Library Subscription: Guest
Begell Digital Portal Begell Digital Library eBooks Journals References & Proceedings Research Collections
Journal of Enhanced Heat Transfer
IF: 0.562 5-Year IF: 0.605 SJR: 0.211 SNIP: 0.361 CiteScore™: 0.33

ISSN Print: 1065-5131
ISSN Online: 1026-5511

Journal of Enhanced Heat Transfer

DOI: 10.1615/JEnhHeatTransf.v18.i2.30
pages 115-125

A COMPARISON OF ARTIFICIAL NEURAL NETWORKS AND SYMBOLIC-REGRESSION-BASED CORRELATIONS FOR OPTIMIZATION OF HELICALLY FINNED TUBES IN HEAT EXCHANGERS

Gregory J. Zdaniuk
Ramboll UK Limited, London W1T 3DA
D. Keith Walters
Department of Mechanical Engineering, Mississippi State University
Rogelio Luck
2Department of Mechanical Engineering, Mississippi State University
Louay M. Chamra
Department of Mechanical Engineering, Mississippi State University

ABSTRACT

Optimization studies were performed for helically finned tubes to investigate the performance of two correlation methods for use within the optimization procedure. Geometric parameters and operating conditions were determined based on maximization of the heat transfer enhancement efficiency index. The first correlation procedure used an artificial neural network (ANN) model found previously to yield highly accurate prediction of the measured friction factor and the Colburn j factor. The second used empirical correlations for the friction factor and j factor developed using symbolic regression based on a genetic programming technique. The optimization performed using the ANN was found to be invalid, producing physically unrealistic results in some regions of the parameter space and leading to a false optimum. The optimization using the symbolic regression correlations performed well and provided reasonable values for optimum geometric and operating parameters. The results showed that the highest efficiency index was obtained using a combination of parameters that promoted a swirling rather than a skimming flow. The optimization study highlights the fact that great care must be exercised when using ANNs for optimization and design purposes, particularly in cases for which the available experimental data are not evenly and densely distributed throughout the parameter space. In contrast, symbolic regression correlations using a relatively small number of additive terms appear to be well suited for optimization purposes due to their increased accuracy over conventional correlations and their robust predictive/generalization behavior relative to ANNs.


Articles with similar content:

SYNTHESIS OF MULTIPERIOD HEAT EXCHANGER NETWORKS WITH A HYBRID META-HEURISTIC APPROACH: SIMULATED ANNEALING AND ROCKET FIREWORKS OPTIMIZATION
ICHMT DIGITAL LIBRARY ONLINE, Vol.0, 2017, issue
Leandro Vitor Pavão, Caliane Bastos Borba Costa, Camila de Brito Miranda, Mauro Antonio da Silva Sá Ravagnani
Maximization of Entropy Flux as a design tool − Application to Heat Exchanger Networks Optimization
Second Thermal and Fluids Engineering Conference, Vol.24, 2017, issue
Ahmad Fakheri, Alexander S. Leanos
THERMALLY DEVELOPING CONCURRENT-FLOW CIRCULAR DOUBLE-PIPE HEAT EXCHANGER ANALYSIS
International Heat Transfer Conference 8, Vol.6, 1986, issue
M. N. Ozisik, Renato M. Cotta
METHODS TO DESIGN OPTIMUM HEAT SINK GEOMETRIES
International Heat Transfer Conference 16, Vol.14, 2018, issue
Kaj Lampio, Reijo Karvinen
LAGRANGIAN CHAOS AND HEAT TRANSFER
International Heat Transfer Conference 11, Vol.16, 1998, issue
Asen Mokrani, C. Castelain , Hassan Peerhossaini