Enhanced tool condition monitoring using wavelet transform-based hybrid deep learning based on sensor signal and vision system

A new approach for enhancing the reliability and practicality of online tool condition monitoring (TCM) is introduced in the current research. This new method is based on analyzing raw force signals and processing cutting tool images. A new vision system based on a CCD camera is used to monitor the tool condition by collecting images of the rotating cutting tool during the milling operation, making it a convenient and feasible process. Firstly, image processing wear extraction based on the projection of the rotating tool is investigated in this study. This method demonstrated a high correlation with the experimental flank wear, reaching 99.37%. Then, the tool wear prediction method was developed by combining a hybrid deep learning algorithm with a raw signal multiresolution analysis method based on the wavelet transform to improve the accuracy of identifying tool wear. This method involves combining a hybrid deep learning algorithm that consists of a Convolutional Neural Network (CNN) and Bidirectional Long-Short Term Memory (BiLSTM) with Maximal Overlap Discrete Wavelet Transform (MODWT) for preprocessing signals. Cutting experiments using different tool sizes and parameters were performed on the vertical CNC milling machine. Finally, to evaluate the performance of the proposed model, its identification accuracy was compared to that of other deep learning and machine learning models. According to the experimental result and in contrast to available TCM methods, the proposed method improves the accuracy of tool wear condition recognition. The proposed model demonstrated the highest regression coefficient R compared with common prediction methods, equal to 99.5% on average.