R&D Focus
Condition Monitoring System for Cutting Tools Yu-Hung Pai and Hung-Tsai Wu
Cutting tool wear has a direct impact on the manufacturing process and has troubled industries for decades. Users often spend considerable cost and maintenance time due to tool breakage. It is therefore critical to maintain the cutting quality and avoid machine damage from abnormal vibration. However, the current pain point is that manufacturers find it challenging to build the criteria for examining tool health conditions based on audio or visual inspection of the cutting tools during machining. Therefore, the best solution so far is to handle the tools with extra care or conduct a conservative evaluation of the tools’ useful life, which leads to additional labor cost, downtime cost, and tool expenses.
CMSCT monitors tool condition, makes prognosis predictions, and provides feedback control.
To overcome these bottlenecks, ITRI developed the Condition Monitoring System for Cutting Tools (CMSCT). This system generates precise criteria for tool condition, makes prognosis predictions, and provides feedback control. Since vibration signals are highly sensitive to wear-related features, CMSCT evaluates the tool condition based on three-axial vibration signals coming from machine spindle. The use of vibration signals also allows the evaluation to be made without stopping the process of production, resolving a dilemma of conventional human inspectors. Furthermore, vibration monitoring is a better solution for most equipment, especially for the old equipment with fixed, unadaptable structure, as vibration detection devices have great installation flexibility and minimum system cost.
To synchronize with the machine tool, CMSCT is connected to the machine tool’s controller to access real-time parameters including feed rate, cutting speed and spindle load to identify the machining status. It prevents the calculating procedure from being interfered by idle signals, and can halt the equipment when failure occurs. Integrating digital signal processing technologies and AI algorithms, CMSCT leverages adaptive self-training to monitor the life cycle of cutting tools.
To avoid building myriads of AI models for different machining conditions, CMSCT adopts a self-supervised deep learning model, that can distinguish data without labeling needed. The learning model quantifies the current tool condition, generates a wear index, and provides a warning threshold upon tool failure. Once the wear index exceeds the threshold or when other abnormalities are detected, CMSCT autonomously halts the equipment and retracts the tool, protecting both workpieces and equipment from potential damage.
The wear index increases when the tool condition deteriorates.
CMSCT reaches an approximately 90% accuracy in predicting the remaining useful life of a cutting tool for milling and drilling. This system provides customized UI for different equipment makers and users. CMSCT has already been implemented in five factories in different fields of two enterprises, one of which is a Taiwan-based heavy equipment manufacturer, with proven performance and excellent user feedback. It reduces millions of dollars spent on tool breakage and saves 10% to 20% maintenance time every year.
CMSCT has been granted a patent in Taiwan in 2022. This cost-effective tool can enhance the reliability and the quality of machining processes, assisting industry players to achieve intelligent manufacturing and energy saving.
About the Authors
Yu-Hung Pai is an associate researcher in the Mechanical and Mechatronics Systems Research Laboratories at ITRI. He received his B.S. and M.S. degrees in Civil Engineering from National Cheng Kung University. He specializes in vibro-acoustics, condition monitoring, and data science, focusing on machining processes and railway engineering.
Dr. Hung-Tsai Wu works as a researcher in the Mechanical and Mechatronics Systems Research Laboratories at ITRI. He received his B.S. and Ph.D. degrees in Communications Engineering from National Chiao-Tung University, Taiwan. His research focuses on predictive maintenance of machinery, and he has published over ten technical papers.