Engine and Oil Condition Analysis Using a Hybrid Supervised Model With a Multilayer Neural Network and Expert Rules

This article relies on using machine learning algorithms for improving maintenance in military vehicles by means of a condition-based approach, allowing more precise maintenance task scheduling. The engine was chosen among several vehicle systems for its faults, effects, and criticality. We adopt a novel approach, using real-time captured operational engine data for condition-based and prognostic methods without additional sensors. Unlike other work that transmits data externally or performs oil chemical analysis, this research leverages data and expert rules for simultaneous analysis of oil use and engine health. The methodology involves data processing techniques, model training, and model testing with operational data. Operational data are used to predict friction power based on several engine parameters. The model combines expert lubrication rules with machine learning algorithms. These rules serve as the foundation for unknown friction power prediction using a multilayer perceptron, among other algorithms. Consequently, the model provides a crucial indicator of engine wear based on friction power. Unlike other models, the proposed model eliminates the need for incorporating viscosity sensors, thus avoiding the complexity of installing such sensors in vehicles. A performance analysis was conducted to identify the most efficient algorithm for developing a lightweight model suitable for installation on an edge computer. Empirical validation revealed a correlation between the cumulative friction power and metal concentration in the oil samples.