Artificial Intelligence Offers “Predictive” Monitoring


University of New Mexico researchers (left to right) Hongbo He, Andrea Mammoli, Dave Menicucci and Tom Caudell have demonstrated the Solar Hot Water Reliability Testbed. Not pictured is Robert Edgar, of Sandia National Laboratories.

The University of New Mexico (UNM) and Sandia National Laboratories are collaborating in ground-breaking research to apply artificial intelligence techniques to improve the reliability of renewable energy generators, a growing and critical need for American utilities.

The goal is to provide smoother balancing of power sources with loads, as intermittent sources (wind, solar and back-up generators) cycle in and out of the grid — plus early warning of status changes for all sources.

When a small commercial or residential system with a backup generator suffers a mechanical or electrical failure, the backup power can mask the fault condition. If the owner is not carefully monitoring the renewable system, it may remain off-line for a long period.

Utilities are also concerned. They want to know how long they can rely on these renewable systems to operate because if the small generators fail, they have a legal duty to supply energy.

Artificial intelligence techniques, such as adaptive resonance theory (ART), hold promise to address these issues. ART algorithms have the rudimentary capability to learn in the same manner as living creatures. For example, ART can be exposed to a solar generator to learn its normal variability. This is called “training” and is similar to the training that humans receive prior to operating equipment. When ART encounters any unusual operational variability, it can flag that condition, just as would a human.

UNM and Sandia researchers are experimenting with ART on renewable generators. They have built the Solar Hot Water Reliability Testbed (SHWRT). It contains a solar hot water generator, similar to ones installed on homes, but with extensive controls and sensors. In testing, ART has been able to identify when a pump is beginning to fail, recognizing the system characteristics leading to the failure.

Predicting component failures with computer algorithms is extremely difficult. What sets ART apart from other methods of fault detection is that it uses only those sensors that are normally available to control the equipment. Thus, ART algorithms can be easily programmed into modern commercial controllers with few or no hardware changes that might increase manufacturing costs.

ART can also be configured to continue self-training after installation. The ART system can tailor its knowledge about a particular system and make it better able to identify and predict failures in the future.

ART technology represents a substantial advance in how renewable systems can be controlled and monitored. Its application is broad, and the next stage of testing will be on a much more complex generator, the solar absorption testbed that is used to partially heat and cool UNM’s mechanical engineering building. ART technology is also being proposed for use on microgrids that include electrical generators, such as microturbines and fuel cells.


Share on Facebook0Tweet about this on TwitterShare on LinkedIn0Pin on Pinterest0Email this to someone

Have a Thought to Share?

Your email address will not be published. Required fields are marked *