Processing quality prediction and multi-objective optimization of polysilicon wire-EDM

Abstract

Wire electrical discharge machining (wire-EDM) has recently been applied in the cutting of polysilicon ingots. However, the challenge of achieving a balance among processing quality, efficiency, and material waste has imposed limitations on its practical industrial application. In this paper, the Dung Beetle Optimizer (DBO) was employed to optimize the weights and thresholds of a back propagation neural network (BPNN) to construct a DBO-BP prediction model. This prediction model was developed to establish a non-linear mapping relationship between process parameters (open voltage, pulse width, pulse interval ratio, wire speed, and feed velocity) and process indicators (surface roughness and material removal rate). Furthermore, a DBO-BP-PSO multi-objective optimization model was developed by coupling with the Multi-objective Particle Swarm Optimization (MOPSO). This model aimed to minimize surface roughness and kerf width while maximizing material removal rate, thereby determining the optimization combinations of process parameters. The Pareto optimal solution set was obtained through solving this model. The results of further verification experiments demonstrated that the DBO-BP-PSO multi-objective optimization model was capable of effectively achieving precise prediction and optimization of the processing results in polysilicon wire-EDM.