The Study Of Structural Parameters Identification Approach Based On Neural Network And ARMA Model

Many civil infrastructures are now deteriorating due to aging, damage accumu-lation, misuse, lacking proper maintenance, and, in some cases, overstressing as a result of increasing load demands and changing environments in service lifetime. Failure of these infrastructures often leads to a high social consequence. It is therefore critical to evaluate their current reliability, performance, and condition for the pre-vention of potential catastrophic events. The structural parameter identification for structural health monitoring, performance assessment and safety evaluation of exist-ing infrastructures has become an increasingly important research topic in civil engi-neering structures.As a non-parametric modelling method, neural network can approximate arbi-trary continuous functions, which have drawn considerable attention in civil engi-neering for identification using time-domain data. On the other hand, the Au-to-regressive and moving average (ARMA) model, as a time-domain method, has been widely employed to identify structures. From the discrete solution of the equation of vibration of engineering structure, the equalility of the neural network based time domain identification and the ARMA model was verified. And then, a novel structural parameters identification methodology by matching neural network and ARMA model was proposed and the accuracy and efficacy of the proposed strategy were validated with numerical simulation and lab test.The main contents of this paper are as follows:1. On the comprehensive review of the traditional parameter identification tech-niques, the basic theory of artificial neural network and ARMA model was introduced. From the discrete solution of the equation of vibration of engineering structure, the equalility of the neural network based time domain identification and the ARMA model was verified.2. In this paper, a novel structural parameters identification methodology by matching neural network and ARMA model was proposed, and the corresponding program was developed to realize the algorithm based on Matlab.3. The accuracy and efficacy of the proposed strategy were validated with nu-merical simulation for a structure under various excitations. The results indicated that the proposed method is able to accurately identify structural parameter matrices. 4. The performance of the proposed novel method for linear multi-DOF structur-al dynamic system was studied with experiment of a 4-DOF frame structure model. The structure model was excited by hammer. The acceleration response of the struc-ture model and the impact force time series were employed to identify the stiffness and damping matrices of the structure model. It can be seen that the proposed method can identify structural parameter matrices with acceptable accuracy.