Embedded Machine Learning Algorithm to Predict the Real-Time Temperature of an Automotive Component

The present work describes the different steps of the development of an embedded machine learning algorithm, which enables to predict the real temperature of an automotive component. This methodology is proposed as an alternative to classical modelling approaches based on simplified physical laws, most often used in engine controls. The present study describes the different steps from data pre-processing to tests on a vehicle, including the model design and the embedding. Two different neural networks are evaluated and prove to be very good and relevant alternatives to classical modelling.