Engine Warm-up Prediction using Machine Learning Methodology.

In the recent times, when emission & CO2 legislations are becoming more stringent, we need to adjust our simulation methods to the engine warm-up for better prediction. A typical fuel economy simulation for a WLTP drive cycle with and without considering warm-up effect differentiates up to 6-8%. Thus, there is a need to consider engine warm-up parameters for drive cycle simulations. The engine warm-up can be predicted in multiple ways but the most common way is detailed Physics based 1D modelling. These thermodynamic based models have a very good accuracy in temperature characteristics compared to actual test but take a lot of simulation time. The current requirement is to have models, which can do the same with 0.3 times real time so that these can be coupled with SiL simulation. The current paper will portrait a machine learning based methodology to predict Engine Oil & Coolant temperature based on driving profile and fuel consumption. ML model is trained by using data from one drive cycle and predicts engine oil and coolant temperature for other drive cycles. This ML model is developed in MATLAB and can be replicated in any other tool. The paper takes us through the process of variable selection with help of correlation matrix. The methodology is able to predict engine warm-up characteristics close to the thermodynamic model prediction with close to 0.3 times the real time. The procedure of this methodology can be applied to many other components apart from engine. The novel approach will talk about merits & demerits of different ML algorithms. The paper also take us through the validation study of the ML model done with the measurement data for different driving scenarios and temperature ranges.