Automatic Electrical Asset Detection with Continuous-Learning AI
Registering each asset by hand held the inventory back. AI took over that step: less labor, faster delivery.
BeforeAI
Drag the handle: nine assets detected in a single photo, from transformer to street lighting.
From manual registration to AI
Inventory for the tariff review (ANEEL, Brazil's electricity regulator) of one of the largest utilities in the State of Rio de Janeiro, carried out by Intellissis.
Before
Manual registration
2 yearsby hand
- Open each of millions of field photos
- Identify poles, transformers, networks, cutouts and nameplates
- Type it all into the system
- Took most of the labor and set the deadline
After
Registration with AI
4 monthsof work
- The AI detects the assets in each photo
- Reads the nameplate and extracts the data (OCR)
- The operator only validates or corrects
- Each correction makes the model more accurate
The Solution: System + Method + Tool
System: continuous learning
Operator corrections retrain the model, and the most accurate one goes to production on its own.
Method: end to end
From data collection to production, in a replicable, auditable flow.
Tool: real integration
REST API and .NET/C# SDK in Intellissis's Windows systems, with no infrastructure migration.
How the model learns, step by step
- Photo archive indexing
- Assisted annotation
- Auditable training
- Automatic promotion of the best model to production
- Continuous feedback from operations to the model
The registration team started using AI without realizing they were dealing with machine learning.
AI in action
Drag each image: original photo on the left, detection on the right.
BeforeAI
Medium- and low-voltage network with transformer.
BeforeAI
Concrete pole and street lighting.
BeforeAI
Urban infrastructure with multiple assets.
BeforeAI
Spacer cable network, an urban infrastructure variation.
BeforeAI
Rare class: medium-voltage switch and nameplate.
The 17 classes, from poles and transformers to reclosers and concentrators.
17 classes and nameplate reading (OCR)
From the nameplate, the system extracts power rating, manufacturer and asset tag number.
BeforeOCR
Nameplate "ROMAGNOLI" read by the system.
Where the system has operated
Each dot is a processed pole: urban, rural and environmentally protected areas of the State of Rio de Janeiro.
By the numbers
Production databases, June 2026.
Beyond AI: management and quality management system at Intellissis
Aug/2025 to Jun/2026, in line with ABNT NBR ISO 9001:2015. Deliverables described in Intellissis's technical capacity certificate of Sep 3, 2026.
Management
- Business management
- Strategic management
- Operational management
Quality system, from scratch
- Organizational diagnosis
- Process mapping
- Quality committee
- Manuals, policies, procedures and forms
- First management review
Production
- Plant layout by flow stage
- 5S
- Production planning and control, with targets and dashboards
- Contract indicators measured in an auditable way, for the first time
Technologies used
Bring us your challenge.
We'll work out how to solve it.
Here the bottleneck was looking at photos; in yours it may be reading documents, checking spreadsheets or scheduling a team. The technology changes, the method does not.
Tell us about your case