Prognostic modelling for residual useful life prediction

Xu, W 2012, Prognostic modelling for residual useful life prediction , PhD thesis, University of Salford.

[img] PDF
Restricted to Repository staff only until 31 July 2022.

Download (2MB) | Request a copy

Abstract

In the context of this thesis, prognosis aims at predicting the residual useful life of components using condition monitoring information. It enables a projection of component condition from the past and present into the future, providing significant assistance to maintenance decision making and asset management. An inaccurate or delayed prognosis might result in unexpected failure of critical assets, thus leading to enormous economic or casualty losses. In order to increase the accuracy and efficiency of prognosis, this thesis studies new approaches for prognostic modelling of residual useful life prediction using condition monitoring information. First, stochastic filtering models are applied for residual useful life prediction, and both failure and censored data are utilized for model parameterization. Then, three types of threshold based models are developed, namely an adaptive Brownian motion based model, an adaptive gamma based model and an adaptive inverse Gaussian based model. The degradation processes of these models are adapted to the history of monitored information, thus providing more realistic models and more accurate prognosis. In addition to these newly developed prognosis models, two developments, namely a threshold zone approach and a multiple failure modes approach, are also presented to complement existing models in order to accommodate more complex situations. Finally, a new proposal of model fusion is presented to combine physics of failure models and data driven models. This type of model fusion is a new trend for condition based prognostics, and possesses the advantages of both combined models. This thesis provides several new methodologies for prognosis modelling of residual useful life. Through comparisons with previously published models, we demonstrate that the proposed models perform reasonably well and generate more accurate predictions. However, more real data are required to evaluate further the prognosis capabilities of the improved models.

Item Type: Thesis (PhD)
Contributors: Wang, W (Supervisor) and Percy, DF (Supervisor)
Schools: Schools > Salford Business School
Depositing User: Institutional Repository
Date Deposited: 13 Aug 2021 13:00
Last Modified: 27 Aug 2021 21:57
URI: http://usir.salford.ac.uk/id/eprint/61523

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)

Downloads

Downloads per month over past year