By Josh Fausset, Director of Customer Success at Sentient Science
As wind energy continues to grow and prosper around the globe, the price for energy has been drastically decreasing. In response, wind operators have been investing in systems to predict failures in their fleets to help lower their cost of producing electricity.
Condition based monitoring is critical if you don’t have an earlier stage of detection, or at best some way to peg a failure before incipient damage with engineered models. For sensor-based warning, vibration seems to be the best technology currently available, however, some failure modes can still go unnoticed. While axial cracks, cage breaks, fluting, and gear tooth liberations are highly detectible, incipient stage pitting from rolling contact fatigue is not. Many would argue that low speed rotation (main bearing) vibration signals are often improperly instrumented or too weak to rely on for early detection. I have personally seen main bearing vibration signals that do work, but I know some in the industry have different experiences. By the time main bearings have reached temperature thresholds, enough to throw faults or the grease is showing glitter or metal particulate, the main bearing damage is beyond mitigation or salvaging with a grease purge or flush. In gearboxes, metal particulate in the oil filter is a huge red flag for technicians that a subcomponent is experiencing excessive wear and/or spalling, but it’s often late in the damage stage.
What operators have learned is that they need to take a comprehensive approach to failure planning and analysis. We cannot rely on one dataset, vibration, temperature, lubricant characteristic, or otherwise to tell the whole story. Comparing all the data on a common timeline of events is important to understand existing failure in the attempt to understand how that impacts the components at a material level; not just applying probabilities given some pattern, but understand the literal “wear” that happens. They need to look at the SCADA, oil reports, accelerometer data, maintenance activities, etc. to make decisions on how to operate and maintain their turbines. They rely on the data to predict when failures might occur to prevent early failure from propagating to late failure stages.
As technology evolves and enhancements are made to the subcomponents, the market dynamics are shifting. Operators and Original Equipment Manufacturers are digitalizing their wind fleets and investing in software that enhances their ability to make strategic business decisions that extend the life of their wind turbines, to better plan their supply and demand needs and to optimize where they focus investment in their fleets. Taking the time and energy to learn the stories the wind turbines tell through correlating data sets is the only way, as far as I can tell, to develop the patterns that preempt failure and how best to prevent them. And not just at the macro-level, but looking at the small stuff. If our hope and expectation is to predict and prevent at a small level, we have to be looking small. The small stuff matters. And the more granular we get with more sets of data, the more complex our analyses become. But how can we expect anything less when we are trying to model reality?
For more information, visit sentientscience.com