Investigation on surface morphology model of Si3N4 ceramics for rotary ultrasonic grinding machining based on the neural network

Si3N4 ceramics parts surface morphology is related with surface friction and wear properties directly. Poor surface morphology will result in friction coefficient increases, strength decreases, and even lead to component failures. In order to improve Si3N4 surface morphology, it is necessary to investigate on the relationship model between the surface morphology and process parameters. In the paper, rotary ultrasonic grinding machining (RUGM) was taken as object to establish the model based on back propagation (BP) neural network. However, the nonlinear relationship of the model is complex, and the traditional algorithm cannot realize satisfying results. So an improved BP neural network algorithm based on Powell method has been proposed. The paper gives the theory and calculation flow of the algorithm. It is found the algorithm can accelerate the iteration speed and improve iteration accuracy. The investigation results provide the support for surface morphology optimization.