How can AI make Construction better?
Construction lags other sectors in innovation and productivity. This post explores applications where AI can help bridge the gap.
The world will need to spend $57 trillion in infrastructure and housing by 2030 to make room for the rural to urban migrants. The infrastructure and construction industry, which employs 7% of the world’s working-age population, will shoulder the bulk of this responsibility. However, the construction sector has an intractable productivity hurdle. Large projects typically take 20 percent longer to finish than scheduled and are up to 80 percent over budget. In the last two decades, the labor productivity of the construction industry has stagnated at 1%. Financial returns for contractors are often relatively low and volatile. Equally worrying is the proportion of worker fatalities (highest of all other sectors).
Traditionally, the construction industry has been making incremental improvements. Each project is unique, because of which it is not possible to scale up new ideas. Adopting new technologies is impractical. In this post, we will look at a few applications where artificial intelligence can help the sector leapfrog across.
Making construction project management predictable.
There are far too many variables that can throw an execution out of control in the Engineering Procurement and Contracts sector. Labor shortages, inclement weather, supply outages, and regulatory clearances are some parameters that contribute to the inherent variability in project management. Each project is considered unique. This lack of standardization has made the process digitization slow. Daily reports, drawings, contracts, continue to rely heavily on paper-based communications. Information flow between front-line contractors and project planners is erratic and slow-paced. Frontline contractor’s planning horizon is often limited to the next 7 days. Most project management is reactive. The mid-senior level staff is engaged in routine firefighting exigencies.
We can use machine learning to implement day- and week-ahead forecasts. Every site gets a list of activities that are predicted to start on that date. A real-time dashboard lists high-risk activities for cost and time overruns. Operational data sources within construction is limited. However, most sites log inventory levels for project costing. Similarly, project progress is monitored on scheduling tools. AI can weave together such data streams and combine them with weather and historical performance to predict future outcomes. It will enable contractors and site managers to take a realistic view and prioritize interventions.
Personalized training for front-line blue-collar workers
Labor-productivity growth in construction has stagnated over the last two decades. It is impossible for the sector to speed up without investing in human capital. Compared to other sectors, construction has a high proportion of migrant workers. Informal recruitment channels on which many construction migrants and labor markets rely heavily are a barrier to a fair and safe working environment. New entrants spend the first five to seven years as an apprentice. During this period they change trades as and when an opportunity arises. Apprentice workers are economically vulnerable and run into migration debt. Most apprentice workers are conversant only in a local dialect, a barrier which limits their ability to upskill themselves, understand their basic rights, or even seek certification for their skills.
Conversational AI has revolutionized training. Personalized language learning apps have taught over 100 million a new language. to the construction sector. Personalized bots can educate a worker to operate a concrete vibrator, for example, leading to a certification from a company that manufactures such a vibrator. We can customize the course to the individual’s pace of learning and his mother-tongue. Training platforms can also be a means to source and recruit replacing the opaque informal channels used at present. It can make the recruitment transparent both to the worker and the employee.
For workforce training collaboration is key: Government, educators, manufacturers and construction firms. AI can play the role of a key enabler by bringing in personalization, scale, and ease of deployment.
Automating safety audits with object detection
20% of all worker fatalities occur in the construction sector. This statistic is even grimmer when you consider only 7% of all workers are employed in the construction sector. A complacent mindset has led to an acceptance of injuries as a part of the job. Safety is often understaffed: a single inspector manages a workforce of 500 spread across a site.
Falls, electrocution, and collapse make up 60% of all workplace accidents in the construction sector. We can use computer vision to detect the underlying conditions that cause such accidents. Most projects have surveillance cameras to discourage theft. We can process this feed to detect individual workmen within a bounding box. Subsequently, we can classify a bounding box based on the presence or absence of a hard hat and safety harness. We can repeat the automated process every minute and log all unsafe events. We can also train image classification models to detect the nature of the activity being performed, ex. bar bending, concreting, etc. We can tune a Mask-RCNN to detect water spillage, obstructions, and other such unsafe conditions. All such events and associated metadata get aggregated across all feeds, and real-time and concerned personnel receives a report or alert for further action.
There are limitations: false detection, misclassifications, obstacles in the visual field and low light. At present, we can not automate completely. However, we can augment auditing and compliance. We can bring down the workload of an inspector by 80%. It will give him time to focus on training and empowering the front line crew.
