Prediction on Geometrical Characteristics of Laser Energy Deposition Based on Regression Equation and Neural Network

Abstract

Abstract A method for predicting the width and height of deposition track was studied to improve the manufacturing accuracy of laser energy deposition. The regression equation of straight arm size was established through the response surface method, and the laser power, scanning speed and powder feeding were analyzed interactively. A 3-10-2 structure of the BP neural network was established by MATLAB, the inputs were laser power, scanning speed and powder feeding amount, the outputs were the width and height of the layer. By comparing the predictive power of the BP neural network prediction model and the response surface model, it can be seen that both two methods have low error rate in size prediction. The results show that the average prediction error of the width and height is 4.39% and 8.96% with the response surface, while the mean relative error of BP neural network is 2.79% and 3.09%. When the precision requirement is low, it is more convenient to choose the response surface method for regression analysis. The neural network method is more advantageous when the precision requirement is high and the experiment is undersigned. This provides a control of geometric properties in the precision fabrication and repair of large structures.