Condition-Based-Maintenance
This work proposes a joint-probabilistic model between the remaining life and inspection observations, which is then used to perform prognostics on currently installed assets. At every new observation, the forward-looking belief on the asset's remaining life is Bayesian updated, granting dynamic estimations on its failure probability. Consequently, inspection times are optimally determined, with increasing frequency as the component deteriorate. The trained model is evaluated using an independent inspection history of a bearing component, and the predicted prognostics are observed to converge towards the actual failure date of the asset.
