High-voltage transmission towers are installed in a wide range of locations and environments and are exposed to all possible weather conditions. Due to this, transmission towers are prone to corrosion and can degrade over time. ​ Energinet needed a solution to effectively spot and quantify the type and level of rust on its transmission towers and to be able to detect changes over time.

In order to match Energinet’s specific needs, customized AI models were developed and trained. Thirty-six annotated pictures were initially made available by Energinet to train the models. The dataset was further enhanced by augmented data techniques and the use of synthetic images from our simulator. The resulting training dataset consisted of 285,025 examples of rust. ​ Our customized model was able to detect rust with high accuracy, quantify the amount of rust on the pictures and easily visualize it for Energinet’s experts.

• Identified rust occurrences in a large dataset of images with 96% accuracy and in a matter of minutes. The solution outperformed manual current practices, which have an 80% accuracy rate. ​
• Gradually captured and transferred the expertise of the Subject Matter Experts from human to machine with annotated pictures. Additionally, map visualization allowed correlations between the rust severity level and the nature of the surrounding environment.
• ​Drastic reduction of time spent reviewing inspection data with better detection accuracy. This represents a huge opportunity to rapidly improve the inspection process and reduce costs.​
• The customized solution enables predictive maintenance and utilizes inspection results to extend the lifetime of individual towers.