The Challenge
Electrical planning is a manual, time-intensive process: electricians analyze building plans, identify symbols for outlets, switches, and luminaires, and convert these into material lists. This workflow is error-prone and leaves little room for quick iterations.
Sonepar Deutschland, the world’s leading B2B distributor for electrical technology, was looking for a solution to automate this process.
What We Developed
innFactory first developed a prototype that impressed Sonepar so much that it became a full product: Sonepar Digiplan - an AI-powered cloud platform for automated electrical and lighting planning.
AI-Based Plan Recognition
- Custom ML models trained on Vertex AI with thousands of labeled electrical plans
- PyTorch-based computer vision pipeline for symbol recognition
- MLflow for experiment tracking and model versioning
- CVAT for collaborative labeling of training data
- Continuous training for constantly improving recognition accuracy
2D-to-3D Transformation
- Automatic translation of 2D plans into complete 3D models
- Integration of lighting design algorithms for optimal illumination concepts
- BIM workflows through ODA (Open Design Alliance) SDK for DWG/DXF conversion
Cloud-Native Architecture
| Area | Technology |
|---|---|
| Cloud | Google Cloud Platform |
| ML Inference | Vertex AI Endpoints |
| ML Training | PyTorch, MLflow, CVAT |
| Container | GKE Kubernetes, Cloud Run |
| Messaging | Pub/Sub |
| Auth | Firebase, Keycloak |
| Backend | Python (Flask), Scala 3 (Pekko, Play Framework, Smithy4Play) |
| Frontend | Svelte |
| CAD/BIM | Open Design Alliance SDK |
| IaC | Terraform |
E-Commerce Integration
- Direct connection to Sonepar merchandise management systems
- Automatic linking of symbols with items from the Sonepar catalog
- One-click shopping cart for immediate ordering
Our Services
- Prototype Development and Proof-of-Concept
- End-to-End Product Development (backend, frontend, AI models)
- ML Engineering: Training and deployment of proprietary computer vision models
- Data Labeling pipeline with CVAT
- Cloud Architecture and Infrastructure as Code
- Ongoing Operations and Site Reliability Engineering
- BIM Integration as Open Design Alliance partner
Scientific Publication
Our research work on AI-powered electrical plan recognition has been scientifically published and is available at IEEE Xplore.
Results
With Sonepar Digiplan, we have created a solution that:
- Drastically reduces planning time through automatic symbol recognition
- Minimizes error rates through AI-powered validation
- Enables seamless ordering processes through e-commerce integration
- Is scalable and future-proof through cloud-native architecture
Related Services: AI & Machine Learning | Google Cloud | Cloud Native Development
