusion speed, columnar water vapor, and columnar liquid water, has been developed recently. Conditional simulation is one of the most rational approaches for modeling unsaturated flow and transport. The presented NN algorithm is a multiparameter one that includes some inputs, one hidden layer, and corresponding outputs. The combined analytical-numerical multilayer Perceptron NN and OMBNN3 approach was developed for the analysis of flow and diffusion transport processes in porous media. The OMBNN3 and the multi-layer Perceptron algorithms, both employing the simultaneous multi-parameter retrieval approach, reduce the bias, and the dependence of the bias, on both water vapor and columnar liquid water concentrations. The OMBNN3 algorithm demonstrates the best performances. The random errors for the OMBNN3 algorithm are significantly smaller and less dependent on the other related medium parameters than those for the other algorithms. A comparison of this algorithm with other algorithms is also presented. The simulation parameters can be used for optimization of the molecular ratio of the polymer sequences to build up networks with imposed diffusion properties according to the desired areas of application. References 1. Tanaka F. and Ishida M. (1994) Physica A, 204 660. 2. Katayama S., Hirokawa Y. and Tanaka T. (1984) Macromolecules, 17, 2462-2463,. 3. Hirokawa Y. and Tanaka T.(1984) J. Chem. Phys., 81, 6379-6380. 4. Ohmine I. and Tanaka T., (1982) J. Chem. Phys., 77, 5725-5729. 5. Tanaka T., Fillmore D., Sun S. T., Nishino I., Swislow G. and Shah A. (1980) Phase Transition in Ionic Gels. Phys. Rev. Lett., 45, 1636-1639. 6. Tanaka T. (1981) Gel Sci. Am., 244, 124-138. 7. Tanaka T., Nishio I., Sun S. T. and Ueno-Nishhio S. (1982) Collapse of Gels in an Electric Field, Science, 218, 467-469. 8. Irie M. and Kunwatchakun D. (1986) Macromolecules, 19, 2476-2480. 9. Suzuki M. (1991) Polymer Gels, Ed. D. DeRossi, Plenum Press, New York, pp. 221-236. 10. Hoffman A. S. (1987) J. Control. Release, 6, 297-305. 11. Freitas R. F. S. and Cussler E. L. (1987) Chem. Eng. Sci., 42, 97-103. 12. Trank S. J., Johnson D. W. and Cussler E. L. (1989) Food Technol., 43, 7883,. 13. Osada Y. and Hasebe M. (1985) Chem. Lett., 9, 1285-1288,. 14. Hamurcu E. E. and Bahattin M. (1995) Macromol. Chem. Phys., 196, 1261,. 15. He X. W., Windmaier J. M., Herz J. E. and Magers G. C. (1992) Polymer, 33, 866. 16. Rusu T., Pinteala M., Iojoiu C., Harabagiu V., Cotzur C., Simionescu B. C., Blagodatskikh I. and Shchegolikhina O., (1998) Synth. Polym., J. 5, 29. 17. Harabagiu V., Hamciuc V. and Giurgiu D., (1990) Makromol. Chem., Rapid Commun., 11. 18. Rusu T., Ioan S. and Buraga S. C., (2002) Euro. Polym. J., 37, 2005, 2001. 19. Rusu T. and Gogan O. M., Proceedings of Franco-Romanian Symposium on Applied Chemistry CoFrRoCA, 129-132. 20. Goodberlet M. A., Swift C. T. and Wilkerson J. C., (1989) Remote sensing of ocean surface winds with the special sensor microwave imager. JGR, 94, 14574–14555. 21. Petty G. W., A (1993) Comparison of SSM/I algorithms for the estimation of surface wind, Proceedings Shared Processing Network DMSP SSM/I Algorithm Symposium, 8-10 June 22. Wentz F. J., (1997) A well-calibrated ocean algorithm for special sensor microwave / imager, JGR, 102, 8703-8718,.