StatSight: Preventing Electrical Fires with AI : Computer Vision Tool for Electrical Meters

There are numerous requirements in the utility sector for photographing maintenance and completed work. As a result, large databases of images are accumulated. Currently, these images are primarily stored to meet regulatory demands and internal demands. The stored images are not always being utilized except for follow-up purposes. This project explores how these images can be leveraged to simplify work supervision and identify deficiencies in installations and maintenance processes. This project is developed in collaboration with IsMobile, a company specializing in field service management. IsMobile provides digital platforms that support large-scale utility operations and industrial maintenance, with a particular focus on the electric power industry. Among IsMobile’s clients are electricity providers responsible for the installation and maintenance of electricity meters-devices that measure electrical energy consumption in homes, businesses, or appliances, typically in kilowatt-hours (kWh). These readings are essential for billing and monitoring purposes. Due to both regulatory requirements and internal policies, IsMobile’s clients must store photographic documentation of meter installations and maintenance for up to ten years. Currently, reviewing these photos is a manual, timeconsuming process that is prone to human error. This project focuses on developing a tool to support IsMobile’s clients by automating aspects of the photo review process. To address this, the project applies machine learning and deep learning techniques to detect anomalies in photographic data. In parallel, a web-based interface is developed to allow users to apply these models, filter and analyze images, and uncover trends within the image databases. The interface is refined iteratively based on user feedback to ensure usability and efficiency. The results of the project were promising, demonstrating the potential of automated analysis in this domain. However, there remains significant room for further development and optimization.