Support vector machine (SVM), which is a new technology solving classification and regression, has been widely used in many fields. In this study, based on the integrated conductivity(including conductivity and tensile strength) data obtained by carbon fiber/ABS resin matrix composites experiment, a predicting and optimizing model using genetic algorithm-least squares support vector regression (GA-LSSVR) was developed. In this model, genetic algorithm (GA) was used to select and optimize parameters. The predicting results agreed with the experimental data well. By comparing with principal component analysis-genetic back propagation neural network (PCA-GABPNN) predicting model, it is found that GA-LSSVR model has demonstrated superior prediction and generalization performance in view of small sample size problem. Finally, an optimized district of performance parameters was obtained and verified by experiments. It concludes that GA-LSSVR modeling method provides a new promising theoretical method for material design.
With the rapid development of society and science, especially electronics and information technology, which means we will need more and more conductive composite materials. Because of their excellent mechanical properties, carbon fiber conductive polymeric composites play more and more important roles in the field of composites. In recent years, they have been widely used in such diverse areas as anti-static materials, self-restrict heating materials, pressure and temperature transmitter and electromagnetic interference shielding materials [1], [2] and [3]. However, during the process of searching for relationship between the technical parameters and desired electrical conductivity of carbon fiber conductive polymeric composites, lots of experiments have to be repeated. Such traditional material design method would be bound to waste lots of manpower and resources. Due to subjectivity of researchers being excessively stressed, it is very difficult to make rational conclusions. Fortunately, theoretical modeling offers a reasonable alternative by which part or total of complex and time-consuming experiments can be replaced [4].
Nowadays, intelligent theory and method have been used to predict, estimate and optimize for material engineering [5], [6] and [7]. Before this work, we had done a number of studies on conductivity and tensile strength predicting model of carbon fiber/ABS composites using neural network (NN). Previous studies had showed that NN model had obtained more satisfied results than those of the symbolic modeling methods; however, the serious disadvantage is that network training last long. Recently, support vector machines (SVM) has been introduced to solve machine 1earning tasks such as pattern recognition, regression and estimation [8]. Compared with NN, SVM provides more reliable and better performance under the same training conditions. Because of such good properties as globally optimal solution and good learning ability for small samples, SVM has received additional consideration. Furthermore, SVM has been successfully used for support vector regression (SVR), especially for nonlinear systems modeling [9], [10], [11] and [12].
In this study, we proposed a new two-stage GA-LSSVR learning system. In stage I, LSSVR model was established by experimental data regression and analysis. In stage II, GA was used to optimize model parameters and improve precision and efficiency.
This paper proposed a GA-LSSVR predicting and optimizing model for integrated conductivity of carbon fiber/ABS composites. In this model, LSSVR was used to develop predicting model and GA was used to optimize model parameters. Based on experiments and investigations carried out, the following conclusions can be drawn:
(1)
GA-LSSVR model characteristics precise approximation performance as well as satisfactory generalization performance. The results of the subsequent experiments show that GA can obtain more precise selection parameters, but take less running time.
(2)
Compared with PCA-GABPNN model, GA-LSSVR model is unnecessary to use any data pretreatment technologies such as data compression, dimension reduction of input data and so on.
(3)
GA-LSSVR theoretical modeling method possesses many advantages: (1) It costs less. It doesnot need more experimental process so that part or total of complex and time-consuming experiments can be replaced. (2) It creates much. In view of materials design, small sample data are at large, support vector machine will built more precise predict model than any other mathematical methods. (3) LSSVR solves linear equations and will lead to important reduction in calculation complexity.
(4)
The main shortcoming of GA-LSSVR theoretical modeling method is that it is one of empirical modeling methods and how to select parameters mainly depends on experience and practice. How to get better choice of the kernel function and parameters has not been solved very well yet.
Hence the GA-LSSVR predicting and optimizing model developed in this study can be effectively used for carbon fiber/ABS composites integrated conductivity prediction and optimization and worth to be popularized.
The future work will focus on how to choose more efficient alternative kernel function and how to provide higher precision parameters selection method.
