A digital twin used to mean a 3D model you could spin around in a viewer. That era is over. The twins that matter now are decision engines: they search possible designs and generate the evidence downstream decisions depend on.

I care about this shift for a specific reason. I spent years selling automated structural design to engineers who have to stamp the output. A pretty model convinces nobody; an engineer signs when the system can show its inputs and its reasoning. That is the bar a twin has to clear. Three areas keep pulling me back. Decision twins that propose and judge designs. Yard operations, where every ETO project looks different. Carbon as a design constraint. Take them in order.

From generative design to decision twins

Generative design is bigger than a new CAD plug‑in: it is an exploration engine. Designers set goals and constraints. Algorithms generate thousands of alternatives; simulation trims the field. The real value comes from the closed‑loop feedback cycle: the twin evaluates each generation in a realistic virtual environment, and the results improve the next.

A decision twin is a digital twin that goes beyond geometry. It contains data schemas, rule sets, acceptance tests and a mechanism to produce audit‑ready evidence packs. Inputs have units and provenance. Rules encode engineering standards and company policies, and acceptance tests define pass/fail. When generative algorithms propose a design, the decision twin evaluates it against these rules. The output is a decision, Go, No Go or conditional, plus a report recording the inputs and assumptions.

The same system that creates concepts also produces the evidence certification or sign‑off needs. And designers can trade weight against cost and carbon instead of minimising mass and calling it done.

Generative design turns into search inside a governed environment. The twin’s job is the explaining.

Digital twins for yard operations and engineer‑to‑order manufacturing

Large yards (modular construction yards or heavy fabrication facilities) operate like small cities. They juggle bespoke product configurations and scarce resources, on schedules that keep moving, under safety rules that do not. Unlike assembly‑line production, ETO yards build one‑off or small‑batch products, so every project looks different. In practice, yards struggle with fragmented IT and manual material flow, and with the basics of locating materials and supporting operators. Four capabilities decide whether a yard twin earns its keep. The first is seamless information flow: design, procurement, production and logistics systems sharing accurate, real‑time data. The second is identification and interconnectivity of objects, from materials to containers, via RFID tags or barcodes. The other two:

• Digitalised operator support, through mobile devices or augmented reality that guide workers and capture as‑built data.

• Automated and autonomous material flow: cranes and AGVs orchestrated by the digital twin.

A yard‑ready digital twin unifies these layers. It holds a live graph of products and activities, so a planner can ask it which blocks are ready, and what delaying a module actually costs. It should generate schedules that are both feasible and explainable (a plan that respects resource capacities and safety rules, with a reason for each decision). Events from the yard (a late delivery, a crane breakdown) must feed back to update the plan, and the twin records why the plan changed, which is essential for audits and lessons learned.

Such twins also need to support engineer‑to‑order logic: when a design change arrives mid‑production, the twin should propagate that change through BOMs, routings, schedules and resource allocations. A yard twin that stops at a dashboard is dead weight.

Sustainability and circularity in twin‑driven design

Many telecom and power-line towers are approaching end‑of‑life, and regulators and investors are pushing for reuse and low‑carbon materials. TowerUP, a Shapemaker AS research project in Norway, points the same way: sustainability‑optimised design software and a modular kit‑of‑parts for tower infrastructure. Optimise for weight alone and you get the wrong tower:

• Embodied carbon and connection mass should feature in the objective function alongside member weight and stiffness. A tower with lighter members but heavy, numerous connections may have more embodied CO₂.

• Dynamic performance (frequency, damping) must meet strict limits for wind or seismic events; optimise only for mass and you get unacceptable sway.

• Disassembly and reuse depend on standardised splices and accessible joints, which can conflict with minimal weight.

• And the supply chain: available section catalogues and transport constraints belong in the design space.

Toward an integrated, sustainable twin platform

Put together, the platform starts from an awkward fact: design, fabrication and operation span companies that do not fully trust each other. So the twin records who changed what, when and why, in a form every party can verify. Around that core, the decision twin holds the rules and acceptance tests and produces audit‑ready evidence for every decision. Generative engines explore design and operational spaces under multi‑objective criteria, leaning on the twin to filter options. A yard‑aware execution layer turns the resulting plans into real‑world operations, with feedback from the field, and sustainability is scored while the design is still cheap to change.

When a designer or algorithm proposes a tower with an unusual taper, or a yard supervisor needs to re‑sequence modules, the decision twin runs the checks, logs the outcome and stamps the evidence. If the result meets the acceptance tests, it becomes a state downstream actors can trust, even the ones who do not trust each other.