Machine learning based flow regime recognition in helically coiled tubes using X-ray radiography
David Breitenmoser, Horst-Michael Prasser, Annalisa Manera, Victor Petrov
Abstract
The understanding of flow regimes in helical coils is essential in the design and operation of modern heat exchangers and has been studied therefore intensively over the past decades. However, due to the subjective nature in defining these flow regimes, previous studies showed conflicting results. The scope of the present study is to perform automatic flow regime recognition for helical two-phase flows by a novel two-step machine learning approach without any prior knowledge. The machine learning algorithms are trained on X-ray radiography based high-resolution high-speed void fraction measurements at adiabatic conditions. The trained machine learning models successfully predict the number of flow regimes as well as the flow regimes themselves. The classifier models show excellent classification accuracy >98%. Comparison of the experimental results with available flow regime models revealed significant discrepancies >2 orders of magnitude in the superficial liquid velocity. The novel machine learning based flow regime recognition methodology presented herein shows not only a significant improvement in the classification accuracy compared to previous studies but offers also a novel data-driven way to define and investigate the very nature of the flow regimes themselves in a more consistent and objective way.