Tool wear prediction in turning using workpiece surface profile images and deep learning neural networks

Accurate prediction of tool flank wear during turning is important so that the cutting tool can be replaced before excessive damage occurs to the workpiece surface. Existing online methods of tool wear prediction using sensor signals can be affected by noise, thus resulting in false alarms. The aim of this work is to develop deep learning regression models to predict tool wear state using features extracted from 2-D images of surface profile of the workpiece. Two models, namely convolutional neural network (CNN) and deep neural network (DNN), were compared in terms of prediction accuracy. Images of the workpiece surface profile were captured using high-resolution camera with the aid of backlighting after each machining pass. Workpiece surface profile images along a distance of two wavelengths were cropped and fed into the CNN network for wear prediction. For the DNN model, the surface height data were extracted to subpixel accuracy from each cropped image and used to train the model. Based on the results, the CNN model was able to predict the wear state with an accuracy of 98.9%, with an average testing RMSE of 2.0969, while the DNN model can predict wear state at an accuracy of 89.1%, with average testing RMSE of 2.5881. The study shows that cropped images of the machined surface profile can be more reliably used to predict the amount of tool flank wear during turning by using the CNN model compared to the height data used in the DNN model.