Internet of Things (IoT) technologies have created the opportunity to deploy new services and products in intelligent buildings, one of which is predictive maintenance. While the idea of predicting building maintenance needs isn’t necessarily new—even now, many technicians put their ear to a machine to detect ominous sound deviations—the concept of using data to gain an in-depth understanding of an asset’s health through a predictive lens is a relatively recent innovation that is steadily gaining ground.
The growth of the predictive maintenance market presents opportunities for both building managers and service providers, but there are still obstacles to overcome to achieve its optimal use. In this article, we talk to two experts, Olga Fink, Chair of Intelligent Maintenance Systems at ETH Zürich, and Brad Pilgrim, CEO of Parity Inc., an energy management as a service (EMaaS) provider that specializes in multifamily buildings, for their perspectives on predictive maintenance and where the industry is headed. This article also draws on information from a special webinar on predictive maintenance and AI held by CABA — the focus of its 2021 large-building research project.
Want access to valuable research on advanced building technologies? Check out the free research library produced by CABA.
What is predictive maintenance?
Predictive maintenance is best explained in comparison to two other types of maintenance: reactive and preventive maintenance.
Reactive maintenance uses a “break and fix” mentality: Something breaks and a phone call gets made; a replacement part is delivered. Preventive maintenance relies on regular inspections to help anticipate, identify, and address imminent problems. Ideally, the choice around which type of maintenance strategy to use should depend on the criticality of the component/system or the function that it fulfills.
In the absence of data, these methods make sense—why fix something if it’s not broken? (With the exception of safety-critical components, which require preventive maintenance sometimes by regulation.) Predictive maintenance completely changes the game. Instead of reacting to failures (or potential failures), predictive maintenance technology uses process data and advanced analytics to predict mechanical failures ahead of time.
“Generally speaking in multifamily buildings, the way maintenance is done currently is not based on operational data; it’s based on time,” says Pilgrim. Conversely, predictive maintenance involves collecting real-time data to evaluate the wear and tear on pieces of equipment. Having this kind of data enables decision‑makers to gain appropriate insights into how a piece of equipment is operating, how long it’s been operating, how the condition of the system has been evolving, and when to conduct maintenance activities. “It’s having the analytics and data insights that enable building owners to make better decisions around the maintenance and when the replacement of capital equipment is warranted.”
As someone whose research focuses on predictive maintenance technology for complex assets, Fink likens predictive maintenance to a healthcare system for business assets: “The idea is to develop algorithms to monitor the health condition of a system [the asset], detect a deviation from the ideal health condition, then diagnose it and find the root cause of the failure. Also, the goal is to predict how the condition of the asset will evolve over time and when the system’s health will reach a critical level. Then, we can even go one step further and start influencing and be proactive with the system condition, prescribing how the system should be operating. This is in fact prescriptive maintenance or prescriptive operation.”
According to Fink, the ingredients needed to perform those tasks (detect, diagnose, predict, and prescribe) are:
- Condition monitoring data (temperature, pressure, vibration, acoustic emissions, etc.) captured by different types of sensing devices
- AI algorithms that need to be developed and adjusted to solve for specific challenges
- Physics-based models, prior knowledge, and domain expertise to add on top of algorithms to help guide and support the learning process
As highlighted in the above-noted CABA webinar, historically the “solutions” to maintenance challenges have come in the form of repairs or other actions implemented by an OEM or other service provider, following their investigation into a support request. A lot of it is reactive, noted Harry Pascarella of Harbor Research during the webinar. To make it less reactive, preventive maintenance programs schedule maintenance activities regardless of whether the equipment is failing or not.
“Obviously these types of strategies leave a lot to be desired in terms of just-in time servicing or most efficient ways of servicing and reacting to equipment failure,” says Pascarella. “This is what AI can start to help with.
“What AI and machine leaning, as well as the other enabling technologies associated with the IoT and smart systems allow us to do, is to start to collect data in real time from machines and systems, analyze that data, and determine what to do, all automatically—driving towards no unplanned downtime and reducing planned downtime as much as possible.”
How does predictive maintenance technology improve asset management and optimization?
Maintenance teams and facilities managers play an important role in making a facility attractive to potential investors, buyers, and tenants; their ability to run a building efficiently directly affects the bottom line, as well as occupant satisfaction. As a previous condominium tenant, Pilgrim knows this firsthand; he started Parity in response to what he saw as a need to control operating costs, which represented more than half the budget and directly impacted his condominium fees.
When he looked into the issue, Pilgrim says, he found there weren’t any good technology solutions available for the multifamily market. “That’s what we set out to do—overcome the barriers of adoption, which were the need for upfront capital, clear and transparent data reporting and to create a product that actually delivered results and could be relied on as a service.”
Parity is primarily an AI-based energy management and control platform for multi-residential building HVAC systems. Part of what the company does is collect process-level data from each piece of machinery in the system. The data it collects from various types of equipment—cooling, heating, and electrical—could provide valuable insights with regard to predictive maintenance. Using the platform means you “have that data to understand how the system as a whole works, how different systems work together, and how they should be working over time. You can then establish baseline operating key performance indicators. Ultimately, that data will enable us to determine how pieces of equipment or processes start degrading over time.” Pilgrim’s team is also in the process of developing capabilities to help multifamily building owners better collaborate to meet their maintenance planning needs.
