M2(HSS)热处理工艺的神经网络优化研究

Research on Neural Network Optimization for Heat Treatment Process of M2(HSS)

采用3×12×2三层拓扑结构,以退火温度、淬火温度、回火温度作为输入层参数,以抗弯强度和磨损体积为输出层参数,构建了M2高速钢(HSS)热处理工艺的神经网络优化模型。并对模型进行了训练、预测验证与现场应用。结果表明:该神经网络优化模型有高的预测能力和预测精度,输出的抗拉强度和磨损体积的平均预测相对误差分别为1.93%、1.89%。与原结果相比,使用神经网络优化工艺参数热处理的M2高速钢的抗弯强度增加35%、磨损体积减小41%。M2高速钢的最佳退火、淬火和回火温度分别为:(810±10)、(1230±10)、(520±10)℃。

Taking annealing temperature, quenching temperature and tempering temperature as input layer parameters, and taking bending strength and wear volume as output layer parameters, the neural network optimization model for heat treatment process of M2 high speed steels （HSS） was built by using three layers topology structure of 3×12×2. And the model was trained, verified and applied on line. The results show that the neural network optimization model has high prediction ability and prediction accuracy. The average relative prediction errors of the output of the tensile strength and wear volume are 1.93% and 1.89%, respectively. Compared with the original results, the bending strength of M2 high speed steel of heat treatment by using process parameters of neural network optimization increases by 35%, and the wear volume decreases by 41%. The optimal annealing temperature, quenching temperature and tempering temperature of M2 high speed steel are （810± 10）, （1230±10） and （520±10）℃, respectively.