A novel algorithm for tool wear monitoring utilizing model and Knowledge-Guided Multi-Expert weighted adversarial deep transfer learning

Abstract

Monitoring tool wear is vital in the cutting process as it guarantees the quality production of intricate aerospace components and boosts manufacturing efficiency. However, traditional monitoring methods may fall short when confronted with varying cutting conditions and a scarcity of data. To address this, the paper introduces an innovative algorithm known as MKWADTL (Model and Knowledge-Guided Multi-Expert Weighted Adversarial Deep Transfer Learning). The primary aim of MKWADTL is to refine the main network’s performance by tapping into the inherent knowledge embedded within the cutting parameters. Moreover, the algorithm capitalizes on the correlation between force signals and the variance in tool wear across different time frames to formulate a loss function that is informed by physical principles. In addition, the paper puts forward a multi-expert weighted adversarial structure. Through this framework, multiple experts can independently learn and identify various signal characteristics. Subsequently, the features extracted by these experts are integrated to ensure more precise data feature extraction, facilitating the monitoring of tool wear across a spectrum of processing environments. The MKWADTL algorithm’s exceptional accuracy in monitoring is exemplified on the custom-crafted dataset, the NUAA dataset and the NASA dataset.