• Mergen Kor

Integrating Digital Twins and Deep Learning

Updated: Jun 2, 2021



We discussed "Integration of Digital Twin and Deep Learning for facilitating Smart Planning and Construction" during a recent Clubhouse Room.

The discussion focused on Construction 4.0, with a variety of professionals interested in debating the opportunity and integration between Digital Twins and Deep Learning.


We began the discussion by defining various terms used to describe these ever-evolving subjects. Digital twin is fairly new terminology, especially for the construction industry, whereas machine learning, and deep learning, are more broadly known terms. To clarify, it was highlighted that deep learning is a subset of machine learning where artificial neural networks, which are algorithms inspired by the way a human brain works, learn from large amounts of data. Deep learning allows machines to solve complex problems even when using a data set that is very diverse, unstructured, and inter-connected. But how can this apply to construction?

The industry already collects an enormous amount of data, much of which remains unstructured.


The question is: can a machine generate insights where our human brain cannot?

Opportunities to apply machine learning could be methods of recognizing objects in a digital environment to automatically label or classify elements from new projects and/or old ones. One obstacle that exists at the moment is not having enough structured data to train the machine on how to label the elements in the models. This not only applies to classifying elements but also to specifying which design systems could work best in some conditions. Although some companies may already have this data organised under diverse scenarios, they still lack full connectivity between databases to algorithms.


Ultimately this limits an algorithm's capability in determining answers to complex questions without needing careful human oversight.

One of the larger issues in the industry is the focus on modeling the geometrical aspect of our built asset, without spending too much time creating elements that are embedded with other meaningful data. This short-sighted approach currently limits the capacity of analysis and extrapolation of useful trends.


There has been a common confusion between digital twins and the connection between deep learning and artificial intelligence. A lively debate sparkled while discussing whether BIM levels are, or are not, necessary to explain the maturity of a digital twin. An argument was made that the BIM levels do not necessarily apply to digital twins, as we can have digital twins with no geometrical information. A highly detailed BIM level may not be required; however, the data is required to be structured. On the other hand, it may not be feasible to create a digital twin of everything; it depends on the use case. Another important aspect mentioned was that data flow must be bi-directional for it to be considered a digital twin.


Additional questions were discussed that warrant further investigation:


• What is an achievable level of interoperability between Digital Twin and Deep Learning towards Construction 4.0?

• What is eventual the added value of integrating Digital Twin and Deep Learning for construction process optimization in construction?


We highly recommend that anyone interested in exploring this topic further should check out the course by Clayton Miller: Data Science for Construction, Architecture, and Engineering; https://www.edx.org/course/Data-Science-for-Construction-Architecture-and-Engineering


Thanks to all that joined and look forward to seeing you in more of our upcoming rooms! https://www.joinclubhouse.com/club/house-of-digital-twi

174 views0 comments