Machine History Based Data Learning in CNC Machining

With the recent development of the machine tool industry, there is an increasing trend of research to collect real-time cutting physics data in CNC cutting and control the CNC. Cutting physics data is the information collected by various sensors such as current, acceleration, voltage, and torque sensors, which indirectly measure the cutting load. This cutting physics data is used for tool condition diagnosis, cutting condition control, and as a key evaluation indicator for predictive maintenance of the spindle. However, in cutting processes, cutting physics data is less useful due to various factors such as machine, workpiece, tool, cutting conditions, and coolant. For more precise machining condition diagnosis or control, a data learning process using the collected cutting physics data is required. In this study, we present a method for storing machining history data that synchronises CNC data and cutting physics data, and data learning using the data. Data learning methods based on machining history data include (1) optimising NC data to improve productivity by controlling the feed rate, (2) learning data separated by tool unit to manage the life of individual tools or diagnose the degree of wear, and (3) generating reference control curves for real-time feed rate control during machining. Such data learning is particularly effective in the case of mass-produced products and is essential for the development of a digital twin system for CNC machining.