Skip to main content
DoneThat

AI Adoption GuideConstructionPlan

Site Logistics Layout Optimisation

ML optimizes temporary works and site logistics layout, including crane radius, laydown, and traffic flow, for a given site footprint.

Construction processBidAwardPlanMobilizeBuildInspectHandoverClose

By Don, DoneThat’s AI coach · updated

What the signed layout has to include

The outcome is a proposed site logistics layout with cites (crane radius, laydown, access) that temporary-works still signs. The model does not produce a PE-stamped drawing. It produces a candidate placement on a given footprint, each placement tied to a source: the legal boundary, the crane data sheet, or an access constraint. You review those cites, change what does not hold, and only then issue the temporary-works drawing.

Treat crane radius as a cited value from the plant data, not a number the model invents. If the data sheet is missing jib length, capacity at radius, or outrigger / counterweight envelope, stop. Do not let the optimiser fill the gap. The same rule applies to laydown: area, height, and what may sit there come from the programme and the permit, not from a default pad size.

A layout that temporary-works will sign names three things at minimum. Where the crane (or cranes) sit and which radius each pick uses, cited to plant data. Where materials, cages, and waste sit, cited to the legal site boundary and to any licence for occupation of the highway or neighbour land. How vehicles enter, wait, and leave, cited to the access plan, including restricted hours. If a cite is missing, that zone is not approved. It is a sketch.

Planning and simulation platforms (Autodesk, Alice Technologies, and similar) are used to test candidate layouts on a given footprint. They do not replace the engineer of record. Use them as a class of tools that propose options. Do not treat a vendor export as the approved temporary-works drawing.

Gather boundary, crane, and access before you run anything

Start with three inputs only. Everything else waits.

Site boundary: the legal red line, plus any licensed occupation (highway, neighbour, railway). If laydown is drawn outside that line without a licence cite, the proposal is already wrong. Pull the same boundary you use for permits, not a CAD convenience polyline that includes the pavement "because we always have."

Crane data: make, model, and the capacity/radius table you will actually hire, plus the base or rail envelope, slew limits, and any oversail agreement. Do not type a radius because it "looks about right on the plot." If two crane options are still in tender, run two constraint sets. Do not average them.

Access constraints: gate locations, vehicle types, turning and waiting, and hours. Restricted-hours access is a hard constraint. If the only gate that can take a low-loader is shared with a school run or a neighbour's loading dock, that window belongs in the model. Ignoring it produces a layout that works at noon on a drawing and fails at 07:30 on site.

Optional but useful once the three are clean: existing structures and overheads, underground services that constrain crane bases, and the first-phase sequence so you are not laying down rebar where the core will rise in a later phase. BIM design clash detection is the right companion when the logistics model starts colliding with structure, services, or the crane's own envelope. Do not use clash as a substitute for the boundary and licence check.

Feed the optimiser only what you would defend in a temporary-works review. Garbage in is not a software problem. It is a cite problem.

Propose a layout with cites, then stop

Run the search. Take the candidate that meets the constraint set, or the shortlist if several are close. For each placed object, keep the cite next to it.

Crane: base coordinates, cited radius for the governing pick, slew / oversail note, and the plant data reference. Do not write a radius that did not come from that sheet.

Laydown: polygon, permitted contents, and the boundary or licence clause that makes that polygon legal. If the best-scoring option puts cages on the footway, reject it unless the licence is in hand.

Traffic: inbound path, holding, unload, and outbound path, each tied to gate, vehicle class, and hours.

Then stop. The proposal is an input to engineering review, not an issued drawing. Do not retitle the sketch "TW-001 approved." Do not send it to the crane supplier as the as-built base location. Do not load it into the site induction as gospel.

Illustrative example. A constrained infill plot: one tower crane, street-only access, a licensed skip bay on the carriageway, and a neighbour gate that may take articulated vehicles only outside school drop-off. The optimiser places the crane to cover the core and the steel, puts main laydown along the rear boundary, and routes unload through the neighbour gate because that path is shorter. The cites fail two checks. The rear laydown polygon crosses the legal red line onto land that is not licensed. The neighbour-gate route ignores the restricted-hours clause, so the short path is illegal for most of the working morning. The usable proposal is the next candidate: slightly longer on-plot haul, laydown fully inside the red line, street skip remaining on the licence, and the neighbour gate marked as out-of-hours only. No radius was invented; the crane stayed on the hired sheet. Temporary-works then checks ground, oversail, and lift plans before anyone calls that sketch a drawing.

CPM schedule generation from scope tells you when the governing picks and the bulk deliveries actually occur, which is what the layout must survive. A plan-stage layout that cannot take the peak laydown implied by the programme is not a quality outcome, even if the crane radius looks tidy.

Review gates temporary-works cannot skip

Engineer review is the quality gate. Walk the proposal against the same three cites.

Boundary: every laydown, cabin, and washout sits inside the legal line or on a named licence. Placing laydown outside the legal boundary is a silent fail that is easy to miss, because the optimiser scores extra pad as better unless the red line is a hard clip.

Access: every inbound class has a legal path in the hours it will actually arrive. Ignoring a restricted-hours access often shows up only when the first steel wagon is turned around.

Drawing status: the sketch is a proposal. Issuing the sketch as the approved TW drawing short-circuits ground investigation, lift plan, oversail, and the engineer's stamp. The model output has none of those.

Also check what the model cannot see: ground bearing for the crane base, buried services, proximity to live railways or overheads, fire and ambulance access that must remain clear, and whether the traffic path still works when a second crane or a concrete pump is on the plot. If those are not in the constraint set, they are not approved by silence.

After the first physical setup, site setup compliance vision check is how you confirm the as-built fences, gates, and laydown still match the signed layout, rather than the sketch that was convenient on day one.

How this layout feeds clash, programme, and deliveries

A signed logistics layout is a planning object, not a one-off drawing. Keep it versioned with the cites.

When structure or services move, re-run clash against the crane envelope and the laydown polygons before you move the crane on paper. When the programme shifts peak deliveries, check whether the pad and the holding bay still fit. critical material delivery ETA prediction is useful only if the layout can actually receive the load in the window the ETA names.

Do not chase a perfect score from the optimiser. Chase a layout a temporary-works engineer will sign: every crane radius cited to plant data, every laydown cited to boundary or licence, every access path cited to gate and hours. That is the quality bar. The model helps you search. It does not stamp.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other. This one is rated high effort to implement, so the baseline matters more than usual.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first