AI and data mining: Solving construction’s other waste issue

Construction generates vast amounts of data, much of it valuable, but often the full potential of this data goes unrealised. A recent WBEF webinar looked how AI and data mining could help solve construction’s other waste issue. Here are some of the things we learned.

Wasted data: Silos and a lack of standardisation

While the construction industry generates vast amounts of data, it doesn’t mean that the industry is data-driven, because around 80% of this generated data is wasted, says Samaneh Zolfagharian Ph.D., CTO of YegaTech. Silos are one of the main causes for this waste. Multiple stakeholders are involved in a construction project, generating data that is unstructured and to their own format, and additionally, these stakeholders are often not comfortable sharing data. Another issue is that data quality is often poor, because it is captured without any business goal in mind, explains Samaneh. She recommends setting a strategy when capturing data to ensure its alignment with the end goal. A lack of standardisation is a common problem within construction, Samaneh says.

Professor Lamine Mahdjoubi, Professor of the Digital Built Environment and the Director of Centre for Architecture and Built Environment Research at the University of the West of England (UWE Bristol). As he explains, standardisation for the construction industry is important because, even among experts, there are differences of opinion. For example, opinions differ regarding how data should be classified: ‘AI is not a magic wand. It relies on the quality of data to learn and to deliver a quality output.’ When we can achieve standardisation, that’s where AI becomes clever and more robust in the future, he says.

AI and humans working as a team

Senior project managers can use their experience to manage things such as risk. But AI has the ability to scale that up and bring collective knowledge across thousands of projects to the table, says Dev Amratia, Co-founder and CEO of nPlan. AI can spot the parts of a project that are the most at risk. AI can also generate solutions, as just highlighting what’s wrong without trying to solve the problem only half answers the question, he says. Algorithms can notice how a particular project is performing against similar projects and analyse what could be learned from this in the same way a human brain does. The difference, however, is that AI can do this with hundreds and thousands of projects – a far greater capacity than the human brain can cope with, he explains.

When you have augmentation by an algorithm onto a project team, you still need to hire experienced people. But those people are becoming smarter at the decisions they are making by using algorithms; checking what the algorithm thinks and using that information in decision making.  ‘And when you do that, the outcomes of the project are very different to projects that are command and control from a central dictator,’ says Dev Amratia.

Lamine Mahdjoubi agrees that humans are central to the AI process. He explains that getting AI to the equivalent learning ability of a person with 20 - 30 years’ experience of dealing with a particular project and data takes time. In a semi-automated AI process, a human must check the robustness of the output, then feeding the result back to the AI so it continuously improves. This mimics the way humans learn, and over time, the AI process can lessen its dependence on humans.

Avoiding bias in algorithms and the role of the domain expert

When used for benchmarking and to compare projects accurately and reliably, it’s imperative to understand the AI approach taken and how robust it is, says Lamine Mahdjoubi. Explainable AI is a processes or method that allows humans to comprehend how the results, decision, or output from an AI process was achieved. Using this method can help to eliminate bias, says Samaneh Zolfagharian.

Programmers should design algorithms without input from domain experts. These domain experts may have opinions on, for example, why projects are delayed, and so introduce bias, suggests Dev Amratia. The value of domain experts, he says, is in translating an insight into an action, and the ability to communicate that complex information to a project team and have it implemented. He says, ‘It might sound easy, but it’s actually the hardest part of the equation – avoiding years of machine learning research coming to a grinding halt when no one in the field cares about it.’

AI and sustainability

The adoption of AI can bring many sustainability benefits to construction; an industry that has the highest impact on global warming, says Samaneh Zolfagharian. For instance, AI can help in optimising building performance. It can adjust and track systems, such as temperature, depending on whether a building is occupied or not, as well as helping owners reduce the consumption of energy during peak hours.

AI can also improve the process of recycling and reuse of materials from buildings at the end of the life, explains Samaneh Zolfagharian. A result being the reduction of the amount of manual labour otherwise required and the quantity of waste sent to landfill. For example, she cites a company, Urban Machine, in California. They utilise a combination of robotics and computer vision machine learning to detect nails in timber so that the wood can be reused.

In a world limited by appetite and risk, there is societal value in using construction data to inform decision-making. It can help to drive support for worthy projects that may not otherwise go ahead because of uncertainty of outcome, says Dev Amratia. We are now seeing projects being considered that would not have been considered previously, he adds.

Innovations in data classification

This AI and data mining webinar featured a presentation by Lamine Mahdjoubi on classifying unstructured data. The presentation was based on an Innovate UK funded initiative as part of the series of Transport Infrastructure Efficiency Strategy (TIES) Living Lab projects.

Without a common standard, it is very difficult to benchmark, says Lamine. The TIES project involved translating construction cost data from different systems into a common standard. A system was developed to classify data into an International Cost Measurement Standard (ICMS) automatically. It is a web-based system where the user enters an item into the system, and the data is processed and classified into an ICMS. In some cases, the system can read batch files.

Looking ahead

Generative pre-trained transformers are machine learning algorithms capable of generating content. For example, these could produce an article or contract from unstructured and poor-quality data. While not yet being able to surpass the quality of a trained professional, it can credibly augment them, says Dev Amratia.

Another new technology, generative design, is an AI-based system that uses algorithms to create building designs, for example, based on specified criteria.  However, at present, Lamine Mahdjoubi believes the technology is not robust enough to compare options, for example, which option will produce a more sustainable building.