Boundary Conditions & Constraints

The Most Important Decision in the Model

A finite element model is often more sensitive to how it is constrained than to how finely it is meshed. Mesh refinement improves the resolution of a solution that the boundary conditions have already defined. If the constraints are wrong, the load path is wrong, and no amount of mesh refinement will recover the correct answer. The constraints tell the solver where the loads go to ground. Everything else — stresses, displacements, reactions — follows from that single decision.

FULLY FIXED DOES NOT MEAN REALISTIC

The most common boundary condition in FEA is the fully fixed constraint — all six degrees of freedom (three translations, three rotations) set to zero. It is easy to apply, it eliminates rigid-body motion, and it is almost never a realistic representation of how a real structure is supported. A fully fixed constraint implies a support of infinite stiffness, a support that does not deform, does not rotate, and does not allow any redistribution of reaction. Real supports — bolted joints, bearings, fillet welds, adhesively bonded interfaces — all have finite stiffness. They deform under load, and that deformation changes the load distribution in the surrounding structure. A fully fixed constraint hides this interaction and produces a model that is artificially stiff at the support.

A fully fixed constraint is a mathematical convenience, not a physical reality. Before accepting any stress result near a fixed boundary, ask: does the real support actually behave as if it cannot move in any direction?

Fixed Constraints

A fixed constraint locks all degrees of freedom at the constrained surface or node. It is appropriate when the support is genuinely rigid compared to the structure being analysed — for example, a component attached to a massive cast iron machine bed, or a test specimen clamped in a hydraulic grip. In these cases the support stiffness is orders of magnitude greater than the component stiffness, and the infinite-stiffness assumption is reasonable. The danger is applying the same assumption to a thin bracket bolted to a thin panel, where the support stiffness is comparable to the component stiffness. The fix is either to model the support structure explicitly, or to replace the fixed constraint with a compliant representation — a spring, an elastic support, or a remote constraint with a realistic stiffness.

• Appropriate: support genuinely rigid relative to the component (massive fixture, rigid test bed)
• Questionable: thin bracket on thin panel — support stiffness comparable to component stiffness
• Inappropriate: bolted joint modelled as fully fixed at the hole — suppresses bearing, bypass, and prying loads

Pinned Constraints

A pinned constraint locks translations but allows rotations. It represents an ideal pin joint — a frictionless hinge that can rotate freely but cannot translate. Like the fixed constraint, it is an idealisation. Real pins have clearance, friction, and finite stiffness. The pinned constraint is appropriate for idealised truss and frame analysis, and for cases where the rotational stiffness of the joint is negligible compared to the bending stiffness of the connected members. It is not appropriate where joint rotation stiffness is significant — for example, a tightly bolted clevis joint, which behaves somewhere between pinned and fixed depending on preload and clearance.

Symmetry Constraints

When the geometry, loads, and material are all symmetric, only half or a quarter of the structure needs to be modelled. The symmetry plane is constrained with symmetry boundary conditions: normal translation locked, in-plane rotations locked. This halves or quarters the model size without changing the result. The critical caveat is that symmetry applies only to the symmetric response. A structure with symmetric geometry under symmetric load can still buckle in an antisymmetric mode. If you are running a buckling or modal analysis on a symmetric model, you must run both symmetric and antisymmetric cases, or model the full structure.

• Symmetry requires symmetric geometry, symmetric loads, and symmetric material — all three, not just geometry
• Symmetry boundary conditions: lock normal translation, lock in-plane rotations at the symmetry plane
• Antisymmetric boundary conditions: lock in-plane translations, lock normal rotation — for antisymmetric load cases
• For buckling and modal analysis, the lowest mode may be antisymmetric — a symmetric model will miss it

Symmetry is a constraint on the response, not just on the geometry. A symmetric structure under symmetric load can still fail in an antisymmetric mode. If you are looking for the lowest buckling load or the lowest natural frequency, do not assume it is symmetric.

Remote Constraints

A remote constraint ties the degrees of freedom at a surface to a single remote point, with a specified behaviour: rigid (the surface moves as a rigid body with the point) or kinematic (the surface can deform but its average motion follows the point). Remote constraints are useful for applying loads or supports at a point that is not on the meshed geometry — for example, applying a load at the centre of a hole without modelling the pin, or constraining a bolted flange at the bolt centreline. The remote point acts as a load introduction or reaction point, distributing the effect to the connected surface.

