Multisensor-based tool wear diagnosis using 1D-CNN and DGCCA

Bad conditions during machining cause tool chatter, wear, or breakage, which affect the tool life and consequently the surface quality and dimensional accuracy of the machined workpiece. Therefore, for the purposes of production efficiency and economics, monitoring and diagnostics of the tool's condition are important. This study presents a one-dimensional convolutional neural network (1D-CNN) and deep generalized canonical correlation analysis (DGCCA) for multiple sensors-based tool wear diagnosis. In particular, 1D-CNN is used to extract features from 1D raw data, such as force, vibration, and sound, whereas DGCCA with attention mechanism is used to fuse the feature output from each 1D-CNN by removing irrelevant or redundant information. Experiments are performed using PHM2010 and NASA data sets. The experimental results show that our proposed approach can achieve satisfactory accuracy of 95.6% and near real-time performance. Results of our study can be implemented in real tool wear diagnosis, and thus identify novel opportunities toward realizing Industry 4.0.