Novel Grey Model for Diesel Engine Oil Monitoring

Electronic sensors and data processing systems are increasingly integrated with mechanical systems for continuous condition monitoring and early warning of pending failure. Various predictive algorithms have been applied to sensor data trends. Traditional grey algorithms have been highly useful for data trend prediction but suffer from serious overshoot at inflection points when applied to fluctuating data curves, that is, curves that cannot be accurately represented by a single grey exponential equation. This paper therefore presents an improved grey model (GM), a two-section residual grey model (2S-RGM), and demonstrates 2S-RGM by applying it to data trends in a marine engine oil quality monitoring system. 2S-RGM constantly evaluates data trends and seeks inflection points at which the curve is broken into separate sections, each being treated by an independent GM. 2S-RGM is tested using two sets of published marine diesel engine oil data, one ferrographic and one spectroscopic. Predictive results of 2S-RGM are compared to results of the same data sets applied to six other well-known predictive methodologies. It is demonstrated that 2S-RGM is significantly more accurate, with the additional benefit of avoiding the setup costs of earlier artificial-intelligence-based GM improvements.