Boundary condition-based machine learning algorithm for fast prediction of supersonic flows

A new neural network-based machine learning algorithm, boundary condition-based machine learning algorithm (BCML), for predicting the flow field around an arbitrarily shaped bluff body at supersonic conditions is reported in the present work. The training data for the physics-guided neural network-based BCML model consists of flow field data generated using an in-house Riemann flux solver-based finite volume method code for a wide range of supersonic gas flow conditions. The new model predicts properties by employing the effect of boundary conditions surrounding the point in the domain. It was found that the flow field prediction by the BCML model was in reasonable agreement with those predicted by the traditional finite volume method solver for a fraction of the computational cost. Further, the BCML model could accurately capture shock waves (shock stand-off distances and strength) and other related wave features for supersonic flow over bluff bodies. The potential of using the BCML model to accelerate computational fluid dynamics simulations and reduce the overall computational cost is also highlighted in the present work.