Research On Product Quality Prediction Based On Improved Support Vector Machine

As a key chain of quality control, Quality prediction is dynamic system engineering. The product quality is influenced by many kinds of high dimensional nonlinear and random factors. Therefore, it is significant to establish a high precision product quality prediction model to improve the quality control ability and reduce the production costs. Currently, there are two ways to establish the product quality prediction model:the physical modeling method and the data mining approach. Due to there is a complicated nonlinear relationship which is difficult to express exactly between the large number of random factors and uncertain factors, making mechanism modeling is more difficult. Therefore, historical data is widely used to establish a quality prediction model. However, the traditional statistical quality control must be based on a certain statistical rules, and the method cannot meet the needs of many kinds of pattern recognition. Therefore, the traditional quality prediction methods have some limitations.In recent years, artificial intelligence algorithms have seen rapid development, at the same time, they have been widely applied in the modeling and differentiation of non-linearity system. As a new machine learning method, Support vector machine based on statistical learning theory of VC dimension and structural risk minimization principle has a strong advantage in dealing with the problems of nonlinear and high dimensions, and has a wide use in pattern matching, nonlinear regression, and time series forecasting and so on. For two types of product quality problems forecasting process, the improved support vector machine prediction models are established according to whether consider the quality of the internal mechanism of the model. Firstly, for the multi-factor model of product quality with specific factors and input-output relationships, the grey relational analysis method is introduced as the attribute preprocessor to select the key quality influencing factors and grid search algorithm is used to optimize the parameters of the model. A multi-factors quality prediction model based on grey correlation support vector machine (GR-SVM) is established. Secondly, the model is unknown, and it is difficult to identify the quality of the product or process is difficult to determine the quality of the model input-output relationship, the product quality prediction model is established based on historical production data. The phase space reconstruction theory is lead into the vector space reconstruction of time series and the advantages of genetic algorithm are introduced into model parameters optimization. By improving support vector machine, the GA-SVM quality prediction model for time series is established. Finally, two quality prediction models are simulated and the validity of the improved model is verified.