A key initiative for UofT's Digital Twin team is finalizing a data portal where UofT's building data and analytics will be readily available. The portal development for UofT's Facilities and Services' data will allow simple access to researchers and subsequently alleviate the burden on F&S.
The data within our portal will include:
- structured data, for example: Energy consumption (electricity, gas, steam)
- Unstructured data, for example: Forms and complaints
Business Process Work Order Analysis
IBDT acknowledges the importance of unstructured data analytics in developing digital twins. To this end, the project aims to develop solutions that enable digital twins to understand the ever changing patterns in unstructured data and help in decision making processes. The key area of focus is work orders submitted through the CMMS portal of F&S. Currently, this breaks down into the following:
- Linking business processes and data models: The main objective of this project is to develop a solution that can capture operational data in different business processes. This is primarily done by establishing a link between data models and business models. The operational data can then be used for developing artificial intelligence based services.
- Artificial Intelligence for work order analytics: The key purpose of this project is to develop AI solutions that can help capture the key concepts and their relationships in work order data. The result is used to develop a conceptual network for work orders that can play the role of a formal data model. This can help provide various services such as classification solutions that can automate business processes by allocating resources, urgency analysis, planning and so on. See ‘Dynamic work order modeling' for further information.
Student Lead Project
One key project proposal led by UofT's Digital Twin team aims to create a nexus between real-world education for our students and promoting the reduction of the carbon footprint of UofT. The project proposes an annual, multidisciplinary student project to facilitate learning from the UofT F&S staff by performing an analysis of the carbon emissions and energy consumption.
In one year, students will evaluate the energy waste source points and how these can be eliminated to make the building more sustainable. The source point can be any mechanical HVAC (Heating, Ventilation & Air Conditioning) equipment, building material, or any faulty building design. In the following year, another group of students can take this analysis further by developing a preliminary life cycle costing analysis and then proposing an action plan. Through using crowdsourcing tools, another group of students can engage a wider section of UofT students or building occupants in the process.
Dynamic Work order modeling
Work orders contain valuable information regarding the overall building performance and occupant satisfaction. However, this knowledge comes in unstructured inconsistent textual data which is challenging to extract. There is a growing interest in applying text-mining techniques for extracting the knowledge hidden in work orders. One key objective of recent studies is to develop automated methods for work order classification. Classifying work orders into different categories is important for several reasons. First, the process of work order management can be a very challenging and resource-intensive process. As a result, it is important to automate the process as much as possible. Work order classification facilitates resource allocation which is critical for efficient facilities management. Furthermore, automatic work order classification paves the way for applying predictive methods that help predict the expected resources needed for each work order. Additionally, by understanding the class of a work order, high-priority issues are better identified.
However, recent studies assume that work order classes can be determined by experts and in advance. As such, there is a tendency to develop static models for managing and classifying work orders. Experience shows that expert-based classification results in two issues:
- As the industry evolves and buildings age, new problems emerge that were not considered in the initial set of classes determined by experts.
- Second, it is a common issue that expert-based classes overlap with each other, and one problem can belong to several classes.
In this study, we propose a dynamic bottom-up unsupervised approach where the data reveals its key patterns rather than being pushed into predefined classes. The key concepts and relationships between them are used to construct a hierarchical model and develop a hierarchical classifier for work order data. The framework is further enabled to identify new concepts over time and dynamically update the model. The method was applied in a case study and achieved a classification accuracy of 88% for multi-class data which is higher than the results reported in the literature. Further the method managed to capture all non-classifiable records in the testing period and successfully updated the work order model.
We used 800 records of an institutional building located in Toronto over a period of 6 months (Jul/2020 – Jan/2021) to develop a model for key work order patterns. The result can be seen in Figure 1.
We then used 200 records of the same building in the 3 following months (Jan,2020 - March,2020). The method managed to classify the new work orders with an accuracy of 88% and updated the model with new set of categories. Figure 2 shows the updated model.