Modern manufacturing emphasizes quality, precision, and accuracy. A revolution of advancements in machinery and tooling has been accomplished over the last decades, including computer numerical control and CNC machining enabling precision manufacturing and enhancing productivity while improving quality. Precision machining takes raw material to fabricate components of desired precise geometry with the help of a cutting tool. During fabrication, the tool erodes progressively and may break as well. Therefore, it is important to assess the state of the cutting tool, as the machine is only aware of the movement of the machine elements and other indirect indicators of the health of the machining process.
This presentation will report an intelligent approach developed and implemented for the assessment and classification of the state of the condition of the cutting tool. Earlier research has mostly dealt with supervised machine learning algorithms that can predict the tool wear using available sensory information of the machine, like power, vibration, and audio signals. Previous algorithms depend on specific features derived from signals that indirectly relate to tool wear. However, in-house experiments suggest that the existing approaches are not generalizable and transferable from machine to machine, which limits their application in industry. Such methods are not applicable to small and medium manufacturers, as they require excessive in-house skills in machine learning for the development and updating of computational algorithms. The research communicated in this presentation will present a simple and effective framework for the estimation and classification of the tool condition using unsupervised learning methods, which make it generalizable, explainable, and intelligent.