摘要详情

ID / 提交时间

27 / 2024-05-31 16:23:09

标题

Intelligent Prediction of the Lateral Velocity and Mixing Characteristics Downstream of the Spacer Grid

关键字

Spacer grid; mixing vane; GPR; backpropagation neural network

主题及专题

TRACK 5: NUCLEAR ENGINEERING

状态

全文录用

作者

quanbo li / Harbin Engineering University

摘要

In the traditional optimization process of the spacer gird with mixing wings, it is necessary to establish a 3D model and conduct CFD simulation . We also need to conduct experimental verification on the simulation model with good performance. The entire optimization process is time-consuming and labor-intensive. In order to optimize the design process and reduce the consumption of manpower, material resources, and computing power, this article introduces artificial intelligence methods to simplify the process of establishing 3D models and simulations in the optimization process. A database of mixing characteristics parameters under different working conditions and different mixing blade structure parameters was established through PIV experiments and CFD simulations. The artificial intelligence model was trained using the database to roughly judge the strength of the mixing characteristics under different working conditions and model parameters. By training the artificial intelligence model using the database and conducting kernel function cross-validation and hyperparameter optimization, conclusions were drawn. It was found that, using the Gaussian process regression algorithm(GPR) algorithm for prediction, the model had good prediction performance for the eddy factor with an determinant coefficients(R2 score)of 0.92, while its prediction performance for the turbulence factor was poor with an R2 score of only 0.31. When using the backpropagation neural network algorithm for prediction, the model had good prediction performance for both parameters after optimization, with a comprehensive R2 score of 0.81, which is acceptable. However, after optimizing each parameter separately, the R2 score for eddy factor prediction was 0.97, while the R2 score for turbulence intensity prediction was 0.93, which demonstrated excellent performance.