Restraints, releases and connectivity

The distinction

A restraint joins the model to the rest of the world. It generates a reaction.

A release describes what happens inside the model, at a connection between members. It generates no reaction.

Both are often called 'pinned', and both begin with R, and that is most of why they get confused. The consequence of confusing them is severe: put pinned restraints at every beam end and the load never reaches the columns — it goes straight into the ground at every floor, and the columns report almost no axial force while the total reaction is correct.

That pattern — total right, distribution wrong — is the signature of a load path that has been short-circuited rather than broken.

Over-releasing

A member is usually split into several elements so results can be plotted smoothly. Release the moment at both ends of every element and the internal nodes have no rotational stiffness at all: the member becomes a chain of links, and the analysis fails or returns kilometres of displacement.

The rule: releases belong to members, not to elements. Release the two ends of each physical member, where the real connection is. Everywhere else the member is continuous and the model should say so.

There is a quick count that catches it. If the number of moment releases is close to twice the number of elements, they have been applied element by element.

Too much restraint

Under-restraint stops the analysis, so it gets found. Over-restraint runs cleanly and changes the structure, so it does not.

Two cases worth knowing:

A truss supported below its neutral axis, pinned at both ends. Under gravity the truss sags, which pushes the supports apart. Restrain both horizontally and it cannot — so it works as an arch instead of a beam. The bottom chord loses nearly all its tension and is dangerously under-designed, and the model looks entirely reasonable.

A line of rigid supports under a continuous base. Each pinned node is infinitely stiff, so the base spans in short bays between them and the moment diagram picks up a sawtooth. Refining the mesh makes it worse, because each new node becomes another rigid support — which is a useful discriminator, since almost every other mesh-related symptom improves with refinement.

The fix in both cases is a spring rather than a pin. A defensible first estimate for a foundation: assume it settles about 10 mm under full unfactored load, which makes the total spring stiffness a hundred times the load.

Infinite stiffness is a modelling choice, and usually a poor one. Wherever a real support has finite stiffness and the structure is indeterminate, a spring is more honest than a pin.

Connectivity

Elements connect through shared nodes and through nothing else. A node that merely lies on another element's edge is not attached to it, however convincing the picture looks.

Coincident nodes — two nodes at the same coordinate that were never merged — draw as one point and separate during the analysis. They arise from copying, moving and importing, which is to say from almost every model-building operation.

Coincident elements — two elements between the same pair of nodes — double the stiffness and double the self-weight in that member.

Hanging nodes — a node on the edge of a coarser element, at a mesh-size transition — produce a hinge line along the transition, visible as a discontinuity in the displacement contour.

All three are found by the same discipline: count things, and check connectivity explicitly after every copy, move or import. Nodes, elements, total length by section, total mass. The graphics cannot tell you.