Hi,
I'm modeling fluid flow through a microfluidic bifurcating network. Among other things, I am calculating the volumetric flow rate for a given applied pressure and comparing it to experiment. The problem I am having is that when I am using a Non-newtonian power law fluid the solution will not converge. I have run the same model/mesh/solver with a newtonian fluid (glycerol) and the solution converges. I have tested smaller models (1 inlet; 2 or 4 outlets) with the power law fluid and the solution converges, but when I do the full scale (64 outlets), it will not converge.
Boundary conditions
This model has one inlet. I set it to either Pressure (no viscous stress) or Laminar inflow (pressure). There are 64 outlets. They are all set to Pressure (no viscous stress) of 0. The walls are no slip. I mostly keep both inlet and outlets at Pressure (nvs).
Solver
The default solver is an iterative solver with multigrid. I got a warning that I needed to increase the "factor in error estimate." So I changed it from the default value of 20 to 40, but this had no effect. I also got another warning that there was an "ill conditioned pre-conditioner." I'm vaguely aware of what this means in a linear-algebra context, but I don't know how I can fix this in COMSOL.
Under Study > Solver config > Solver 1 > Stationary Solver > Fully coupled, I changed the "Nonlinear method" from the defuault of "Linear (Netwon)" to "Highly non-linear," but the solution still did not converge.
I noticed that in the model library there is a model that uses a non-newtonian Carreau model fluid.
www.comsol.com/showroom/gallery/171/
In this model, the solver is a Direct solver - PARDISO. I tried this solver on my model, and after ~8 hours the solver appeared to make little progress.
Mesh
The mesh I'm using is a free tetrahedral mesh. I vary the mesh fineness by a factor I call "C", where the maximum element size is (the width of the channel)/C. In this model I uploaded, C=5 (coarse), but I typically have C=7 or 8 (finer). I'm wondering if my meshing is part of the problem. The mesh elements are all about the same size—they are not finer closer to the walls of the channels. Though this may be a problem for the non-newtonian fluid, it seems to be OK when I am using glycerol as the fluid.
I'm also wondering if I should do some sort of swept mesh to create a hexahedral mesh since my geometry is composed entirely of rectangular prisms.
I'm hoping to gain a better intuition of what's going on and how to work this out.
I would be very grateful for any tips/insights into this problem, thanks!
I'm modeling fluid flow through a microfluidic bifurcating network. Among other things, I am calculating the volumetric flow rate for a given applied pressure and comparing it to experiment. The problem I am having is that when I am using a Non-newtonian power law fluid the solution will not converge. I have run the same model/mesh/solver with a newtonian fluid (glycerol) and the solution converges. I have tested smaller models (1 inlet; 2 or 4 outlets) with the power law fluid and the solution converges, but when I do the full scale (64 outlets), it will not converge.
Boundary conditions
This model has one inlet. I set it to either Pressure (no viscous stress) or Laminar inflow (pressure). There are 64 outlets. They are all set to Pressure (no viscous stress) of 0. The walls are no slip. I mostly keep both inlet and outlets at Pressure (nvs).
Solver
The default solver is an iterative solver with multigrid. I got a warning that I needed to increase the "factor in error estimate." So I changed it from the default value of 20 to 40, but this had no effect. I also got another warning that there was an "ill conditioned pre-conditioner." I'm vaguely aware of what this means in a linear-algebra context, but I don't know how I can fix this in COMSOL.
Under Study > Solver config > Solver 1 > Stationary Solver > Fully coupled, I changed the "Nonlinear method" from the defuault of "Linear (Netwon)" to "Highly non-linear," but the solution still did not converge.
I noticed that in the model library there is a model that uses a non-newtonian Carreau model fluid.
www.comsol.com/showroom/gallery/171/
In this model, the solver is a Direct solver - PARDISO. I tried this solver on my model, and after ~8 hours the solver appeared to make little progress.
Mesh
The mesh I'm using is a free tetrahedral mesh. I vary the mesh fineness by a factor I call "C", where the maximum element size is (the width of the channel)/C. In this model I uploaded, C=5 (coarse), but I typically have C=7 or 8 (finer). I'm wondering if my meshing is part of the problem. The mesh elements are all about the same size—they are not finer closer to the walls of the channels. Though this may be a problem for the non-newtonian fluid, it seems to be OK when I am using glycerol as the fluid.
I'm also wondering if I should do some sort of swept mesh to create a hexahedral mesh since my geometry is composed entirely of rectangular prisms.
I'm hoping to gain a better intuition of what's going on and how to work this out.
I would be very grateful for any tips/insights into this problem, thanks!