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

2026-26-0386

To be published on 01/16/2026

Authors Abstract
Content
This study presents a novel methodology for benchmarking Hydrogen Internal Combustion Engine (H2E) emissions against diesel vehicle configurations, emphasizing Real-Drive Emission (RDE) test procedures. By leveraging Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) models, emission profiles for legal cycles and RDE scenarios are predicted. Integrated data pipelines and physics-based modeling enable virtual evaluations of Selective Catalytic Reduction (SCR) system performance, ammonia dosing accuracy, and exhaust temperature dynamics. Key results demonstrate high prediction accuracy across models, 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 = 85%, dosing accuracy = 90%). This approach significantly reduces the need for extensive on-road driving tests, as the model performs most of the work, thereby lowering development costs and supporting OEMs in meeting stringent emission standards through efficient virtual testing of Exhaust Gas Treatment (EGT) systems.
Meta TagsDetails
Citation
S, M., Shah, J., Ratnaparkhi, A., and H, S., "Data-Driven Modeling for Hydrogen ICE Emission Prediction and ATS Performance Benchmarking," SAE Technical Paper 2026-26-0386, 2026, .
Additional Details
Publisher
Published
To be published on Jan 16, 2026
Product Code
2026-26-0386
Content Type
Technical Paper
Language
English