Research Progress of Performance Prediction and Optimization Design of Polymer Materials by Artificial Neural Network

March 2001 æ›° Computer and Applied Chemistry mputers Artificial Neural Networks Research Progress in Performance Prediction and Optimization Design of Polymer Materials Yang Fanwen Zhao Yaoming School of Materials Science, South China University of Technology. Guangzhou 510640 discusses the artificial neural network, the performance of polymer composites and the methods and effects of polymer composition, self-contained artificial neural networks for the control of village group processing, and the optimization of conditions for research and development. , Research instrument Ding human neurons for 1; 1 differentiation of the artificial neural network in the field of polymer science applications and the need to solve the question.

The Sino-French classification No. 1322 artificial neural network, also known as the neural network, is a new type of information processing that is gradually developed and mathematically abstracted and simulated on the basis of modern neuroscience research achievements. And computing system. The artificial neural network has the following four characteristics. The artificial neural network is a mega-complex system that is widely connected. The artificial neural network has a parallel structure and a parallel processing mechanism. 3 The distributed structure of the artificial neural network makes it human-like fault-tolerant and associative. Competence 4 artificial neural network has self-learning self-organizing adaptability compared with traditional information and data processing. The artificial neural network can be distributed and stored in parallel, and it is a nonlinear dynamic process. Therefore, it has obvious advantages in dealing with complex multi-dimensional nonlinearity.

Hawthorn artificial neural network several self 74 high fault tolerance height 1 linear description of the advantages. Now it has been widely applied to economical robotics and automatic control of military medical chemistry, and has achieved many results. This article will mainly discuss artificial neural network prediction of polymer energy in polymer materials. Research on quality control and optimization of polymer village molding processing, optimization of polymerization reaction and computer-assisted molecular design, achievements, 1 performance prediction artificial neural network with self-learning self-organization self-adaptive ability and human brain-like fault tolerance Sex and association capabilities, it is applied to the material performance prediction has the advantages of convenience and speed. At present, its application mainly involves the prediction of the performance of composites based on the physical and mechanical properties of homopolymers and the prediction of the composition of polymers.

Received date 200008, Received revised date 200,18 Computers and Applied Chemistry 18 Volume 1.1 Performance prediction of homopolymers Ding et al. 6 used rape 15 programs to predict the conductivity of 34 polymers and the conductance of 20 polymers. Rate, using a neural network model with 31 contacts in the input layer and 5 contacts in the output layer and 1 contact in the output layer, the average error of the predicted dielectric constant is 9.2, and the average error of the predicted conductivity is 36.1. The reason for the large average error is analyzed. The physical structure and mechanical properties of the 295 filament samples of the lamp were studied. The neural network was used to predict the forming conditions of the filaments and the physical structure versus the performance stress-strain curve. The maximum value of the stress-strain curve for the first time The elongation at the extreme value of the stress-strain curve The effect of the elongation at break The rate of heat shrinkage and the heat shrinkage force, it was found that the human Lafa method can effectively determine the molding processing conditions. And the physical structure of the performance, can effectively improve the optimal formulation and l art conditions Remaklo; lufk.1H mining jj neural network method, according to the dynamic rheological parts of polyene iene can be calculated against its molecular test distribution. Take ; and; as input parameters. Using a 121 model, the molecular weight distribution curve obtained from the clothing network using the neural network agrees well with the gel permeation chromatography 30 determination curve, which is superior to other methods such as tear 5 models.

The advantage of this method is that its molecular weight distribution is calculated based on the rheological properties of the polymer. Cross gel permeation chromatography 3 is inexpensive and convenient.

Xin Xianming et al. applied the synergy between synthesis reaction space and product space of the neural network 1 called synthesizing thermosetting phenolic resin, and the electron affinity of the metal ion in the catalyst. The yield of o-paramethylol under the action of each catalyst The raw material aldehyde phenol molar ratio input parameter, the weightlessness temperature at the weightlessness 5102030 as the output parameter, establishes a model that can be used to quantitatively predict the thermal properties of the resin cured product, the error between the experimental value and the predicted value is less than 25, the clear neural network method is The synthesis performance of the complex polymer system is quantitatively related to the prediction of the performance of the 1.2 composite materials. The impact resistance properties of the two carbon fibers and the strong polymer complex media CFT are determined. When the acoustic emission parameters are input parameters, the prediction result of 1300 is better than 1800. The reason for this may be that the choice of parameters is not ideal or the signal is diverging in the material. River Sapporo 3! The others used a type 1 neural network to evaluate the stress and impact resistance of composite materials and achieved good results. 1.3 et al.13 used neural networks for the real-time monitoring of low-energy impacts of plates. The position and size of the signal impact energy obtained by the piezoelectric sensors was analyzed using the Trunkbow Network, and a set of impact monitoring and monitoring systems was erected. The low energy impact properties of the steel ball impact graphite epoxy laminates were successfully predicted.

In addition. Et al. Using the dynamit network to predict the size of the styrene and quaternary amine cations. 1. Cheng Chongren 14 Using infrared spectroscopy and neural network interfaces, the mean diameter of polystyrene suspension polymerization products was measured. Online evaluation and monitoring.

