Skip to content

02.convergence.md

Latest commit

 

History

History
148 lines (109 loc) · 4.79 KB

02.convergence.md

File metadata and controls

148 lines (109 loc) · 4.79 KB
Raw

Convergence Monitoring in Flow360

This document describes how to monitor and analyze convergence in Flow360 simulations through the graphical user interface.

Available Plots

Plot Type Description Purpose
Nonlinear Displays absolute or relative residuals for continuity, momentum, energy equations Primary convergence indicator
Linear Shows solver iteration convergence for each equation Solver performance analysis
CFL Displays Courant-Friedrichs-Lewy number evolution Stability monitoring
Minmax Shows minimum/maximum values of flow variables Solution bounds monitoring

Detailed Descriptions

Each convergence plot provides specific insights into the simulation:

Nonlinear

  • View Options: Absolute or Relative scaling
  • Variables tracked:
    • cont: Continuity equation
    • momx/y/z: Momentum equations
    • enrg: Energy equation
    • nuhat: Turbulence model - modified viscosity (SA)
    • k: Turbulence model - turbulence kinetic energy (SST)
    • omega: Turbulence model - specific dissipation rate (SST)
  • Logarithmic scale display

Linear

  • Variables tracked:
    • cont: Continuity equation
    • momx/y/z: Momentum equations
    • enrg: Energy equation
    • nuhat: Turbulence model - modified viscosity (SA)
    • k: Turbulence model - turbulence kinetic energy (SST)
    • omega: Turbulence model - specific dissipation rate (SST)
  • Logarithmic scale display

CFL

  • Variables tracked:
    • NavierStokes_cfl: Main flow equations
    • SpallartAllmaras_cfl: Turbulence (when applicable)
  • Linear scale display

Minmax

  • Variables tracked:
    • min density
    • min pressure
    • max velocity magnitude
    • min/max modified viscosity (SA)
    • min/max turbulence kinetic energy (SST)
    • min/max specific dissipation rate (SST)
  • Logarithmic scale display
  • Obtain more information in a tabular form by clicking on a point in the plot

Interactive Features

  • Toggle visibility of individual variables
  • Select time range using the bottom timeline
  • Export plots as images
  • Hover for detailed values

📊 Example Convergence Patterns

  1. Good Convergence:

    • Monotonic residual decrease
    • Smooth CFL ramping
    • Stable MinMax values
    • Clear asymptotic behavior

  2. Poor Convergence:

    • Oscillatory residuals
    • Erratic CFL behavior
    • Diverging MinMax values
    • Stalled residual reduction

💡 Tips for Convergence Analysis
  • Monitor nonlinear residuals dropping at least 3-4 orders of magnitude
  • Check for oscillatory behavior in residuals that might indicate instability
  • Verify CFL number stability and ramping behavior
  • Examine MinMax values for physical reasonableness
  • Use logarithmic scale for better visualization of residual drops

Advanced Monitoring Tips
  • Cross-reference force coefficients with residual convergence
  • Look for plateauing of residuals indicating steady-state
  • Check for correlation between CFL changes and convergence behavior
  • Monitor individual equation components for potential source terms

❓ Frequently Asked Questions

  • What indicates good convergence?

    A drop of 3-4 orders of magnitude in residuals, stable force coefficients, and physically reasonable MinMax values typically indicate good convergence.

  • Why are my residuals oscillating?

    Oscillations can indicate physical unsteadiness, numerical instability, or too aggressive CFL numbers. Try reducing CFL or switching to time-accurate simulation if physical unsteadiness is expected.

  • What should I do if convergence stalls?

    Check CFL numbers, examine boundary conditions, verify mesh quality, and consider solution initialization. Reducing CFL or implementing solution limiting might help.

  • How do I interpret the linear residuals plot?

    The linear residuals show solver performance within each physical timestep. Steeper slopes indicate better convergence rates, while flattening might suggest preconditioning issues.

🐍 Python Example Usage

Below is a Python code example showing how to access residuals:

import flow360 as fl

case = fl.Case(id="case-XXXXX") # provide a valid case id
case.wait() # wait for the case to finish running
results = case.results

# Non-linear residuals
nonlinear_residuals = results.nonlinear_residuals
print(nonlinear_residuals)

# CFL
cfl = results.cfl
print(cfl)