Artificial Intelligence (AI) and the Cases of Intelligisation of the Engineering and Construction Industry



A study of 800 major projects (those of value over $US1 billion) found that, on average, projects were one year behind schedule and 30 percent over budget. But what’s true for major projects is similarly true for projects of various sizes, even projects as small as $US10-$US20 million.

One avenue to prevail over missed deadlines and budget overruns is the intelligisation of the sector through the adoption of Artificial Intelligence (AI).

Even without the imperative of AI-driven productivity, engineering and construction (E&C) firms would be wise to futureproof themselves by embracing AI and harnessing AI to augment capabilities and break new ground.

Research firm Gartner expects the global AI economy to increase from about US$1.2 trillion last year to about US$3.9 Trillion by 2022, while McKinsey sees it delivering global economic activity of around US$13 trillion by 2030.

Intelligisation of the E&C industry

With science and technology advancing by leaps and bounds, the E&C industry is also constantly evolving, from the traditional E&C industry to construction 4.0. And its level of automation and intellectualisation has a continuous improvement, it went into a new stage of development, thus, the combination of AI and E&C has become a hot topic.

What is Artificial Intelligence (AI)?

Put simply, Artificial Intelligence (AI) is a field of study that is concerned with the intelligent automation of tasks.

The level of intelligence of automation can be below or beyond human abilities. A purchaser of burial urn received an urn recommendation by an AI built by Amazon for product recommendation for months after the death of the customer’s mother, despite urn is needed once in a blue moon and it is insensitive to suggest such products. On the other hand, Google’s AI AlphaGo learned to play chess under four hours and defeated the world’s best chess-playing computer programme.

Approaches to the Intelligisation of the E&C sector

There are two techniques of intelligisation, namely rule-based intelligisation and learning-based intelligisation.

Rule-based Intelligisation

Rule-based intelligisation is the classical approach of AI, whereby human specialised knowledge is represented in an explicit declarative form of rules and facts. Rule-based AI is also known as symbolic AI, as well as coined by John Haugeland as the Good Old-Fashioned Artificial Intelligence (GOFAI). Symbolic Intelligence is a variant of intelligence gained from rule-based intelligisation.

Symbolic Intelligence

Symbolic intelligence is the kind of intelligence derived from rule-based intelligisation. It is based on the physical symbol system, which consists of a set of entities, called symbols, which are physical patterns that can occur as components of another variant of an entity called an expression (or symbol structure).

Symbolic Intelligence and the E&C sector

Expert system is a popular form of symbolic intelligence. Expert systems are designed to address complicated problems by reasoning through bodies of knowledge, represented mainly as if–then rules rather than through conventional procedural code. Expert systems were among the first truly successful forms of artificial intelligence (AI) software.

Expert systems are used in designing structures safe for human beings. DAMSAFE is an expert system tool for the monitoring of dams. In this system, the behaviour of dams is represented as a network of processes. Features extracted from monitoring data are mapped onto processes (that describe behavioural states) by means of rules defined by experts.

DAMSAFE is an AI expert system which helps in constructing safer dams
Source: The role of AI technology in the management of dam safety: The DAMSAFE system

Learning-based intelligisation

Intelligence is incredibly complex. Oftentimes it is impractical, if not impossible, to hard code the intelligence into machines. Enter learning-based intelligisation, or commonly known as Machine Learning: the subfield of AI that gives machines the ability to learn without being explicitly programmed.

Learning-based intelligence is broadly composed of Computational Intelligence, Statistical Intelligence and Experiential Intelligence.

Computational Intelligence

Computational Intelligence attempts to mimic nature-inspired problem-solving methodologies. This form of intelligence is a set of biology-inspired computational approaches learned from data and/or experimental observation to address tough real-world problems.

Computational Intelligence and the E&C sector

Neural Networks

Scientists drew inspiration from the neurons found in animal’s brains and other parts of the nervous system and invented neuronal networks. Neurons are connected to each other and they receive and send impulses throughout the animal’s body, or in the case of computing, the network. Neural networks are models that are composed of layers (at least one of which is hidden) consisting of simple connected units or neurons followed by nonlinearities.

Some of the actual neurons of the 100 billion neurons in a human brain

Neural networks are particularly suited for deciphering complex and/or hidden patterns. Neural network predicted the compressive strength of concrete modelled from 425 specimens.

