A CONTINUAL LEARNING MODEL FOR COATINGS HARDNESS PREDICTION BASED ON ARTIFICIAL NEURAL NETWORK WITH ELASTIC WEIGHT CONSOLIDATION
Abstract
In order to continuously update the prediction model based on the ever-expanding data set solely, this study established a continual learning model, i.e. the elastic weight consolidation (EWC)-based artificial neural network (ANN) model to predict the hardness of Ni–Cu–CrBN coating that could be used in tribology field. The results showed that after being trained by the ever-expanding dataset, the determination coefficient of the normal ANN model on old data decreased to 0.8421 while that of the EWC-based ANN model was still 0.9836. It was indicated that the EWC-based ANN model presented good performance on both new and old data after being trained by the ever-expanding dataset solely, which saved time and was more in line with practical application.
