AI needs context to work

Burak Boyaci (left), Senior Director of Software Development
Alan Esguerra (right), Product Expert
Bentley Systems

Two experts from Bentley Systems, a US-based infrastructure software company, explain how digital twins give AI the data and context needed to support design, maintenance, and decisions across a building’s lifecycle.

Gavin: How is Bentley using technology to improve the performance of smart buildings?

Alan: Bentley has developed a digital twin technology, called iTwin, to house the entire lifecycle of anything built, from factories to freeways. The technology encompasses planning, design, construction, and maintenance.

Burak: That information can be stored in a single data environment and used over time. With sensors connected to the digital twin, building performance can be monitored continuously, reducing the need for manual inspections. Historical data, design standards, and analysis can also be used to simulate future scenarios, helping operators identify risks early and plan maintenance more effectively.

Alan: Predictive analytics is a key part of this. Our iTwin provides context for data from multiple sources, which AI can then analyze. For example, if cracks appear in a building 20 years earlier than expected in its lifecycle, finding the root cause of the problem is much easier. In the past, you'd have to go through libraries of information to find out where the concrete was sourced and mixed, who the contractors and designers were, and so on, to assess where the fault lay. With the digital twin and the large amounts of data that AI can help us sort through, engineers can identify where there may be other areas in the building that need to be looked at and address them very quickly. This technology is available today, and its capabilities will be far more advanced in five years.

Burak: AI needs two major things to work well: vast amounts of data, and context. Without context, the AI may produce unreliable output, such as hallucinations or made-up answers. Our digital twin provides ample data with the contextual framing of engineering expertise. That enables more effective inspections, predictive maintenance, and better planning of budgets and resources.

Alan: We believe AI will support engineers, rather than replacing them (although some engineers fear this). AI can help augment human expertise by analyzing past building performance and suggesting better design options, such as alternative materials or construction methods.

Burak: AI can also compare data across buildings, using past performance to predict what might happen in a new project.

Gavin: It's a rolling accumulation of knowledge. What impact do you think it has on sustainability?

Burak: It improves sustainability across all stages of the life cycle. In design, it saves energy because the same engineer doesn't need to drive back and forth to the office, making the design stage shorter and more efficient. During construction, AI can help you consume less concrete or other carbon-generating materials. For operations, AI can reduce energy use and extend the life of the asset.

Alan: A Korean energy company used our digital twin with a wider data context, linking a plant model with near-real-time operational data. In the past, data would be in spreadsheets or tables that are not easily digestible. Now, we can digest it much more easily. We can visualize it and include more contextual data, such as other energy sources like offshore wind. AI can help us optimize how energy is used, and when. One simple idea of “when” is scheduling your heavy workloads at night, when there's less energy use and power demand. We do this today, but AI can help us improve upon this with near-real-time operational data.

Gavin: Are companies responding positively to digital twins and AI?

Burak: Adoption is growing, but engineers are typically conservative and many still use 2D plans. There are also concerns about storing critical data in the cloud. Clients want to know where it will be kept, who will have access to it, whether it could be hacked. We make sure our customers understand that their data is their data. We don't use it unless they explicitly tell us we can. Those concerns need to be addressed.

Gavin: How do you do that?

Burak: It requires both a top-down and bottom-up approach. Technology providers need to show clear productivity gains to engineers. Governments, meanwhile, should set direction and standards.

Alan: Few of us use digital twins anywhere near their full capacity, and that’s largely down to a people challenge. We need to work harder to build confidence, providing a safe and sustainable path forward that is in some cases even conservative, to ease people’s concerns. Some companies say they’ll use digital twins, but maybe just for one building. But we want to provide data on the building, the site it's on, the roads leading up to it, even something like traffic flows. We have traffic simulation software that can analyze and model different scenarios. You could make better decisions about traffic scheduling during construction. This could be analyzed over weeks or parts of the day. In operations, schedule scenarios like seasons or sports events. You can use digital twins for whole context for things like lighting angles and visibility scenarios.

Currently, one of our biggest concerns is what’s underground. We don't know what’s under that borehole, or whether there’s an ancient ruin that someone built on 100 years ago. We dig, we find something, and it results in huge delays. We have the technology to provide that context today. We just need to get everyone to use it, to input the data and collect all that information.

Gavin: So, the main thing preventing broader adoption of digital twins is the natural human tendency to be cautious?

Burak: Right. It’s like self-driving cars. The technology exists today and is continuously improving, so why isn’t it more common? First, a lack of government regulation; second, people don't feel comfortable using it. There’s the human aspect to it. It’s the same with digital twins: the tech is there, but people have to start using it to trust it more.