2 Quality Control and Optimization The application of artificial neural networks in the molding and processing of molecular materials is of great significance to improving product quality and optimizing process conditions. 1. Others 13 will like the tax bar and withered by the Internet 1. Set up a mixed control model for injection into the bar, to control the defective rate is less than 1 million, the results show that the results of the use of mixed models Better than other models. 1 spit 0. and others 1 The prediction of the quality of polypropylene or injection molding products, the method is to establish the tether between the weight of the product and the temperature and pressure of the solution at the nozzle. Accurately predicting the quantity of products is the potential quality of on-line quality supervision. Youfa et al. applied it to melt polymer gas atomization to establish the melt pressure and pressure The relationship between the type and the particle size distribution of the powder, the results show that the correlation coefficient 0.98 obtained using the neural network prediction results is significantly better than the minimum bias multiplication method 15 correlation coefficient of 0.82.

(3) Simulation of the polymerization process, and others called a neural network-based hybrid model to combine the neural network and the phenomenological model to simulate the polymerization process of nylon 66 in a twin-screw reactor, and to study the polymerization temperature, vacuum pressure, flow rate, and extrusion. The machine head pressure screw speed, etc., are related to the relative viscosity of the product, the number of amino end groups, and the number of carboxyl end groups. The phenomenological model is good for predicting the number of carboxyl end groups of ammonia end groups. However, when predicting the relative viscosity of products, it can not consider the degradation reaction in the polymerization process and cause large deviations; using the mixed model not only works well. The predicted product predicted the number of carboxyl end groups of the amino terminal group and was able to predict the relative viscosity predicted by the product well. The model 2 is used by artificial rushing networks such as Ming Yang, Fan Wen et al., for the prediction and optimization of helium 1 for high-segment tree materials, and for the polymerization of the 133-gauge tract, and the temperature distribution curve and low-density polycondensation of the high-pressure tube reactor are studied. The relationship between the performance density of ethylene melt index average molecular weight, the results of the clear network can be used to simulate the polymerization system. In 5, et al. 2 who applied it to the random polymerization reaction 4 computer-aided molecular design 0. people. 31 On the basis of a thorough study of the relationship between the structure and properties of the series of polymers, the author has established the glass transition warm-work expert system 181 to decompose the thermal system and the system. Secondary Relaxation Expert System 10 Refractive index expert system 1 Milk 6, pull-strength special-purpose system; fracture body length cattle 1 seven-for-one system Yan Yun 1 heart; card shrinkage strength expert system. 0 hardness specific system also. Defective strength concrete system 6 respectively Rice model 1; 10 pairs of different structural buckle thermal properties 00 Mechanical properties Tensile strength Compression strength Rockwell hardness Elongation at break 1 Nil impact strength and optical properties Refractive index into the hall The predictions; among them, the correlation coefficients between the sum of the 1 and the experimental values ​​of the shuttle-on-sniff predictions were 0.9920.998 and 0.990, respectively. According to the prediction results, they were considered to be 3.28 derivatives containing 3 pyridine 3.20, pyridine. The derivative material and 3.2571 soil, such as the derivative of 1 also has better impact resistance properties, and successfully designed a new generation of polycarbonate with improved performance. Must be 331 state. Et al. studied the density and structure temperature of 5 kinds of polyester plasticizers of Indomethacin, ester 10 oxime monodecanoate 1 decanoate 2 and methyl pentamethyl acetonate. For the relationship, a neural network model of the 241 structure was used to establish the relationship between the density of the polyester and plasticizer and the temperature of the methylene group in the structural ester. At present, computer simulation technology has achieved many research achievements in the design of low-molecular-weight organic compounds and drugs. It is possible to learn from the research methods and achievements in the molecular design research of macromolecules, and it is expected to combine the relationship between the structure and properties of macromolecular materials. Opens new ways for the molecular design of polymers.

5 conclusions In summary, the neural network has the following characteristics in the performance prediction and optimization design of polymer materials. 1 Through the training network, you can learn to hide the relationship between the input side and the process conditions and output performance, establish the input and output The non-linear relationship between search and corpse optimization formulas has a strong search and bound 2 error. , The law and noise in the research process can be highly effective, and the results of supplementary experiments can be used to improve the training of the network to achieve better learning results; 4 Neural network using matrix to increase formula factor test items or test times Without the need for large structural changes, it is applicable to multi-input multi-output multi-factor multi-performance ANNs. It shows a wide range of application prospects in the field of polymer science and engineering, which promotes the theory and practice of polymer science. Such as the number of hidden nodes in the network and over-fitting of the network. Future studies should improve the theoretical deficiencies and advance the application of neural networks in the field.

1 Liu Zengliang. Liu Youcai. Research and Exploration of Fuzzy Logic and Neural Network Theory . Beijing Beihang University Press, 19962 Zhang Jixian, Yu Xia. Withered network and its application in engineering. Beijing Mechanical Industry Press, 19963 Modern Neural Network Application. Xing Chunying, translation. Beijing Electronic Industry Press, 194 Xu Bing earned, Zhang Bailing, Wei Gang. Neural Network Theory and Application. Guangzhou South China University of Technology Press, 19945 Shi Hongbao. Neural network and its application. Xi'an Xi'an Jiaotong University Press, 1993 Fei Sheng 2, How, Life, Boots, 4 Accounts, M, and 5 Change Notifications The Chinese Chemical Society's Computer Chemicals Application Committee will be renamed in 2001 as the Information Technology Application Committee. .

Since its establishment in the 1980s, the China Chemical Society’s Committee for the Application of Computer Chemicals has successfully hosted the 7th National Conference on Computer Chemicals Applications. At the beginning of the 21st century, the 8th annual meeting to be held in 2001 will also be renamed with the professional committee, and will be renamed as the 8th annual meeting for the application of information technology in chemical industry.