The Defects in Power Distribution Components dataset is presented by the authors as a dataset comprising 708 images and corresponding labels of defects found in components within an electrical distribution system. In the context of their work, an electric system constitutes a complex infrastructure designed to supply power to various consumers, including residential, industrial, and railway facilities. Within an electrical distribution network, various components come into play, such as arresters, transformers, and cross arms, among others. These components are typically situated within power posts and are susceptible to faults stemming from routine operation, weather conditions, or acts of vandalism.

The high frequency of electric system failure is of considerable significance, as it can translate into substantial costs for maintaining a company of the power infrastructure. To mitigate the impact of these faults on consumers, the implementation of a contingency plan or, even better, preventing failures before they manifest becomes imperative.

One commonly adopted approach to prevent these failures involves visually inspecting the components of the distribution network. However, due to the extensive network of power posts and the numerous elements involved, human inspections are susceptible to errors. Consequently, automating this inspection process emerges as a promising solution to address these issues. Automation is often employed to identify visible defects, such as cracks, burns, and cable issues, using techniques like the inspection of images captured by drones. Nevertheless, using drones for inspection presents certain limitations, including potential degradation in image quality due to external variables like distance, balance, excessive vibration, or inadequate lighting conditions.

Moreover, the identification of objects during automatic inspections proves challenging. This is attributable to the varying sizes of the objects, the distance between the observer and the objects, and sometimes the presence of intersections between them. Notably, larger objects, such as the posts on the right side of the image, are more easily discernible, while smaller components pose a greater challenge. Consequently, devising a robust solution for automatic object identification is imperative to achieve superior outcomes compared to conventional methods.

The energy lab is purpose-built for training in the maintenance of power distribution network components. The lab environment replicates various structures typically encountered along a railway. Notably, most of the lab’s components are installed at a height of approximately 1.5 meters. This deliberate choice was made to provide hands-on training for company employees. Moreover, this proximity to the ground enabled us to capture images for constructing the dataset from various distances and angles. As mentioned earlier, the main challenge in identifying objects and their defects arises from the small size of these components.

The authors’ study specifically focuses on two types of components, namely, insulators and electrical cables, and three common types of failures: (1) cable out of insulator; (2) cable out of spacer; and (3) insulator without ring.

Furthermore, it’s worth noting that these three common defects are frequently encountered along our power distribution network adjacent to the railway. These components are manually repositioned under the supervision of a railway expert with an electrical engineering background, effectively replicating real-world scenarios found within the company’s railway network. This approach allows us to capture images for constructing a dataset that closely mimics real situations.