Talk about a lessons learned database. U.S. construction and engineering giant Bechtel has been in business for 120 years, with some 25,000 construction projects under its belt—many of them megaprojects. Yet all of that deep history wasn't doing much to move the needle on productivity gains, which across the sector had held steady recently at low single-digit gains, according to Justin Leto, PMP, Bechtel's senior big data architect and engineer.
The organization wanted to jump-start productivity rates but realized it couldn't get there with incremental innovation. “We needed something that was more disruptive,” he told iTnews.
So the company created the Big Data & Analytics Center of Excellence (BDAC) in partnership with data science firm Miner and Kasch. Part of the center's aim was to create a way to apply deep learning and AI to the complicated task of sequencing for construction megaprojects.
The model works much like a strategy game, with a design similar to DeepMind's AlphaGo Zero. It has a 3D neural network, with each space on the grid characterized by its density, while pipes and steel act as “game pieces.” A project team is able to test out different sequences virtually until it finds the right move, or sequence, that maximizes the project's productivity.
“If you think about some of these megaprojects that span decades and cost billions of [US] dollars…there is no deterministic way to order the sequence of construction,” Mr. Leto said. “And if you take two people building a similar plant with the same workers, we found that we would get very different results. That variability adds a lot of risk, and that risk adds a lot of cost. [We thought:] Let's compare how humans sequence construction compared to the machines.”
—Justin Leto, PMP, Bechtel, to iTnews
With the BDAC model, which was completed in September, the team is able to draw on lessons learned from past projects and also explore new sequencing approaches not yet attempted.—Amelia Garza