This article addresses the design of sensor-based tool-wear monitoring systems and their implementation, and specifically focuses on interpretation of signals from multiple sensors in terms of tool-wear level. Keeping in mind that the absence of a well-accepted reliable methodology and the ignorance of practical issues in implementation are two major reasons to prevent a successful on-line monitoring system from being realised in industries, four critical issues are raised and discussed, which are: (1) expensive information on correct tool condition; (2) troublesome off-line preselection of features from original sensor signals; (3) impractical fresh/worn dichotomy; and (4) fallible signal-interpretation with stationary sensor information. We provide a detailed suggestion of methodology to handle each issue effectively; adoption of an unsupervised clustering-type method (issue 1), development of a customised feature evaluation algorithm (issue 2), tendency of our methodology to avoid serious misinterpretation errors with a finer distinction of wear level (issue 3), and cooperative interpretation with sensor information from multiple time points close to each other (issue 4). Our methodology is evaluated with machining data from a simplified monitoring situation, and proves to be able to bridge the gap between academic enthusiasm and industrial needs for a practical and reliable sensor-based tool-wear monitoring system.
All Science Journal Classification (ASJC) codes
- Control and Systems Engineering
- Signal Processing
- Civil and Structural Engineering
- Aerospace Engineering
- Mechanical Engineering
- Computer Science Applications