Illustrations by DAQ
The Toronto District School Board (TDSB) in Canada owns almost 600 schools, a large proportion of which are more than 40 years old. With a limited annual budget for asset repairs and maintenance ($312m in 2020-21) it faces an ongoing challenge to accurately prioritise and allocate funds.
Teams of external surveyors are deployed as part of a five-year cycle to monitor and inspect schools and individual components. They gauge the condition of the schools and pinpoint those most in need of funds for capital renewal.
However, the data contained in inspection reports typically lacks granularity and ‘critical’ priority is often assigned to many more assets than can be realistically funded. This forces asset managers on the TDSB to rely largely on subjective judgment to prioritise those most in need.
Keen to develop a more accurate and objective form of assessment, researchers at the University of Waterloo in Ontario worked with the TDSB. They created a data-driven tool, able to provide detailed and unbiased insights on how different assets should be assessed and ranked for renewal. A paper on their work, Data mining of school inspection reports to identify the assets with the top renewal priority, appears in the Journal of Building Engineering.
The prototype AI and data-mining solution focused on roofing. It provided a deeper understanding of inspection reports and a higher resolution analysis of the most urgent damages.
Surveyor inspection reports are an important first step in maintaining healthy school buildings in Toronto. External consultants periodically inspect buildings and submit individual school reports with data on the condition of all systems and subsystems.
Senior managers at TDSB use the reports, in combination with feedback received from sources including maintenance and operations teams and end users, to devise a three-year capital plan. This allocates a percentage of the total funds to each system, for example roofing, mechanical or electrical. Then managers decide which assets at which schools to replace or push back for future years.
The surveyors’ inspection data is contained in spreadsheets and typically only identifies the most ‘urgent’ or ‘high priority’ assets for renewal, explains Kareem Mostafa, research assistant at the University of Waterloo: “Giving everything the same treatment or the same assumption of criticality means that individuals in charge of the budget are forced to exercise personal judgment on which schools to fund, making it highly subjective and inconsistent.” Furthermore, it requires considerable time to analyse and categorise each of the 600 separate school inspection reports every five-year cycle, he adds.
The researchers applied rule-based text mining tools to dig deeper into the inspection data and more accurately and speedily distinguish the urgency of potential roofing defects.
Text mining is a class of data mining used to extract useful knowledge from unstructured textual data, such as full-text documents. The tools scanned the reports for keywords related to effects like damage or deterioration or the need for repairs – the frequency of which gives an indication of the severity of the roofing damage.
Data clustering techniques, that classify data into groups with similar attributes, were also applied to group roofs into different levels of severity, with the top level being the most eligible for funding.
Through this process, the researchers were able to identify the top 10% of schools in Toronto requiring immediate attention. If the budget was not sufficient to complete all the repair work, an additional optimisation algorithm provided an estimate, based on the money available, of which schools to target to improve the overall asset portfolio condition.
AI and advanced algorithms are disrupting a variety of processes across property and real estate. But the innovation of this methodology is, according to the researchers, the ability to overlay AI onto traditional school condition assessment.
“We're not changing the way inspections are being done or imposing a new method of data collection,” says Mostafa. “We’re using good old-fashioned reports and descriptions, but introducing a systematic and objective method for their analysis.”
Facilities management personnel at the TDSB “were satisfied with the results” generated using the model, says Mostafa, who is hopeful it will become a regular feature of the funding allocation process.
The potential economic benefits of more accurate budget allocation are matched by social benefits. The system provides an objective justification for why certain buildings should be repaired before others. This means school board managers need no longer defend their decisions to schools that missed out on funding.
Although the software was developed to assess the need for roof repairs, it could be tweaked. Adding different keywords and parameters would prioritise other kinds of work for organisations with budget limitations and many buildings to maintain.
Office personnel in charge of budgeting are the intended end users for the tool, but surveying firms could also benefit. Large numbers of condition reports run through the model could give better insights into commonly experienced defects. Analysis of surveys across an estate of properties could give an overview of the most urgent work and related costs.
The research by the University of Waterloo with TDSB represents the first phase of a multi-pronged research project with new funding in place.
"Once you’ve identified the top 10% of schools that urgently need work, an algorithm can analyse how to schedule the repairs between July and August” Kareem Mostafa, University of Waterloo
The plan is to develop a computer vision-based system, which uses algorithms to automatically analyse photos of assets. It assesses their condition and rehabilitation needs, further improving efficiency and consistency.
The AI/data mining tool is also likely to be expanded to cover other parts of a school, including parking lots and heating, ventilation and air conditioning (HVAC). “We're going to be fine-tuning our model to give it a different set of keywords to look into, then we’ll reapply it and check the results with the plant operations people to see if it’s working correctly,” says Mostafa.
A final avenue of research will expand the tool to enable it to prioritise schools based on a schedule of available time, not just available funding. Schools in Toronto, as in other regions in the northern hemisphere, typically want to carry out repairs in the summer when classrooms are empty and the weather is more suitable for building work.
“Once you’ve identified the top 10% of schools (maybe 50 or 60) that urgently need capital renewal, an algorithm can analyse how to schedule the repairs between July and August,” says Mostafa. “If the timeframe is too tight, it can work out a suitable number of projects, for example only the top 35.”
The ultimate ambition is to build an end-to-end software platform that can detect and monitor the most pressing repair issues. It can then create an optimised schedule for renewal based on available funding, time and resources, such as construction labour.