1774 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 56, NO. 5, MAY2009 Predicting Future States With n-Dimensional Markov Chains for Fault Diagnosis Ian Morgan, Student Member, IEEE, and Honghai Liu, Senior Member, IEEE Abstract—This paper introduces a novel method of predicting future concentrations of elements in lubrication oil, for the aim of identifying possible anomalies in continued operation aboard a large marine vessel. The research carried out is supported by a discussion of previous work in the field of fault detection in tribo- logical mechanisms, although with a focus upon two stroke marine diesel engines. The approach taken implements an n-dimensional Markov chain model with a singular weighted connection between layers. The approach leverages the computational simplicity of the Markov chain and combines this with a weighted decision calcu- lated from the correlational coefficients between variables, with the notable assumption that interconnectivity between elements is not constant. The approach is compared to an established method, which is the Kalman filter, with promising results for future work and extension of the method to include expert knowledge in the decision making process. Index Terms—Engines, fault diagnosis, marine equipment. I. I NTRODUCTION M AINTENANCE is often thought of as a necessary evil rather than a competitive edge, partly because of slow adoption of new technology and, more recently, little enthusiasm for high cost retrofitting of diagnostic systems for large fleets. There are two broad categories that maintenance programs fall into: corrective or preventative maintenance [1]. Corrective maintenance, which can be thought of as reactive, only occurs when a machine effectively stops functioning. This is aimed at trimming maintenance costs as much as possible; however, in practice, it is the more expensive option as cost is measured in lost production or operational time, as well as the net sum of maintaining the mechanism [2]. This approach has also been discounted in the maritime industry, as it is not practical due to the high cost of repairing a marine engine while still at sea, coupled with the safety and environmental impact of a drifting vessel. As such, schedule or time-based maintenance is preferred, where a vessel has repairs carried out while in port, which is appropriate to the expected life span or mileage of a particular part [3]. Although this is the simplest approach, it does not account for a number of external factors which govern wear upon engine components. This is notable, Manuscript received April 9, 2008; revised November 12, 2008. First pub- lished January 6, 2009; current version published April 29, 2009. This work was supported by the Engineering and Physical Sciences Research Council, U.K., under Grants CASE/CNA/05/67, EP/G041377/1, and Open Automation and Control Group. The authors are with the Institute of Industrial Research, University of Portsmouth, Portsmouth PO1 3QL, U.K. (e-mail: ian.morgan@port.ac.uk; honghai.liu@port.ac.uk). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIE.2008.2011306 as there is evidence to suggest that large marine engines have unique wear rates and tendencies, even when compared to those of the same model [3]. As a result, although time-based maintenance is effective, an approach is required, which first does not require significant retrofitting costs by utilizing data from one source and secondly identifies possible issues with a particular engine, rather than an average over many sources. Therefore, this paper will focus upon another approach, that of predictive maintenance, which can be considered as fault- driven maintenance, without the drawback of having to wait for a fault to occur. The principles of predictive maintenance apply to all machine and industrial applications, as for example, in [4] and [5]; however, the domain of the marine industry will here be used to illustrate these due to the extreme conditions that have to be faced. A. Elemental Analysis Previous work in engine diagnostics has recorded indicators of a mechanism’s state from a number of sources, otherwise known as “sensor fusion” [6]. These are generally known under the general term of “tribology” or “the science of rubbing” [7] and can refer to the study of any mechanism that has moving parts. In this case, the specific factors relate to elemental analysis, as it is the degradation or lack of lubrication oil that generally results in failure. Edwards et al. [7] list three main causes: oxidation, where the oil loses hydrogen and, hence, its molecular properties are altered; fluid contamina- tion, where silicon or water is mixed with the oil; or solid contaminants, where oil is mixed with metallic fragmentation from mechanical wear, although importantly, all three increase the speed of degradation of a given mechanism. This can be subdivided into further indicators of wear; physical indicators include vibration [8], thermography, and acoustic emission [9], although it has been suggested by both Zhang et al. [10] and Liu et al. [11] that vibration detection is not suitable for large tribological mechanisms. Chemical indicators include ferrog- raphy, and compositional indicators from oil analysis include both spectrometry and viscosity [3], [11]–[17]. Ferrography has typically been the method of choice and involves pumping oil across a thin membrane or glass slide while applying a magnetic field, whereupon the size, shape, and number of collected particles can be identified. This has more recently been complemented with spectrometry, which measures particles’ characteristic electromagnetic wavelengths when introduced to a high energy source. One issue or benefit, depending upon the application, of spectrometry is that a max- imum particle size for accurate measurement is 5 μm, whereas 0278-0046/$25.00 © 2009 IEEE