Development and Comparison of Virtual Sensors Constructed using AI Techniques to Estimate the Performances of IC Engines 2022-01-1064

Alternative propulsion systems such as renewable fuels and electric powertrains are expensive; thus, efficient internal combustion engines (ICE) with hybrid powertrains still play significant roles in the transportation fleet in the coming decades. Modern engine technologies have been adopted to meet stringent emissions and fuel economy standards. As a result, engine control systems are becoming more complex. Furthermore, as ICE control parameters increase exponentially, engine calibration and design become bottlenecks in the development process. While a map-based feed-forward control method is a current de facto standard in combustion control, online closed-loop feedback control can improve engine performance and robustness. However, adding physical sensors to measure the various data for the online feedback control and calibration increase the vehicle cost. Therefore, this research proposes a method to replace various physical sensors with virtual ones, developed using Artificial Intelligence (AI) techniques. Other advantages of the data-driven-based virtual sensors over the physical ones are their high robustness and accuracy. This work develops AI-based virtual sensors to estimate engine performance and emission parameters, including NO_x and CO₂, from in-cylinder pressure data. Multiple AI techniques in the Keras library, such as Artificial Neural Network (ANN), Extreme Gradient Boosting (XG-Boost), Random Forest (RF), Support Vector Machine (SVM), and Decision Tree Regression (DT), are used to construct the virtual sensors. The results confirm that the computation time for estimating the combustion parameters depend on the AI techniques. However, the computational load should be low for the feedback control using in-cylinder pressure data. Therefore, the data-driven techniques were compared to optimize the computational load and prediction accuracy. The results show that the XG-Boost outperforms other AI techniques regarding model accuracy and prediction time.