1. ¹University of Salford, Salford, UK
2. ²Harbin Institute of Technology, Harbin, China
3. ³Universiti Teknikal Malaysia, Melaka, Malaysia

Correspondence: W Wang, Centre for OR and Applied Statistics, Salford Business School, Salford, Greater Manchester, M5 4WT, UK. E-mail: w.wang@salford.ac.uk

Received October 2007; Accepted April 2008; Published online 4 June 2008.

Abstract

This paper presents a model and methodology for estimating the residual time of a plant item. This plant item can be an engine or any complex technical system monitored by a regularly spaced oil analysis programme. Typically in the oil samples taken, two groups of observed variables are available, namely, metal concentrations and variables related to the condition of the lubricant and contaminants. We term the former as internal variables and the latter as external variables. External variables are those that may cause the change of the underlying condition of the plant item and therefore the residual time, while internal variables are those variables that only reflect the residual time but cannot change it. We modelled both variables in an oil-based monitoring case, but the principle can be generalized to other monitoring situations. The main techniques used are stochastic filtering for residual time prediction and the maximum likelihood method for parameters estimation. The model established was fitted to the real data of marine diesel engines monitored by an oil analysis programme and the results are presented.