Machines break down, often at the least convenient time. But today’s forward-looking leaders, helped by torrents of data from IoT, are using predictive maintenance to get ahead of equipment breakdowns, and better attune productive capacity to market needs.
Definition, benefits of predictive maintenance
Modern predictive maintenance takes a proactive approach to equipment servicing using technologies such as IoT sensors and data analysis. Machine learning brings a level of prediction and accuracy to the effort that was not possible before.
“Predictive maintenance is the result of monitoring operational equipment and taking action to prevent potential downtime or an unexpected or negative outcome,” said Mike Leone, an analyst at IT strategy firm Enterprise Strategy Group.
For example, manufacturers can use sensor data and real-time analytics to uncover the overall equipment effectiveness (OEE) of machines used throughout the manufacturing process, Leone said. Manufacturers can gather information on issues such as current health levels, machine performance and location of hold ups in the production line.
One of the main benefits of predictive maintenance technology is that by taking action before equipment fails, organizations can mitigate downtime risk and ensure high levels of consistent operational efficiency and product quality, Leone said. For the business, that means better insight into internal process timelines, which leaders can then use to more accurately set customer expectations.
Mike LeoneAnalyst, Enterprise Strategy Group
Predictive maintenance relies heavily on data collection and analysis. And the speed at which that data can be collected and analyzed is critical.
In general, operational data is collected from equipment via sensors, Leone said. That data helps establish baselines for optimal or peak operation, and the technology team can use this to establish acceptable ranges for future operations.
Real-time data analysis comes next, Leone said. The technology team can compare equipment data to the established and acceptable ranges, and when something falls out of the band indicating a trend toward failure or downtime, alerts sound so the appropriate people can take action.
Predictive maintenance more achievable today
As with other elements of digital transformation, predictive maintenance is now in the realm of possibility for more of today’s organizations.
Two things have changed to make predictive maintenance more achievable than it was 10 to 15 years ago, said Felipe Parages, senior data scientist at Valkyrie, a data science consultancy. First, sensor technology has become much more widespread, and organizations can monitor factors such as temperature and pressure of industrial machines — from manufacturing equipment to freight trucks and locomotives — in real time. The other change is that in most cases, not only has the volume of data grown exponentially but all that new data is generally going to the cloud, which can often make it more accessible.
All these factors are having or will have a big organizational impact, Parages said. Before predictive maintenance, uncovering when a machine was operating suboptimally or when it was failing required the assessment of a skilled professional with a direct knowledge of the equipment and its operating principles.
“Nowadays, with the amount of data you can leverage and the new techniques based on machine learning and AI, it is possible to find patterns in all that data, things that are very subtle and would have escaped notice by a human being,” he said.
Thanks to that amplified power and the “always on” nature of IoT sensor technology, one person can now monitor hundreds of machines. In addition, companies may have accumulated several years of historical data, which can enable deeper trend analysis and detect patterns that people might miss.
Predictive maintenance “is a very powerful weapon,” Parages said.
Example of predictive maintenance
Predictive maintenance is not yet common, but there are many examples, including a promising one from Italy.
Italy’s primary rail operator, Trenitalia, adopted predictive maintenance for their high-speed trains, said Forrester analyst Paul Miller. With maintenance spending of around 1.3 billion Euros annually, Trenitalia is expecting to save 8% to 10% of that amount through predictive maintenance, according to the Forrester report, “IoT Transforms A 200-Year-Old Industry.”
“They can eliminate unplanned failures which often provide direct savings in maintenance but just as importantly, by taking a train out of service before it breaks — that means better customer service and happier customers,” Miller said.
Predictive maintenance strategy
Creating a successful predictive maintenance initiative is not easy, and business leaders and other stakeholders need to focus on a proactive predictive maintenance strategy.
“A data scientist might make a compelling case that predictive maintenance technology can save money and improve reliability,” said Chris Bergh, CEO of DataKitchen, Inc., a DataOps consultancy and platform provider.
However, the model itself is only a fraction of the overall machine learning system, Bergh said. Moving a model from development into operations can involve further tasks like provisioning infrastructure, installing and configuring software, preparing data, and testing both code and data.
In other words, all the strategic factors that go into making a project successful — or not.
“Enterprises rarely [take the time to] talk about these challenging tasks upfront,” he said.
A number of software packages incorporate specific domain expertise, such as packages targeted at aircraft operation and oil field operations, Miller said. But getting started with predictive maintenance isn’t a matter of just flicking a switch.
Those embarking on the predictive path should start with a pilot project. You will need sensors and an IoT rollout of some sort and you will need data, he said.
“Just having a machine learning model doesn’t help on its own unless you understand what’s normal,” Miller said. “Models can break down if look only at the data.”
Domain knowledge is critical as well.
“[For example,] if an organization is using predictive on a wind turbine, they need to know how carbon fiber behaves so they can assess blade wear properly,” Miller said. “If they just look at the sensor data, they don’t really know what it means.”
Combining domain knowledge with predictive maintenance involves a process of learning.
“Like any other machine learning job, you probably have a learning or training set to start and then you run that against a test set to build a credible model, but you must continue to refine that over time,” Miller said.
Ultimately, there are no magic bullets.
The trick is building the right combination of computing and machine learning to match the specific needs of a company or industry, Miller said.