Research On Data Processing And Analysis For Harmless Production Process Of Injection Foaming

Foamed plastics have many useful properties, including low weight and highstrength as well as good sound insulation, heat-shielding and impact absorption, andthus are widely used. But traditional foaming processes produce an adverse impact onthe environment, so the environmentally friendly production is desirable goal.Processing and analyzing the data from harmless production processes helps toachieve the reliable forecast of product quality, which can provide guidance for newprocess improvements, directly benefiting the actual production.The harmless production of injection foaming is a complicated, multi-variable,nonlinear process: quality of products is the combined result of many factors and therelationship between quality of products and influencing factors is very complex. Thiscomplexity determines the difficulties of creating a mathematical model of product’squality. This paper aims to find a data processing method that will facilitate theprediction of product quality based on the parameters of the foaming process. Thedata used in this paper are all derived from orthogonal array testing of the injectionfoaming process. More specifically, the following steps were taken to build apredictive model.(1) Main parameters of the foaming process were collected using a dataacquisition system based on Virtual Instrument technology.(2) Product weight reduction was chosen as an indicator of quality. Stepwiseregression was used to establish a predictive model, but the predictive accuracy of themodel is not very high, because the harmless production is a nonlinear process.(3) Another quality prediction model based on a BP neural network wasestablished. It was trained and tested using MATLAB7.0software. Forecast errors ofthe model are mainly within a reasonable range, with larger error at only a few points.(4) There are several shortcomings of BP neural networks, such as their low ratesof convergence and the ease with which they fall into local minima. Geneticalgorithms, on the other hand exhibit strong global optimization capabilities. Thispaper uses a genetic algorithm to optimize the BP network weights and thresholdvalues. The neural network configured using the GA-BP algorithm is called a GA-BPneural network. A quality prediction model based on GA-BP neural network wasestablished and compared with the two models based on multiple linear regressionand BP networks, the prediction error of the GA-BP neural network model is the minimum, prediction effect is the best, thus its application to forecasting productquality is most feasible. Thus, for experiments that have not been conducted, thispaper uses the GA-BP network-based model to forecast weight reduction.