基于BP神经网络算法的316L不锈钢极薄带热处理力学性能预测  

Prediction of Mechanical Properties of Heat-treated 316L Ultra-thin Strip Based on BP Algorithm

作　　者：张志雄 付永伟[1] 王涛 王彬 熊晓燕[1] 王天翔[3] ZHANG Zhi-xiong;FU Yong-wei;WANG Tao;WANG Bin;XIONG Xiao-yan;WANG Tian-xiang(College of Mechanical and Delivery Engineering,Ministry of Education,Taiyuan University of Technology,Taiyuan 030024,China;Engineering Research Center of Advanced Metal Composites Forming Technology and Equipment,Ministry of Education,Taiyuan University of Technology,Taiyuan 030024,China;Shanxi Taigang Stainless Steel Precision Strip Co.,Ltd.,Taiyuan 030006,China)

机构地区：[1]太原理工大学机械与运载工程学院,太原030024 [2]太原理工大学先进金属复合材料成形技术与装备教育部工程研究中心,太原030024 [3]山西太钢不锈钢精密带钢有限公司,太原030006

出　　处：《精密成形工程》2023年第2期68-77,共10页Journal of Netshape Forming Engineering

基　　金：国家自然科学基金(51974196,51901151);山西省科技重大专项(20181102015);中国博士后科学基金(2020M680918,2021T140503)。

摘　　要：目的 为了预测不锈钢极薄带热处理后的力学性能、优化热处理工艺以及实现热处理工艺的智能控制,构建基于BP算法的神经网络模型。方法 以316L不锈钢极薄带为研究对象,进行热处理试验和拉伸试验,通过以热处理的退火温度、保温时间和取样方向作为输入层参数,以屈服强度、抗拉强度、断后伸长率作为输出层参数,采用BP算法构建了316L不锈钢极薄带力学性能预测的思维进化算法优化BP神经网络模型,并进行模型的预测和应用验证,考虑不同隐含层节点数及不同BP神经网络模型对性能的影响。结果 思维进化算法优化的BP神经网络模型测试集的屈服强度、抗拉强度和断后伸长率的平均相对误差分别为8.92%、5.21%和9.28%,训练集相关系数为0.980 94。思维进化算法优化BP网络单、双隐含层误差总和最低分别为0.578 6和0.546 9,BP网络与思维进化算法优化的BP网络误差总和最低分别为0.579 9和0.546 9。结论 思维进化算法优化BP神经网络模型具有较好的预测能力和泛化能力,以及较高的预测精度。与企业现用生产工艺相比,采用模型优化后热处理工艺的综合力学性能有显著提高。The work aims to predict the mechanical properties of heat-treated stainless steel ultra-thin strip, optimize the heat treatment process, and achieve intelligent control of heat treatment, and constructs a neural network model based on BP algorithm. The heat treatment experiment and tensile experiment were carried out on 316L stainless steel ultra-thin strip. The annealing temperature, holding time and sampling direction of the heat treatment of 316L stainless steel ultra-thin strip were taken as the input layer parameters. The yield strength, tensile strength and elongation after fracture were taken as the output layer parameters. BP algorithm was used to construct the BP neural network model optimized by the mind evolutionary algorithm for predicting the mechanical properties of 316L stainless steel ultra-thin strip and the prediction and application verification of the model were carried out. The effects of different hidden layer nodes and different BP neural network models on performance were considered. The average relative errors of yield strength, tensile strength and elongation of BP neural network model testing set optimized by mind evolutionary algorithm were 8.92%, 5.21% and 9.28%. In addition, the correlation coefficient of training set was 0.980 94. The minimum error sum of single and double hidden layers of BP network optimized by mind evolutionary algorithm was 0.578 6 and 0.546 9 respectively, and the minimum sum of BP network and BP network optimized by mind evolutionary algorithm was 0.579 9 and 0.546 9 respectively. The BP neural network model optimized by mind evolutionary algorithm has a good prediction ability, high prediction accuracy and good generalization ability. Compared with the current production process of enterprises, the comprehensive properties of the production process after model optimization is significantly improved.

关 键 词：BP神经网络 思维进化算法(MEA) 316L 极薄带 热处理 综合量化 

分 类 号：TG156[金属学及工艺—热处理]