BY: Karthik Venkatasubramanian, Vice President of Data and Analytics, Oracle Construction and Engineering
Artificial Intelligence Is Enabling Construction To Tap Into New Digital Opportunities
Machine learning (ML) from a construction perspective has come a long way in terms of understanding what can be achieved using data science. But it’s still got a long way to go before adoption of things such as predictive analytics is commonplace.
In general, ML capabilities have increased especially in areas such as computer vision, use of neural networks and deep learning.
We’ve seen increases in the ability to process Big Data through distributed computing as well as the emergence of Machine Learning as a Service (MLaaS) from all leading cloud providers including Oracle to democratise artificial intelligence (AI) and make it available to as many organisations as possible.
Additionally, there have been improvements in existing open source frameworks such as Tensorflow, Keras, PyTorch etc, and new ones that are making AI more accessible. All of this means that the next five years will probably see a large uptake of ML technologies in construction.
Data is the key
Data is the lifeblood for any AI and ML strategy to work. Many construction businesses already have data available to them without realising it.
This data, traced from previous projects and activities, and collected over a number of years, can become the source of data that ML models require for training.
Models can use this existing data repository to train on and then compare against a validation test before it is used for real world prediction scenarios.
Model accuracy can improve with not just more of the same data but more of different data that is now becoming increasingly possible due to digitisation of many aspects of construction. Often termed feature selection, the vast amount of data from different systems allows identification “markers” of project
success and delays and therefore contributes to building ML models with better accuracy than was possible before.
Developments in automated machine learning (AutoML) means that a large part of the grunt work typically required in identifying the right data set for prediction (called features) and the right model that produces the best prediction is getting simpler. These developments mean this technology is getting into the hands of people that weren’t able to develop and deploy ML models prior to this.
Machine learning in construction
There are many emerging use-cases of ML in the industry including ideas that will positively impact the metrics that the industry has always cared about – schedule, budget, quality, safety and risk. Data and ML is being used to change the status-quo across all these key dimensions.
Computer vision is being used to solve problems such as identifying progress on site, tracking delivery of materials, understanding movement of labour and material on-site and more recently, application of social distancing rules on construction site.
The application of ML techniques to unstructured data coming from videos and photos is becoming progressively pervasive in solving several use-cases that were often tricky to solve previously.
Use of natural-language processing (NLP) is now being deployed for use-cases that should reduce manual error, improve productivity and mitigate risks. For example, NLP is being used to track submittals required for different jobs, identify non-standard terms in a contract, highlight a potential HSE issue or escalate a risk of an upcoming change request. These allow contractors and owners to better plan and respond to situations.
Use of ML for predictions about schedule delays and cost-blowouts is another area where ML really scores, as there is a lot of prior data on schedule and budget performance which can be used as training data to make predictions.
Schedules and budgets are becoming smart by incorporating ML driven recommendations, supply chain selection is becoming smart by using data across disparate systems and comparing performance and risk planning is getting smart by using ML to identify and quantify risks from the past that might have a bearing on the present.
Oracle Construction and Engineering is concentrating on what machines are really good at as opposed to humans–the ability to separate signal from noise in large amounts of non-visual data.
Our products are used across many different phases of construction. We’re currently focused on building ML powered predictive insights using Oracle Aconex, Oracle Primavera Cloud, Oracle Textura Payment Management and Oracle’s Primavera Unifier data. Some of these use cases will include:
- Predicting the probability of schedule and activity delays
- Surfacing early warnings of underlying risks across plan, design and build phase activities
- Predicting budget performance and sources of potential cost over-runs
- Providing actionable insights and recommendations to help ensure that projects are delivered ahead of time and budget
- Enabling supply chain selection based on historical performance
As smartification drives datafication, AI-driven transformation will naturally happen as companies begin to question how they leverage all the data that they have. The cost of ML is already decreasing with infrastructure that leverages pay-as-you-go models in the cloud.
New tools are democratising ML to the non-data scientists by way of dragdrop modelling, visualising predictions and simplifying the creation of easy to consume insights. Digitisation is going to disrupt every part of the supply chain and this is creating data that can change how built assets are delivered:
■ Schedules and budgets auto-created as a starting point optimised for the client and their requirements
■ Risks identified and mitigated through machine powered recommendations before issues arise
■ Sub-contractors dynamically selected based on their historical performance and the type of job being done
■ Change requests and variations predicted well in advance to ensure minimum disruption
■ Work sites made safer by completely eliminating work site HSE incidents.
As processes get digitised for improving productivity, the data that these systems generate will drive a paradigm shift in the process itself.