Machine learning application with Bayesian regularization for predicting pressure drop in R134a's annular evaporation and condensation

• Published in: Journal of the Brazilian Society of Mechanical Sciences and Engineering Vol. 47; no. 5
• Main Authors: Çolak, Andaç Batur, Bacak, Aykut, Karakoyun, Yakup, Koca, Aliihsan, Dalkilic, Ahmet Selim
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.05.2025; Springer Nature B.V
• Subjects: Artificial intelligence; Artificial neural networks; Bayesian analysis; Condensation; Data points; Deviation; Differential pressure; Engineering; Evaporation; Heat exchangers; Machine learning; Mechanical Engineering; Network analysis; Neural networks; Pressure drop; Regularization; Technical Paper; Vaporization; Evaporation; Bayesian regularization; Pressure drop; Condensation; Machine learning
• ISSN: 1678-5878; 1806-3691
• DOI: 10.1007/s40430-025-05527-8

The study of condensation and evaporation in plain pipes is a significant area of engineering and scientific inquiry, as it has relevance for enhancing and developing various industrial procedures. This study utilized a Bayesian artificial intelligence method to determine the pressure drop in a vertically oriented plain copper pipe during R134a's annular condensation and vaporization. The heat exchanger uses R134a and water in the tube and annulus sides, in turn. The tube has an inner diameter of 8 and a length of 500 mm. The training sets for artificial neural networks comprise R134a mass fluxes within the interval of 260–515 kg/m 2 s for in-tube condensation and 200–405 kg/m 2 s for evaporation. The obtained data on pressure drop during condensation and evaporation experiments were utilized in artificial neural network analysis using a differential pressure transducer in the test section. The study randomly divided 368 and 50 data points into training (85%) and testing (15%) sets for condensation and evaporation. The Bayesian method, primarily applied on this subject, effectively forecasts pressure drop during experimental condensation and evaporation. Regarding margin of deviation analyses, the models' condensation and evaporation conditions display variations of approximately  ± 7.5 and  ± 8.9%, respectively. The artificial neural network training technique was optimized to achieve minimal mean squared error values of 1.4211E-06 after 32 iterations for condensation and 1.3511E-06 after 40 iterations for evaporation. The predicted pressure drop data has a deviation of  ± 10% with the experimental one for both condensation and evaporation. The artificial neural network model produced R-values of 0.99903 and 0.98451 for condensation and evaporation, correspondingly.