Research On Damage Diagnosis Based On Time Series Analysis And Artificial Neural Network

Developed for studying long sequences of regularly sampled data, time series analysismethods have been increasingly investigated for the structural health diagnosis.Meanwhile, Artificial Neural Networks (ANN), an emerging smart detecting method,have received increasing attention for detecting damage of structures based onvibration. Silo structure have been wildly used in industrial production as one of thesimple structures. During the servicing period, the structures suffer different degree ofdamages due to various complicated factors. It may cause severely loss of lives andproperties if not paying attention to these damages promptly. Hence, it’s verynecessary to diagnose structure damage. In order to improve the accuracy and robustof damage detection, the above-mentioned methods are combined, and a newapproach based on time series analysis and Artificial Neural Networks is proposed todiagnose structural damage. Simulating and analyzing of silo structures to white noiseverify the validity of the method.The content and methods contained in this thesis are mainly as follows:By motivating the structures by white noise, the acceleration responses are usedto create the time series model, and the first four-order model coefficient and firstperiod of structure are considered to be damage sensitive factor and used as inputsinto a neural network, and the outputs of neural network are established according tospecific cases. After completion of the training network which can be used to detectstructure damage, the position of damage is detected by a trained BP network, whilethe extent was detected by a trained RBF network.The finite element model of silo structure is built by SAP2000. On singledamage condition, the extent was classfied by a RBF network after position wasidentified by BP network. Then several measurement noise was added into theacceleration responses to determine whether this method is valid in noise condition.The identification of damage extent and position under muti-damage condition wasthe same as single damage condition.The result showed that the combination of time series analysis and artificialneural network is a efficient tools for damage classification，no mater under single or double damage condition.