IIOT Power

Real-Time Wind Turbine Monitoring: Data Challenges, and Rewards

With the advantages of clean, renewable energy, wind power has become one of the fastest-growing energy sources and one of the most economical solutions for electricity generation.

There are already 350,000 wind turbines installed globally, with more than 650,000 MW of  installed generation capacity. In light of climate change, demand continues to grow. However, to remain competitive, the industry needs to operate as efficiently as possible.

According to the Institute for Energy Research, the cost of onshore wind power is already low compared to other energy sources at $45 to $92 per megawatt hour. Although this is quite reasonable, operators are looking for ways to further reduce costs to move into an era in which wind energy is no longer subsidized.

Life Expectancy of a Wind Turbine

Wind turbines are generally required to last 20 to 25 years, with failure rates that vary over that life span. Failure rates are higher in the first three years. Sometimes this is due to working through issues with a new turbine model, or perhaps adjustments need to be made to operating controls.

Once the initial “break-in” period occurs, wind turbines typically run reliably for the next 15 years. After that, wind turbines start to wear out and failure rates climb. However, if operators can manage to prolong the useful life of a turbine, it’s valuable from a financial perspective. In all three stages, it’s imperative to reduce failure rates.

Obviously, wind turbines are complex pieces of machinery, with electrical components, sensors, hydraulics, yaw motors, rotor blades, mechanical brakes, gearboxes, generators, and so on. Failures with any of these components cause varying durations of downtime. A gearbox failure, for example, doesn’t happen often but causes on average six days of downtime—a costly proposition.

Electrical failures are more common, but the duration of downtime is much shorter. Operators mainly focus on preventing gearbox, generator, and drivetrain failures, because they cause the longest periods of downtime.

Adding to the challenge, every wind turbine is different. Each is in a unique site, and most have different types and brands of components.

Failures Lead to Downtime—and High Costs

To optimize turbine performance, operators must conduct both planned and unplanned maintenance. Regardless, the more time technicians spend maintaining the turbine, the more expenses are incurred. The key is to be able to predict failures before they occur. Perhaps the oil in the gearbox is heating up or the gearbox is vibrating, which could signal that a failure is about to happen. Instead of installing additional hardware sensors on various parts of the turbine, operators try to use software as much as possible because additional hardware components tend to be more expensive and labor-intensive.

Turbine maintenance is by far the most complex and costly aspect of wind energy production. Global onshore wind operations and maintenance costs reached nearly $15 billion in 2019. Of that number, $8.5 billion was spent on unplanned repairs and correctives caused by component failures, according to new research by Wood Mackenzie Power & Renewables. This has sharpened the focus on operational expenditures for wind power plants and asset owners are searching for new solutions to improve their bottom lines.

Reducing Costs, Improving Performance

Turbit Systems recognizes the potential to apply machine learning (ML) to reduce the cost and complexity of operation and maintenance for operators in the wind energy industry. This is just one example of how machine learning and real-time data analysis can be applied in the energy industry. Other applications include optimized energy production, weather forecasting, building-energy management, demand response, and predictive maintenance for equipment, among others.  

For example, utilities, grid operators, transmission companies, and other electric industry entities have long been users of predictive analytics to improve operations, reduce downtime, and maximize revenues. Recently, a perfect storm of conditions is allowing them to take these efforts to a much-higher level. The wide-scale use of the Internet of Things (IoT) sensors, stream processing technologies, and supercharged analytics based on artificial intelligence (AI) methods is allowing energy producers and providers to boost efficiencies and better meet customer and regulatory demands.

Turbit develops condition monitoring software for wind turbines, such as these at the Cedar Creek Wind Farm in Colorado, that is specially designed to detect events such as unwanted shutdowns, throttlings, and other power losses. Source: POWER archives

External shutdowns by the direct marketer or the grid operator are not immediately logged in the status messages of the wind turbines. By monitoring with Turbit, operators can directly monitor and account for yield losses. Turbit learns and predicts the normal power output and deviations can easily be detected.

Even long-term changes are detected. For example, a change in the manufacturer’s software parameters, or nacelle and pitch malpositions, can lead to performance losses. Long-term changes in performance are detected and reported.

