Data-Driven Modeling for Emission Prediction and ATS Performance Benchmarking for Hydrogen and Diesel ICE Applications

This study presents a comprehensive methodology for benchmarking hydrogen and diesel internal combustion Engines, with emphasis on virtual Real-Drive Emission (RDE) test procedures for diesel and hydrogen application. Emission profiles for legal cycles and RDE scenarios are accurately predicted through integration and development of Artificial Neural Networks (ANN) based on Long Short-Term Memory (LSTM) models. Virtual evaluations of Selective Catalytic Reduction (SCR) system performance, Diesel Exhaust Fluid (DEF) dosing accuracy, and exhaust temperature dynamics enabled by integrated data pipelines and physics-based modeling are also explored for holistic prediction of output. Across models, validation demonstrates good prediction accuracy including temperature (R2 > 0.94, RMS error < 25°C), air flow (92% accuracy, RMSE = 28 kg/h), upstream NOx (93% accuracy, RMSE < 10 mg/s), and SCR (TP NOx accuracy = 91.82%, dosing accuracy = 87.73%). This approach has the potential to offer significant reduction in the need of extensive on-road driving tests, as the model provides capability to emulate the same, thereby lowering development costs and supporting OEMs in meeting stringent emission standards through efficient benchmarking of Aftertreatment systems (ATS).