Multi-objective optimization of fiber laser cutting based on generalized regression neural network and non-dominated sorting genetic algorithm

Abstract An integrated model based on generalized regression neural network (GRNN) and non-dominated sorting genetic algorithm (NSGAII) with elite strategy is proposed to predict and optimize the quality characteristics of fiber laser cutting stainless steel. An orthogonal experiment has been conducted where laser power, cutting speed, gas pressure, defocus are considered as controllable input parameters with kerf width and surface roughness as output to generate the dataset for the model. In GRNN-NSGAII model, the cross-validation method was performed to train the network to obtain the optimal GRNN. Significance of controllable parameters of laser on outputs is also discussed. The GRNN model is determined as the fitness function for prediction and calculation during the NSGAII optimization process. NSGAII generates complete optimal solution set with Pareto optimal front for outputs. The prediction relative error of GRNN model is within ± 5%. Experimental verification error of optimized output less than 5%. Characterization of the process parameters in Pareto optimal region has been described in detail.