Assume we have two domains $\Omega_a$ a acoustic domain with boundary $\Gamma_a$ and $\Omega_s$ a domain of a solid body with boundary $\Gamma_s$. $\Omega_a$ and $\Omega_s$ have the common interface $\Gamma_{as}$. In $\Omega_s$ we have the equation of motion for a solid: \begin{equation} \rho_s \frac{\partial^2 u}{\partial t^2} - \nabla \cdot \sigma = 0 \end{equation} And in $\Omega_a$ the wave equation: \begin{equation} \frac{1}{c_a^2} \frac{\partial p^2}{\partial t^2} - \nabla \cdot \nabla p =0 \end{equation} where $u$ is the displacement of the solid and $p$ is the acoustic pressure. Further $\rho_s$ is the density in $\Omega_s$ , $\sigma$ the stress-tensor and $c_a$ the speed of sound in $\Omega_a$. The coupling at $\Gamma_{as}$ can be formulated by the following equations: \begin{align} -\frac{1}{\rho_a} \nabla p \cdot n = \frac{\partial^2 u}{\partial t^2} \cdot n\\ -pn = \sigma \cdot n \end{align} where $\rho_a$ denotes the density in $\Omega_a$ and $n$ the normal vector at $\Gamma_{as}$.

Now I want to derive a finite element formulation such that I can have a non-matching grid at $\Gamma_{as}$ and I want to do this by the Mortar method.

The first step is to derive the weak form of the two equations on their domains with the test functions $p'$ and $u'$:

\begin{align} \int_{\Omega_s} u' \cdot \rho_s \frac{\partial^2 u}{\partial t^2} dx + \frac{1}{2} \int_{\Omega_s} \nabla u' : C : (\nabla u + (\nabla u)^T ) dx -\int_{\Gamma_{as}} (u' \cdot \sigma) \cdot n \ ds =0 \\ \int_{\Omega_a} p' \frac{1}{c_a^2} \frac{\partial p^2}{\partial t^2} dx + \int_{\Omega_a} \nabla p' \cdot \nabla p dx - \int_{\Gamma_{as}} p' \nabla p \cdot n \ ds =0 \end{align}

Here I'm stuck. What I think I should do next is:

For the Mortar method one needs to define a lagrange multiplier as $t=-pn=\sigma\cdot n$. How to incooperate it into both equations?

The second interface condidtion is then added in the weak form with the test function $\mu'$: \begin{align} \int_{\Gamma_{as}} \mu'(\frac{1}{\rho_a} \nabla p \cdot n +\frac{\partial^2 u}{\partial t^2} \cdot n) ds = 0 \end{align} Is this correct?

Edit: Here are two papers about the subject that I found so far:

[1] Triebenbacher, S., Kaltenbacher, M., Wohlmuth, B., & Flemisch, B. (2010). Applications of the mortar finite element method in vibroacoustics and flow induced noise computations. Acta Acustica united with Acustica, 96(3), 536-553.

[2] Walsh, T., Reese, G., Dohrmann, C., & Rouse, J. (2009). Finite element methods for structural acoustics on mismatched meshes. Journal of Computational Acoustics, 17(03), 247-275.

They use the coupling conditions that I have given above. A pity is that they dont show how a mortar formulation is done for the interface between the structural and acoustic domain.


After some digging I could finally find a paper that more or less answered my question:

Bermúdez, A., Rodrıguez, R., & Santamarina, D. (2003). Finite element approximation of a displacement formulation for time-domain elastoacoustic vibrations. Journal of computational and applied mathematics, 152(1-2), 17-34.

It turns out that the equation for the acoustic pressure from above might not be suited for a Mortar coupling. Instead, we can alternatively describe the behaviour of the fluid in the domain $\Omega_a$ by a displacement based formulation. The governing equation are then:

\begin{equation} \rho_s \frac{\partial^2 u_s}{\partial t^2} - \nabla \cdot \sigma = 0 \quad \text{on } \Omega_s\\ \rho_a \frac{\partial^2 u_a}{\partial t^2} -\nabla(\rho_a c^2 \nabla \cdot u_a) =0 \quad \text{on } \Omega_a \end{equation} Here we have now the displacement-vector of the solid by $u_s$ and of the fluid in the acoustic domain by $u_a$. For this new formulation we get the coupling conditions on $\Gamma_{as}$: \begin{align} u_s\cdot n = u_a \cdot n\\ \sigma \cdot v = \rho_a c^2 \nabla \cdot u_a v \end{align} where $v$ is any vector at the interace $\Gamma_{as}$. This coupling conditions allow us to use the Mortar method as usual, which is to define a vectorial-lagrange multiplier as $tv:=\sigma \cdot v = \rho_a c^2 \nabla u_a v$ and then derive the weak formulation :

\begin{align} \int_{\Omega_s} u_s' \cdot \rho_s \frac{\partial^2 u_s}{\partial t^2} dx + \frac{1}{2} \int_{\Omega_s} \nabla u' : C : (\nabla u_s + (\nabla u_s)^T ) dx +\int_{\Gamma_{as}} (u_s' \cdot \sigma) \cdot n \ ds =0 \\ \int_{\Omega_a} u_a' \cdot \rho_a \frac{\partial^2 u_a}{\partial t^2} dx + \int_{\Omega_a} \rho_a c^2 \nabla \cdot u_a\nabla \cdot u_a' dx -\int_{\Gamma_{as}} \rho_a c^2 \nabla \cdot u_a u_a' \cdot n \ ds =0 \end{align} In these equations we can now easily insert the Lagrange multiplier $t$ and enforce the other coupling condition in a weak sense by a third equation. The final weak formulation is then:

\begin{align} \int_{\Omega_s} u_s' \cdot \rho_s \frac{\partial^2 u_s}{\partial t^2} dx + \frac{1}{2} \int_{\Omega_s} \nabla u' : C : (\nabla u_s + (\nabla u_s)^T ) dx +\int_{\Gamma_{as}} t u_s' \cdot n \ ds =0 \\ \int_{\Omega_a} u_a' \cdot \rho_a \frac{\partial^2 u_a}{\partial t^2} dx + \int_{\Omega_a} \rho_a c^2 \nabla \cdot u_a\nabla \cdot u_a' dx -\int_{\Gamma_{as}} t u_a' \cdot n \ ds =0 \\ \int_{\Gamma_{as}} \mu'(u_s - u_a )\cdot n ds = 0 \end{align} Note that $u_a'$ and $u_s'$ are the test functions for the two domains $\Omega_a$ and $\Omega_s$ and we have to choose their Sobolev space such that $u_a'\cdot v = u_s'\cdot v$ for an vector $v$ on $\Gamma_{as}$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.