Context
Switching operations in low‑voltage networks are commonly performed by plugging and unplugging fuses in cable distribution cabinets.
Challenge
There is limited visibility into the actual switching status of low‑voltage networks, as switching activities are often not documented or not documented accurately. Lack of reliable information leads distribution system operators to make incorrect assumptions during planning and operation, resulting in higher costs.
AI Solution
The solution enables a mobile application to assist service engineers in capturing images of cable distribution cabinets, which are analysed by an AI‑based image processing system that classifies the actual switching status of fuses and automatically updates the digital twin within a network information or management system. The solution enables on‑the‑fly classification of switching states and improves efficiency by reducing manual field effort and operating costs. Immediate in‑field image recognition supports accurate switching state classification.
Impact & evidence
Impact is evidenced by the Artificial Neural Network being trained on approximately 3,000 images of cable distribution cabinets and achieving an F1‑score close to 1 for different switching states (e.g. on, off, open), demonstrating highly accurate visual classification and contributing to more reliable network operation and reduced operational risk.
Implementation / Adoption
Implementation requires the existence of a network information or management system with appropriate interfaces and a consistent network data model or digital twin of the distribution grid. Adoption within the IT/OT environment of a distribution system operator depends on the interfaces provided by these systems to enable integration of the AI solution and alignment with existing grid data and operational workflows.
Data
Solution requires relevant data models / digital twin of the low voltage networks and a sample of (annotated) photos of the cable distribution cabinets.
Results of the projects are documented in a common conference paper. See proceedings from: Protection, Automation and Control Conference, Virtual Conference, 2021; S. Plötz et al.: Artificial Intelligence-based Visual Detection of Switching States in Low Voltage Networks