Defect prediction of low pressure die casting in crankcase production based on data mining methods

Abstract

Low Pressure Die Casting (LPDC) is widely used in crankcase production. Owing to the unstable production situation and the complex shape of crankcase, porosity defect usually occurs in LPDC process which leads to loss in quality and productivity. In this paper, we apply data mining methods to predict the porosity defect in advance, so that we can take actions to prevent it from reoccurring in the next production part, thereby increasing the quality and productivity. To do it, firstly, we collect production data from a real-world casting line. Secondly, we prepare the data for prediction by feature extraction and feature selection. Thirdly, we apply an ensemble algorithm named Forest of Local Trees (FLT) for defect prediction. Finally, we present a thorough experimental study of the proposed method. The results show that our method outperforms other five algorithms on five real-world datasets in terms of three indicators, recall, precision and F-measure.