TY - GEN
T1 - Optimizing the swiss-roll recuperator of an innovative micro gas turbine by a surrogate neural network and the multi-objective direct algorithm
AU - Chang, Yau Zen
AU - Hung, Kau Tin
AU - Shih, Hsin Yi
PY - 2008
Y1 - 2008
N2 - Micro-turbines are promising high power-density engines for distributed generation. In this paper, an optimization process is proposed to design a Swiss-roll type recuperator used to recover the exhaust heat of a micro gas turbine. The recuperator is a counter-flow spiral plate heat exchanger, composed of two flat plates wrapped around each other. There are several design parameters to be optimized, including the number of turns, channel width, plate thickness, and mass flow rate. The complex interconnections of these design parameters make it difficult to analyze the process and select adequate parameter combination to build a recuperator with the highest effectiveness and lowest pressure drop. In order to reduce the number of numerical analysis in the optimization process, a neural network is employed as surrogate model, and a multi-objective DIRECT (Dividing RECTangle) algorithm, named as MO-DIRECT, is developed. After merely 5 iterations, with 3 representative sets selected from the Pareto front for convergence test during each iteration, we were able to find a min-max solution with prediction error lower than 4 %. Also, only 24 numerical simulations are required to achieve the results, and only 2,313 steps were conducted in the MO-DIRECT search, rather than 35,343 required in an exhaustive search.
AB - Micro-turbines are promising high power-density engines for distributed generation. In this paper, an optimization process is proposed to design a Swiss-roll type recuperator used to recover the exhaust heat of a micro gas turbine. The recuperator is a counter-flow spiral plate heat exchanger, composed of two flat plates wrapped around each other. There are several design parameters to be optimized, including the number of turns, channel width, plate thickness, and mass flow rate. The complex interconnections of these design parameters make it difficult to analyze the process and select adequate parameter combination to build a recuperator with the highest effectiveness and lowest pressure drop. In order to reduce the number of numerical analysis in the optimization process, a neural network is employed as surrogate model, and a multi-objective DIRECT (Dividing RECTangle) algorithm, named as MO-DIRECT, is developed. After merely 5 iterations, with 3 representative sets selected from the Pareto front for convergence test during each iteration, we were able to find a min-max solution with prediction error lower than 4 %. Also, only 24 numerical simulations are required to achieve the results, and only 2,313 steps were conducted in the MO-DIRECT search, rather than 35,343 required in an exhaustive search.
KW - Heat exchangers design
KW - Multi-objective optimization
KW - Recuperator
UR - http://www.scopus.com/inward/record.url?scp=69949148451&partnerID=8YFLogxK
U2 - 10.1115/GT2008-50762
DO - 10.1115/GT2008-50762
M3 - 会议稿件
AN - SCOPUS:69949148451
SN - 9780791843116
T3 - Proceedings of the ASME Turbo Expo
SP - 713
EP - 721
BT - 2008 Proceedings of the ASME Turbo Expo
T2 - 2008 ASME Turbo Expo
Y2 - 9 June 2008 through 13 June 2008
ER -