Structural Damage Identification of Low-Carbon Energy Infrastructure Using Convolutional Neural Networks: A Case Study of Wind Turbines.

The development of low-carbon energy technologies, along with the modernization of energy infrastructure, plays a vital role in the transition towards a net-zero future. Energy infrastructures are susceptible to major disruptions caused by extreme weather events, natural disasters, technical failures, and man-made accidents. The structural monitoring of energy infrastructures is therefore necessary to ensure the health and safety of systems and secure a reliable power supply. In recent years, many remote-sensing technologies such as satellite and drone imagery have been introduced to monitor large geographical areas. These technologies can provide high resolution images from energy infrastructures such as power plants, pipelines, and wind turbines. Collecting and analyzing such data can help to identify any potential damage to the infrastructure before it turns into a major incident. This paper proposes an Artificial Intelligence (AI) enabled tool that can correlate image data from various sources to identify and locate faults as well as provide some useful details about the damage such as size, shape, and orientation. Our model utilizes a ResNet50 convolutional neural network (CNN) model to classify the faults and a Region CNN model to localize the faults such that the classified faults can be singled out, labelled and given a short description to aid with the maintenance process. The model is tested on a dataset containing hundreds of images taken by a drone during an inspection of wind turbine blades. The results show that the proposed methods improve the detectability of faults, reduce failure rates, and consequentially cut down on repair expenditures. The effect of cost reduction is derived in terms of levelized cost of energy (LCOE) for six wind farms across the UK and Europe, and an average reduction of 1.2% in LCOE was achieved.