Physics-Informed Machine Learning model for Emission Prediction in Exhaust Aftertreatment Systems: Prediction of NOX, CO, HC and NH3 Emissions for Volvo Penta Off-Road Diesel Engines

Abstract

Modern industrial diesel engines rely on advanced exhaust aftertreatment systems (EATS) to comply with increasingly stringent emission regulations such as EU Stage V, reducing emissions of nitrogen oxides (NOx), hydrocarbons (HC), carbon monox ide (CO), and ammonia slip (NH3). Among these systems, selective catalytic re duction (SCR) plays a central role in NOx reduction through urea-based ammonia dosing. However, SCR development, calibration, and catalyst sizing still depend heavily on expensive and time-consuming physical engine and rig testing. This cre ates a strong industrial need for reliable predictive models that can complement physical testing through early-stage virtual evaluation.

This thesis investigates the use of physics-informed feature engineering combined with machine learning for predicting system-out emissions in Volvo Penta off-road diesel engine platforms. The objective is to develop a supervised regression frame work capable of predicting system-out emissions based on engine-out conditions and exhaust aftertreatment parameters, while improving robustness and physical inter pretability compared to purely data-driven approaches. Particular focus is placed on NOx and NH3 prediction, since these emissions are most strongly linked to SCR behaviour and catalyst dynamics.

The study uses existing experimental test data from Volvo Penta engines in the D5 D13 platform range, representing industrial and off-road applications. Separate XG Boost regression models were developed for each target variable using both directly measured signals and derived physics-informed features related to SCR behaviour, catalyst thermal conditions, flow dynamics, and ammonia availability. Model per formance was evaluated using ShuffleSplit cross-validation together with validation on unseen datasets, primarily using the coefficient of determination (R2) and Mean Absolute Error (MAE).

The results show that physics-informed feature engineering improved both model ac curacy and robustness compared to baseline models using only directly measured in puts. The strongest improvements were observed for NOx prediction, where physics informed features improved model accuracy and robustness across different operating conditions and engine platforms. The final models also achieved strong predictive performance for HC and CO, while NH3 prediction remained more challenging due to the complexity of ammonia storage and slip behavior. Feature importance analy sis further confirmed that the learned relationships were consistent with known SCR physics, improving confidence in model interpretability and engineering relevance. The developed framework demonstrates that physics-informed feature engineering combined with machine learning can provide a practical and computationally effi cient approach for EATS emission prediction. The results also show clear potential for supporting early-stage virtual evaluation, SCR catalyst sizing, and aftertreat ment system optimisation within industrial engine development.