Carnegie Mellon University
May 15, 2015

Assessment Process Improves Wind Turbine Maintenance

Assessment Process Improves Wind Turbine Maintenance

The operation of wind farms is an important consideration in the development of more sustainable forms of energy. The costs of operating and maintaining a wind farm, which account for 25-30% of the life cycle costs of the facility, can keep it from being a competitive option. This is because wind turbines are made up of highly instrumented electro-mechanical components that are subject to fatigue-induced degradation and aging.

Carnegie Mellon researchers—Assistant Professor of Civil and Environmental Professor Matteo Pozzi, Assistant Professor of Computer Science Zico Kolter, and CEE doctoral student Milad Memarzadeh—are working with Pennsylvania-headquartered EverPower Wind Holdings to overcome this challenge.

EverPower Wind Holdings is a developer, owner, and operator of utility grade wind projects with four wind holdings in Pennsylvania. The team focused their research on the company’s Highland Wind Farm in Cambria County, PA, which operates 25 Nordex N-90 wind turbine generators that create enough electricity to power over 15,000 homes annually.

They have developed an assessment tool to process the monitoring data of a wind farm, at the system level, that detects anomalies and predicts the residual life of the components that are used in wind turbines. The overall goal is to eventually reduce the overall costs of manufacturing, operations, and maintenance. Their approach is based on probabilistic analysis using a Dynamic Bayesian Networks probability model and also using a partially observable Markov decision process.

“Being able to predict the residual life of mechanical components and to detect damages is of key relevance for wind farm operation,” explained EverPower Regional Asset Manager Kevin Wigell. “Down-time due to malfunctioning is a cause of relevant economic opportunity losses in the wind power industry. The Carnegie Mellon research team is helping us improve the effectiveness of our detection process.”

They focused specifically on developing a model that uses monitoring data about the turbine gearbox and yaw system (the component that helps orient the wind turbine rotor toward the wind), both of which have failure problems.

 “We intentionally focused on wind farm systems because they are comprised of very similar components, so their information can be processed at the system level,” described Pozzi. “This allows us to now work on adapting the process we have created for assessing the gearbox to also assess other turbine parts.”

Article reprinted with permission by PITA.