基于贝叶斯正则化的多源/连续冲击载荷识别及试验研究  

Identification and test study on multi-source/continuous impact load based on Bayesian regularization

作　　者：龙旭[1] 胡运涛 林华刚[1] 马锐磊 常晓通[1] 苏昱太 LONG Xu;HU Yuntao;LIN Huagang;MA Ruilei;CHANG Xiaotong;SU Yutai(School of Mechanics,Civil Engineering and Architecture,Northwestern Polytechnical University,Xi’an 710072,China;The 705 Research Institute,China Shipbuilding Industry Corporation,Xi’an 710077,China;Science and Technology on Underwater Information and Control Laboratory,Xi’an 710077,China)

机构地区：[1]西北工业大学力学与土木建筑学院,西安710072 [2]中国船舶重工集团公司第七〇五研究所,西安710077 [3]水下信息与控制重点实验室,西安710077

出　　处：《振动与冲击》2024年第21期55-63,共9页Journal of Vibration and Shock

基　　金：航空科学基金(20220015053005);陕西省自然科学基金(2021JQ-072)。

摘　　要：针对冲击载荷识别中存在病态矩阵求逆的不适定问题和噪声敏感问题,给出一种增广Tikhonov正则化技术的改进贝叶斯方法。通过引入小波阈值方法解决高噪声水平下多源/连续冲击载荷识别精度不佳的问题,在识别过程中自适应地确定最优正则化参数,并有效地剔除噪声对冲击载荷识别的影响。通过开展飞机壁板结构在不同冲击载荷和信噪比噪声下的数值仿真分析,以相关系数和相对误差作为评价指标,对比讨论了基于L曲线法和广义交叉检验法的Tikhonov正则化方法、贝叶斯正则化方法以及该文方法的识别效果,结果表明该文方法兼顾了曲线的光滑性与峰值识别的准确性,在20 dB高噪声水平连续冲击载荷识别时峰值平均误差不超过14%。开展了典型加筋壁板结构的冲击试验,验证了该文方法对实际工程中典型多源/连续冲击载荷的识别能力,峰值平均误差控制在18%以内,为解决工程应用中的载荷识别问题提供了有效途径。Here,aiming at ill-posed problems of ill-posed matrix inversion and noise sensitivity in impact load identification,an improved Bayesian method for augmented Tikhonov regularization technique was proposed.By introducing wavelet thresholding method to solve the problem of poor accuracy in identifying multi-source/continuous impact load under high noise level,the optimal regularization parameters were adaptively determined in identification process,and effects of noise on impact load identification were effectively eliminated.By performing numerical simulation analysis of aircraft wall panel structures under different impact loads and noise with different signal-to-noise ratios,taking correlation coefficients and relative errors as evaluation indexes,recognition effects obtained with Tikhonov regularization method based on L-curve method and generalized cross validation method,Bayesian regularization method and the proposed method,respectively were discussed contrastively.The results showed that the proposed method takes into account both curve smoothness and peak recognition accuracy,and the average peak error doesn’t exceed 14%when identifying continuous impact load at a high noise level of 20 dB.Impact tests were conducted for typical reinforced wall panel structures to verify the recognition ability of the proposed method for typical multi-source/continuous impact loads in practical projects.It was shown that the average peak error can be controlled within 18%;the proposed method can provide an effective way for solving load recognition problems in engineering applications.

关 键 词：载荷识别 贝叶斯正则化 小波阈值法 不适定问题 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3