ResearchInfrastructure Damage Classification and PrognosisClassifying damage types and determining the remaining useful life (RUL) form a higher level of demands for infrastructure sustainability. Bayes' theorem characterizes how the prior confidence is revised given the data observation. Adopting the Bayesian decision theory, damages are detected and classified in a supervised learning fashion. With sufficient number of iterations, the posterior probability of selecting the correct damage type converges to 1, and the wrong types to 0. Compare to the traditional likelihood approach, Bayesian framework is more sensitive and specific to the group classification.
As the need for condition-based maintenance scheduling, tracking and predicting the deterioration of infrastructures is necessary in order to prognose the defects. Bayesian framework is employed to fuse the damage indexes observation with a fatigue model. As an approximation of the Bayesian process, particle filter provides the most plausible outcome considering both physical model and data measurements, and forecasts the future condition features to predict the RUL.
|
