|
| 基于移动主成分分析敏感特征的智能损伤识别方法 |
| A Novel Intelligent Damage Identification Method Utilizing Sensitive Features Extracted by Moving Principal Component Analysis |
| 投稿时间：2023-05-24 修订日期：2023-07-25 |
| DOI： |
| 中文关键词: 结构健康监测 移动主成分分析 损伤敏感特征 损伤识别 机器学习 |
| 英文关键词:Structural health monitoring moving principal component analysis damage sensitive feature damage identification machine learning |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目）,中国博士后科学基金 |
|
| 摘要点击次数: 1504 |
| 全文下载次数: 0 |
| 中文摘要: |
| 本文基于主成分分析(moving principal component analysis，MPCA)发展多阶PCA损伤敏感特征，并结合机器学习算法开展结构损伤识别研究。首先，考虑不同强度噪声干扰，基于方差累积率自适应地确定PCA特征向量阶数；其次，利用多阶特征向量内积获得PCA敏感特征；更进一步地，将新型PCA敏感特征与机器学习算法结合，对结构进行损伤定位和损伤定量研究。结合双跨连续梁的计算结果表明：相比其他PCA特征，所提出的新型损伤敏感特征具有更好的灵敏度和精度，结合新型PCA敏感特征与人工智能算法在损伤定位和定量任务中具有较高的准确率和稳健的鲁棒性，即使在信噪比为SNR10的高强度噪声干扰下，所提出的方法在损伤定位和损伤定量方面的准确率依旧高于80%。 |
| 英文摘要: |
| A novel structural damage identification method is proposed in this paper. In this method, multi-order eigenvectors are extracted by moving principal component analysis (MPCA) and cooperated with for machine learning algorithms for damage identification. Firstly, based on the cumulative variance ratio principles, PCA eigenvector orders are determined adaptively. Secondly, the PCA sensitive features are obtained by the inner product of multi-order eigenvectors. Additionally, the proposed PCA sensitive features are combined with machine learning algorithms for damage localization and quantification. The results of a double-span continuous beam, demonstrate that the novel proposed damage-sensitive features have higher sensitivity and enhance the prediction accuracies for machine learning(ML) algorithms, compared to other PCA features. Furthermore, ML algorithms with novel PCA-sensitive features have superior performance in damage localization and quantification. Even under strong noise of SNR10, the prediction accuracies of the proposed method for damage localization and quantification are still higher than 80%. |
| 下载PDF阅读器 |
| 关闭 |
