Structural Damage Identification Based On Hidden Markov Model

Damage identification is an important component of structural health monitoring system.Nowdays big data and artificial intelligence bring new opportunities and challenges for damage identification,and such new technologies as neural network,genetic algorithm,etc.,have been widely applied in damage identification and model updating of complex structures,showing their great potential in these aspects.Due to the uncertain relationship between the response of structure and damage state,in this paper,a time series probability model,the Hidden Markov Model(HMM)is established a probability relationship between the former two and employed for structural damage existence,damage localization and severity identification.The main contents are as follows.After an introduction to the basic theory of HMM,several HMM based structural damage identification schemes are proposed,including a damage existence identification method based on a HMM-lnKL index,the model library method using the maximum likelihood Bayesian probability and the implicit status sequence recognition method using the Viterbi algorithm.Both of the latter are for damage localization and severity identification.The three-level structural damage identification is investigated based on the Discrete Hidden Markov Model(DHMM).A simple support beam model is used to generate the sample sets for training of the DHMM.The HMM-lnKL index is used to indicate existence of damage in the structures.For damage localization and severity identification,both the model library method and the implicit status sequence recognition method are studied.For the first method,a model library for the single damage,two damages and multi damages cases are trained,and then damage identification is performed.The classification capability and generalization capability of the HMMs are discussed.For the second method,by constructing a specific state transition structure,a HMM is trained.The Viterbi algorithm is used to find the highest probability of hidden state sequence to identify the damage location and severity.Comparison of the two methods shows that the model library method is more applicable in damage identification of large structures.A Continuous Hidden Markov Model(CHMM)is further investigated for three-level structural damage identification.The requirement that observation values in DHMM have to be integers is removed in the CHMM,which is proved to improve the performance of damage identification.To improve the performance of CHMM for damage localization and severity identification,a CHMM-SVM method is presented by introducing the Support Vector Machine algorithm.Combination of the strength of SVM in generalization and the strength of HMM in dynamic signal processing leads to the improvement of damage identification.Validation of the methods presented in this paper is made with the test data of a laboratory truss bridge model and the field measurement data from the Xi Houmen bridge.It shows that HMM based damage identification method is promising in real applications.