The limited availability of real images of rare anomalies represents one of the main challenges in developing artificial intelligence models for the inspection of electrical infrastructures. To address this need, Elewit, together with Red Eléctrica and in collaboration with Unusuals, has promoted a project based on the generation of synthetic images to enhance and strengthen visual anomaly detection models in the Transmission Grid.

In the field of high-voltage infrastructure maintenance, early detection of such anomalies is essential to ensure system reliability. However, many of these incidents occur infrequently, making it difficult to gather sufficient real data to train artificial intelligence models with the required level of accuracy.

To overcome this limitation, the project proposes the use of synthetic images as a complement to real data. Through advanced generation and simulation technologies, developed in collaboration with Unusuals, it is possible to realistically recreate multiple operating scenarios and failure typologies, generating larger, more balanced and controlled datasets.

This approach makes it possible to reproduce complex or uncommon situations, as well as introduce variations in lighting conditions and capture angles. It also enables the full generation of certain anomalies in simulated environments, significantly expanding the scope of detection models.

The integration of synthetic images into the training of artificial intelligence models has led to a cross-cutting improvement in their performance. Key advances include improvements in core metrics such as precision and recall, a reduction in false positives, and the development of more robust, faster models with greater generalization capabilities.

In addition, the project has enabled the development of hybrid models that combine real and synthetic data. This approach has made it possible to address use cases that previously could not be developed due to a lack of available information, as well as to progress in the detection of specific anomalies such as corroded counterweights, cracked joints, misaligned insulator strings, or deteriorated spacers, among others.

In certain cases, it has even been possible to generate and train complete models using anomalies created entirely in synthetic environments, achieving levels of accuracy suitable for deployment in operational environments.

Overall, the project demonstrates the value of synthetic images as a lever for advancing the digitalization of critical infrastructure maintenance. Their application expands analytical capabilities, improves decision-making, and supports the transition towards more comprehensive and scalable predictive models, without relying exclusively on real incident occurrences in the field.

This work is part of the initiatives developed in 2025 between Elewit and the Transmission and Operations divisions, aimed at incorporating technological innovation into the management of the Transmission Grid.