基于改进BP神经网络的激光选区熔化表面粗糙度预测  

Selective Laser Melting Surface Roughness Prediction Based on Improved BP Neural Network

作　　者：丁燕[1] 王磊[1] 王远 DING Yan;WANG Lei;WANG Yuan(School of Electrical Engineering,Yellow River Conservancy Technical Institute,Kaifeng 475004,China)

机构地区：[1]黄河水利职业技术学院电气工程学院,河南开封475004

出　　处：《电加工与模具》2025年第1期62-68,共7页Electromachining & Mould

基　　金：河南省高等学校重点科研项目(24A510008)。

摘　　要：为提升激光选区熔化表面粗糙度预测的精确度,提出改进BP神经网络模型。首先依据参数建立指数模型,利用灰色关联度分析各因素,求解获得各因素的指数值;然后建立BP神经网络模型,改进粒子群算法优化包括自适应惯性权重更新和动态调节学习因子,同时指数模型预测结果作为特征输入到BP神经网络模型;最后给出算法流程。实验显示,改进BP神经网络在较少的隐含层节点下达到了更低的平均相对误差,激光选区熔化表面粗糙度预测更接近真实值,改进BP神经网络决定系数相比EM、BPNN、GABPNN分别提升了6.40%、1.14%、0.07%,均方根误差相比EM、BPNN、GABPNN分别降低了0.0363、0.0627、0.0668,评价指标较优。In order to enhance the precision of laser selective melting surface roughness prediction,an improved BP neural network model is proposed.Firstly,index model is established based on parameter,and the grey relational degree is used to analyze the correlation degree of each parameter,obtaining the index value of each parameter.Secondly,BP neural network model is established,and the improved particle swarm optimization algorithm is included adaptive inertia weight updating and dynamic adjustment of learning factors,the predicted result of the index model is input the BP neural network model as features.Finally,the process is given.Experiments show that the improved BP neural network has a lower average relative error with fewer hidden layer nodes,selective laser melting surface roughness prediction is close to the true value.The determination coefficient of the improved BP neural network increase 6.40%,1.14%,0.07% compared to EM,BPNN,and GABPNN.The mean square error reduce 0.0363、0.0627、0.0668 compared to EM,BPNN,and GABPNN,so it is better than other algorithms.

关 键 词：BP神经网络 激光选区熔化 粗糙度 粒子群 精确度 

分 类 号：TG665[金属学及工艺—金属切削加工及机床]