A Physics-Guided Neural Network Approach for EGR Cooler Temperature Estimation: Addressing ECU Model Discrepancies

<div class="section abstract"><div class="htmlview paragraph">As the automotive industry explores alternative powertrain options to curb emissions, it is pertinent to refine existing technologies to improve efficiency. The Exhaust Gas Recirculation (EGR) system is one of the pivotal components in emission control strategies for Internal Combustion Engines (ICE). The EGR cooler is crucial in thermal management strategies, as it lowers the temperature of recirculated exhaust gases before feeding it along with fresh air, thereby reducing nitrogen oxides (NOx) emissions. Precise estimation of the EGR cooler outlet temperature is crucial for effective emission control. However, conventional Engine Control Unit (ECU) models fall short, as they often show discrepancies when compared to real-world test data. These models rely on empirical relationships that struggle to capture precisely the transient effect, and real time variation in operating conditions.</div><div class="htmlview paragraph">To address these limitations and improve the accuracy of ECU based model, various signal processing techniques, such as noise reduction filters and bias correction within control logic were attempted. While these enhancements improve stability and consistency, they could not capture complex thermal interactions and real-time dynamic variations in EGR cooling.</div><div class="htmlview paragraph">In this study, a Physics-Informed Neural Network (PINN) approach is used to enhance the accuracy of EGR Cooler outlet temperature estimation. Unlike conventional deep learning models, which rely solely on data, PINNs incorporate fundamental thermodynamic and fluid dynamics principles into the learning process ensuring physically consistent and interpretable outputs. The proposed framework integrates the Transient Heat-Transfer equations in addition to the primary inputs. By embedding domain-specific knowledge into the neural network architecture, significantly reduces the data dependency and improves the reliability of EGR temperature estimation, supporting fault diagnostics, and making this approach highly suitable for real-time monitoring applications.</div></div>