Simulation method enhances wind turbine reliability testing efficiency without compromising accuracy

Wind power, a key source of renewable energy, relies on large turbines to generate electricity. When designing and maintaining turbines, reliability testing helps engineers prevent dangerous system failures, like a rotor breaking under stress and dropping a blade. A research team led by the University of Michigan developed a method that has the potential to make virtual testing of system components for turbines—and other large-scale structures—less expensive and more accessible.

With limited testing facilities available, the typical physical testing process for large turbine components can be time-consuming and expensive. Digital simulations, like those developed by the National Renewable Energy Laboratory (NREL), provide a more accessible alternative while still generating crucial data. Specifically, stochastic simulations—a simulation type that can handle random changes in variables like wind speed—are crucial to ensuring wind turbine reliability.

However, digital reliability tests using models like these still require considerable time and computational resources. The new method, called “optimization-guided and tree-based stratified sampling” or OptiTreeStrat for short, improves model efficiency to make digital testing less resource-intensive, without sacrificing accuracy.

“Our approach successfully recognizes important variables that impact system reliability, and decides effective test conditions to save digital test time,” said Eunshin Byon, a professor of industrial and operations engineering at U-M and corresponding author of the study published in Technometrics.

When analyzing system performance, too much variance in the data can reduce how precise a simulation can be. Stratified sampling is one key method used to reduce overall data variance, by prioritizing the most important data and leaving out information less critical to the model. In addition to improving model precision, this helps to cut down the time and resources needed to run the simulation.

This type of sampling works by dividing model input into subsets called strata, and then taking samples from each stratum. By drawing on new algorithms that identify critical variables and then using these to optimally design strata, OptiTreeStrat significantly reduces estimation variance in these digital simulations, lessening the computational burden.

While effective in principle, stratified sampling isn’t scalable—in other words, it’s not capable of expanding to accommodate larger workloads for high-dimensional problems. OptiTreeStrat, however, is highly scalable because it deals with variables one by one without considering more complex functions.

Additionally, while the study was motivated by a need to evaluate wind turbine reliability using digital modeling, this method can be readily applied in other contexts.

“We demonstrate the effectiveness of the proposed approach using wind turbines, but it can potentially be applied to any large-scale structures, such as bridges,” said Jaeshin Park, a doctoral student of industrial and operations engineering at U-M and lead author of the study.

Methods like OptiTreeStrat may be key to the more widespread use of well-designed virtual testing, allowing physical tests to be reserved for the final stages of prototype development. Allocating testing resources this way could notably reduce the overall costs of developing wind turbines, paving the way for more wind power.

Pohang University of Science and Technology and North Carolina State University also contributed to this research.

Additional co-authors: Young Myoung Ko of Pohang University of Science and Technology, and Sara Shashaani of North Carolina State University.