A Michigan elementary school is kicking off a new era of generative design – News

When Mt. Hope Elementary in Lansing, Michigan, opened its doors early this month, the ribbon-cutting wasn’t the only first. Beneath its floors and behind its walls lies 11,000 feet of electrical raceways precisely routed by artificial intelligence.

For Francesco Iorio, the project caps a six-year-long journey to prove that AI can tackle the scale, complexity and real-world messiness of construction. Iorio co-founded Augmenta Inc. in 2019 to bring AI to a construction field that has long resisted automation.

“We’ve been doing generative design for years, but always on small objects like car parts that you can hold in your hand,” Iorio said. “A building is orders of magnitude more complex.”

Hard for AI

Designing electrical systems involves laying out intricate webs of conduits, bends, junctions and load points that must obey both rigid engineering standards and the quirks of each project. Designers must weigh such factors as safety codes, constructability and long-term maintenance while juggling cost and schedule pressures.

There are also many unwritten rules, or workarounds and efficiencies that veteran electricians pick up over years in the field. “You don’t learn those in school, and they’re not written down anywhere,” Iorio said. “But they’re critical to making something that’s actually buildable.”

Generative AI excels at turning out images, text or music, but it struggles with creating outputs that require extreme precision and adhere to a set of constraints. Generative design is a process that uses algorithms and AI to create designs based on a set of specified requirements and constraints. Essentially, the designer defines the output and the limitations and lets software figure out what design works best.

Even a midsized construction project such as an elementary school can have “hundreds of thousands of interconnected parts, physical constraints, building codes, and decades of unwritten best practices that only live in people’s heads,” Iorio said. “Capturing all that in software has never been done before.”

Iorio became hooked on generative design during his time as a research director at Autodesk Inc. Augmenta’s approach combines generative AI and physics-based generative design. The former learns patterns from thousands of past designs, including elements of human intuition about layout and routing. The latter places every part exactly where it belongs, ensuring compliance with both codes and constructibility.

Common-sense filter

Using generative AI as a “common sense filter” before running optimization slashed an otherwise infinite search space for possible designs to a manageable few. Without it, “you’d have to wait until the end of the universe to find a solution that fits all the rules at this scale,” Iorio said,

Designers upload a full 3D building information model created in Autodesk’s Revit design software showing elements like walls, ceilings, beams and columns. A high-level electrical specification shows the locations of rooms, load requirements and system types. The spec also includes contextual constraints, such as “never run conduit through the atrium” for aesthetic reasons.

Augmenta’s proprietary cloud-based AI model ingests all these factors, runs the optimization routines and spits out a fully detailed BIM model of the electrical system showing the exact recommended placement of every panel and conduit. The model can be used for procurement and installation.

For this first project, the system yielded only a modest 25% speed improvement, in part because both the customer and Augmenta were learning the ropes of working with an AI design partner. Subsequent projects are seeing productivity gains in the 50% to 70% range, Iorio said.

Waste dividend

The model also delivered a 15% cut in material waste. “In a hospital or a data center, that’s thousands of pounds of copper and steel that don’t have to be made, shipped, and dumped in a landfill,” Iorio said. “Even state-of-the-art projects procure a large amount of construction materials that end up in a ditch as excess material.”

Augmenta’s AI doesn’t recommend a single “perfect” design but produces a menu of viable options with the tradeoffs clearly stated for each. For example, a layout might minimize installation cost but limit future expansion or take longer to execute but make maintenance easier. The ultimate choice is always up to a human.

Though an elementary school is a manageable proving ground, Iorio said Augmenta’s system is already at work in data center and hospital projects spanning hundreds of thousands of square feet of construction space. Such designs can take 10 to 14 hours to run on Augmenta’s cloud platform, but it’s a drop in the bucket compared with the weeks of work such designs typically require.

Electrical system design is just the start for Augmenta, which has raised more than $25 million in funding. It plans to move into mechanical, HVAC and plumbing systems, eventually covering the full stack of building engineering.

“We chose to start with the hardest problem first,” Iorio said. “If we can validate in the most complex case, then we have a trajectory to target the other trades without painting ourselves into a corner.”

Photo: The Christman Co./LinkedIn