Investment casting defect prediction using neural network and multivariate regression along with principal component analysis

Investment cast parts can contain different types of internal defects like ceramic inclusion, flash, misrun, shrinkage, and slag inclusion. Their occurrence can be prevented by prediction from a given composition and process parameter set. In this work, we explore the application of artificial neural network (ANN) and multivariate regression (MVR) to predict a wider range of investment casting defects based on real-time industrial data. The data of 24 parameters related to process, chemical composition, and defects was collected from about 500 heats in an industrial investment casting foundry. This was reduced to ten principal components using principal component analysis (PCA). Different six ANN training models were used to train a portion of the data and used for prediction. The MVR was also employed on the same data for prediction, and its results were compared with ANN models. The best results were obtained by ANN with Levenberg and Marquardt algorithm. [Received 16 August 2016; Accepted 25 October 2016]