A new programmed method for retrofitting heat exchanger networks using graph machine learning

Xin Chen,JiangBo Huang,Lin Zhang
DOI: https://doi.org/10.1016/j.applthermaleng.2024.122427
IF: 6.4
2024-04-01
Applied Thermal Engineering
Abstract:Unsupervised graph machine learning provides a powerful framework for modeling and analyzing heat exchanger networks (HENs). This paper proposes a graph-based thermally guided path search (TGPS) method that systematically identifies and evaluates retrofit options to enhance the thermal performance of HENs. The method represents the HEN as a bipartite graph and uses automated algorithms to search for feasible heat integration paths. Thermodynamically inconsistent paths are filtered out based on temperature feasibility rules. The resulting retrofit options are evaluated using graphical metrics like betweenness centrality and cluster coefficients, as well as thermal performance indicators. A re-routing technique is introduced to address temperature mismatch issues for serial heat exchanger connections. When applied to a Kalina power cycle, the thermal efficiency of the optimum configuration is increased by 9.7%. This method is compared with both pinch analysis and the Energy Transfer Diagram approach, and it is thoroughly tested and verified for an ammonia-water absorption refrigeration cycle as well. This methodology provides an effective tool for retrofitting existing HENs while adhering to thermodynamic principles.
energy & fuels,engineering, mechanical,thermodynamics,mechanics
What problem does this paper attempt to address?