Aggregating contracts collaboratively to drive excellence
We can broadly split the construction industry into two segments. Large multinational firms involved in big infrastructure and housing projects and smaller fragmented specialty builders. Productivity within the fragmented units is often half that of the larger ones. Financial returns and timelines of projects under execution of smaller firms remain uncertain.
One of the critical components of project execution is contract management. We should assess contractors against criteria, such as past performance, financial status, certifications, and HSSE compliance. Smaller firms lack access to a network of suppliers, let alone having in-house skills to manage complicated relationships and contracts. Inefficient contracting causes firms to lose between 5% to 40% of value on a deal.
Retail e-commerce has been growing at 20% year on year. It’s likely to exceed 4.2 Trillion $ by 2020. Two-sided platforms buyers-seller platform can transform the contracting landscape. Startups, publicly traded companies, multinational giants are trying to enter this domain to leverage their logistics, network presence, and in-depth knowledge about products. However, the key aspect missing in such platforms is the support of back-end processes involved in contracting. Small organizations and departments need tools to budget, pre-select, evaluate, manage bids, issue purchase orders to sellers. Some larger firms might have an existing ERP and would need integration to an e-commerce platform. None of the existing platforms offer such functionality.
Construction firms approach each project as a unique endeavor. A platform derives its strength from its ability to organize and structure information. Lack of shared standards creates an additional hurdle. We can not wish away this problem by having elaborate PO forms and asking understaffed departments to fill them.
Natural Language Understanding algorithms today are a key part of several applications from autocomplete to processing insurance claims to the spam filter. Other than the financial data, most construction contracts exist as unstructured information. If we need to organize this information, we should be able to extract information out of unstructured text documents. We are already seeing some applications augmenting the para-legal teams’ due-diligence ability on legal documents. Similar approaches will help us organize messy construction contracts across projects by extracting named entities and relationships. Hassle-free and transparent transactions on platforms will increase efficiencies by a notch and drive the cost down. The platform can use its knowledge derived over millions of similar transactions to assist writing contractual specifications, establish guidelines for assessing, match buyers and sellers, and weed out incomplete and ambiguous orders.
B2B e-commerce is huge, it’s about twice the size of B2C. It will reach 6 Trillion $ by 2020. Construction contracts are an under-tapped opportunity in this segment.
Surveying and Inspecting with photogrammetry
One of the first activities in any construction project is land surveying. Surveyors manning theodolites record the terrain as a grid of points. In large and dispersed infrastructure projects, this is a labor and time-intensive activity in difficult conditions. Computer vision, namely photogrammetry has revolutionized this field. Drone borne cameras capture a sequence of images that are subsequently processed using positional meta-data to stitch it together as an array of 3D coordinates. In an 8hour shift, a drone (with additional batteries) can cover as much terrain as a surveyor could capture in 2 months. In a pre-design phase, time is of the essence and computer vision approaches are two orders of magnitude faster, produce a higher resolution 3D grid and do not suffer from manual errors.
Photogrammetry application extends beyond surveying. A 3D camera mounted on fixtures can generate internal views of buildings. We can compare the measured as-built dimensions with the original design and automatically detect deviations. In a retrofit construction, we can read dimensions directly from the 3D view. We also use photogrammetry to perform maintenance audits in assets such as wind farms and bridges. We can compare structural dimensions across time and detect anomalous strain.
Some exciting work in this domain has been towards generating layout out of RGB panorama images. Others are using stereoscopic and monocular photographs to estimate the depth of the field. Accuracy levels in some state-of-the-art approaches are in the order of ±5–10 cm per meter. We cannot use these dimensional inspections yet. However, we can build applications to use a panoramic image from a smartphone, derive the room layout, use the layout to present furnishing options for a homeowner to select.
The construction industry is ripe for disruption. However, there are substantial challenges. Digital transformation across companies has been slow. Adoption of innovation beyond pilot-scale implementations has been slower. Data transparency is limited. In the b2c segment, solution providers have to step in and create an entire vertical.
Despite this construction sector offers an unparalleled scale. If we are patient to fine-tune user experience and reach the product-market fit, applications can transform the lives of millions. Artificial intelligence will play a key role in these products.
Have you been through some of the challenges of the construction sector outlined in this post? I would love to hear your perspective on how to prioritize solutions. Are you working towards some of these challenges? It would be good to hear your story. Please drop a message on my LinkedIn.
