A CFD Data-based Cavity Flow Surrogate Model with Machine Learning Algorithm

The aeroacoustic study of the cavity flow is closely related to the stealth technology of the modern military flight vehicle and has attracted intensive attention from both the research community and industry since the 1950s. Traditionally the cavity flows were studied with experimental or computational fluid mechanics techniques. However, due to the great progress in computational hardware capability and the development of artificial neural network algorithms, machine learning technology has widely been applied to various areas of fluid mechanics. In this paper, a CFD data-based cavity flow ML surrogate model has been proposed. The development of the model began with the extraction of the dataset from the CFD result, an RNN algorithm called LSTM was then applied to predict the pressure fluctuations in the cavity flow domain, transforming the prediction to the SPL spectra and comparing it with wind tunnel data. Three test points in the frequency domain were selected to assess the prediction accuracy of the LSTM algorithm by comparing the results with the CFD results of the same case. The results show that the present proposed surrogate model is highly efficient and reasonably accurate in predicting pressure fluctuations, thus is very suitable for engineering applications during the initial concept design stage, during which time is critical.