Abstract
In a large, low-head pump system, the outlet conduit is typically so wide in width that a partition wall is required to ensure structural strength. To investigate the internal flow and energy consumption mechanisms of a siphon outlet conduit due to the partition wall and the inlet flow conditions, a data-driven machine learning method based on the numerical three-dimensional flow field of the pump system is introduced. Specifically, the method employs a supervised residual neural network to train on the inlet velocity distribution and the partition wall starting positions, subsequently, it uses the gradient-weighted class activation mapping method to evaluate the effect of the inlet cross section, and the flow field and entropy production in the outlet conduit are analyzed. The findings include: The interplay between circulation and secondary flow leads to imbalanced flow on the left and right passages of the outlet conduit, especially at high net heads. A partition wall with a proper starting point eliminates the flow imbalance and improves the flow patterns within the outlet conduit, thereby reducing hydraulic losses, and the pump system efficiency is enhanced by 2.82 percentage points compared to no-wall configurations. The flow pattern of different regions at the inlet cross section has varying degrees of influence on different parts of the outlet conduit. The effect of the outlet side of the inlet cross section on the hydraulic loss in the outlet conduit is weakened by the partition wall. This study provides guidance for analyzing the internal mechanisms in external characteristics and for optimization strategies.
Funding Information
  • National Natural Science Foundation of China (52179091)
  • Water Conservancy Science and Technology Project of Jiangsu Province (2024029)
  • Shandong Water Transfer Maintenance Center Project (SDGP370000000202302003985A 001)