中文摘要：

针对环境激励下结构健康检测信号存在非平稳性和损伤样本非常有限的问题，提出一种基于时变自回归滑动平均（Auto—Regressive MovingAverage，ARMA）模型和支持向量机（Support Vector Machine，SVM）相结合的桁架结构损伤识别方法．应用该方法对结构进行损伤识别主要有4个步骤：首先应用数值仿真计算方法以及时变参数ARMA模型构建SVM的训练样本和测试样本；其次应用训练样本和测试样本在SVM中进行训练和测试建立结构损伤识别网络；再次采集环境激励下结构的振动信号，并应用小波分析技术对信号进行降噪处理；最后应用所建立的损伤识别网络对预处理后的信号进行计算分析，从而识别出结构的损伤位置和程度．应用该方法对工作状态下WS-PA矿井架进行损伤识别，识别结果验证了该方法识别环境激励下结构损伤的可行性．WS-PA矿井架损伤识别误差都在25％范围内证明所提出的损伤识别方法具有较高的识别精度．本方法为桁架结构的损伤识别提供了新的途径．

英文摘要：

Non-stationary and limited damage sample were two main problems for health detection signal of structure under ambient excitation. To solve the two problems, a method of structural damage identification was proposed based on the model of time-varying auto-regressive moving average （ARMA）and support vector machines （SVM）. There were four main steps to identify structural damages applying the method. Firstly, training sample and testing sample for SVM were established applying numerical simulation method and the time-varying ARMA model. Secondly,structural damage identification network was built with SVM using training sample and testing sample. Thirdly, the structural vibration signal under ambient excitation was collected and the noise of signal was reduced by wavelet analysis technology. Finally, the network was used to analyze the preprocessed signal in order to identify the location and extent of structural damages. The proposed method was used to identify the damages of WS-PA mine frame in working condition. The feasibility of the proposed method to identify structural damages under ambient excitation was verified. The average identification errors of the mine frame in the range of 25% proved that the damage identification method has higher recognition accuracy. The method provides a new way to identity the damage of truss structure.