Numerical Assessment of Plenums Intersected by Four Baffles Using the Boundary Element Method,Genetic Algorithm,and Neural Networks

Abstract

There has been much research on partitioned plenums in the industrial field. However, maximum noise reduction of a plenum within a constrained space, which frequently occurs in engineering problems, has been neglected. Therefore, the optimum design of multi-chamber plenums becomes essential. In this paper, five-chamber plenums intersected by four baffles within a fixed space are assessed. In order to select the appropriate design parameter sets used in the shape optimization of a five-chamber plenum, three kinds of design parameter sets (Case I: L₁* and L₂*; Case II: L₁** and L₂**; Case III: L₁***, L₂***, and L₃***) are proposed.

In order to simplify the shape optimization of plenums intersected by multiple baffles, a simplified objective function (OBJ) is established by linking the boundary element model (BEM, developed using SYSNOISE) with a polynomial neural network fitted with a series of real data - input design data (baffle dimensions) and output data approximated by BEM data in advance. Before optimization is performed, accuracy of the boundary element method (BEM) for a one-chamber and three-chamber plenum is checked using analytical and experimental data and found to be accurate. To assess the optimal plenums, a genetic algorithm (GA) is adopted. Consequently, optimal results reveal that the depths of the two upper baffles and the two lower baffles play essential roles in minimizing the noise level of the lower frequencies (400∼800 Hz). Moreover, the horizontal span of the baffles will influence the acoustical performance of the higher frequencies (1200 Hz and above).