A CNN architecture for tool condition monitoring via contactless microphone: regression and classification approaches

Machine learning (ML) techniques combined with Internet of Things (IoT) sensors for tool condition monitoring (TCM) have emerged as a great potential for the online monitoring of the milling process. Monitoring tool degradation is necessary in modern manufacturing industries to ensure production safety, workpiece quality, and cost reduction. An automatic TCM system can be realized, based on the combination of ML and IoT sensors. It provides a live report of tool conditions, identifies the best time for tool replacement, reduces machine downtime, ruins the surface quality of machined parts due to elevated tool degradation, and, consequently, increases the sustainability of the process. A major challenge in developing such a system is utilizing a low-cost and non-invasive monitoring tool that can make timely and accurate decisions about cutting tool replacement, without interfering with the machining process. Audio signal analysis can meet this need, by using a contactless, low-cost microphone for the acquisition. A convolutional neural network was developed and compared with various ML techniques to solve the TCM problem. The issue was approached as a regression and a classification problem, and a thorough analysis was performed between the two approaches. Results demonstrated that addressing the problem as a regression enables industries to adapt the desired results to their production policies such as preventing surface quality or using the cutting tool as much as possible. Conversely, the classification results are more straightforward for operators and maintenance personnel to interpret, thereby simplifying decision-making on tool replacements.