This paper presents an efficient neural net approach to turbulence prediction that has the potential to yield real-time prodiction hardware. Neural nets are inherently paralel algorithms and, as such, can be made to evaluate very quickly once trained for a specific task. A nonadaptive predictor must be designed to recognize and predict a broad range of flow conditions. However, an adaptive predictor can be simplere and potentially less sensitive to parameter variations since it evolves as flow conditions change. It will be shown that a neural net can be trained to predict turbulent flow and so probable hardware real-time predictor is feasible in the near future.
展开▼
