基于IWOA-BP算法的金属结构弱磁检测缺陷量化研究  

Quantitative study on weak magnetic detection defects of metal structure based on IWOA-BP algorithm

作　　者：樊梦 童博[2] 高晨 姚中原 张宇 胡博[1] FAN Meng;TONG Bo;GAO Chen;YAO Zhongyuan;ZHANG Yu;HU Bo(Key Laboratory of Non-Destructive Testing of Ministry of Education,Nanchang Hangkong University,Nanchang 330063,China;Xi’an Thermal Power Research Institute Co.,Ltd.,Xi’an 710054,China;Jiangsu Clean Energy Branch,Huaneng Power International Inc.,Nanjing 210015,China)

机构地区：[1]南昌航空大学无损检测技术教育部重点实验室,南昌330063 [2]西安热工研究院有限公司,西安710054 [3]华能江苏清洁能源分公司,南京210015

出　　处：《机械强度》2025年第3期113-120,共8页Journal of Mechanical Strength

基　　金：中国华能集团总部科技项目(HNKJ20-H72);南昌航空大学研究生创新专项(YC2022-088);江西省重点研发计划(20243BBG71005);江西省主要学科学术和技术带头人培养项目(20243BCE51052)。

摘　　要：金属结构被广泛用于工业界,在役金属结构受拉压疲劳载荷易产生裂纹缺陷,为实现金属结构裂纹缺陷的定量化检测,研究了一种基于反向传播(Back Propagation,BP)神经网络的金属结构弱磁检测缺陷定量分析方法。针对BP神经网络在参数调整时的效果欠佳、效率低等问题,采用基于Sine混沌映射的改进鲸鱼优化算法(Improved Whale Optimization Algorithm,IWOA)对BP神经网络参数调整方式进行优化,兼顾全局寻优的同时提高局部寻优的能力,进而将IWOA搜索到的最优参数赋值给BP神经网络,提高网络初始参数的质量。以人工矩形槽模拟裂纹,对矩形槽的长度、宽度、深度进行反演定量。结果表明,IWOA-BP神经网络预测的平均精度均在80%以上,深度、长度、宽度预测精度分别提高了106.72%、9.68%、6.86%。Metal structures are widely used in industry.Metal structures in service are prone to crack defects under tensile and compressive fatigue load.In order to realize quantitative detection of metal structures’crack defects,a quantitative analysis method of metal structures’weak magnetic detection based on back propagation(BP)neural network was studied.In view of the poor effect and low efficiency of BP neural network in parameter adjustment,the improved whale optimization algorithm(IWOA)based on Sine chaotic mapping was adopted to optimize the BP neural network parameter adjustment mode,giving consideration to global optimization while improving the local optimization ability,and then the optimal parameters searched by IWOA were assigned to BP neural network,improving the quality of initial network parameters.The length,width and depth of the artificial rectangular slot were quantified by inversion.The results show that the average prediction accuracy of IWOA-BP neural network is above 80%,and the prediction accuracy of depth,length and width is improved respectively by 106.72%,9.68%and 6.86%.

关 键 词：弱磁检测 金属结构 BP神经网络 鲸鱼算法 IWOA-BP神经网络 

分 类 号：G306.0[文化科学]