Limit Analysis And Concrete Plasticity

Practical Guide to Limit Analysis of Reinforced Concrete Structures

Most engineers encounter limit analysis in graduate school and then never use it again until a project demands something more than what a standard FEA package spits out. The gap between textbook theory and a working model is wider than people expect. This is a walkthrough of how I actually set these up, not a literature review. Start by understanding what the method is actually doing. Lower bound theorem says if you can find a statically admissible stress field where equilibrium is satisfied everywhere and stresses never exceed the yield criterion, the corresponding load is a safe lower bound on the true collapse load. Upper bound theorem works in reverse. Pick a kinematically admissible mechanism, compute internal and external work rates, and the resulting load multiplier is an upper bound. The true collapse load sits between them. That's the entire framework. Everything after that is numerical implementation. I used to run into persistent problems with reinforcement modeling. Concrete is compressive and weak in tension. Steel carries all the tension. Most off-the-shelf limit analysis modules treat reinforced concrete as a homogeneous material with some adjusted strength. That approach misses the fundamental asymmetry. What I ended up doing was separating the two materials explicitly. The concrete domain uses a compression-only yield criterion with a friction angle calibrated to the concrete's compressive strength. The reinforcement is modeled as discrete truss elements embedded in the concrete mesh, each carrying tension only. This matters because a lower bound solution for a simply supported beam will completely mispredict the moment capacity if it assumes the concrete can resist any meaningful tension. It can't. The reinforcement yield stress and area are what define the plastic moment, not the concrete's nominal tensile strength.

Here's a realistic example from a project I worked on last year. We were analyzing a flat plate system with heavy column capitals. The initial model used a standard Drucker-Prager criterion for the concrete with isotropic hardening and predicted a collapse load factor of 4.2. My gut said that was wrong. I checked the mechanism. The solver had found a flexural collapse pattern that relied on concrete carrying tension across the entire span, which is physically impossible for a flat plate. The actual failure would be a punching shear mechanism around the columns, not a global bending yield line pattern. I switched to a modified Coulomb criterion with zero tensile strength and added discrete truss elements for the top and bottom reinforcement mats. The new collapse load factor came out to 3.1. A difference of nearly a full point, and it was in the unsafe direction with the original model. I've since learned to always inspect the mechanism before trusting the number.

Implementing Limit Analysis And Concrete Plasticity in Your Own Models

The workflow breaks down into five steps, though they don't always happen in that order. The first is defining the yield surface. For concrete, I use a modified Mohr-Coulomb criterion in the principal stress space. The friction angle is set to approximately 35 degrees for normal strength concrete and increases to about 42 degrees for higher strength grades. The cohesion is derived from the uniaxial compressive strength divided by a factor that accounts for the friction angle. Tensile strength is set to zero or a very small value, typically around 10% of the compressive strength. Some implementations use a tension cutoff, others use a cap. The tension cutoff is more conservative and more accurate for plain concrete members. The cap is necessary when you're modeling confined concrete in plastic hinges. The second step is mesh generation. Quadrilateral elements are preferable for most concrete limit analysis problems. Triangular elements introduce too much numerical diffusion in the stress field and tend to smear yield zones across multiple elements instead of localizing them properly. Eight-node quadrilaterals with reduced integration work well. Full integration causes shear locking in thin members. I typically use a mesh size of 50 to 100 millimeters for slab systems and 25 to 50 millimeters for concentrated load regions. Refining further doesn't improve accuracy in a limit analysis context because the solution is governed by the yield criterion and equilibrium constraints, not by stress gradients. Boundary conditions and loading come next. Support conditions should be applied as displacement constraints on the appropriate degrees of freedom. For a simply supported slab, that's vertical restraint at the edges. For a fixed support, both vertical and rotational restraints. Loads are applied as nodal forces or distributed pressures. Point loads should be distributed over at least three elements to avoid artificial stress concentrations that the yield criterion will interpret as premature failure.

