Using a Hybrid Neural Network to Predict the Surface Morphology of Laser Surface Textured Ni-coated MoS2 40Cr Alloy Steel

A method for the prediction of surface texture morphology after the laser surface texturing (LST) of a biomimetic sinusoidal texture based on scal-lop shells was established. The method was based on the back -propaga-tion (BP) neural network optimized using hybrid metaheuristic algorithms. The number of scanning times, pulse frequency, laser power and scanning speed were used as input parameters, while the texture morphology parameters were used as the output parameters. A hybrid approach com-bining the genetic algorithm (GA) and particle swarm optimization (PSO), which complemented the advantages of both algorithms, was applied to optimize the BP neural network (GA-PSO-BP). The resulting model was then compared to two traditional neural networks, GA-BP and PSO-BP neural networks, and its prediction accuracy was evaluated using several different metrics. Metrics included the maximum relative error (MRE), mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). The results have shown that the GA-PSO-BP neural network has better prediction performance of the tex-ture width and depth in terms of lower error, while also being more stable. Such behaviour demonstrated that the established GA-PSO-BP neural network can be a powerful tool for predicting the sinusoidal texture mor-phology parameters obtained in laser processing.