基于压电阻抗测试的结构诊断神经网络方法

A neural network method for structural diagnosis based on piezoelectric impedance testing

采用单片压电陶瓷片PZT粘贴于某工程结构表面，用阻抗分析仪测量其不同结构状态引起的阻抗特性变化，测试扫描频率范围为1～700kHz．将得到的导纳曲线数据作为神经网络的输入参数，建立一个BP神经网络模型，进行结构状态的识别．该方法与传统的模态方法等相比具有许多优点，如PZT安装方便，小受结构限制，PZT的激励和测试能够利用一组电缆同时进行，测试操作简单．同时，由于直接将测试数据作为神经网络模型的输入参量，凼此无需对测试数据进行复杂的特征提取处理用实测数据对构建的BP神经网络进行了训练和测试，应用结果表明，训练收敛的网络可以较好地识别不同的结构状态．

A single piezoelectric ceramics （PZT） patch is bonded on a structure＇s surface, and the coupled electro-mechanical impedance is measured in different mechanical conditions by an impedance analyzer. A sweep frequency range from lkHz to 700kHz is selected during the testing. Using the measured electro-mechanical admittance curve data as the artificial neural networks＇ input parameters, a BP artificial neural network is designed. The aim is to identify the structure＇s different mechanical states. This method has many advantages compared with the conventional dynamic approaches such as modal-based method. The testing system is very simple and easy to be operated. Meanwhile, it is not necessary to pretreat the testing data to obtain the characteristic parameters as usual. The BP neural network is trained and tested by different measured electro-mechanical admittance data. The testing results show that the network can exactly identify the tested structure in different states.