基于GA-BP和PSO-BP神经网络的SLM GH3625高温合金残余应力预测研究

Study on residual stress prediction of SLM GH3625 high temperature alloy based on GA-BP and PSO-BP neural networks

采用PSO-BP和GA-BP混合算法的人工神经网络模型预测了选区激光熔化成形GH3625高温合金的残余应力。通过响应面法为实验设计生成样本集,以激光功率、扫描速度和扫描间距作为模型的输入层,以残余应力作为模型的输出层进行预测优化。采用相关系数R^(2)和平均绝对相对误差e_(AARE)评价指标对预测模型进行了验证和对比分析。结果表明:BP、 GA-BP和PSO-BP神经网络模型均能够较好地预测不同工艺参数下GH3625高温合金的残余应力,且通过算法优化后的BP神经网络具有更高的预测精度。其中GA-BP神经网络对选区激光熔化成形GH3625高温合金残余应力的预测精度最高,模型性能更优越,其相关系数R^(2)和相对平均绝对误差e_(AARE)分别为0.909和2.06%。

An artificial neural network model with PSO-BP and GA-BP hybrid algorithms was used to predict the residual stress of GH3625 superalloy by selective laser melting forming.The sample set was generated for the experimental design by response surface method,and the laser power,scanning speed and scanning space were used as the input layer of the model,and the residual stress was used as the output layer of the model for prediction optimization.The correlation coefficient R^(2) and the average absolute relative error e_(AARE) evaluation indexes were used to validate and compare the prediction models.The results show that BP,GA-BP and PSO-BP neural network models can well predict the residual stress of GH3625 superalloy with different process parameters,and the BP neural network optimized by the algorithms has higher prediction accuracy.Among them,the GA-BP neural network has the highest prediction accuracy and superior model performance for the residual stress of GH3625 superalloy formed by laser melting forming in the selected area,and its correlation coefficient R^(2) and relative average absolute error e_(AARE) are 0.909 and 2.06%,respectively.