Applying Machine Learning, Real-Time Analytics

Turbit develops condition monitoring software for wind turbines that is specially designed to detect events such as unwanted shutdowns, throttlings, and other power losses. For each turbine and for each site, performance behavior is trained from the historical SCADA (supervisory control and data acquisition) data. In particular, weather data such as air pressure, temperature, wind direction and turbulence intensity are included in the calculation of the models. With the help of these models, performance is predicted with an accuracy of more than 99%.

With all of the variables that contribute to wind-farm energy output, turbine maintenance is a data-intensive endeavor. Wind turbines generate a large amount of SCADA data, outputting values for up to 500 different metrics each second. Turbine SCADA metrics include energy output, weather conditions, temperatures, nacelle positions, blade angles, gearbox and generator accelerations, fault codes, and other system statuses and control values.

The industry standard is to aggregate turbine SCADA data at the source and report 10-minute minimums, maximums, averages, and standard deviations. Turbit built its original machine-learning system to work at the industry-standard pace, collecting 500 data points per 10 minutes per wind turbine in a database, and re-running the machine learning algorithms on an hourly or daily basis.

In 2019, Turbit began work on a real-time version of the system. It is capable of creating a learning-feedback loop that feeds machine-learning outputs back into the turbine control algorithms to continuously improve turbine behavior.

Increasing Output Up to 5%

This new software runs machine-learning algorithms against the turbine-generated SCADA data in real time, comparing measured data with expected behavior. Relying on a large database of failure events, Turbit can automatically detect the slightest technical fault and recommend corrective measures. This insight helps wind farm operators and service providers avoid costs that are related to damage or operational stops and can improve wind-farm energy output by up to 5%. For instance, real-time condition monitoring and control offers the ability to detect a gust of wind at one end of a wind farm, and precisely adjust the position of the turbines to capitalize on that gust before it reaches the other end.

However, once data was being monitored in real time, there was the need for a better way to manage data. Here are some of the main reasons why real-time condition monitoring on a second or sub-second level is an extremely challenging data management task:

●          150x more data. Real-time monitoring collects 30,000 data points per minute per turbine, or 150x more data than processing 10-minute aggregates requires.

●          300x faster processing. Algorithms must run within two seconds, a 300x smaller response-time window than typical 10-minute cycles.

●          Data quality. Missing data gaps arising from data lag and connectivity problems (e.g., from a turbine with a bad internet connection) must be detected and filled on the fly in order to avoid skewed insights.

●          Hundreds of customers, thousands of turbines. Turbit Systems needs to provide fast analytics to many concurrent users looking at data generated by thousands of wind turbines.

Overcoming a Data Bottleneck

The previous database was not able to meet the real-time data requirements. Turbit looked at options; we knew we did not want an expensive, proprietary database. We decided to try free, open source PostgreSQL accelerated by Swarm64 DA, a performance-acceleration software solution that enhances PostgreSQL with better parallel processing, data compression, I/O reduction, and, if present, support for FPGA coprocessors.

The acceleration enables PostgreSQL to analyze data orders of magnitude faster than usual, even while data is streaming into the database. In less than two months, Turbit was able to develop a working prototype of our real-time, condition-monitoring system.

When the performance of this configuration was compared to the previous database, we found that in many cases we achieved a lot better performance. There were sub-second responses, even as the number of concurrent users increased from one to 30. This is essential for real-time control of turbine performance. As the number of concurrent clients increased, the performance of advantage increased from 2x to 6x. There was also more consistent performance at scale. Now, if a turbine manufacturer needs real-time data for a turbine control system, we can say “Yes, no problem.”

More Data, Faster Answers

Turbines are often located in remote areas and receiving a steady stream of SCADA data from them can be unpredictable. As data streams in, Turbit can detect missing data gaps, and raise connectivity issue alerts to system operators.

Answers are now developed in seconds, giving our data scientists the ability to test new algorithms more quickly and against larger amounts of data. It’s gratifying to be able to help wind-farm operators and service providers observe the performance of their wind turbines in order to maximize energy output and minimize turbine wear and repair costs.

The energy industry has an enormous opportunity to transform itself through technologies such as ML, the IoT, and supercharged data processing. These new technologies are critical to bringing utilities and energy operators into today’s digital era and establishing a more affordable, reliable, accessible, and sustainable infrastructure.

Michael Tegtmeier is CEO of Turbit Systems, a company headquartered in Berlin, Germany, that monitors the performance of wind farms.

SHARE this article