A Data-Driven Modeling Approach for Accelerating Dynamics Simulation of Feed Drive System

In the field of CNC machine tools, the feed system typically utilizes servo control for motion control. Throughout the motion control process, the ideal motion control command undergoes correction by the motion controller before reaching the motor controller. Subsequently, the motor controller modulates the motion control command through pulse width modulation to generate pulse signals that regulate the motor. The computational complexity associated with comparator-based digital pulse width modulation techniques poses challenges to the efficiency of dynamic simulations in feed systems. To tackle this challenge, this paper proposes the integration of LSTM neural networks to model the pulse width modulation component of the motor controller, creating a direct mapping between the output of the motion controller and that of the motor controller. This direct mapping relationship significantly reduces the generation time of motor motion commands, thereby enhancing the overall efficiency of dynamic simulations in feed systems.