Research on the control of the grinding force of spiral bevel gear by neural network system based on model reference adaptive algorithm

Abstract

This article designs a neural network model based on model reference adaptive (MRA) control, which outputs the control voltage of the X, Y, and A-axis permanent magnet synchronous motor (PMSM) of the machine tool, so that the motor speed always follows the expected value. By changing the grinding speed, the goal is to control the main grinding force and reduce the roughness of the gear engagement surface. Firstly, a main grinding force model for spiral bevel gears was established, and the height parameters of the gear meshing surface roughness were scanned. The analysis indicates that the Pearson correlation between the main grinding force and roughness is 81.58 %. To reduce tooth surface roughness, set a grinding force threshold and calculate the expected angular velocities of the axes. Secondly, the state equation of the PMSM is established, and the Lyapunov second method is applied to design an MRA control algorithm. It is found that the model output can follow the reference model well and adapt to changes in load torque. However, there is an overshoot, and the model requires many feedback signals. Finally, to further optimize the control system, a generalized regression neural network (GRNN) was established. Founded on the output voltage of the MRA control system, training samples were established to complete the speed control of the machine tool PMSM. The results indicate that there is a strong correlation between grinding force and tooth surface roughness, and the GRNN system has good force control performance, which can indirectly improve grinding quality.