• Rigid remote: surface moves as a rigid body with the remote point — adds stiffness, can overconstraint
• Kinematic (deformable) remote: surface follows the remote point on average but can deform — does not add stiffness
• Use for load introduction at a point, applying bolt preload, or representing a pin without modelling it

Springs and Elastic Supports

Springs and elastic supports provide a finite, tunable stiffness at a boundary. They are the most honest way to represent a compliant support when you know or can estimate its stiffness. A spring to ground represents a support that resists displacement with a known force per unit deflection. A bushing or grounded spring with six independent stiffnesses (three translational, three rotational) represents a joint with known stiffness in each direction. The challenge is knowing the stiffness value — it is often not readily available and must be estimated from the support structure, measured, or iterated. An elastic support (a distributed spring per unit area) represents a compliant foundation, such as a structure resting on soil or rubber.

• Spring to ground: point stiffness, force per unit deflection — for discrete supports
• Elastic support: distributed stiffness per unit area — for continuous foundations
• Bushing: six independent stiffnesses — for joints with directional compliance
• The stiffness value is the hardest input to obtain — estimate from the support structure, measure, or iterate

Coupling and Multipoint Constraints

Coupling and multipoint constraints (MPCs) tie the degrees of freedom of multiple nodes together. A coupling constrains a set of nodes to move together in selected directions — for example, coupling the radial displacement of all nodes on a hole circumference to enforce a circular deformation. An MPC defines a linear relationship between nodal degrees of freedom — for example, a rigid link, a beam connector, or a distributed constraint. These are powerful tools for representing idealised connections without modelling the physical connector, but they introduce constraint forces that can be unrealistic if the idealisation is too aggressive.

• Coupling: nodes move together in selected DOFs — simple, can overconstraint if too many DOFs are coupled
• MPC (rigid link): nodes connected by a rigid bar — rigid, adds stiffness
• MPC (beam): nodes connected by a flexible beam — can tune stiffness to represent a real connector
• Distributed coupling: load distributed from a point to a surface with a weighting function — more realistic than rigid

Rigid Elements

Rigid elements — rigid links, rigid beams, rigid surfaces — connect nodes with infinite stiffness. They are useful for representing components that are much stiffer than the surrounding structure without modelling them in detail: a steel bolt in an aluminium panel, a rigid mounting plate, or a stiffener represented as a rigid link. The danger is that rigid elements introduce infinite stiffness into the model, which can cause numerical problems and can overconstraint the connected nodes. They also concentrate load transfer at their endpoints, which can produce artificial stress concentrations. Use rigid elements only when the represented component is genuinely much stiffer than its surroundings, and be wary of stress results near their attachment points.

Overconstraint and Underconstraint

Overconstraint occurs when the boundary conditions are stiffer than the real support. The model becomes artificially stiff, stresses near the constraint are unrealistically high or unrealistically low (depending on whether the constraint suppresses a real deformation or forces a real deformation to zero), and the load path is distorted. Underconstraint occurs when the boundary conditions are insufficient to prevent rigid-body motion. The solver either fails to converge (in static analysis) or produces meaningless displacements (the structure flies off in the direction of the unconstrained degree of freedom). The challenge is that overconstraint does not produce a solver error — it produces a plausible-looking but wrong result.

Overconstraint rarely produces a solver error. It produces a result that looks reasonable but is wrong. The only reliable defence is to check reactions, displacements, and stress distributions against expectations — hand calculations, test data, or engineering judgement.

Artificial Stiffness

Every constraint that is stiffer than the real support adds artificial stiffness to the model. A fixed constraint adds infinite stiffness. A rigid link adds infinite stiffness. A coupling that locks too many degrees of freedom adds stiffness. The cumulative effect of multiple overconstraints can be a model that is significantly stiffer than the real structure — lower displacements, higher natural frequencies, higher buckling loads, and a load path that does not match reality. The remedy is to replace idealised constraints with compliant ones wherever the support stiffness is known or estimable, and to model the support structure explicitly where it is not.

Real Support vs Numerical Stabilisation

Every static FEA model needs enough constraints to prevent rigid-body motion — six constraints for a 3D model (three translations, three rotations), or three for a 2D model. Sometimes a structure is genuinely self-equilibrated (internal loads only, no external reactions) or is supported in a way that does not fully constrain all rigid-body modes. In these cases, the analyst adds weak springs or minimal constraints — "soft" supports with very low stiffness — to stabilise the model without significantly affecting the load path. These are numerical stabilisation, not physical supports. The distinction matters: a physical support carries real reaction load; a numerical stabilisation carries negligible load and exists only to prevent the solver from seeing rigid-body motion.

