How to teach a computer how to ensure the quality of wind turbine blades?

The AE module (assisted evaluation) is a result of the GTS institute FORCE Technology’s work with the RK project “Automatic evaluation of complex sensor data”. Among other things, the project has worked with quality assurance of composite materials, and here the wind turbine industry is a major consumer of this for their wings.

There is currently no standardised quality assurance of wind turbine blades.

The wind turbine industry has a desire to automate the quality assurance of wind turbine blades. The challenge is that there are no common standards for the quality assurance of wings. Danish wind turbine manufacturers are at the forefront of development, where the wings are constantly getting longer, and the production methods are still partly manual.

The wind turbine manufacturers therefore use automated ultrasound for quality assurance and thereby check the blades for many different types of defects. What defects are there? Where are the defects located and whether the defects are so serious that they need to be repaired now? Or can the repairs wait?

Bottlenecks in production can be avoided

A wind turbine blade that is 100 m long naturally takes a long time to quality-ensure if it is done manually. The production can end up in a bottleneck.

FORCE Technology has therefore for a number of years developed automated ultrasonic solutions for wind turbine blades, which are fast and provide high-resolution images. Then an inspector has to review and evaluate the large amounts of data manually.

What if one could minimise the time for the evaluation considerably and still achieve the same result? It will soon be an option with a new AE software module.

Man vs. machine (learning)

“Today we have inspectors with experience and certification to evaluate collected data. It is not a simple task to transfer their knowledge to an algorithm, and it is necessary to have a method to validate that the computer is able to handle the task,” says Steen Arnfred, project manager and R&D coordinator at FORCE Technology.

“It is a prerequisite for all use of artificial intelligence and Machine Learning (ML) that algorithms have been trained and validated. Validation of the algorithms is a very important factor in the efforts to move ML from assisted evaluation (AE) to fully automatic evaluation,” Steen Arnfred continues.

Currently, there are several commercial solutions for Machine Learning, but due to the complexity of the wings, it is ultimately the inspector who has to make the final and thus subjective decision.

Software module must learn to recognise defects

So how can one evaluate data collected with automated ultrasound in the future? With the new software module, the wind turbine manufacturer can omit large parts of the manual evaluation of collected sensor data, because the software module helps to check the condition of the blade objectively.

FORCE Technology has manufactured some composite laminates with artificially inlaid defects. Thus, we know in advance what defects are found and where they are in the laminate.

The focus of the RK project is the software module, which contains both components for reading NDT volume data and models for automatically evaluating the detected defects.

“The software module provides a structured data set, where data is segmented/annotated with the types of defects found. In other words, all types of defects are provided with comments and positions, so that the algorithm can learn to recognise defect types and sizes in future scans,” says Steen Arnfred.

Once the algorithm has located defects in the wing, a specialist can carry out the final verification of the defects. So, the crucial thing for the assisted evaluation is to get the defects and their locations described as accurately and clearly as possible. The RK project is currently working with three different types of defects in the wind turbine wing, but the software module can be easily developed to include several types of defects.

The digital future strengthens the green transition

“In the long run, digitalisation of the entire value chain is desired from initial component testing and production of blades until they are put into operation on a wind farm. Overall, it provides more efficient production and extends the life of the wind farm,” concludes Steen Arnfred from FORCE Technology.

Quality control is both improved and time-optimised within the wind turbine industry. The software module can eventually be used by all industries and their subcontractors that work with composite materials. Examples of this includes the aircraft, ship, and car industries.

This is an important step towards a more profitable green transition and increased digitalisation in the industry.