US Offshore Wind

Free online virtual conference

Using AI and Neural Networks to Reduce OW Costs


Advanced computer techniques allied to sensors can forecast stresses on components and failure rates, helping O & M to predict outages and reduce maintenance bills for offshore wind farms.

Wind turbines are huge pieces of equipment and normally are installed in locations characterized by powerful winds to exploit the best energy potential. Regular on-site inspection and preventative maintenance of these arrays are required to keep the plant operating productively. As well as regular maintenance duties, unforeseen electrical and mechanical failures can cause prospective breakdowns and outages, leading to machine downtimes and loss of energy output. All of these can be an obstacle to the financial performance of the wind turbine array.

Offshore wind turbines in ocean locations are hard to reach and may pose problems for maintenance crews. The cost of repairs is a significant pressure on turbine operating companies. The emerging solution is to utilize remote monitoring and diagnostics together with AI and neural networks so the computer system can learn and predict failures, thus ensuring that problems are pinpointed and sorted out before they become a critical issue.

Wind turbines have generated a massive amount of data over the years and companies have also accumulated much-needed records of weather conditions. Sensors deployed on OW equipment routinely capture data regarding the direction and speed of the wind, temperatures, electric currents and voltages, as well as vibrations produced by major components such as the generator and the rotor blades. This data coupled with advanced computing techniques promises to make systems better, more reliable, more efficient, and thus more profitable.


Machine Learning and  Neural Networks


Volkmar Sterzing is a machine learning specialist at Siemens Corporate Technology. "Previously sensor parameters were used only for remote maintenance and service diagnostics. But now they are also helping wind turbines generate more electricity," says Sterzing. By developing neural networks, researchers at Siemens spent four years analyzing and modeling various dependencies and interrelationships. Neural networks are the key to successful machine learning in wind turbines. "Neural networks are computer models whose operations are similar to those of the human brain," explains Sterzing. They learn from examples, recognize patterns, and use past measurement data to make forecasts and ideal models regarding the future behavior of complex systems.

This is particularly applicable to wind turbines. On the basis of past measurement data, software can calculates optimal settings for various weather scenarios that involve a variety of factors such as sunshine duration, hazy conditions, and thunderstorms. The data is transmitted to the wind turbines' control units, which take it into account from then on as they adjust the functions. If familiar wind conditions arise, the control units immediately use the optimal settings that were ascertained as a result of machine learning. This can result in the adjustment of rotor blade angles, for example. "As a result, turbines become more and more efficient and produce more energy," says Sterzing.

Results of DNV-GL tests measuring real and predicted loads

Machine learning systems usually rely on two main data sources: inputs and outputs. There are two approaches to building a useful application which can be utilized in reality:


  • Supervised learning: The algorithm is trained with example inputs and associated desired outputs and the goal is to learn a general rule that maps inputs to outputs

  • Unsupervised learning: In this scenario, labels are not provided to the learning algorithm, meaning it discovers own classification to find structures in the inputs. Unsupervised learning is mainly used for discovering hidden patterns in data.


In the case of wind turbines, supervised learning is the most appropriate as historic sensor and corresponding fault occurrence data can be used as predictors and the responses can be used to train fault detection algorithms.

Usually one turbine in an array is used to provide the sensor data and then this is extrapolated by the machine learning system across all the turbines. DNV GL has conducted numerous tests on operating turbines over time. The results show that with only 30 days of data gathered from a wind turbine measurement campaign, the prediction has an error close to 3% compared to the measurement campaign itself. This can be extrapolated to the whole turbine fleet in the array, saving costs and time.

The wind industry will be using machine learning a great deal in the future to understand and efficiently manage the operations and maintenance of a wind farm. These tools can be combined with real wind farm measurements to ensure optimal power production, extend the life of the equipment and determine the optimal use of maintenance and repair over the whole wind farm's life cycle.

Image: Visualization of Neural Network, Wikimedia Commons


The US offshore wind market is probably the most rapidly-developing renewable energy sector in the world.

Join us at the #USOW20 Virtual Conference - Your Gateway into the U.S. Offshore Wind Industry - June 18-19, 2020, Eastern Time. Free to Register

US Offshore Wind 2020 Virtual Conference

The premier conference for businesses that are looking to invest, find partners or secure contracts in upcoming U.S. offshore wind projects. With all the major decision-makers present, this is the most compelling event for anyone wanting in-depth insight into the expanding U.S. offshore wind market. Click here to find out more.


By Julian Jackson – writer on technology, arts, blockchain and cryptocurrencies

View More Articles

Download the Brochure Banner

Download the Brochure Banner