How can Natural Language Processing help us achieve sustainability?

My name is Soroush Sobhkhiz and I am a data scientist at Adaptis. This is the story of how and why we plan to utilize natural language processing (NLP) to make sustainable practices more cost-effective, efficient, and achievable.

Every building tells a story, not just through its architecture but through the vast array of textual data it accumulates over time: condition assessments, inspection reports, maintenance logs, and more. Unfortunately, this wealth of information seldom comes in an easy-to-analyze, structured format. Instead, we're often left with a puzzle of text documents, each piece holding key insights into the building's lifecycle.

This data becomes crucial, especially when it's time to consider the end-of-life phase of a building. We need to answer questions such as:

❓What materials are hidden within the structure?

❓Are they still in good shape, or are they damaged beyond repair?

❓More critically, can these materials be recycled, or are they contaminated with harmful substances that prevent recycling?

These insights are typically buried in the documents produced over decades of a building's operation. When conducting an end-of-life analysis, sifting through this information efficiently, without missing key details, is crucial. Therefore, capitalizing on NLP to facilitate and automate access to this information holds great potential.

One of my main missions at Adaptis is to develop solutions that seamlessly integrate textual data into our circularity assessment product. I am leveraging my previous research on this subject where I linked maintenance requests to objects of a building information model (BIM). In my research I essentially transformed textual documents and BIM Models into interconnected graphs, providing a holistic view of a building's narrative.

In simple terms, the approach constructs semantic networks from the descriptions of building elements. This way the key terms that represent a component (a window or an HVC system) are captured into graph nodes. For example, figure A below represents the key terms used to describe doors and windows in the context of BIM. You will find nodes such as ‘IfcWindow’ or ‘IfcDoor’ because IFC is the name of the data schema used in BIM models.

Similarly, we can extract these nodes and relationships from any other document such as a Building Condition Assessment Report (BCA) or an inspection document. For instance, figure B below represents a network of concepts captured from maintenance request tickets of an institutional building.

(A) Network of building elements (B) Network of concepts

Sobhkhiz & El-Diraby (2022)

At this point, we can link the networks together using graph theory and analytics to establish connections between building components and key concepts in our textual documents. For instance, if a section of a BCA is describing some findings on the external walls of a building, we can potentially, link the descriptions to the external walls of our BIM model. This is a relatively complex process that requires the use of machine learning algorithms. But once trained, these models can help us automate the time consuming process of document review which is a significant step towards analysis automation and therefore, scalability.

Establishing connections between building components and key concepts in our textual documents

Sobhkhiz & El-Diraby (2022)

If you are interested to know more about the approach have a look at my papers here and here.