This neural network takes in seven inputs to estimate the compressive strength of the concrete.
Source: Model the compressive strength of high performance concretes using Neural Designer

Deep Neural Networks are central to the technology behind self-driving trucks for carrying E&C materials.

Scene perception trained by Deep Learning techniques enables real-world driving.
Source: A Survey of Deep Learning Techniques for Autonomous Driving

Convolutional Neural Network is able to decode patterns of workers’ brain waves acquired from wearable Electroencephalography (EEG) devices worn by E&C site workers.

Deep Learning architecture for classifying the stress levels of E&C works.
Source: Houtan Jebelli, et. Al., Mobile EEG-Based Workers’ Stress Recognition by Applying Deep Neural Network: Proceedings of the 35th CIB W78 2018 Conference: IT in Design, Construction, and Management

Machine Vision

Machine Vision technologies involve the capture, processing and analysis of digital images, essentially decoding their meaning and context. There are many Machine Vision technology areas: machine vision, optical character recognition, image recognition, pattern recognition, facial recognition, edge detection and motion detection, all of which support the overall Machine Vision technology spectrum.

Convolutional Neural Network excels in image recognition.
Source: Yen-Wei Chen, Lakhmi C. Jain – Deep Learning in Healthcare: Paradigms and Applications

Machine Vision can serve numerous purposes, including:

  • measuring dimensions and placements of E&C materials
  • detecting hazardous objects and activities for occupational safety management;
  • counting quantity of E&C materials;
  • tracking E&C workers, equipments and materials;
  • analysing site images and videos for productivity;
  • etc.

Counting shear studs with Deep Learning
Source: Diffgram launches operating system for visual deep learning

Evolutionary computation

Evolutionary computation is a class of algorithms for global optimization inspired by biological evolution, and the subfield of computational intelligence. Evolutionary computation is used by The National Aeronautics and Space Administration (NASA) to design antennas for radio communication which produces unusual looking but very high efficiency antennas called Evolved Antenna.

Joris Laarman Lab employs evolutionary computation to design E&C structures. One of which is a 3D printed steel bridge. The partially AI-designed bridge was constructed by Arup

The bridge designed with the application of evolutionary computation
Source: Project Update: World’s First 3D Printed Steel Bridge

Under the Evolutionary Computation, there is a technique that mimics Darwinian natural selection—Genetic Algorithms. Genetic Algorithms is a computational approximation to how evolution performs search, which is by producing modifications of the parent genomes in their offspring and thus producing new individuals with different fitness. From an engineering perspective, Genetic Algorithms is an optimization technique that evaluates more than one area of the search space and can discover more than one solution to a problem, in other words, better-performing individuals get selected more often, which leads to outstanding solutions. Genetic Algorithms can be considered as a type of stochastic direct search method.

In the E&C sector, Genetic Algorithms can be used to optimise E&C material transportation routes to save costs and time.

Researchers make use of genetic algorithms to find the shortest routes from one point to another.
Source: Genetic algorithm and a double-chromosome implementation to the traveling salesman problem

Statistical Intelligence

Statistical Intelligence is the intelligence derived from statistical and probabilistic modelling of vast quantities of input data using Machine Learning algorithms. The algorithms can be classified as supervised or unsupervised. In supervised algorithms, ground truths are associated with the estimator inputs so the machines can generalise from the associations and infer the ground truth estimates when new estimator inputs are fed into the machines. In unsupervised algorithms, no such associations are provided and machines infer the ground truth estimates from the estimator inputs themselves.

Unlike traditional statistics, learning-based statistics can be more accurate as depicted in the graph below.

Source: Leanne Luce – Artificial Intelligence for Fashion: How AI is Revolutionizing the Fashion Industry

Statistical Intelligence and the E&C sector

The categories of applications of statistical intelligence comprise prediction, classification, clustering, anomaly detection and dimensionality reduction. In the E&C sphere, they come in handy in predicting bid price, factor analysis, clustering of clients and employees, just to name a few. Logistic regression is employed in identifying the most important factors contributing to successful bidding of projects.

