Hybrid ANN approach for prediction of lubricant retention in compressor suction lines based on analytical model and parametric study
Hybrid ANN approach for prediction of lubricant retention in compressor suction lines based on analytical model and parametric study
复制标题
DOI：
10.1080/23744731.2022.2058843
复制
发表时间：
2022-03
影响因子：
1.9
通讯作者：
Weijie Zeng;Bo Gu;Zhiting Zhang;Jinting Hu;Yuxiong Sha
中科院分区：
文献类型：
--
作者：
Weijie Zeng;Bo Gu;Zhiting Zhang;Jinting Hu;Yuxiong Sha
本文提出了一种预测压缩机吸入管路中含油量的模型研究。提出了一种基于流型的双循环模型（DCM），用于评价吸油管的吸油量。建立了一个新的统一的保油数据库，以开发界面摩擦因子关联。采用力平衡分析方法进行参数研究，确定影响输油的因素。通过数据库验证了DCM和参数研究的正确性。最后，提出了一种混合人工神经网络（ANN）方法，该方法分别使用DCM和参数学习来产生外推数据和确定输入参数。采用对数S形（logsig）传递函数和Lavenberg-Marquardt（L-M）算法对6-13-1构型的混合ANN进行了优化。混合人工神经网络模型预测的平均相对误差（MRE）和R2分别为6.25%和96.86%。
A modeling study for prediction of oil retention in compressor suction lines is presented in this article. A new flow-pattern-based analytical model called the double-circle model (DCM) was developed to evaluate the oil retention amount in suction lines. A new consolidated oil retention database was established to develop interfacial friction factor correlation. Parametric study was conducted by means of force balance analysis to determine the influence factors of oil transport. The DCM and parametric study were validated by the database, and the results showed that they display reliable accuracy. Finally, a hybrid artificial neural network (ANN) approach was proposed, which used the DCM and parametric study to produce extrapolation data and determine the input parameters, respectively. The hybrid ANN was optimized for the 6-13-1 configuration with logarithmic sigmoid (logsig) transfer function and the Lavenberg–Marquardt (L-M) algorithm. The hybrid ANN model prediction yields a mean relative error (MRE) and R 2 of 6.25% and 96.86%, respectively.