Tool wear classification based on machined surface images using convolution neural networks

Among several factors that are having a profound impact on the overall machining process efficiency, cutting tool wear is the most significant one. Monitoring and identification of cutting tool wear state well before to its failure is important to achieve superior machining quality and profitable production. With the recent advancements in computational hardware, significant amount of research is being carried out on using deep learning techniques, in specific, convolution neural networks (CNN) for developing cutting tool wear monitoring system. Although, few researchers reported the use of CNN as a pathway to tool wear classification problems with significant results, the fundamental methodology adopted by these techniques still needs to be investigated. Hence, in the present work, a deep CNN architecture is designed by choosing appropriate hyper-parameters and a CNN model is developed by selecting proper training parameters for cutting tool wear classification. Machined surface images acquired during turning operation performed on mild steel components under dry condition by uncoated carbide inserts as cutting tool are used as input data to the CNN model for predicting the tool condition. The proposed model, whose classification performance is independent of machining conditions, has capability to extract the features and classify the cutting tool among the two classes (i.e., unworn and worn classes). Accuracies of 96.3% and 99.9% are realized for classification of tool flank wear from raw and minimally pre-processed (contrast enhanced) machined surface images, respectively.