Bayesian Framework for System Identification and Structural Health Monitoring:Theory,Algorithms and Applications
SI(system identification)methods may be used to update a parameterized model of a structure based on features extracted from its measured vibrations such as modal parameters.A common goal in applying SI to SHM(structural health monitoring)is to infer damage-induced substructure stiffness reductions by updating stiffness parameters in a structural model.However,since no structural model is an exact representation of a structures behavior,(ⅰ)there are no true parameter values,and(ⅱ)an uncertain model prediction error will always exist.Further,parameter estimation often gives non-unique results,raising the issue of model identifiability.A Bayesian framework has been developed that addresses these difficulties.It views probability as a multi-valued conditional logic for quantifying plausible reasoning under uncertainty.In this framework,there is no need to postulate the existence of inherent randomness.Instead,the relative plausibility of each model in a parameterized model class is quantified by the posterior PDF(probability density function)for the model parameters coming from Bayes Theorem.Then model predictions that are robust to modeling uncertainty can be produced where the probabilistic predictions of all models in the model class are integrated weighted by their posterior probability.Many computational tools have been developed to sample or approximate the posterior PDF and to perform integrations over the parameter space.Another powerful feature of the framework comes from application of Bayes Theorem to a set of candidate model classes for a structure; the posterior probability of a model class is governed by a trade-off between its data-fit and how much information it extracts from the data during updating.Generally,the data-fits of model classes that are more “complex in the sense of having more adjustable parameters are penalized more,so the procedure is a Bayesian version of the Ockham razor that can automatically avoid over-fitting of the sensor data.This regularization is utilized in Sparse Bayesian learning(SBL),which can be exploited for robust Bayesian compressive sensing of SHM signals.We have also introduced SBL into Bayesian SI for improved damage detection and assessment by exploiting the knowledge that damage usually induces spatiallysparse substructure stiffness reductions.Illustrative examples will be given.
System Identification Structural Health Monitoring Bayesian Updating Bayesian Ockham Razor Sparse Bayesian Learning
James L.Beck Yong Huang
George W.Housner Professor,Emeritus,Division of Engineering and Applied Science,California Institute Associate Professor,School of Civil Engineering,Harbin Institute of Technology,Harbin,China
国际会议
The 7th World Conference on Structural Control and Monitoring(7WCSCM)(第七届结构控制与监测世界大会)
青岛
英文
1-20
2018-07-22（万方平台首次上网日期，不代表论文的发表时间）