Application of classical clustering methods for online tool condition monitoring in high speed milling processes

Tool Condition Monitoring (TCM) is a necessary action during end-milling process as worn milling-tool might irreversibly damage the work-piece. So, there is an urgent need for a TCM system to provide an evaluation of the tool-wear progress and resulted surface roughness. Principally, in-process tool-wear and surface roughness measurements requires costly stopping of the milling machine. However, to implement the condition monitoring system, resulted signals of milling process are utilized to form a reference model that detects the performance of the system non-intrusively. Therefore, the needed milling-process reference model have to apply more beneficial feature extraction and AI techniques. Since the signals are continuous, their time-frequency analysis are applied for feature extraction. Also, proper AI-based modeling techniques have to be joined together to form a repeatable and generalizable reference model. As one of the available AI techniques that can make an insightful change in traditional AI based modeling techniques for the process, clustering methods are applied on the wavelet features of milling signal as an interpretation layer between the sensor signals and the next artificial intelligent blocks. This paper illustrates the consistency and repeatability of different clustering methods on wavelet features of force and vibration signal as well as a comparison in accordance to their performance and possible generalization for online condition monitoring and sequential clustering. Finally, fuzzy C-means clustering method is shown to be a useful AI-based block for a noise-robust and generalizable ball-nose milling reference model while it provides suitable platform for further investigations regarding online fault diagnosis and prognosis and sequential clustering.