Research on Performance Evaluation Model of Diesel Engine SCR System Based on CFD Flow Field Characteristics and Machine Learning

In the new on-board SCR (Selective Catalytic Reduction) system, it is of great significance to realize the real-time prediction of system performance and early diagnosis of system faults through data driving. This research aims to develop a machine learning model that can analyze and integrate historical simulation and experimental data, explore correlations among the data, and enable rapid prediction of system performance. Based on the experimental test results and CFD flow field characteristics of the SCR system, the key indicators affecting system performance were analyzed. Based on the multimodal simulation data and experimental data, the machine learning model was trained to accurately predict the NOx conversion efficiency and accelerate the development process of the SCR system. Finally, the performance of the model was tested using the test set, and the average percentage error was below 1%, with all residual errors falling within the 3σ confidence interval.