A Quality Prediction Hybrid Model of Manufacturing Process Based on Genetic Programming

Abstract

The design of manufacturing parameters in the initial stage is backed by quality prediction to realise intelligent manufacturing. Accurate prediction translates to better quality, lower costs and more flexibility. However, the real production is a complicated and variable process, most of which involved multiple parameters simultaneously. The data on the basis of feature construction can filter the impurities of data, accuracy of predictive model can be satisfied. Existing approaches to provide results are useless when the insufficient mining of the relationship between the data or the some case without adequate manufacturing data and expertise. In this paper, a two-stage hybrid approach with genetic programming is proposed for quality prediction. The feature construction is realized by genetic programming in the first stage, and the new features are utilized as additives to subsequent stage of the extreme gradient boosting. The comparison experiments indicate that the two-stage hybrid model outperforms the existing methods in overall performance.