Potential for using the ANN-FIS meta-model approach to assess levels of particulate contamination in oil used in mechanical systems

Abstract

Abstract Mechanical systems need to ensure high levels of quality. Today, greater generic reliability in systems makes it difficult to base any failure prognosis on previous system failures. Predicting the condition of a mechanical system needs to be based, instead, on monitoring the degradation of a system's components. Diagnostic signals can be identified and used as data to estimate the rate of degradation. A key driver for this work is the need to understand the performance of lubricants in systems involving mechanical contact. This article presents methods for studying field data collected with regard to oil. It focuses, in particular, on contaminated oil as this is an excellent source of diagnostic signals and information. However, data on oil present a degree of uncertainty in terms of both their collection and their use in the laboratory. Analysis of oil contaminants was, therefore, performed by applying a fuzzy inference system (FIS) and neural networks. The multilayer perception network was found to be an effective tool. The concentrations of iron and soot particles in used oil were selected as being both illustrative and the most significant model variables. The aim of this study is to acquire information about the condition of both lubricants and the mechanical systems, along with the development of degradation in mechanical equipment and the estimation of residual useful life (RUL). The results obtained will be useful in organizing effective operation of the mechanical systems being studied and modifying their maintenance.