Abstract | Microchannel heat sinks have attracted considerable attention in thermal management applications owing to their high heat transfer capabilities and compact size. Among the various cooling techniques, flow boiling in microchannels has emerged as a promising method for efficient heat dissipation. However, the intricate flow patterns in microchannels present challenges for accurate classification, pattern recognition, and inefficient data handling practices. This paper presented a comparative analysis of flow boiling classification techniques for pattern recognition in microchannel heat sinks. Three different clustering algorithm-driven convolutional neural networks (CNNs) were analysed and compared alongside a base CNN to establish a data pipeline capable of agile flow boiling pattern recognition. The Gaussian Mixture Model Clustering-based CNN exhibited the best performance, achieving an overall mean accuracy of 88\% for the test set validation. Thus, this study lays the groundwork for improving the performance of flow boiling pattern recognition in microchannel heat sinks. |
---|