Exploring Human Perception in Machining Monitoring

Monitoring CNC machining processes ensures the satisfaction of dimensional tolerances and surface finish requirements, while minimizing tool wear and downtime caused by accidents. Sensor-based automated monitoring systems driven by Artificial Intelligence (AI) trained on extensive datasets have demonstrated high accuracy and consistency in monitoring tool wear progression. However, the performance of AI-driven monitoring systems heavily relies on the available training data, often struggling with scenarios not covered in training, such as variations in tool path, feed rate, and machine models. In contrast, experienced machinists excel in monitoring machining operations across diverse tool paths, machining parameters, and machine tool models, often without the need for specific case-by-case training. Limited research has been dedicated to understanding the underlying reasons for the flexibility exhibited by humans in machining monitoring. This thesis aims to investigate the human perception process in tool wear monitoring. Applied machining tasks and sensational signal source focused experiments were conducted to understand utilization of human senses as well as effectiveness of sensory signal sources. The method of constant stimuli, a common approach in perception research to analyze perception and judgment trends of participants is employed. Results show vision, hearing, and touch are utilized to gather sensory information from sources including machining noise, finished parts, end mills, and chips. Vision, auditory and touch perceptions with corresponding signal sources have different wear thresholds, sensitivity and consistency. Insights gained could serve as inspiration for efficient sensor applications and data processing algorithms for future sensor-based automated machining monitoring systems.