Abstract
Reduction of exhaust noise combined with the need to improve the design of exhaust systems (e.g. by reducing mass) requires better prediction tools. The existing computational technologies for exhaust noise analysis are based on either purely acoustic transfer calculation (e.g. by electrical analogy) or by non-linear one-dimensional compressible Euler equations. The first approach neglect all non linear effects including viscosity, turbulence, shock waves..., whereas the latter does not consider 3D aspects of acoustic behavior as well as viscosity and turbulence. None of them offer the possibility to predict internal aero-acoustic noise sources (blow noise), nor the coupling to the structure, which may directly affect the environment and the passenger comfort. It is only by bringing new perspectives in understanding the underlying phenomena of exhaust noise that innovation will be attained. The current work is an attempt to use Computational Fluid Dynamics (CFD) as well as Computational Aero-Acoutics (CAA) to solve as much as possible all physical phenomena involved in the exhaust system. One important aspect is to take into account local turbulent noise cause by macroscopic turbulent structures. Using Large Eddy Simulation (LES), which is becoming available for such industrial applications thanks to the increasing computer performances, can theoretically solve this. The current paper describes our attempt to validate this technique on a simple exhaust component. Identifying flow mechanisms creating aero-acoustic noise sources inside a real exhaust system is by nature challenging. Besides the experimental difficulties of flow visualization in such a system, the flow pulsation as well as the acoustic propagation and cavity resonances may interfere with the local flow unsteadiness and turbulence creating the blow noise. Fortunately, a non-pulsating flow may produce similar mechanisms, at least those creating annoying noise sources in the medium/high frequency range. An exhaust system is here experimentally tested with a forced flow imposed by a flow tester. Particle Velocity distribution upstream is measured through LDV technique and the aero-acoustic noises created by one specific component recorded. Additionally, CAA simulations are also performed for the component. They involve an explicit transient Finite Element solver, which takes into account: flow unsteadiness through a simplified compressible LES, acoustic wave propagation, impedance of fluid boundaries and noise transmission to the structure. A model with approximately 400 000 elements is built and run with the code until both flow and acoustic convergence are obtained. Numerical results are analyzed and compared to the experimental data. The noise spectra compare reasonably well showing that the method is able to predict those noises. Flow structures creating noise are visualized through classical animation techniques as well as through modal decomposition, allowing to identify for each relevant frequency the noise source patterns as well as the resonance. This methodology is targeted to be applied at early design stages to estimate quality of exhaust components in terms of aero-acoustic noise and to imagine solutions for improvements.