Predicting novel failures in an asset fleet
Failure of industrial assets can have a massive impact on the industrial operations. And with the increasing servitisation of the manufacturing sector, the manufacturer bears the maximum costs.
An important class of failures is that of novel failures- failures which were not experienced before. We not only have limited data for such failures, but also lack a method to transfer/ amend its knowledge for remaining assets in the fleet. As a consequent safety measure, the manufacturer has to pull off remaining operating assets. However, using state-of-the art Machine Learning, and Collaborative Prognostics techniques manufacturers can quantify the differences between the assets. The differences can be used as a measure to modify the predictive models of the failed asset, and incorporate them into those of the operating assets to increase their reliability. The manufacturer can therefore conserve their resources. Concretely, this PhD focuses on transferring knowledge from failed assets to the assets which have not encountered that failure before. Maharshi tries to minimise the asset failures in the fleet before an accurate prediction can be made.
Siemens Turbo-machinery UK