Data-driven estimation of printed circuit heat exchanger internal geometric data via hybrid genetic algorithm and machine learning method

Printed Circuit Heat Exchangers (PCHEs) are used in advanced thermal systems, including supercritical CO₂ (s-CO₂) cycles for nuclear reactors, waste heat recovery units, and concentrated solar power (CSP) plants due to their superiority on compactness and effectiveness. However, the internal geometric configurations of commercially available PCHEs are often confidential, limiting accurate modeling, simulation, and control model development in experimental test rigs and operational facilities. To address this issue, this study proposes a reverse engineering methodology to estimate the unknown internal configuration of an existing PCHE from experimental data by using a penalty function-based genetic algorithm (GA) coupled with K-Means clustering. An indigenous steady-state discretized logarithmic mean temperature difference (LMTD) heat exchanger code NODEX which can calculate the performance of a variable heat sink arrangement of s-CO₂ heat exchanger for any given set of boundary conditions was used in GA simulations. Unlike conventional optimization studies aiming efficiency, size, or cost, the proposed methodology minimizes the discrepancy between experimentally measured and numerically calculated data by modifying the geometric data. The validated GA analysis resulted in 102 possible PCHE configurations. The K-Means clustering grouped theses into 5 clusters. The cluster with the highest number of configurations and the lowest penalty function was selected as the final configuration. The results showed that the discrepancies in key variables such as channel diameter and the number of hot and cold channels were found to be <1%, demonstrating a high level of accuracy of the proposed methodology.