Hysteresis modeling and experimental research of pneumatic constant force grinding system based on GRU neural network

The pneumatic constant force grinding system has attracted a lot of attention in the fields of robot grinding. However, the hysteresis behavior of the pneumatic constant force grinding system has multi-valued mapping characteristics and dynamic characteristics related to the frequency and amplitude of the input signal. Aiming at the hysteresis problem of pneumatic constant force device, a gated recurrent unit neural network (GRU) modeling method based on Bayesian optimization is proposed, and the effectiveness of the modeling method is verified by experiments. The ability of the model to describe the hysteresis behavior of the system and the force control performance and polishing effect based on the model are analyzed by experiments. The results show that the GRU model has better description ability for the hysteresis behavior related to amplitude and frequency than the rational Bezier curve fitting method (BCM). Through the grinding test, it can be seen that the output constant force fluctuation range based on the GRU neural network model is −1 ~ 1 N. With the increase of the grinding force, the surface roughness and morphology of the specimen are improved, and the average surface roughness is reduced by 23.79 % at most.