FORTE AI DRIVEN MAINTENANCE HUB

Corrective maintenance AI for plant technicians, planners, and managers

Your browser does not support the video tag.
FORTE logo
History

AI Maintenance Assistant

Automated Metadata Extraction

Transforming Documents into Actionable Data

Objective

To automatically read complex technical documents and convert their valuable, unstructured information into organized, searchable, and structured data that can be queried using natural language.

The Process Flow

1. Input Documents

Starts with an inventory of all file types.

2. AI Analysis Engine

Interprets text and visual elements like diagrams.

3. Rule Application

Applies business logic, like skipping PDFs over 5 pages.

4. Structured Output

Saves clean, organized data as .json files.

Source Data Types

41

PDF Documents

38

TIFF Images

2

Excel Spreadsheets

2

Database Files

1

PNG Images

1

Word Documents

Interactive Analysis

Natural Language Query

"Analyze the document image or diagram or picture and describe in detail what you see. Describe the layout of the document's images and describe the text structure."

Instant, Summarized Answer

Found in: COLUMN_STROKE=800mm.PDF

Layout: Multi-view technical illustration (A, B, C, D, E) with a component table on the right.

Text Structure: Hierarchical with a main title block, component list, and detailed view annotations.

Example Result Snippets

Successful Extraction

// From: BRIDGE_PALLETE_CHANGER.PDF
{
  "document_type": "technical_drawing",
  "machine_model": "HELLER H6000 MC",
  "components": [
    {
      "component_name": "Pallet arm",
      "part_number": "04.112011"
    }
  ]
...
}

Handled Exception

// From: ALARM__ERROR_LIST.pdf
{
  "file_name": "ALARM & ERROR LIST.pdf",
  "file_size_mb": 12.57,
  "analysis_results": [],
  "error": "PDF has 774 pages, exceeds limit of 5"
}

Deep Dive: Full JSON Output Structure

Below is a detailed example from a successfully processed technical drawing. This showcases the full depth of information extracted, including visual analysis, component breakdowns, and complete text recognition.

{
  "file_name": "COLUMN STROKE=800mm.PDF",
  "document_type": "technical_drawing",
  "machine_model": "HELLER H5000",
  "description": "This technical drawing illustrates the assembly and components of a machine column...",
  "detailed_visual_analysis": {
    "overview": "The drawing is a multi-view technical illustration showing the column assembly...",
    "layout": "The drawing is organized into multiple views (A, B, C, D, E)..."
  },
  "components": [
    {
      "component_name": "Column",
      "part_number": "02.008200",
      "manufacturer": "HELLER",
      "location": "Central vertical structure in all views (A, B, C, D, E)"
    },
    // ... 10 more components extracted
  ],
  "connections": [
     {
      "connection_type": "Pneumatic",
      "from_component": "Column",
      "to_component": "Pneumatic Cylinder"
    },
     // ... 2 more connections extracted
  ],
  "dimensions": [
    {
        "dimension": "Column Stroke",
        "value": "800",
        "unit": "mm"
    }
  ],
  "text_structure_analysis": {
    "document_layout": "The document is structured with multiple views of the column assembly on the left...",
    "text_grouping": "Text is grouped into sections: title block, component list, and view annotations..."
  }
}

Performance Snapshot

100% Files Processed

Key Metrics

  • All 85 target files were processed by the system.
  • High success rate on critical image & schematic files.
  • Business rules correctly skipped large documents to save cost.

File Explorer

Metadata Files

Not yet extracted

Uploaded Documents

Local staging only — not in AI knowledge until you index. Files stay in this browser.

No documents uploaded yet

Recomendación de Refacción

Recent changes this session

Part Stock

JSON Viewer


                    
File information will appear here
Microsoft

Dynamics 365 · Asset Management (sandbox)

SIMULATION / MOCK — not connected to live D365
Work Orders
Inventory
Finance
Deploy
OT-———— Open

What a real deploy looks like

  • Dynamics 365 Work Orders
  • Dynamics 365 Inventory
  • Dynamics 365 Finance

Plant owns master data; the AI layer recommends and drafts work orders

Assistant → Dataverse / D365 APIs (not live in this demo)

Technician chat → AI recommendation → D365 WO + reserve part → Inventory deduct → Finance post

AI Settings

Endpoint, API key, and model choices are stored in this browser. The default OpenRouter URL can still use server /api when available.

Examples: OpenRouter, Azure OpenAI, Ollama (http://127.0.0.1:11434/v1/chat/completions), LM Studio, vLLM.

Leave blank to use the server key or the built-in demo fallback.



File Search Store
Not indexed yet

Voice off — configure API / Voice on the landing page.