Aggreko powers predictive maintenance with Azure machine learning

Power generator manufacturer Aggreko has begun using Microsoft Azure machine learning to support technicians at its remote monitoring centre in Louisiana, US.

The Glasgow-headquartered company makes power and cooling equipment that is often used to supply temporary power at major sporting and music events and to support power generation in the developing world.

Its rental business has been progressing on a path from reactive maintenance to proactive, and now predictive maintenance to increase the uptime of its customers’ power generation equipment.

Steven Faull, applications development manager at Aggreko, said the company collects telemetry data from 40 control points on its generators.

The data is processed at Aggreko’s global technology centre in Glasgow and passed to the data analytics team at its remote operations centre in Louisiana.

“We use Azure to surface telemetry to remote operations for the proactive monitoring of generators in the field,” said Faull.

The system was initially available only in North America, but Azure’s global reach has enabled the service to be available globally.

Aggreko’s technology team is now enhancing the proactive monitoring service. “Our team is good at scanning for new technology,” said Faull, who added that machine learning is one of the areas now being developed.

“We looked at taking data from telemetry, to move from proactive to predictive maintenance,” he said. This would enable the technology centre to provide advance notice of any potential problems to the generators’ remote operations centre (ROC). “We could alert the ROC that there is an 85% probability of a problem due to high temperature alarms,” Faull added.

Aggreko worked with Microsoft’s services arm to build a microservices and a microfabric architecture for its generator telemetry application. “This allows us to build services we can reuse multiple times,” said Faull.

For example, he said, Aggreko can adapt the application deployed at its remote operations centre to build a portal for customers and an IoS and Android mobile customer app, built on open source .net platform Xaramin.

Aggreko has also partnered with Microsoft and the University of Strathclyde to gain expertise to develop machine learning models for proactive monitoring applications. “We signed up to a knowledge transfer programme with Strathclyde,” said Faull.

The company has taken a steady approach to rolling out machine learning, despite business pressures to go faster. “There has been a real push from the remote operations centre to solve every problem at once, but we have taken a structured approach and have built models for specific fault types,” said Faull.

One of the generator problems that machine learning is now being applied to is overheating, he said. “We can get a two-week warning on high temperatures in the generators, which avoids downtime.”

There are many more areas where machine learning could be used in Aggreko’s business, said Faull, adding: “It is not just for reliability. There are massive opportunities in other parts of business, such as to target prospects or future product development.”

Although the company has worked with Microsoft and Strathclyde University on the machine learning platform that now supports its rental business, Aggreko is open to using other products where they can do the job better, said Faull.

While the ROC was being set up in 2016, Faull visited GE’s Atlanta remote monitoring centre to benchmark Aggreko’s facility, because GE was among the early adopters of remote telemetry with Predix, a platform that provides predictive analytics.

Faull expects such technology to be used in certain parts of Aggreko’s business, especially given that it recently formed a partnership with GE. But Aggreko also plans to build out its own predictive systems. “We are building capability internally, but we are looking to take advantage of best in class when we can,” said Faull.