基于移动主成分分析与集成学习的结构损伤识别方法

Structural Damage Identification Method Based on Moving Principal Component Analysis and Ensemble Learning

为了提高结构损伤定量和定位的准确率,提出基于移动主成分分析与集成学习的结构损伤识别方法;利用移动主成分分析对原始应变响应数据进行特征分析,得到包含损伤信息的第一、第二特征向量,将两者相结合所得的组合特征向量作为损伤指标输入集成学习模型,进行结构损伤程度和损伤位置预测;采用双跨平面梁的仿真应变监测数据,对所提出的结构损伤识别方法的有效性进行验证,对比分别以第一、第二、组合特征向量作为输入的分类模型的损伤定量和定位的准确率。结果表明:在一定强度的噪声条件下,组合特征向量能同时具备第一、第二特征向量的优点,并且能克服单个特征向量的局限,获得优异的损伤识别性能和抗噪性;在信噪比为40 dB的弱噪声情况下,将组合特征向量输入集成学习模型进行损伤定量和定位,准确率分别可达98.9%、99.0%,在信噪比为10 dB的强噪声情况下准确率仍分别可达82.3%、73.2%。

To improve accuracy of structural damage quantification and localization,a structural damage identification method based on moving principal component analysis and ensemble learning was proposed.Moving principal component analysis was used to analyze characteristics of original strain response data,and the first and second eigenvectors containing damage information were obtained.The combined eigenvector obtained by combining the first and second eigenvectors was input into an ensemble learning model as a damage index to predict damage degree and damage location of the structure.Effectiveness of the proposed structural damage identification method was verified by using simulated strain monitoring data of a double-span planar beam.Accuracy of damage quantification and localization of classification models was compared with the first,second,and combined eigenvectors as inputs,respectively.The results show that the combined eigenvector takes advantage of the merit of each eigenvector and is capable of overcoming their respective limitations under a certain intensity of noise,resulting in excellent damage identification performance and noise resistance.With the low noise of signal to noise ratio of 40 dB,the ensemble learning model with the combined eigenvector as a input can achieve damage quantification and localization accuracy of 98.9% and 99.0%,respectively,and for signal to noise ratio of 10 dB,it can reach 82.3% and 73.2%.