Workflow Description

This n8n workflow tests predictive maintenance for manufacturing equipment, using machine learning to predict failures based on IoT sensor data. A multi-agent system processes data, generates predictions, and logs results for POC validation, integrating with Qdrant for maintenance protocols.

Technical Specifications

• Platform: n8n (cloud-hosted, AWS).
• Trigger: Cron node for periodic data pulls.
• Multi-Agent System:
  • Gatekeeper: Routes tasks (data preprocessing, prediction, reporting).
  • Specialized Agents: Handle preprocessing, ML predictions, and reporting.
• Vector Database: Qdrant for maintenance protocol retrieval.
• Integrations:
  • IoT platform API (e.g., AWS IoT Core).
  • PostgreSQL for logging.
  • Slack for alerts.
• Output: JSON with predictions and reports.
• Error Handling: Error Trigger node with Slack notifications.
• Security: API keys in AWS Secrets Manager, HTTPS.

API Payload (Input):

{
  "equipment_id": "EQ123",
  "sensor_data": {
    "vibration": 0.15,
    "temperature": 75.2,
    "timestamp": "2025-05-28T19:07:00+05:30"
  }
}

API Payload (Output):

{
  "equipment_id": "EQ123",
  "prediction": "Failure risk: 85% in 48 hours",
  "maintenance_action": "Inspect bearings",
  "timestamp": "2025-05-28T19:07:00+05:30"
}

Workflow Configuration

1. Cron Node (Trigger):

   • Settings: Every 10 minutes (0 */10 * * * *).
   • Output: Initiates data pull.

2. Gatekeeper AI Agent Node:

   • Settings:
     • Model: Grok (xAI)
     • Prompt: “Route task: preprocess data, predict failure, or generate report. Return { task: string }.”
   • Output: { "task": "preprocess" }.

3. Switch Node:

   • Settings: Routes for task == 'preprocess', task == 'predict', task == 'report', Else.
   • Output: Passes equipment_id, sensor_data.

4. Preprocessing AI Agent Node:

   • Settings:
     • Model: Grok
     • Prompt: “Clean sensor data, normalize values. Return cleaned JSON.”
   • Tools:
     • Set Node: Normalizes vibration (0-1 scale), temperature (°C).
   • Output: Cleaned JSON.

5. Prediction AI Agent Node:

   • Settings:
     • Model: Grok
     • Prompt: “Predict equipment failure risk using sensor data and Qdrant protocols.”
   • Tools:
     • Qdrant Node:
       • Collection: maintenance_protocols
       • Query: Embed sensor_data, top_k=2
     • HTTP Request Node:
       • URL: https://iot.modnexus.in/data/{equipment_id}
       • Method: GET
     • PostgreSQL Node:
       • Table: predictions
       • Columns: equipment_id, prediction, timestamp
   • Output: { prediction: string, maintenance_action: string }.

6. Structured Output Parser Node:

   • Settings: Schema as per output payload.
   • Output: Formatted JSON.

7. Slack Node:

   • Settings: Channel #maintenance-poc, Message: “Prediction: {{ prediction }}”.
   • Output: Confirms delivery.

Visual Workflow Diagram

[Cron] --> [Gatekeeper AI]
              |
              v
           [Switch]
         /   |   |   \
        /    |   |    \
[Preprocess] [Predict] [Report] [Default]
    |           |          |
    v           v          v
[Qdrant, HTTP, PostgreSQL] --> [Structured Output] --> [Slack]

Diagram Explanation: Cron triggers data pulls, Gatekeeper routes to preprocessing, prediction, or reporting agents, which use Qdrant and PostgreSQL, outputting to Slack.

Graphical Mockup Description

• Layout: Vertical flowchart, 800x600px, white background.
• Elements:
  • Nodes: Rounded rectangles (blue for agents, green for tools, red for output).
  • Connections: Black arrows with labels (e.g., “task: predict”).
  • Annotations: Text labels for nodes (e.g., “Gatekeeper AI: Grok”).
  • Title: “Predictive Maintenance POC Workflow” at top, bold 24px.
• Colors: Blue (#1a73e8) for ModNexus branding, grey for connectors.
• Tools: Qdrant, PostgreSQL icons in bottom nodes.