Prediction of 18Ni300 laser cladding topography based on back-propagation neural network and particle swarm optimization

Abstract

Abstract Laser cladding is an advanced new surface manufacturing technology, which has a wide range of applications in production, repair and surface strengthening of parts in many industries. Aiming at the problem that the forming quality cannot be guaranteed after the Q235 substrate, a full factorial design method was used to conduct the experiments. Then a multiple linear regression model was established to obtain the functional expression between the cladding layer morphology parameters (cladding layer height, cladding layer width and dilution ratio) and laser power, scanning speed and powder feeding rate. And then back-propagation neural network (BPNN) and particle swarm optimization (PSO) were used to establish prediction models between the morphology quality of the laser cladding layer and the process parameters. Finally, the prediction results of two models were compared with the measured value, respectively. The results showed that the average relative errors of the multiple linear regression model for the prediction of cladding layer width, height, and dilution ratio were 4.525%, 6.018%, and 5.659%, respectively, while the average relative errors predicted by the optimized neural network model were 3.177%, 7.242%, 4.679%, respectively. The goodness of fit(R2) in linear regression model were 0.8034, 0.8407, and 0.8297 respectively, while the R2 in PSO-BPNN model were 0.8134, 0.8336, and 0.9576 respectively. In this paper, the feasibility of the PSO-BPNN model in theory and practice was verified, and it provided theoretical guidance for the realization of the subsequent laser cladding closed-loop control system.