Research on laser cladding layer morphology prediction based on multi-objective optimized BP neural network

Laser cladding offers significant potential in industrial manufacturing, yet complex parameter interdependencies limit its broader adoption. While current research predominantly targets roughness reduction for precision applications, components like brake systems require controlled roughness for enhanced friction. Addressing the inherent multi-response nature of laser cladding, this study develops a Taguchi-Grey relational analysis method for multi-objective optimization of surface morphology (width, height, roughness). Experimental validation identified optimal parameters: laser power (1600W), scanning speed (6mm/s), and powder feed rate (13r/min). The neural network model achieved average prediction errors of 5.72% (width), 7.72% (height), and 12.86% (roughness), with experimental errors aligning closely with training-phase values. This demonstrates effective capture of nonlinear parameter-morphology relationships while maintaining balanced multi-objective performance.