While data-driven predictive maintenance is rapidly becoming the norm, the next step is already available. Prescriptive maintenance delivers real-world benefits and proves both value and potential.
Data centers are increasingly complex and subject to an expanding list of environmental goals. Pinpointing the root cause of failure or inefficiencies often requires more time, experience, and skill than detecting such issues in the first place. Whether the issue lies in faulty hardware, simple wear and tear, or even the dynamics that make operations hard to predict, efficient data center operations depend on prompt diagnosis. Another challenge arises from the departure of seasoned professionals from the workforce, as directly replacing their skills and experience is seldom possible.
AI (Artificial Intelligence) has played an important role in partly alleviating these challenges. Sophisticated analysis can dramatically improve the detection rate of failures and can even warn of increased likelihood of impending failure. However, such predictive maintenance has limits, especially when drilling down on where failures actually originate. More importantly, it lacks any direction on the correct course of action. For that reason, we at Coolgradient have been developing and finetuning our AI models that enable prescriptive maintenance.
What is Prescriptive Maintenance?
While predictive maintenance can tell you that a component or system is about to fail, prescriptive maintenance offers direct, actionable insight into why something is failing. For prescriptive maintenance, equipment condition data are analyzed through sophisticated machine learning algorithms that offer recommendations and probable outcomes. While predictive maintenance can warn of impending failures, it doesn’t highlight the root cause or how to mitigate or prevent them. Prescriptive maintenance offers more comprehensive insights by predicting failures and providing direct, actionable recommendations. Prescriptive maintenance enables truly proactive measures to optimize asset performance and longevity.
Crucially, this goes beyond recommendations such as ‘replace this’ or ‘repair that’. Prescriptive maintenance also involves advice on how to adjust use to increase the longevity of your assets while retaining peak operating performance.
Prescriptive Maintenance in Practice
When we at Coolgradient started to work on our prescriptive machine learning models, the concept was mostly consigned to proof of concepts and pilots. But we’ve been running our models in production environments for several years now, and we’re seeing excellent results from our customers. To illustrate, here are three real-life scenarios we’ve seen of prescriptive maintenance in practice.
Filter Cleaning
When our model told one of our customers that a temperature anomaly over the buffer vessel in the drycooler chiller system was caused by filters needing maintenance, they were adamant that there had to be a mistake. After all, the filers were brand new. Still, the prescriptive analytics showed a clear temperature difference caused by back pressure.
Next to the recommendation to clean the filters, the underlying AI models also calculated the amount of energy wasted due to this issue alone. The mixing of hot and cold water resulted in additional chillers having to be turned on while others were working at a higher capacity. To top it off, the filters put additional strain on the pumps.
It turned out the prescriptive analytics were right: the filters were extremely dirty despite being used for a relatively short time. Making the adjustments ultimately saved this particular data center 75MWh per month.
Water nozzle maintenance
Like power, water is becoming a premium resource for data centers. WUE (Water Usage Effectiveness) is increasingly a key metric for data center sustainability, especially where adiabatic cooling designs are involved. Coolgradient has developed models to follow, predict, and remediate water usage issues. Water nozzles can get damaged or clogged for a variety of reasons, including vibrations, water pollution, hardness, and more. Historically, it has been a time-consuming manual process to detect such issues, let alone remediate them.
At one of our clients, our AI constantly validates the performance of every single nozzle within the adiabatically cooled data center environment, immediately pointing out any possible issue and giving recommendations for nozzle inspections. This is one way Coolgradient assists them in maintaining optimal adiabatic control and power usage and avoiding unnecessary water waste. For this customer, it resulted in a 22% reduction of water consumption.
Beyond the Specs
Most importantly, prescriptive maintenance demands a holistic tracking of technical assets and resources. While certain equipment has specifications, a host of factors influence whether it works according to those specifications or not. These include dynamic conditions, such as weather, but also static elements that are not immediately obvious.
One example of the latter is the placement of units relative to one another. If you have three cooling units next to each other placed on a roof, the two on the outside can reasonably be expected to perform according to specs. But what about the middle one? If that one must deal with the air coming from the neighboring cooler unit’s heat exchange, it will not perform the same. Our prescriptive models consider all contexts that impact actual system efficiency, providing an accurate prediction for energy consumption in all conditions.
By examining the role of individual components during actual use rather than vendor-provided specs based on theoretical conditions, engineers know when and how to react. In short, clients get maintenance recommendations based on what is actually happening, not on what is supposed to happen. The result is an immediate productivity increase.
Prescriptive Maintenance is Here
As data centers become more complex and skilled staff become scarcer, now is the time to leverage the power of AI to make your teams more efficient. Prescriptive maintenance surpasses predictive maintenance by directly improving resource usage and reducing troubleshooting time. It helps maximize efficiency and streamline operations.
Crucially, prescriptive analytics have proven themselves in the field, something that our engineers see daily. If you want to know more about how prescriptive maintenance can make your data centers more reliable and sustainable, please contact us: info@coolgradient.com. Our experts are always happy to discuss your needs and what Coolgradient can do to meet them.
