Maximizing vehicle uptime and reducing maintenance costs are critical objectives in modern automotive systems, making efficient resource utilization a top priority. One key factor is engine oil degradation, which directly affects engine performance, longevity, and overall vehicle efficiency. Current Conventional oil monitoring approaches are real time based on dedicated sensors—can be inefficient, costly, or poorly adapted to varying usage conditions.
While earlier studies have explored sensor-based or chemical analysis methods for oil health monitoring, these solutions often lack scalability and cost-effectiveness for widespread adoption.
This paper introduces a Machine Learning-based Engine Oil Degradation Detection System that operates non-intrusively using existing vehicle CAN signals, requiring no additional sensors. Key engine parameters such as RPM, Torque, Gear Position, Engine Power, Coolant Temperature, and Odometer readings are trained with tri-bological properties (kinematic viscosity, TAN, TBN, oil content) based on real-time load conditions and modeled using an XGBoost algorithm deployed on a surrogate ECU. A physics-based model has been integrated to augment the algorithm, improving its precision and ensuring greater authenticity in the results.
To deliver actionable outcomes, the system provides a user interface displaying Oil Deterioration Level (%) and Remaining Useful Life (RUL in KMs), along with alerts for timely replacement.
Validation under varied real-world driving scenarios demonstrates the system’s robustness and adaptability, offering a scalable solution for condition-based engine maintenance.