Imagine a self-learning or machine-learning system for fault detection. One where algorithms find patterns and correlations between a range of data sources to work out the most efficient order for fixing faults. A major data science research program on this by Buildings Alive has produced some promising results.
Over time, buildings have become complex creatures kept occupiable by an intricate web of systems – electricals, HVAC, security, lighting, energy management and more.
With greater complexity, the chance of system errors and failures increased, creating a burdensome workload for maintenance technicians and facility managers to manually track down issues and fix them.
This led to the rise of modern automated fault detection and diagnostics tools, which effectively automate the fault detecting capabilities of engineers and facility managers.
This software abides by rules written in accordance with existing engineering knowledge to identify abnormalities – when a room is too hot, or when a valve has clammed shut – and then alert the user.
“Essentially, in this setup, machines are doing what they do well – exactly what they are told to do by a human operator,” Buildings Alive chief executive officer Craig Roussac explains.
The problem with this “bottom-up” automated fault detection method, according to the building energy performance expert, is that it’s hard to prioritise faults in order of importance. And often important “outside the box” questions are overlooked.
“If you start from the bottom, you find a bucket of little faults, but you don’t necessarily see the complete picture.
“In all walks of life, the most effective people are strategic thinkers. They concentrate on their goals and what they need to do to achieve them. Sometimes they get drawn into the minutiae, but only when they know that’s the best use of their time. High-performance buildings need to be managed with the same purpose-driven mindset.”
There’s a better way: empowering smart people with smart computers
Hao Huang, senior data scientist and building systems engineer at Buildings Alive, has been looking into the shortcomings of modern fault detection for some time.
He’s spearheaded research at Buildings Alive in next generation fault detection that works from “top to bottom, rather than bottom up”.
Rather than programming in pre-established rules, the system zooms out on a building and acts as a “search engine” with machine learning algorithms finding patterns and correlations between a range of data sources, including weather data, interval metering data, and building management system data.
The result is a list of faults compiled in an order that allows facilities managers to see what benefits will be gained by fixing them, for example, how much energy will be saved or how air quality might be improved.
Research conducted by Huang and his colleagues at Buildings Alive pitted the “self-learning” method against common rule-based fault detection platforms used across a diverse range of office buildings in Australia over a period of 18 months, with promising results.
He says the new method consistently beat rules-based systems by detecting faults that were materially affecting energy performance and provoked fewer false alarms.
Perhaps most importantly, engineers and facility managers were happier when empowered with the machine-learning detection tool to make their own decisions about how to spend time and money.
The other benefit of the self-learning model is that it leverages whatever building data is available, meaning no labour-intensive set up.
A top shelf energy performance tool
Buildings Alive is now offering its self-learning building optimisation tool to its clients either as an alternative or “add-on” to more conventional fault detection technologies, which the company also deploys.
For the company, this system is a logical progression in its existing building optimisation service that empowers building managers to close the gap between a building’s efficiency “potential” and how it typically operates.
Roussac says once you have a good idea of how your building is performing, and how far it is from “perfection”, the next logical step is guidance on how to improve through sophisticated, targeted fault detection and diagnostics.
He says operators of high performing buildings are typically attracted to this offering because it unlocks those less obvious tuning opportunities that wouldn’t be picked up otherwise. These are the buildings that have had a taste of high performance and want to go the extra mile.