Grid-Scale, Predictive Maintenance Can Be Done Better

This article is part of our exclusive IEEE Journal Watch series in partnership with IEEE Xplore.

Energy grids are about to get a lot more complex as renewable energy is integrated into the system. To support these more dynamic grids, researchers have proposed a novel, predictive maintenance system that anticipates when critical components—in this case switchgears—may fail and need to be replaced.

Currently, distribution grid operators wait until a component of the grid fails before replacing it. Or, they may replace components at set intervals, when the lifespan of the technology is estimated to be at its end. However, this approach can lead to unnecessary replacements of still-functional components as well as components operating beyond their ideal “expiration dates”—and, in either extreme, unnecessary difficulties and higher costs.

The researchers’ prototype system, called DigiGrid, analyzes data capturing the general operations of the grid. For example, it tracks when a switchgear is installed, and then DigiGrid follows geographic information systems (GIS) data on the location of the equipment. Also, DigiGrid can analyze data from a range of sensors, including those to determine the flow of electricity, thermal sensors to identify overheating, air-quality sensors to determine contamination levels (for example, from dust or soot), and camera sensors to identify animal intruders that may have damaged equipment.

“DigiGrid utilizes this information, including sensor data, to calculate a current health status and a prediction of how long the asset will function without errors,” says Philipp zur Heiden, a postdoctoral researcher at Paderborn University, in Germany.

Testing such a system in real-world settings would be challenging, zur Heiden says, so instead the researchers surveyed six grid-distribution operators and studied their willingness to pay for smarter grid components.

“Our willingness-to-pay analysis showed that distribution grid operators are willing to spend money on switchgears as smart products,” says zur Heiden. He adds that the researchers found that grid components that can reliably report back their current condition were particularly valued. By contrast, he says, grid operators did not prioritize nearly as much those components that might attempt to then predict future errors.

Since developing the DigiGrid prototype, zur Heiden says he has teamed up with new partners to use AI to assist distribution-grid mechanics with their maintenance tasks, as part of a project called AProSys.

Predictive maintenance technologies such as DigiGrid and AProSys, zur Heiden says, will become increasingly valuable as increasingly diverse and renewably powered grid systems become more complicated in the future. Such predictive systems, he adds, reduce unnecessary costs, personnel activities, and downtimes for grids.

Moreover, he adds, renewables also increase the degree of bidirectionality in a grid, with multiple different sources and voltages flowing in different directions, at different times of the day. “Thus, central components of the distribution grid will be used more and need to be more often maintained,” he says.

The researchers published the results of their study last month in IEEE Transactions on Engineering Management.