Client Background

  • Client Name: A leading AI Data Center firm in the USA
  • Industry Type: Engineering Analytics & AI Solutions
  • Products & Services:
    • Engineering document analysis
    • Computer vision-based automation
    • AI-driven validation systems

About Client:

  • The client works with complex engineering datasets including:
    • Electrical panel images
    • Architectural drawings
    • Multi-page technical PDFs
  • Their workflows required high accuracy, scalability, and automation
  • Objective:
    • Reduce manual inspection
    • Improve validation accuracy
    • Automate structured data extraction

Project 1: Electrical Panel Audit Automation System

The Problem

  • Manual inspection of electrical panels involved:
    • Identifying switch states (ON/OFF)
    • Verifying breaker labels and amp ratings
    • Checking wiring correctness
  • Challenges:
    • Time-consuming for large-scale audits
    • High chance of human error
    • Subtle visual differences between switch states
    • Variation in lighting and panel layouts
  • No automated way to convert panel images into structured audit data

Our Solution

  • Developed a computer vision-based automated audit system
  • Key capabilities:
    • Detect panel components using YOLOv8
    • Extract text using OCR (Gemini)
    • Convert visual and textual data into structured outputs
    • Validate electrical correctness using rule-based logic
  • System processes:
    • Panel images
    • Engineering drawings
    • PDFs

Solution Architecture

  • Input:
    • Panel images / PDFs
  • FastAPI layer for API handling
  • Modal serverless compute:
    • CPU for preprocessing and tiling
    • GPU for model inference
  • YOLOv8 detection:
    • Breakers, switches, wires, labels, amp ratings
  • OCR layer:
    • Extract breaker numbers and amp ratings
  • Data association engine:
    • Groups components into breaker-level records
  • Validation engine:
    • Phase calculation
    • Wire-phase mapping
    • Rule-based verification
  • Output:
    • Structured JSON / Excel dataset

Deliverables

  • Automated electrical audit pipeline
  • Structured breaker-level dataset
  • Validation system for wiring correctness
  • Exportable audit reports

Technical Challenges

  • Detecting small components in high-resolution images
  • Differentiating visually similar switch states
  • Associating multiple detected elements correctly
  • Handling missing or unclear data

Solutions

  • Image tiling and preprocessing for better detection
  • Precise annotation of multiple object classes
  • Geometry-based association logic
  • Phase mapping algorithms (L1, L2, L3 logic)

Business Impact

  • Reduced manual inspection effort significantly
  • Improved audit accuracy and consistency
  • Enabled scalable processing of large datasets

RELATED ARTICLESMORE FROM AUTHOR