System Modeling

With the linked-model structure from Step 1 in place, each discipline now models its actual building systems — and how carefully that's done determines whether Step 3's clash coordination finds real problems or just noise.

For this 18,000 sq ft, 2-story mixed-use building, each discipline models within its own linked file: architectural builds walls, floors, roofs, and room boundaries; structural models the column grid, beams, and foundation system; mechanical routes ductwork and places HVAC equipment; electrical models panels, conduit runs, and lighting/power circuits; plumbing routes supply, waste, and vent piping plus fixture placement. Every discipline models against the shared levels and grids established in Step 1 — a duct routed at the wrong level relative to the structural model isn't a modeling error that shows up immediately, it's exactly the kind of hard clash Step 3 exists to catch.

A critical modeling-quality decision at this stage: how much detail to model, and when. Modeling every duct fitting and every conduit bend at full fidelity from day one is often wasted effort at early design stages — most firms model to a level of development (LOD) appropriate to the project phase, with mechanical and electrical routing getting progressively more detailed as the design moves from schematic through design development into construction documents. Modeling too little detail too late makes Step 3's clash detection meaningless (nothing routed yet, nothing to clash); modeling too much detail too early wastes hours on geometry that will change anyway.

The mechanical room and the space above the corridor ceiling — where ductwork, conduit, and piping from all three MEP disciplines typically converge in a building this size — are where system modeling quality matters most, because that's exactly where Step 3's coordination effort concentrates. A well-modeled mechanical room with real duct sizes, real conduit diameters, and real pipe routing (not placeholder geometry) is what makes the clash detection in the next step actually productive instead of a list of false positives from underspecified models.