Data mining system for predicting quality of polymeric films

Abstract

A computer data mining system for predicting the quality of polymer films in international, large-scale and multi-assortment industries is described. This article presents a library of statistical and data mining methods that allows, using statistical tests, to test data for distribution normality, predict the quality of polymer film materials for various line configurations and different types of film and includes the following methods: recurrent neural networks, neural network with long short-term memory and convolutional neural network. Analysis of methods of predicting the quality of polymer films was carried out and an algorithm was developed that allows selecting the most appropriate method for predicting the quality of polymer films based on the type of film, line configuration and requirements for film quality. The system includes interfaces that display trends in the process characteristics of the process class. The system was tested using the example of industrial data of the corporation on production of polymer film in plants of Russia and Germany.