Machine-Learning-Based Model Parameter Identification for Cutting Force Estimation

Abstract

Recently, there has been an increased demand for precise monitoring of the milling process using machine tools through a simple and cost-effective method. Accurate estimation of cutting forces is highly effective for this monitoring, and one approach is the modeling of tool spindles and tables of a machine tool. To model machine structures, well-known methods involving the use of impulse hammer response or structural analysis exist. However, the complex modeling is hard to achieve when using the impulse response. Moreover, it is often considerably difficult to achieve the modeling with structural analysis because the preparation of the accurate model and highly complicated calculations are required. Therefore, in this study, we propose a new monitoring method to identify model parameters of the machine structure and estimate cutting forces. First, a simplified assumed structure is prepared based on locations where sensors can be mounted. Next, measurement data during actual milling process are collected through the acceleration sensors mounted on the tool spindle and the dynamometer for the cutting force attached to the table. Subsequently, model parameters are identified from these data using machine learning. A 3-axis NC milling machine was used to evaluate the application range of the model parameters by changing cutting conditions, milling direction, cutting tools, and materials. The model parameters identified using the proposed method were equivalent to those using the impulse response. Furthermore, even in cases where the impulse response was difficult to identify, suitable model parameters were identified using machine learning. Finally, we confirmed that the proposed method can accurately achieve in-process monitoring of cutting forces in the X, Y, and Z directions.