17 factors were analysed with logistic regression, a supervised Machine Learning algorithm, to determine their significance.
Source: Decision Making Modelling with Logistic Regression Approach

 Support Vector Machines Algorithm

Support Vector Machines algorithm, a kernel methods based learning, is a popular Machine Learning tool. Data scientists apply Support Vector Machines on linear or nonlinear classification, regression, and outlier detection on small- or medium-sized datasets. The goal is of Support Vector Machines algorithm is to find the hyperplane that effectively divides the class representation of data

Support Vector Machines algorithm is versatile; it is applicable in a wide variety of fields; for instance textual classification, medical diagnosis, image recognition. In the domain of E&C, the Support Vector Machines algorithm is adapted to detect structural damages on buildings.

Structural damages on buildings detected by Support Vector Machines algorithm
Source: Identification of Structurally Damaged Areas in Airborne Oblique Images Using a Visual-Bag-of-Words Approach

Random Forest Algorithm

Random Forest Algorithm is a type of decision tree based learning. The algorithm is an ensemble of multiple decision trees. Random Forest performs well as a Machine Learning model, thanks to the concept of the wisdom of crowds—“A large number of relatively uncorrelated models (trees) operating as a committee will outperform any of the individual constituent models.”

Random Forest is the combination of multiple decision trees.
Source: Gollapudi Sunila, Practical Machine Learning

For classifying urban damages for subsequent reconstruction, a Random Forest model is used. Visual data in the form of high-resolution satellite imagery was employed to train the algorithm.

Urban damaged classified using Random Forest model
Source: Detection of Urban Damage Using Remote Sensingand Machine Learning Algorithms: Revisiting the 2010 Haiti Earthquake

Experiential Intelligence

In Experiential Intelligence, machines gain intelligence through Experiential Learning, or widely known as Reinforcement Learning, i.e., learning from lots of experiences. The intelligence is reinforced (hence the name Reinforcement Learning) from numerous trials and errors or rewards and punishments.

Experiential Intelligence and the E&C sector

Reinforcement Learning algorithm is used to define the state of learning of the E&C workers with respect to the instructional contents administered via the virtual instructor. The reinforcement learning algorithm will choose the type of instructional content to adapt the virtual instructor when the postures are deemed unsafe ergonomically.

Experiential Intelligence in optimising the occupational health of the E&C workers
Source: Anil Sawhney, Construction 4.0: An Innovation Platform for the Built Environment

E&C materials can be discovered by making use of Reinforcement Learning, whose goal is to maximize return when interacting with an environment.

Policy gradients, an algorithm under the family of Reinforcement Learning, and Monte Carlo Tree Search (MCTS) are used in generating new materials.
Source: Inverse molecular design using machine learning: Generative models for matter engineering


Robotics is an interdisciplinary scientific field concerned with the design, development, operation, and assessment of electromechanical devices used to perform tasks that would otherwise require human action. Robots typically consist of at least three parts: a mechanical structure (most commonly a robotic arm) that enables the robot to physically affect either itself or its task environment; sensors that gather information about physical properties such as sound, temperature, motion, and pressure; and some kind of processing system that transforms data from the robot’s sensors into instructions about what actions to perform.

Various types of robot for carrying out different functions. The following are the taxonomy of robots and the respective tasks they perform.

Source: Anil Sawhney, Construction 4.0: An Innovation Platform for the Built Environment

The Semi-Automated Mason (SAM) is a commercially available brick-layer robot. It is created by New York-based Construction Robotics.

The Semi-Automated Mason (SAM) in action.
Source: High-Tech SAM Bricklaying Robot Can Build Massive Walls 6-Times Faster Than Humans


Before concluding this piece, we would like to bring to your attention that AI techniques are dime a dozen. We have barely scratched the surface of the methods that are presently in application.

The intelligisation of the E&C industry to rid the E&C industry of inefficiencies and/or as an innovation strategy is by no means a given. Data in the E&C industry is poorly FAIR – Findable, accessible, interoperable and reusable. Furthermore, the “If It Ain’t Broke, Don’t Fix It” mentality is ingrained in many of the players of the E&C sector; as such, they have a tendency to default to familiar tried-and-tested methods. And not to mention a host of other issues: deficiency in resources and expertise, privacy issues, immature technologies, lack of the right culture.

There can be little doubt that the future of the E&C sector will be defined by AI. The opportunities from the diffusion of AI will be enormous and it is without question that, in 50 years’ time or so, the industry will look markedly different to the one we see today. We are at an exciting stage as we journey to the future in E&C.

By Primercon Research Team