采用实验与神经网络预测相结合的方法, 对基于温度控制的激光相变硬化工艺参数进行了研究。首先, 使用基于温度可控的大功率半导体直接输出激光加工系统对45#钢进行设定温度下的激光相变硬化实验。然后, 通过改进的BP神经网络算法构建神经网络模型, 并采用所获得的实验样本数据训练该人工神经网络模型。模型中所采用的改进BP神经网络算法是遗传算法和基于新型误差函数的批量训练神经网络算法相结合的混合算法。为验证改进算法的性能, 在Lab Windows/CVI软件上应用C编程语言实现了该算法。通过运行程序发现, 采用此算法的收敛速度比传统算法提高了约80%, 预测输出的指标与实际值之间的偏差小于4%。

In this paper, experiment and neural network prediction are combined to study the process parameters of temperature controlled laser transformation hardening. Firstly, the laser phase transformation hardening experiment of 45# steel was carried out by using a high power semiconductor direct output laser processing system based on temperature controlled. Then, an improved BP neural network modelwas constructed by a hybrid algorithm which integratedgenetic algorithm and batch training neural network algorithm based on a new error function. The experimental data was used to training the artificial neural network model.In order to test the performance of the model, the algorithm was implemented by a C language compiler in Lab Windows/CVI. The calculation resultsshow that the convergence rate of the improvedalgorithm is about 80% higher than that of traditional algorithm, and the deviation between the predicting output values and the actual valuesare less than 4%.