Machine learning based approaches for medium-thick plate stress analysis feature extraction and product defect prediction

In order to accurately predict the plate production process defects, and increase the rate of finished products, and improve enterprise profits, on the base of large-scale industrial data accumulated in medium-thick plate production process, this paper proposes using machine learning and data analysis theories and methods to study the data-driven stress and product defect prediction model of plate. After selecting the data features which have significant effect on the stress of the plate, we establish the logistic classification and forecasting model, and use cross-validation to train and validate the model. The experimental results show that the feature extraction and prediction model can accurately predict the stress defect classification of the medium-thick plate production process.