“Application and evaluation of multi-sensor fusion technology in diesel engine oil analysis” – A review

Engine oil analysis is becoming one of the core subjects for Condition monitoring of the Diesel engines. During engine operation, Lubricating oils are used to lubricate the engine, and hence they serve as the lifeblood of the Diesel engines. During lubrication, a considerable amount of worn metal particles from the engine's moving parts moves and settle in the oil and contaminate it. Therefore, a lubricating oil flowing cyclically in a Diesel engine carries much information about the working condition of the engine parts. The contaminants in the oil indicate the health of Diesel engines. Oil properties like corrosion, oxidisability and pollution content can be diagnosed by different physiochemical parameters such as total acid number (TAN), viscidity, total base number (TBN), water content, etc. Condition monitoring of Diesel engines requires the use of various sensors and instruments. Each sensor and instrument generate some output data during the engine operation. The combination of sensors' data helps in an accurate understanding of the engine parameters. Multi-sensor fusion technology works by combining sensors' data. It is used to improve the accuracy of the sensors' information and the fault tolerance ability of the diesel engine sensors. Sensor fusion is generally performed using statistical methods like classical interference, Bayesian inference, and others, which are not accurate for dynamic systems. Statistical methods are not used in diesel engines because of the engine's dynamic behaviour (the engine parameters change suddenly). Therefore, In this study, an Artificial Neural Network (ANN) was deployed. Automated decisions can contribute to help making right decision and at right time for predicting condition of the engine oil. This research describes the oil characteristics extracted from experimental test results, which is used to create ANN model. 60 samples of engine oil of different vehicles working in similar condition are tested using standard methods. The engine oil is monitored in seventeen parameters and company expert decides when the oil needs to be changed. ANN modelling is done with different hidden layers and the optimal ANN model is chosen which shows good performance, predicting decisions similar to the company expert decisions.