Recent years have seen a rise in the number of predictive maintenance solutions and services that are helping to support customers in making the shift from the old strategies to the new. Pilgrim believes that the digitization of maintenance is the first step.
“Maintenance right now has been in a very manual format in the multifamily sector. We still walk through a lot of HVAC rooms with plastic sleeves and paper record cards on equipment—these are hundred-thousand-dollar-plus pieces of equipment that have a recordkeeping system like an old library book. That’s where Parity is playing a big role—we’re collecting that process-level data, looking at key performance indicators on pieces of equipment, and keeping a digital record of them over time to show collaborators what the performance of the piece of equipment has been during its lifespan.”
For example, take an electric motor. If you can determine the key performance indicator for its degradation and measure it, you can eventually map the degradation of that asset over its performance life. Going forward, you would then be able to understand the motor’s useful asset life and know in more detail where each motor is in its life cycle, thereby better informing capital planning and having greater certainty in maintenance schedules.
What are some of the challenges for the predictive maintenance industry?
OEMs and service providers stand to benefit by being able to offer new types of data driven service opportunities—evolving into much more than simply help-desk and break/fix service providers. As software and service capabilities expand, they can enable tighter integration of equipment, software, and services across customer operations, offering more value and account controls.
Through advances in connectivity, and integration of historical part and failure data within predictive models, as well as changes in end-user preferences, solution providers can effectively share product ownership and maximize uptime, CABA’s recent webinar noted.
In short, industry participants are poised to access new revenue models and offer more to their customers by leveraging the value of asset health and systems knowledge, and achieving this shift is actually “the next big step for mission-critical equipment providers and their sub-system and component providers,” Pascarella highlighted in the CABA webinar.
For the world of building maintenance, this means that the arrival of broad, software-enabled asset management and uptime solutions has just begun. In the last few years, more building owners and managers have begun to show interest in the concept, which has been largely borne out by predictive maintenance statistics as determined by current case uses. (The statistics run far and wide, but as an example, one company reduced unplanned downtime of its machinery by up to 20%; another estimates that it will save approximately $34 million in unscheduled repairs over the course of five years.) But there are still challenges to overcome.
From a customer perspective, Pilgrim acknowledges that a lack of quantifiable results sometimes stands in the way of adoption. He feels there isn’t always a clear ROI behind predictive maintenance and that the services offered by some providers are costly yet produce no quantifiable results (or produce results that are very difficult to measure). Pilgrim says the Parity platform pays for itself through energy savings; the building is saving more in any given month than it’s paying for the services provided and the gains would be further quantified through predictive maintenance service features. The rich datasets that building owners get is an additional benefit.
On the technology side, Fink points out limitations with regard to data, specifically a lack of sufficient fault patterns to learn from and a lack of “time to failure” trajectories, which makes it difficult to predict the remaining useful life of any piece of equipment. “We also need to deal with the high diversity of the systems we’re actually monitoring—none are quite the same, even if they were produced by the same manufacturer. This prevents us from transferring models directly that we trained on one system to another that operates slightly differently,” she says.
To address those challenges, Fink develops AI algorithms that are robust to changes in operating conditions, as well as different degrees of noise. She is also working on AI models that can be transferred between different operating conditions and different units of a fleet. This enables her to develop models also for new assets that haven’t collected sufficient data to train a dedicated AI algorithm. Such approaches from the field of transfer learning enable a better scalability and generalizability of AI algorithms. Fink’s team demonstrated such transferability, for example, for a large fleet of gas turbines.
Fink believes that, in the future, systems are likely to be more interconnected, and predictive maintenance systems will need to evolve to match the increasing level of complexity. “We will no longer be able to consider systems in isolation because they will be interconnected, communicate with each other, and mutually influence one other.” She sees the need to take a system-of-systems approach that still maintains the ability to preserve information on the specificity of components but can combine information and aggregate it at different hierarchical levels.
With regard to the lack of data, one idea Fink has been advocating for in recent years is data sharing. If one stakeholder doesn’t have sufficient data, it’s likely that others don’t as well. “Why not start collaborating across company borders? If it’s not possible to share data directly, there are algorithms that can be trained in a distributed way, so every one of the stakeholders can start benefiting from the data across company borders.
“In the economy of things, sharing means dividing. But in the economy of ideas and data, sharing means multiplying. So let’s start sharing, and let’s start multiplying.”
Looking for an authoritative source of information about other intelligent building technologies?
To learn more about the intelligent building technologies market, visit the CABA website. CABA is the leading international association for the intelligent buildings market. Our nearly 400 member companies are actively involved in designing, manufacturing, installing, and retailing of products related to intelligent buildings and contribute to our growing library of in-depth, specialized research and industry trends. Check out our extensive free resources, or become a CABA member today to access even more valuable insider information.