基于神经网络的卧螺离心机螺旋流场流速建模

Neural Network-Based Modeling of Flow Ve-locity in the Spiral Flow Field of a Decanter Centrifuge

摘要: 提出了一种基于深度学习方法的卧螺离心机螺旋流场流速预测模型。对于流体力学问题的数值模拟主要依赖于求解离散化的控制方程，深度学习由于其处理强非线性和高维性的能力，在解决流体问题方面显示出了新的前景。然而，现成的神经网络架构大多用于简单流场，对于稍复杂些的流场未得到充分应用。首先，使用一个结构化的深度神经网络，将流场控制的偏微分方程(即Navier-Stokes方程)纳入DNN的损失来驱动训练；然后，根据螺旋流场特征参数施加边界条件建立流速预测模型；最后，利用所构建的模型预测卧螺离心机螺旋流场流速分布并与数值模拟结果做对比实验。结果表明：通过对流场不同压力、角速度下的流速进行预测，构建的模型与数值模拟结果具有很好的一致性。

Abstract: A flow rate prediction model for the spiral flow field of a decanter centrifuge based on a deep learn-ing approach is proposed. Numerical simulation of fluid dynamics problems mainly relies on solving discretized governing equations, and deep learning shows new promise in solving fluid problems due to its ability to handle strong nonlinearities and high dimensionality. However, off- the-shelf neural network architectures are mostly used for simple flow fields and are underutilized for slightly more complex flow fields. First, a structured deep neural network with partial differential equations for flow field control (Navier-Stokes equations) is used to drive the training by incorpo-rating the losses of the DNN; then, a flow velocity prediction model is built by imposing boundary conditions based on the characteristic parameters of the spiral flow field; finally, the constructed model is used to predict the flow velocity distribution in the spiral flow field of a decanter centrifuge and experiments are conducted to compare the results with those of numerical simulations. The results show that the constructed model is in good agreement with the numerical simulation results by predicting the flow velocity under different pressures and angular velocities of the flow field.