Real-Time Capable Identification of Spindle Bearing Loads Using Computer Simulation, Feed Drive Currents and Machine Learning Methods

The main spindle bearings of machine tools are subjected to significant loads resulting from cutting forces. These loads make the bearings susceptible to damage and eventual failure, leading to costly downtime. In addition to sudden failures caused by crashes, gradual wear is a common issue that can ultimately lead to complete bearing failure when the maximum allowable bearing load is exceeded. Consequently, ensuring that this load threshold is never surpassed can significantly extend the bearing’s service life. Key factors in assessing bearing loads include the maximum surface pressure between rollers and bearing rings, ball advance, and the spin-to-roll ratio. However, direct measurement of these parameters in real time is either extremely challenging or not feasible. A promising approach to obtain these values is through computer simulation of the spindle. One widely used tool in the industry is the simulation software by MESYS AG, a standard solution for spindle and bearing manufacturers during the design phase. This software can calculate all relevant bearing loads based on tool forces, spindle rotational speed, and temperature. In this study, machine learning models were trained on a large dataset of simulation results, using cutting forces, spindle speed, and temperature as inputs. These models are capable of predicting bearing loads in a fraction of the time required by the simulation software. As model input cutting forces were calculated using a current-to-force model and compared with measurements obtained via a dynamometer. Additionally, the simulation was corroborated through force tests and eddy current sensors, which measured shaft deviations.