EI、Scopus 收录
中文核心期刊
肖盛鹏, 朱宏博, 周岱, 包艳. 基于机器学习的弯管固液两相流流动特性研究. 力学学报, 2024, 56(3): 273-285. DOI: 10.6052/0459-1879-23-356
引用本文: 肖盛鹏, 朱宏博, 周岱, 包艳. 基于机器学习的弯管固液两相流流动特性研究. 力学学报, 2024, 56(3): 273-285. DOI: 10.6052/0459-1879-23-356
Xiao Shengpeng, Zhu Hongbo, Zhou Dai, Bao Yan. Study on flow characteristics of solid-liquid two-phase flow in bend based on machine learning. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(3): 273-285. DOI: 10.6052/0459-1879-23-356
Citation: Xiao Shengpeng, Zhu Hongbo, Zhou Dai, Bao Yan. Study on flow characteristics of solid-liquid two-phase flow in bend based on machine learning. Chinese Journal of Theoretical and Applied Mechanics, 2024, 56(3): 273-285. DOI: 10.6052/0459-1879-23-356

基于机器学习的弯管固液两相流流动特性研究

STUDY ON FLOW CHARACTERISTICS OF SOLID-LIQUID TWO-PHASE FLOW IN BEND BASED ON MACHINE LEARNING

  • 摘要: 管道水力输送是工业中常见运输方式, 具有运输距离长、安全性高、运营和维护成本低、环境友好和布置灵活的优点, 当前关于管道系统内水平、倾斜和垂直管道的两相流流动特征研究较多, 而关于系统内弯曲段管道研究较少, 亟需明确该段的两相流流动机理及辨明该段的磨损机制, 并作出准确预测. 本研究首先采用欧拉−拉格朗日框架下的CFD-DEM耦合方法, 针对弯曲管道压降和磨损率, 探究了弯曲角度、弯曲半径、输入速度、颗粒直径和颗粒浓度等5个因素的影响; 并基于上述5个变量, 通过Pairwise配对法进行工况组合并进行数值模拟计算, 得到数百个可用数据; 基于此数据集, 开发了6个机器学习模型进行训练, 比较了各自模型的准确率并得到各特征对于预测结果的相对重要性. 结果表明, 弯管的压降随输入速度、颗粒浓度、颗粒直径和弯曲角度增大而增大, 与弯曲半径关系较小; 磨损率随输入速度、颗粒浓度、颗粒直径和弯曲半径的增大而增大, 随弯曲角度的增大, 在90°前先略有下降, 在90°后增大. 最佳的机器学习模型对压降和磨损率的预测准确率评估指标R2(越接近1越准确)分别在0.96和0.99左右, 具有较好的预测能力, 可用于多参数影响下的弯曲管道固液两相流水力压降及管壁磨损率的预测, 且计算发现输入速度和颗粒浓度分别是对压降和磨损率预测的影响程度最大的因素.

     

    Abstract: Pipeline hydraulic conveying is a common mode of transportation in industry, which has the advantages of long transportation distance, high safety, low operation and maintenance costs, environmental friendliness and flexible layout. At present, there are many researches on the flow characteristics of two-phase flow in horizontal, inclined and vertical pipelines in the pipeline system, while there are few researches on the bend in the system. It is urgent to clarify the two-phase flow mechanism of this section, identify the erosion mechanism of this section, and make an accurate prediction. In this study, firstly, the CFD-DEM coupling method under the Euler-Lagrange framework was used to investigate the influence of five factors including bending angle, bending radius, input velocity, particle diameter and particle concentration, on the pressure drop and erosion rate of bend. Based on the above five variables, the Pairwise method was used to carry out the combination of cases and numerical simulation calculation was done, and hundreds of valid data were obtained. Based on this data set, six machine learning models were developed for training, the accuracy of each model was compared, and the relative importance of each feature for the prediction results was obtained. The results show that the pressure drop increases with the increase of input velocity, particle concentration, particle diameter and bending angle, and has little relationship with the bending radius; The erosion rate increases with the increase of input velocity, particle concentration, particle diameter and bending radius. With the increase of bending angle, it decreases slightly before 90° and increases after 90°. The prediction accuracy evalution index R2 (the closer to 1, the more accurate) of the best machine learning model for pressure drop and erosion rate are about 0.96 and 0.99, respectively, which have good prediction ability, and can be used to predict the pressure drop and erosion rate of solid-liquid two-phase flow in bend under the influence of multiple parameters. It is found that the input velocity and particle concentration are the most influential factors on the pressure drop and erosion rate prediction, respectively.

     

/

返回文章
返回