Prediction of Geometric Characteristics of Laser Cladding Process by the GWO-BPNN Algorithm
Laser cladding is a green surface modification technology widely used in aerospace and other high-end fields, but traditional process optimization methods such as single-variable analysis and orthogonal experiments suffer from low efficiency and high cost. The geometric characteristics of the cladding layer—dilution rate, forming coefficient, and wetting angle—directly determine service performance. Existing machine learning models often fail to achieve multi-objective optimization and comprehensive prediction. This study proposes a hybrid algorithm combining Grey Wolf Optimizer (GWO) with Backpropagation Neural Network (BPNN) to predict geometric quality indicators. Full-factorial single-track laser cladding experiments were conducted on 316L stainless steel with 316L alloy powder. A polynomial regression model predicted clad width and height with relative error below 4.2%. The GWO-BPNN model predicted dilution rate, forming coefficient, and wetting angle with an average coefficient of determination (R²) of 95.28%, a 12.4% improvement over traditional BPNN (82.93%). Experimental and inverse validation confirmed stable predictive performance across different parameter ranges, meeting engineering tolerance requirements. The method provides a quantitative basis for multi-dimensional optimization of cladding quality and demonstrates practical applicability in industrial scenarios.