Research On Structural Health Condition Prediction Based On The Extreme Learning Machine Markov Model

The construction of civil engineering has made a breakthrough 、 a landmark development since the 20 th century at home and abroad, the safety and durability of civil engineering structures is paid close attention. At the same time the research of structural health monitoring(SHM) techniques becomes hot points in civil engineering. It has important significance to take effective measures to make diagnosis 、evaluation and prediction for the health condition of structures. For the purpose of SHM and damage diagnosis, the damage feature extraction and trend prediction for engineering structure are studied in this paper.Summarized as follows:(1)In order to extract the effective structure damage feature, the methods of damage feature extraction are developed based on Ensemble Empirical mode decomposition(EEMD) and Hilbert transform(HT). The original signal processed by using EEMD, the intrinsic mode function(IMF) which contains structural damage information are selected; then calculating the instantaneous frequency using the HT transform. The main results are summarized as: It is shown that the instantaneous frequency(IF) is obviously changed before and after the structure damage occurrence. The IF can effectively reflect the structure status change and the tendency of structure rigidity change. So it can be taking as a feature index to represent the developing trend of a structure health condition.(2)Due to the prediction accuracy of the Extreme learning machine(ELM) algorithm is not high, the methods of prediction are developed based on ELM and Markov Model. Firstly, the IF of Structural Damage is predicted by ELM, fitting error is calculated; then the prediction error of ELM is predicted by Markov Model, and the prediction value of ELM is corrected. The main results are summarized as: the prediction accuracy is not high because of the extreme learning machine Markov Model’s own algorithm, the prediction of structure health is not effectively; different number of hidden layer neurons and Markov state will affect prediction accuracy, so we need to choose the number of neurons and states for Markov on specific issues.(3)Because of the prediction accuracy of extreme learning machine Markov Model cannot meet the requirements, the optimization methods are developed based on Particle Swarm Optimization(PSO). This method uses PSO to adapt optimization the state prediction value of extreme learning machine Markov Model. The main results are summarized as: PSO can optimize the state prediction value of extreme learning machine Markov Model, so the prediction value of Extreme learning machine Markov Model is closer to the true value and more accurate.