A telling fact: while preparing this closing panel at the Tech For Industry Show, Xavier Fodor, Editor-in-Chief (Twin+), the session's moderator, found that none of his five speakers defined the digital twin the same way. An operational tool for one, a way to connect systems and data for another, a scientific model for a third. Rather than a single definition, the session hosted by buildingSMART France explored what connects industry, infrastructure and construction — with Mouaad Aabibou, Digital Twin Program Director (TDF), Bertrand Vanden Bossche, Product Manager (Suez Smart Solutions), Marc Barthelemy, Research Director (Institute of Theoretical Physics, CEA), Dominique Chevillard, Board Member (bSFrance / FNTP) and Guillaume Vray, Director AVEVA France and Numeum representative(AVEVA and Numeum).

TDF — DTT and radio broadcasting, hosting mobile operators, indoor coverage, data centers — operates nearly 9,000 sites: towers, rooftops, concrete masts. Its 250 technicians travel 5 million kilometers a year across mainland France and overseas territories — a figure that doubles once contractors and clients are included. And often for a single purpose: capturing data. "Kleenex data," used once, poorly structured, poorly stored.

Hence the digital twin program, led by Mouaad Aabibou: scan the infrastructure, produce a point cloud, a virtual tour and a BIM model, interfaced with the information system, project management, and IoT (energy consumption) — for design (simulating adding a 5G antenna and the resulting load on the tower), maintenance and operations. Users: technicians, but also design offices, project managers, and site portfolio managers. Gains: time, kilometers — and safety, since every avoided tower climb is one less risk. Across 9,000 sites, all different, what gets standardized is the methods: capture, processing, model production — while the equipment itself (a 5G antenna is a 5G antenna) becomes standardized too.

Bertrand Vanden Bossche (Suez Smart Solutions) starts from a simple observation: a plant takes 2 to 2.5 years to build, operates for 15 to 30 years — and the structure itself can last 100 to 150 years. The real challenge: keeping the data alive long after construction ends. The first platform, dating back to 2018-2019: managing the sewer network of the Hauts-de-Seine department, operated with Sevesc — 650 km of network, including 450 km that can be physically inspected (surveyed by humans and modeled in 3D), and around sixty production sites, within an original Île-de-France setup where municipalities, the department and SIAAP share collection, transport and treatment, "just like in the energy world."

Suez's challenge: connecting assets (networks in the GIS, plants in the CMMS) to operational tools, on a single platform open to the end client — with automatic links recreated every week, "because things are constantly moving." And vertical optimization twins, such as aeration at wastewater treatment plants — a major energy expense — handled through mixed physics + AI approaches, checked against actually measured savings. The hardest part? "Getting organizations to agree" more than the technology itself: as a service operator, Suez takes on assets it doesn't own, with existing naming conventions.

Marc Barthelemy (Institute of Theoretical Physics, CEA) draws a decisive distinction: a complicated system (a watch) has many components but precise interactions and no surprising behavior; a complex system (a city, an ecosystem) sees collective behaviors emerge that are hard to predict — traffic jams, ice crystallization. The consequence for city digital twins: missing just a few interactions is enough to throw everything off, and — a generally false belief — more realism doesn't produce better forecasts. Physical modeling shows an optimum: too simple, the model predicts nonsense; too detailed, it's also wrong. Enough to raise questions about entire "digital China" type projects.

Having become a "necessary buzzword" in European research projects, the digital twin risks making decision-makers believe in a magic tool while sidestepping the question of validation: to what extent can I trust the model — especially outside "business as usual" conditions? Echoed by Dominique Chevillard (FNTP/bSFrance): "a simulation is only accurate within a defined field — outside that, it's an extrapolation." And an example of frugality from a 2022 InfraBIM Open study: measuring a beam's deflection once a week is more than sufficient; doing it every second costs more, emits more CO₂, and adds nothing. A useful model isn't an exhaustive one — although, as Guillaume Vray nuances, industrial processes (chemicals, pharma) genuinely do require second-by-second, real-time data: it all depends on the use case.

The figures cited by Guillaume Vray (AVEVA, representing Numeum): around 26% more data collected every year — and nearly 70% of industrial data unused today. Around 40% of companies say the digital twin is strategic. The condition for everything else, including AI: data governance — "no good digital twin without quality data," and without it, "adoption fails."

Case cited: a global industrial group with around 70 plants where every piece of equipment carries the same name everywhere and reports the same information — making it possible to deploy a structured twin at global scale and multiply use cases across sites. That's exactly buildingSMART's fight: interoperability, openBIM, data dictionaries — knowing that the term "digital twin" itself isn't yet standardized (ISO TC 442 WG9 working group, underway for 2.5 years). With two safeguards reiterated by the FNTP: respecting protected know-how (not everything gets shared) and national sovereignty and security concerns.

Industry retains a lead — more data, longer experience managing it, and mastery of brownfield challenges: making old PDF plans intelligent, correlating laser scans with P&ID diagrams. But construction has its own specifics: 1.1 million kilometers of roads and 200,000 to 250,000 bridges in France — "not knowing whether it's 200,000 or 250,000 worries me," Dominique Chevillard notes — demand frugality rather than blanket instrumentation. Five-year visions: borderless IT/OT convergence and the spread of real time and AI (Vray); integrating asset knowledge with operations, all the way to predictive maintenance (Vanden Bossche); standardization of twins across TowerCos, already under discussion, driven by client demand (Aabibou); interoperability, open tools and "no naivety" (Chevillard); and Marc Barthelemy's requirement: never give up on understanding — behind the curves the system spits out, always look for the dominant mechanisms and the meaning.

This closing panel shows that convergence between industry and construction around digital twins doesn't hinge on a shared definition — none exists yet — but on shared challenges: data quality and governance, the unresolved question of model validation, and the need to stay frugal rather than over-instrument. Standardized interoperability, championed by buildingSMART, emerges as the most concrete common ground between these two worlds.

Watch the full panel, along with the other Tech For Industry Show replays, on our dedicated page.