Derivative free optimisation in engineering acoustics

Abstract

This paper describes recent work on the optimisation of very complex or computationally expensive systems such as those found in many engineering acoustics applications. It uses optimisation techniques that require no knowledge of the derivative of the objective function with respect to the input variables, and hence is suitable for application to problems where the derivative is potentially noisy or expensive to calculate by a finite difference approximation, difficult to calculate analytically, or simply unavailable as is the case in many commercial codes. The paper begins with a brief review of optimisation methods as applied to acoustic problems, and discusses the limitations of traditional techniques. The theory of two derivative free optimisation methods, a parallel genetic algorithm and a surrogate optimisation technique called Efficient Global Optimisation (EGO) are then described. Two example cases are then discussed: the optimisation of the position and design parameters of vibroacoustic absorbers mounted on the interior of a rocket payload bay to reduce the payload interior pressure fluctuations on launch; and the shape optimisation of an audio loudspeaker to improve sound quality. Finally future directions and challenges in this field are discussed.