In Part I of our "Trailblazers" interview with Dr. Martin Fischer, Kumagai professor of engineering and director of CIFE at Stanford University, he shared the background of his career in academia and his insights into the AEC industry as the director of the Center for Integrated Facility Engineering (CIFE). In Part II, Fischer explores which technologies present the most promise in our industry as well as how AI and machine learning can help us.
Dr. Burcin Kaplanoglu, executive director, innovation officer at Oracle Construction and Engineering, leads the discussion.
MF: I put them in two buckets: 1) Technology that establishes the demand the project should meet 2) Technology that establishes what was done so that the plan, or the demand, can be compared with the actual and improvements can be identified on the basis of facts.
For demand, do we have the right design? Are we using the right structural system for that project? Does it offer the client the flexibility, strength, and durability needed for the particular purpose of say, building a bridge, tunnel, or whatever the structure is?
The second major bucket of demand is the production process in the design and construction phases. Do we have the right resources available at the right time? Are they likely going to be productive?
What’s the demand and supply for the type of structure we need as well as the production process? When we match those, we have a good project, and everybody feels they’ve contributed and made money. And when we don't have a match between the supply and demand, things get very challenging.
I look for technologies that help me understand: What demand should be expected? For example, what’s the schedule to establish the demand for resources?
I also explore technologies that establish: What is the supply? How are we meeting the demand? The big task is closing the feedback loop between our predictions in terms of the planned performance of buildings and infrastructure, the expected performance of the production system that creates them, and what is completed and achieved.
Everybody allocates their resources, whether that's materials, say, in a mechanical system or some other system, or whether that's people working in design, management, construction, etc., based on a prediction. We need to find a way of making those predictions better; the schedule, estimate, and design are all predictions.
We should be able to say, “In 2020, our predictions are better than last year’s in this way.” And so far, I haven't met a single company that can say something like this. The feedback loops are key so that we can figure out how to make demand and supply more predictable.
I'm very excited about the technologies that help us understand how a project is going. For example, this may be in the design phase in terms of documenting how a team and the design it creates changes over time and how information flows.
In addition, I’m excited about technologies that show us how equipment is used on site and how crews are working. Startups and established companies are making lots of advancements here. Are the workers performing their work safely? Are they being productive? Are they creating high quality work?
We need technologies that help us to simulate and predict the expected performance using the observed or measured data to the level of detail at which the data is collected.
On the design side, we have good methods in terms of parametric design with computational design optimization. These technologies also help establish “the production system” to create the design, construction, and the operations and maintenance approach.They also help us leverage the data about actual performance that we can collect to assess how things are working against how we thought they should work.
These days, I look at design and construction in terms of matching demand and supply: the product we design vs. make, the process we plan to have vs. have, and the challenges for making the predictions and measurements of demand and supply reliable.
When they aren’t reliable—which they often aren’t— project teams must deal with lots of variability. If you look at operations science, this results in extra project duration, extra work in progress and inventory, and tied up cash.
Many of the problems we see in the industry are symptoms of this fundamental challenge. The problems you see in the industry come from the difficulty of matching demand and supply. That’s why I'm excited about technologies that help us understand supply and demand much better.
MF: These technologies are very useful in developing a design, a construction schedule, and an estimate in terms of running different scenarios and exploring different risks.
AI and machine learning help explore the expected repercussions if you go with "Approach A" versus "Approach B", including: How expensive will the project be and how long will it take? What range can you expect given the variability and the data?
I see AI and machine learning as key technologies that allow us to understand the project much better—and earlier—because we played with so many scenarios.
AI and machine learning can help us understand productive or unproductive patterns in how a project is proceeding so that the project management staff can intervene in a timelier manner. Combining data-driven insights, assuming the data are of good quality, and the gut-driven experience of professionals will be very powerful.
However, I haven't found a company that’s able to do this because it requires connecting multiple data sets about contracts, schedules, estimates, and design evolution as represented in changes to the project’s BIM, payments, etc.
Companies have these data sets, but they are not easily linkable today. The kinds of insights we could hopefully have in the future might be, “For this type of project, whenever we have seen so many changes in BIM this early on, that’s rarely turned out well.”
Or you might say, “Well, when we've never seen any schedule changes early on, that hasn’t turned out well because people haven’t really thought about the project or explored enough options.”
The patterns that we could measure early in the project—and throughout, of course—but particularly early on, could put the project on a different trajectory than it’s likely on if the trajectory isn’t good based on the history.
Finding these patterns is entirely impossible to do by ourselves. And that's where we absolutely need to employ machine learning. I've seen the power of those patterns in other industries and sectors where they’ve had the data to make this analysis.
I believe this is the direction we need to go. It’s still a bit of a challenge for technology to truly grasp everything that matters about the context now, because many factors come into play in a situation, but the experienced engineers and managers will know how to account for that.
But what we cannot do with just our branis is understand the broader context—the history of how things went and what factors were most important. We cannot understand what the pattern is over the last hundred or thousand projects. Or if we proceed in this way, in other words, if we simulate the likely path forward, what are the expected patterns and where could things go?
That's totally beyond the cognitive capabilities of humans. That’s where the combination of machine learning and AI with human understanding, experience, and knowledge becomes potentially very powerful. I'm very excited about that potential.
Explore how you can deliver project success with Oracle Construction and Engineering.
Related "Trailblazers" posts: