Prediction of Laser Cladding Layer Cross-Sectional Morphology—Based on SSA-BP Neural Network
L.J. Cui, Y.X. Liu, S.R. Guo, Y. Zhao, Y.Q. Chen, X.L. Li, Y.H. Cui, B. Zheng and S.Y. Liu
The cross-sectional morphology of the laser cladding layer significantly impacts its quality by influencing both flatness and overlap rates. However, laser cladding faces challenges including difficulties in producing high-quality layers, high costs, and computational complexity. In this paper, the process parameters and cross-sectional morphology of laser cladding layers were used as datasets. A prediction model of cross-section morphology of single-channel laser cladding based on an error back propagation network was proposed. To enhance the performance of the error Back Propagation (BP) network, we applied the Sparrow Search Algorithm (SSA) for optimization, resulting in the development of the SSA-BP network model. The model’s effectiveness was evaluated using R2 (Coefficient of Determination), RMSE (Root Mean Square Error), and MAPE (Mean Absolute Percentage Error) as metrics. SSA-BP was compared against the Marine Predators Algorithm (MPA-BP), the Seagull Optimization Algorithm (SOA-BP), and Particle Swarm Optimization (PSO-BP). Additionally, to assess the model’s applicability beyond the initial process parameters, further experiments were conducted with extended input ranges. The results revealed that the SSA-BP model delivered the best predictive accuracy with the lowest error rates. In summary, incorporating the Sparrow Search Algorithm significantly improved the neural network’s predictive capabilities and general applicability. It provides a reference for the preparation of high-quality cladding layers in laser cladding technology.
Keywords: Laser cladding, cross-sectional morphology, predictive models, neural networks