• Physical support: carries real reaction load, defines the load path
• Numerical stabilisation: prevents rigid-body motion, carries negligible load, does not affect the load path
• Check: if a stabilisation spring carries significant load, it is not a stabilisation — it is a support, and it needs to be justified

If a "soft spring" used for stabilisation carries more than a negligible fraction of the applied load, it is not a stabilisation — it is a support. Its stiffness is now part of the load path, and it must be justified like any other boundary condition.

Rigid-Body Motion

A modal analysis can help verify whether the applied constraints reproduce the intended physical support conditions. For a free-free three-dimensional structure, six rigid-body modes are normally expected at or near zero frequency. For a properly restrained structure, these modes should be suppressed by the intended physical supports. The absence of near-zero rigid-body modes does not, by itself, indicate overconstraint. The analyst should compare the calculated mode shapes and natural frequencies with the expected physical behaviour, examine the constrained degrees of freedom and assess whether the support representation introduces artificial stiffness.

free-body

Saint-Venant's Principle

Saint-Venant's principle states that the effect of a localised load or constraint diminishes with distance from the point of application. Far enough from the constraint, the stress field depends only on the resultant force and moment, not on the detailed distribution. This principle is the justification for replacing a complex constraint with a simpler one — for example, replacing a detailed bolted joint model with a fixed constraint on the bolt hole, provided the region of interest is far enough away. The critical question is always: how far is "far enough"? A common rule of thumb is two to three times the characteristic dimension of the constraint (the hole diameter, the bolt spacing, the contact width), but this is a guideline, not a guarantee. The only reliable check is to refine the constraint representation and confirm that the stress at the region of interest does not change.

• Saint-Venant's principle: far from a constraint, the stress field depends on the resultant, not the distribution
• Rule of thumb: two to three characteristic dimensions away — but verify, do not assume
• The principle does not apply at or near the constraint — local stresses are always constraint-dependent
• Verify by changing the constraint representation and confirming the result at the region of interest is unchanged

Poor vs Better Constraint Practice

The difference between poor and better constraint practice is not usually the constraint type — it is whether the constraint represents the real load transfer. A detailed example illustrates this.

The better practice is not always the most detailed practice. The goal is to represent the real load transfer at the support, not to model every bolt and weld. Sometimes a compliant spring with a justified stiffness is better than a full contact model, because it captures the essential compliance without the computational cost and convergence sensitivity.

Constraint Strategy: A Practical Checklist

A constraint strategy should be chosen deliberately, not by default. The following questions help ensure the constraints represent the real structure, not just prevent rigid-body motion.

• Is the constraint stiffness representative of the real support? — If not, use springs, elastic supports, or model the support structure
• Are all six rigid-body modes constrained? — Check with a modal analysis — first six modes should not be zero
• Are any constraints overconstraint? — Compare reactions and displacements to hand calculations or test data
• Are symmetry conditions valid for the response of interest? — Geometry, loads, material, and response mode must all be symmetric
• Is the region of interest far enough from the constraint for Saint-Venant to apply? — Verify by changing the constraint and checking the result is unchanged
• Do stabilisation springs carry negligible load? — If they carry significant load, they are supports, not stabilisation
• Are stress results near constraints treated with caution? — They may be artefacts of the constraint, not real stresses

Key takeaways

• A finite element model is often more sensitive to how it is constrained than to how finely it is meshed.
• A fixed constraint represents infinite stiffness — it is almost never a realistic representation of a real support.
• Overconstraint artificially stiffens the model and suppresses real load paths; underconstraint leaves rigid-body motion and produces meaningless results.
• Symmetry constraints are only valid when geometry, loads, and material are all symmetric — and when the response mode of interest is also symmetric.
• Numerical stabilisation to prevent rigid-body motion is not the same as a physical support — it should add negligible stiffness to the real load path.
• Saint-Venant's principle allows a constraint to be moved or distributed if it is far enough from the region of interest — but "far enough" must be justified, not assumed.
• FULLY FIXED DOES NOT MEAN REALISTIC. The goal is not to prevent motion; it is to represent how the structure is actually supported.