Dynamic Automatic Transmission Oil Life Prediction Using Machine Learning

Conventional resource conservation for automotive traction and its utilization to maximize efficiency is not only important but also essential for climate change. Automatic transmission fluid (ATF) is one such component, which has enough potential considering the current methods of servicing. It also plays a crucial role in the performance and longevity of the transmission system. Predicting the actual life of the ATF can be challenging due to various factors such as its application, driving conditions, driving behavior, oil grade and maintenance schedules, which can help prevent costly repairs and improve the overall performance of the vehicle. Present work is focused on developing/validating a process, which can predict critical properties of oil at any given instance and give indication to the driver/fleet owner. Data is collected for various oil properties and vehicle parameters as vehicle's usage patterns such as shift density, vehicle load, torque, current gear, lock up state, input/output shaft speed, oil temperature etc. These data is acquired from a test vehicle over a period of time. The proposed approach involves data preprocessing, feature selection, model selection, model training and then evaluating the model. An ML model is built using data obtained from the test to classify oil quality and predict the remaining mileage. The prediction can be done at any given point of time and independent of vehicle operating conditions and vehicle mileage. While this model was deployed over raspberry pi for live computations (classification - ok/not ok and remainder mileage) to simulate end user real time monitoring. The future work will focus on real time testing of a heavy automatic transmission integrating both experimental and ML approaches to predict the life of transmission fluid coupled with dynamic scenarios and potential fluid failure modes for informed decisions about ATF replacement schedules and maintenance.