The Influence of Optimization Algorithm on Suction Muffler Design
The refrigeration segment is becoming more and more important in new applications present on the medical, food and technology markets, requiring high energy efficiency standards both in commercial and household refrigeration system. This requires an improvement of the performance of the most important compressor components involving thermodynamics, mechanical and electrical engineering.
One of the most important components in the compressor is the suction muffler. The main suction muffler functions are the thermal insulation of gas from the evaporator and noise attenuation. The increase of muffler performance requires the modification of the length and diameter of tubes, the geometry and volume of chambers.This paper analyzes the performance of three different algorithms in a suction muffler optimization process. Using the same variables and objective functions for all algorithms, the performances of algorithms are evaluated to define the best strategy to optimize the suction muffler. The objective functions elected were cooling capacity and coefficient of performance as a minimization-maximization problem. The compressor model simulated uses R134a as refrigerant, has 6.75cm3 of displacement and achieves 262W of cooling capacity for -23.3°C / 40.5°C test condition. The muffler is composed by two chambers and two tubes. Eight geometric variables have been parameterized, being four of them the diameters and the lengths of the tubes.
DOE methods selected to create initial population for genetic algorithms were random and incremental space filler with fifty and ten individuals, respectively. Considering suction muffler optimization response and computational time an initial population of sixty individuals was deemed suitable. Twenty independent runs with different random seeds were performed inmodeFRONTIER with NSGA-II, MOGA-II and MOGT and the quality of obtained Pareto frontiers has been measured. It was possible to observe certain similarities between the performances of single algorithms, as well as differences (in particular, MOGT was strongly influenced by the initial population, whereas NSGA-II and MOGA-II were not). Furthermore, genetic algorithms have proved to be much faster: in fact, NSGA-II performed calculations 2,7 times faster per minute than MOGT.