电磁声多传感多特征融合的构件应力评估研究  

Electromagnetic-Acoustic Sensing-Based Multi-Feature Fusion Method for Stress Evaluation

作　　者：邱发生 刘刚 肖树坤 郭朝阳 章俊燕 李栋 QIU Fasheng;LIU Gang;XIAO Shukun;GUO Chaoyang;ZHANG Junyan;LI Dong(Centre for Inspection and Testing,Jiangxi Hongdu Aviation Industry Group,Nanchang Jiangxi 330096,China)

机构地区：[1]江西洪都航空工业集团有限责任公司检验检测中心,江西南昌330096

出　　处：《传感技术学报》2025年第3期518-525,共8页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金(62201241);江西省自然科学基金青年基金(20224BAB214057);赣鄱俊才-主要学科学术和技术带头人青年项目(20232BCJ23092);中国博士后科学基金面上项目(2023M741097)。

摘　　要：针对传统单一传感和单一特征应力评估稳定性差和准确度低的问题,提出一种多传感多特征融合的应力检测方法,该方法对微观磁畴运动产生的电磁和声学信号进行同步获取并进行智能化合成。从时域和频域两个角度,提取了传感信号的多阶统计特征,构建了不同应力下的特征图谱,为了解决特征之间多重共线性问题,提出主成分分析降维的方法,有效降低了特征维数和应力预测模型的复杂度。进一步,提出一种粒子群优化的BP神经网络算法,构建并求解应力预测模型,与传统的应力回归预测模型相比,该模型准确度达到0.98。同时,提出的多传感多特征融合应力检测方法的性能优于单传感多特征融合方法,能够有效提高应力检测的准确度。多传感多特征检测方法在检测、监测和目标识别等方面具有广泛的应用前景。As the accuracy and stability of the stress measurement by using single sensing and feature is unsatisfactory,a multi-sensing-based multi-feature method for stress characterization is proposed.The method combined the electromagnetic and acoustic signals from the micro level of magnetic domain wall dynamics for fusion.From the time and frequency domain,multi-order statistical feature is ex-tracted to create the feature maps under different levels of stress.To solve multiple co-linear relationships between different features,the principal component analysis is proposed to reduce the dimension of the variable and the complexity of the stress prediction model.Fur-thermore,hybrid of particle swarm optimization(PSO)and BP neural network method is proposed to build and calculate the stress pre-diction model.Compared with conventional regressive predictive model,the proposed model has higher accuracy and robustness.Be-sides,the performance of the proposed method is better than multi-feature fusion of single sensor.The multi-sensing and multi-feature method has promising application prospect in the field of testing,monitoring and target identify.

关 键 词：应力检测 多传感 特征降维 粒子群优化 BP神经网络 特征融合 

分 类 号：TH878[机械工程—仪器科学与技术]