基于潜在成分和概率神经网络的时变结构系统的损伤识别

Damage Identification by Probabilistic Neural Networks Based on Latent Components for Time-Varying Structure System

介绍了基于潜在成分(LC)分析和概率神经网络的损伤识别方法,并应用于一个实验室模型的损伤识别。结果表明,基于潜在成分(LC)分析和概率神经网络的损伤识别方法能在正常的时变质量情况下以较高的成功率对位于A或B处的某一损伤程度未知的损伤进行归类,为时变结构系统的定量损伤识别作出了有益的尝试。

A novel method of damage identification for health monitoring of a time-varying system is presented. The functional-series time-dependant automation auto regressive moving average （FS-TARMA） time series model is applied to vibration signal observed in time-varying system for estimating TAR/ TMA parameters and innovation variance. They are the time function represented by the group of projection coefficients on certain functional subspace with specific basis functions. The estimated TAR/ TMA parameters and innovation variance are further used to calculate the latent components （LCs） as the more informative data for health monitoring evaluation based on an eigenvalue decomposition tech- nique. LCs are then combined and reduced to numerical values as feature sets, which are input to proba- bilistic neural networks （PNNs） for damage classification. For evaluation of the proposed method, numerical simulations of the damage classification for a time-varying system are employed, in which different classes of damage are modeled by the mass or stiffness reductions. It is demonstrated that the method can discriminate the time varying nature of system parameters and damages occurring in the course of operation and causing the change of parameters. By using the proposed method, the success rate of classification will be enhanced compared with non-reduced and ordinary feature extraction methods.