Connection Designer & joint FEA

The Connection Designer is a dedicated workspace for one joint: pick the template, edit every geometric parameter, read the capacity checks and the detailing audit side by side, and - when you want a second opinion from a different method - run the joint as a 3D shell FEA (CBFEM-style) model.

It complements the connection schedule: the schedule sizes every joint at once, the designer is where you interrogate one of them properly.

Creating a joint

Three ways in, all ending at the same classified joint:

• Wizard - a four-column dialog: Class → Topology → Design → Parameters. The taxonomy is the connection library itself, so the eleven templates (fin plate, web cleat, flexible / flush / extended end plate, welded moment, pinned / fixed / shear-key base, bolted splice, gusset) appear here automatically.
• From selection - select two or more members that meet and create the joint from them. If the selection doesn't meet at exactly one shared node, the action refuses and says why.
• From the schedule or inspector - click a row to open that joint in the designer with its analysis forces attached.

The workspace

• Left rail - the template, the members framing in, and the fabrication operations list (the sequence a fabricator reads: crop, drill, weld, etc.). A refused joint lists no operations.
• Centre viewport - three tabs: Model, Mesh, Stress (below).
• Bottom table - the load cases; the design actions per case come from the solved analysis, floored at the AS 4100 Cl 9.1.4 minima. Unsolved cases show blank, never zero.
• Right panel - flips between Parameters (every geometric input: plates, bolts, welds, pitch, gauge, edge distances) and Checks (below).

Editing any parameter re-runs the checks immediately and drops any FEA result - a stale stress field would look like an answer.

Checks panel

One panel, three instruments answering three different questions:

1. Refusals lead, at fail severity. If the joint is outside its template's declared validity envelope, the refusal is stated with its remedy, and a named alternative template is a one-click control (its label discloses that switching discards hand-edited geometry). A refused joint shows the AS 4100 minima as "not assessed", never as a green pass.
2. Component checks - the closed-form AS 4100 utilisations with clause references: bolt group, ply bearing/tear-out, block shear, plate shear/yield/fracture, weld, tension bolts + plate bending (moment connections), concrete bearing / anchors / shear key (bases), Whitmore section (gussets). These are the source of truth.
3. Detailing audit - the buildability findings (asymmetric bolt groups, single-ply grips, overhangs, coincident plies, spacing violations, bolts in webs). A joint can pass every capacity check and still be unbuildable; this is the instrument that catches it.

Model tab

The Model tab draws the joint as 3D solids built from the same layout primitives the checks and the FEA mesher read - plates, bolts with heads and nuts, weld beads, member stubs at 1.5 section depths. What you see is what was checked. A refused joint draws nothing, captioned with the reason.

Scripting

Every control routes through a single action table, so a script and a click cannot diverge. The workspace is exposed as __fem.connectionDesigner in the browser console and to automation - the same actions the buttons fire.

Joint FEA (3D CBFEM)

The designer can solve the whole joint as one shell-FEA model - the CBFEM-style counterpart to the component method:

• Members are meshed as MITC shells (web + flanges + the welds that make them a section), carrying N / Vy / Vz / Mx / My / Mz at the far end of their stubs; the bearing member is held at both ends.
• Plates are meshed from the same layout primitives the Model tab draws.
• Bolts are connectors: a chain of short Timoshenko beams down the shank, one per consecutive ply pair - shear as a fixed-fixed shank in series with EN 1993-1-8 Table 6.11 bearing, tension as axial stiffness over the elongation length.
• Welds are rigid ties that transmit moment across the half-thickness offset.

The solve runs in the existing solver worker, off the main thread (typically 8k-30k shell elements, a few seconds to ~20 s). The Mesh tab shows the real mesh - and declines with reasons when a joint can't be meshed (a refused joint, or geometry outside what the mesher converts). The Stress tab paints the von-Mises field on the deformed mesh that produced it, with the contour scale clipped at fy.

What the FEA is for - and what it is not

Analysis-only - the component checks decide

The joint FEA is a linear elastic instrument for understanding load paths and stiffness. It is not a design verdict:

• No contact - the joint cannot open, prying is not captured, and a negative bolt tension means the plies bear on each other (only the tension side is readable).
• No pretension, no slip, no bearing nonlinearity, no plasticity - and no weld or bolt strength in the FEA at all.
• Peak elastic stress does not converge - it is a mesh-dependent singularity at junctions, not a utilisation. That is why the contour clips at fy, and why CBFEM's real acceptance criterion (5% plastic strain, EN 1993-1-5 Annex C) stays gated until plasticity exists.
• The closed-form component checks remain the source of truth for every verdict.

The FEA has been validated against beam theory (stub end forces within ~0.5%), the classical elastic bolt group (bolt forces within 4-6%), equilibrium to machine precision, and mesh refinement (answers hold to 5% across a 2× refinement). It has not yet been verified against a published CBFEM benchmark (IDEA StatiCa verification examples or AISC worked joints) - treat its pictures as insight, not evidence.

A stiffness class (rigid / semi-rigid / pinned) is computed from the FEA but is not yet fed back into the global frame analysis - the frame keeps the restraints you gave it.

Proposals: what auto-sizing would have done to what you set

Auto-sizing leaves a value you have set alone. That is the point of setting it - but the silent version of that promise is a joint that stays under-sized and never says so.

The Proposals tab in the results panel is where those withheld intentions are read, with a count on the tab. It is a panel rather than a dialog on purpose: a dialog after every solve on a 300-joint model gets dismissed by reflex, and a badge that stays until it is dealt with cannot be lost.

Rows come in three groups, worst first, and the caption is true of every row under it:

• Needs attention - a value auto-sizing is not allowed to change, on a joint the check reads OVER. Sorted by utilisation. These are the ones to act on.
• Wants more, not required - auto-sizing would have gone bigger, but the joint is not over as it stands: something else on the ladder covered it, or nobody has checked it yet.
• Efficiency - auto-sizing would use less than the joint carries. Often a deliberate decision, listed so it can be seen rather than because it is wrong.

There are no accept buttons. Making the disagreement visible is the half that was missing; applying a proposal is a second act with its own undo question.

Into the report

The joint page of the PDF report prints the geometry as solved, the component-check table with clause references, the detailing audit findings, and - if the FEA was run - a joint-FEA section with an isometric of the analysis model, the painted stress field, the bolt-force table and every caveat above. A refused joint prints "AS REQUESTED, NOT BUILT" with its remedy instead of a check table.