The solver setup is where most people hit walls. The lower bound approach leads to a linear programming problem. The objective is to maximize the load multiplier subject to equilibrium equations at each node and yield inequality constraints at each integration point. The upper bound approach leads to a similar LP problem but with work rate equations instead of equilibrium. Commercial solvers like MOSEK or Gurobi handle the LP efficiently. Open-source alternatives like IPOPT work too but require more setup. The computational time is usually measured in minutes for a typical slab model and hours for a full building frame, depending on the number of elements and constraints. Verification is the step most people skip. Run the same model with both lower and upper bound approaches. If the gap between the two solutions is more than 10 to 15 percent, something is wrong. Either the mesh is too coarse, the yield criterion parameters are inconsistent, or the boundary conditions aren't properly constraining the mechanism. I also run hand calculations for simple cases. A simply supported beam with uniform load should give a collapse moment of wL²/8. If your limit analysis model gives 0.7wL²/8 or 1.3wL²/8, check the model setup before proceeding. One thing that surprises beginners is how sensitive the results are to the assumed reinforcement layout. Limit analysis is deterministic once the yield criterion and boundary conditions are fixed. But if you model the reinforcement incorrectly, the answer is correctly wrong. I learned this the hard way on a post-tensioned slab where the tendon profile wasn't properly captured in the model. The lower bound solution predicted adequate capacity because the stress field assumed the tendons were providing uniform confinement. They weren't. The tendons were draped in a parabolic profile, and the actual collapse mechanism involved slip at the deviators, not flexural yielding. The model missed it entirely because the deviator capacity wasn't included as a constraint. I added discrete link elements at each deviator location with a slip threshold based on the deviator friction and normal force. The collapse load dropped by 18 percent. That's a significant difference in design terms.

Another edge case that caught me off guard involved asymmetric loading on a continuous slab system. The lower bound solution found a collapse mechanism that only activated spans 2 and 4, leaving spans 1, 3, and 5 essentially elastic. That turned out to be correct, but only because the load combination was asymmetric. If I had run the model with symmetric loading, the solver would have found a different mechanism spanning all bays. The key insight here is that limit analysis doesn't assume a mechanism. It finds the worst case mechanism for the given loading. That's both the strength and the limitation. It won't predict secondary effects like redistribution due to cracking or time-dependent creep. Those require a nonlinear incremental analysis. The practical limitation of limit analysis for concrete is that it assumes a rigid-perfectly plastic material. Concrete doesn't behave that way. It has pre-peak hardening, post-peak softening, and significant strain rate dependency. For static collapse analysis of well-designed structures, this is usually acceptable because the relevant behavior occurs near ultimate capacity where hardening effects are small. For structures where serviceability or ductility is a concern, limit analysis alone is insufficient. You need a complementary elastoplastic analysis with hardening and damage to assess deformation capacity and crack widths. I also recommend keeping a library of verified benchmark cases. A simply supported beam, a fixed-end beam, a simply supported slab, a continuous two-span slab, a cantilever wall. Run these through your setup every time you change the model parameters or update the solver. The benchmarks catch configuration errors before they propagate into production models. I spend about 20 minutes on benchmarks at the start of each project. That 20 minutes has saved me days of debugging incorrect results.

The tools available today make this more accessible than it was ten years ago. Commercial packages like ANSYS and Abaqus have limit analysis capabilities, though they require manual setup of the yield criterion and reinforcement model. Specialized software like YLD and LAPSS is designed specifically for concrete limit analysis but has a steeper learning curve. For quick hand calculations, yield line theory remains the most efficient approach for slab systems. A properly set up limit analysis model takes roughly 30 to 45 minutes for a standard slab and 2 to 3 hours for a frame structure, not including verification time. If you're approaching this for the first time, start with the lower bound approach. It's more intuitive, the constraints are easier to formulate, and the results are conservative by definition. Move to upper bound once you're comfortable with the setup. Trying to do both simultaneously early on adds complexity without adding value. The methodology works. It requires careful attention to the material model and boundary conditions, and it has clear limitations that every practitioner should understand before relying on it for design decisions.