Optimizing Tool Wear Prediction in Intelligent Manufacturing: A Multi-Sensor Approach Enhanced by RealNVP

Abstract As intelligent manufacturing advances, tool condition monitoring (TCM) becomes crucial for maintaining the accuracy and efficiency of cutting tools. Traditional methods, such as physics-based and data-driven models, often fail to keep pace with the rapidly evolving manufacturing environments, resulting in efficiency losses and potential equipment damage. In response, this study introduces a novel tool wear monitoring model called deep convolutional neural network-multichannel bidirectional long short-term memory enhanced by RealNVP (DCNN-McBiLSTM-ENVP). This integrated model combines the robust feature extraction of CNNs with the sequential analysis capabilities of BiLSTMs and includes RealNVP data augmentation to enhance real-time applicability in smart manufacturing settings. The effectiveness of the DCNN-McBiLSTM-ENVP model is proven by its excellent performance on the PHM2010 challenge dataset across multiple key indicators.