Structural Health Monitoring of Plates Without Reference Data: Accurate Crack Identification using a Model-Based Bayesian Framework
Date of Completion
Structural health monitoring has been an active area of research over the last half-century. During this time the dominant theme of academic research has been to avoid use of physical understanding of the structure in favor of pattern-recognition based approaches which require copious reference data. Countless incremental improvements have been published and then found wanting by industry. A radical new approach is needed if this logjam is to be broken. The method must be accurate, inexpensive, robust, and independent of reference data. Such a method is presented here. In this thesis a model-based Bayesian approach to crack identification in plates is described and demonstrated experimentally. First the sensitivity of the method to variation of the crack parameters and to sensor noise is examined. Then, because of the highly complex structure of the parameter space, the standard Markov chain Monte Carlo approach to solving Bayes theorem is extended by using a population-based technique that adds the power of an evolutionary algorithm to the Bayesian framework. In order to validate the technique experimentally, the free response of a cracked plate subjected to an impact is recorded by three resistive strain gages. The resulting time series is then used to estimate the crack parameters that characterize the damage by using a Bayesian method built upon an efficient finite-element model of the plate. The approach is demonstrated to be effective in identifying crack location, orientation and length. The Bayesian results are also compared to the results produced by a genetic algorithm. A benefit of using a Bayesian methodology, as compared to a frequentist method such as the genetic algorithm, is that rigorous confidence intervals can be assigned to the estimates. The results show that even with limited, noisy vibration data accurate information regarding the damage state can be successfully estimated. In addition, a novel method for selection gage locations and orientations and the location of excitation impact is explored and the benefit to estimation accrued by careful location is quantified. ^
Moore, Edward Zrenda, "Structural Health Monitoring of Plates Without Reference Data: Accurate Crack Identification using a Model-Based Bayesian Framework" (2011). Doctoral Dissertations. AAI3485432.