Machine Learning–Based Prediction of Underhood Airflow in Passenger Vehicles

Engine performance is affected by cooling airflow onto the engine cooling module. During initial design, frontal openings, grills, cooling module size, placement, and location are optimized to ensure sufficient airflow onto the cooling module. Currently, design concepts are validated using 3D computational fluid dynamics (CFD) simulations performed iteratively on full vehicle models to predict and optimize cooling airflow onto cooling modules. Each design concept iteration consumes significant time and resources. This study introduces a machine learning (ML) model to streamline underhood airflow prediction, reducing reliance on iterative CFD. Previous CFD simulation data is used to create a training dataset, which calibrates the ML model, describing underhood airflow as a function of input parameters. The relevant ML algorithm is used to calibrate the model, perform data fitting of the training values, after which a testing dataset is created to validate the model for a range of design parameters and vehicle conditions. Upon achieving the target testing accuracy (90% accuracy target in this particular case), the ML model is ready for implementation. The ML model is used to predict initial estimates of airflow and refine the design iterations, while CFD simulations are performed for the finalized concepts. This eliminates the need for expensive and lengthy design analysis iteration loops, effectively replacing them with a highly flexible model capable of predicting underhood airflow for even minor design changes quickly. Use of this model can decrease the time required per iteration by more than 90% compared to conventional CFD, thus enabling analysis of more designs in a given time frame.