Creep FEA Convergence & Numerical Stability

Nonlinear convergence at each step

At each time step, the creep FEA solves the nonlinear equilibrium equations iteratively. The Newton-Raphson method (or a variant) is typically used. The iteration proceeds until the residual forces are below a specified tolerance. The convergence may fail if the time step is too large (the creep strain increment is too large, causing a large stress change that the iteration cannot accommodate), if the creep law is very sensitive to stress (a high Norton exponent n amplifies the stress sensitivity), or if the material approaches tertiary creep (the accelerating creep rate may cause convergence difficulties). The convergence should be monitored at each step — the number of iterations, the residual force and the convergence rate. If convergence fails, the time step should be reduced, or the creep law parameters should be checked.

Time-step sensitivity

The results of the creep FEA should be insensitive to the time step size. If the results change significantly when the time step is halved, the step is too large and the results are inaccurate. The time-step sensitivity should be checked by a convergence study: run the analysis with the baseline time step, then with half and quarter, and compare the key results (creep strain, stress, displacement at the critical location). If the results converge, the baseline is adequate. If not, the step must be reduced. The time-step sensitivity is particularly important in the primary creep stage (where the creep rate changes rapidly) and in the stress redistribution phase (where the stress changes rapidly). The adaptive time stepping (if available) should be verified against a fixed-step solution.

• Time Stepping in Creep FEA

Constitutive model sensitivity

The results of the creep FEA are sensitive to the constitutive model and its constants. The Norton stress exponent n has a strong effect: a small change in n produces a large change in the creep rate (because the rate is proportional to sigma^n). The Arrhenius activation energy Q has a strong effect on the temperature dependence. The primary creep exponent m affects the early-time redistribution. The sensitivity to the constitutive constants should be assessed by varying each constant within its uncertainty range and observing the effect on the key results. The sensitivity analysis identifies which constants most affect the result and where the calibration effort should be focused. If the result is very sensitive to a constant that is poorly known, the uncertainty in the result is large and should be quantified.

• Sensitivity Analysis in Creep Assessment
• Creep Model Calibration

Mesh sensitivity

The mesh must be fine enough to capture the stress concentrations, the stress gradients and the creep strain localisation. In regions of stress concentration (notches, fillets, holes), the stress gradient is steep and a fine mesh is needed. In welds and heat-affected zones, the property mismatch creates local stress concentrations that require a fine mesh. In regions where creep damage localises (e.g. the HAZ of a weld), the mesh must be fine enough to resolve the damage gradient. The mesh sensitivity should be checked by refining the mesh in the critical region and comparing the results. If the peak stress or the creep strain at the critical location changes with mesh refinement, the mesh is too coarse. The mesh should be refined until the results converge. The mesh refinement should be localised to the critical region to manage the computational cost.

Output checks

• Equilibrium check: the residual forces at the end of each step should be below the tolerance — verify that the solution has converged
• Time-step independence: the key results should not change with smaller time steps — run a time-step convergence study
• Mesh independence: the key results at the critical location should not change with mesh refinement — run a mesh convergence study
• Stress redistribution: the stress at the high-stress regions should decrease over time (relaxation) and the stress at the low-stress regions should increase — verify that the redistribution is physical
• Creep strain monotonicity: the creep strain should increase monotonically with time (it should not decrease or oscillate) — oscillation may indicate a time-step or convergence problem
• Total strain consistency: the total strain should equal the sum of the elastic, thermal, plastic and creep components — verify the strain decomposition
• Energy balance: the work done by the loads should equal the strain energy plus the dissipated energy (creep dissipation) — a significant energy imbalance may indicate a numerical problem

Common convergence problems and remedies

Several common convergence problems arise in creep FEA. If the solution fails to converge at a step, the most common cause is a time step that is too large — reduce the step size. If the solution converges but the results oscillate (the stress or the creep strain alternates between steps), the time step or the convergence tolerance may be too large. If the solution converges slowly (many iterations per step), the creep law may be very sensitive (high n) or the problem may be nearly incompressible (large creep strain with small elastic strain). If the solution diverges (the residual grows), the creep law parameters may be incorrect or the material may be entering tertiary creep with a damage model that is not properly formulated. In each case, the remedy starts with reducing the time step and checking the constitutive model parameters.

• Finite Element Analysis of Creep
• Creep Model Verification

Separating equilibrium convergence from physical accuracy

A converged Newton iteration only shows that the discrete equations for that increment have been satisfied to the chosen tolerance. It does not establish that the constitutive integration, time resolution or mesh are accurate. Creep analysis therefore needs several separate convergence questions: are residual forces acceptably small, is the local constitutive update stable, are time increments sufficiently fine, and are the spatial stress and strain gradients adequately resolved? These checks should be evaluated independently. A model that converges easily with very large increments may still under-resolve rapid early relaxation, while a finely stepped model can still be wrong if a coarse mesh regularises a local concentration. Report convergence controls together with time-step and mesh sensitivity so that numerical robustness is not mistaken for physical verification.

Diagnosing cutbacks, stabilisation and difficult contact

Repeated increment cutbacks are useful diagnostic information. They often coincide with rapid stress redistribution, contact status changes, strong temperature dependence or a constitutive region outside the calibration range. Rather than simply increasing the iteration limit, identify where and when the residual originates and inspect the local state. Numerical stabilisation, artificial damping or softened contact can sometimes help a difficult solve, but these devices change the equations and must not be allowed to carry a significant share of the structural response. Compare stabilised and unstabilised energy or force measures where available, and reduce the numerical aid until the engineering outputs are insensitive to it. If convergence deteriorates because deformation or damage is genuinely accelerating, the difficulty may be physical rather than numerical; forcing the solver through that point can obscure the onset of collapse or rupture.

A creep FEA result is not trustworthy until it has been verified for convergence — nonlinear convergence at each step, time-step independence, mesh independence and physical output checks. Always perform these checks and document them. A result that has not been verified may contain significant numerical error.