基于数据驱动管道流体湍流模型的系数修正及流动特性分析

Flow Characteristic Analysis of Pipe Flow and Turbulence Model Coefficient Correction Based on Data-driven

为提高RANS湍流模型数值模拟的精度,基于人工神经网络对RANS中标准k-ε湍流模型系数进行预测与修正,并对受壁面射流扰动的管道流场进行分析。首先建立管道流的有限元模型,同时考虑壁面射流对管内流体流动状态的影响,对管道流的流动进行有限元仿真,得到管道流受扰动后的速度场分布。其次,搭建预测标准k-ε湍流模型系数的神经网络模型,预测描述管道内速度场变化趋势的标准k-ε湍流模型系数。最后,将新的系数带入有限元模型计算,并与实验数据对比,发现修正后的湍流模型对管道各部分速度场的模拟精度均有较大提升,对标准k-ε湍流模型系数的预测与修正可以提高模型对管道内速度场变化趋势的模拟精度。

To improve accuracy of the Reynolds-averaged Navier-Stokes(RANS)turbulence model simulation,relevant parameters in the standard k-εmodel control equation are predicted and modified based on artificial neural networks(ANN).An investigation is analyzed for how the fluid field is affected by the wall injection as part of accuracy improvement processes.After initializing the finite element(FE)model of the fluid field,a standard k-εturbulence model is used to perform relevant turbulent calculations.Meanwhile,circumstances of how wall injection affected the fluid field in the pipe are also added in the FE model.Transformations of the fluid are analyzed by the FE method.Thus,velocity field distribution after the wall injection is interpreted.Moreover,a neural network(NN)intelligent algorithm is established to predict standard k-εmodel parameters.The NN model′s input is considered as a ratio of the velocity component in the streamwise direction and the average velocity of the fluid in the pipe.Parameters in the turbulence model control equation are network′s outputs.Thus,the network predicts standard k-εmodel control equation parameters to describe the velocity field trend.The final step is retrieving the outcome parameters into the FEM calculation.Comparing with experiment data,simulation accuracy of the velocity field is significantly improved with the modified turbulence model.It shows that predicting and adjusting the standard k-εmodel control equation parameters improves simulation accuracy of the velocity field trends.