Industrial Equipment Health — AI-Based Anomaly Detection with ThingsBoard

• Architecture
• Prerequisites
• Telemetry Input and Output
• Calculated field configuration
• Rule chain configuration
• Testing
• Performance & Cost Controls
• Troubleshooting
• Next steps

This document describes a ThingsBoard solution that turns raw device telemetry into actionable maintenance insights:

Workflow:

1. Devices send vibration (mm/s), temperature (°C), and acousticDev (% deviation from baseline).
2. Calculated Fields maintain a rolling window of the last N values (default: 100) and/or last M minutes for each metric.
3. The rolling window is forwarded to an AI Rule Node (OpenAI or another LLM provider).
4. If the AI detects an anomaly, the rule chain creates a ThingsBoard Alarm and optionally sends a notification.

Architecture

Prerequisites

• ThingsBoard version 4.2+
• Device(s) capable of sending the required telemetry values (for this guide, we will emulate them)
• LLM provider credentials (OpenAI, Azure OpenAI, etc.)

Telemetry Input and Output

Expected telemetry keys (per device):

• vibration — mm/s (float)
• temperature — °C (float)
• acousticDev — % deviation from baseline (float)

AI Output:

• anomaly — short label (e.g., "Bearing Wear")
• summary — concise human-readable recommendation (e.g., "Vibration has reached 7.4 mm/s and temperature is at 86°C accompanied by irregular acoustic patterns, indicating bearing wear. Recommend immediate bearing inspection and replacement to avoid catastrophic failure.")

Calculated field configuration

Purpose: Maintain a rolling window of the last N readings (default 100, configurable) efficiently and forward them directly to the AI node.

Steps:

1. Download and import the EquipmentSensor device profile into your ThingsBoard instance.

2. Download and import the calculated field into the EquipmentSensor device profile.

Key notes:

• Rolling window time range is set to 1 day.
• The number of stored values is set to 100. The default maximum is 1000, configurable in system settings.
• The TBEL script outputs the rolling values in a single message to the rule engine:

1
2
3
4
5
6
// Sample script to output raw values of the rolling arguments;
return {
    "acousticDevRecords": acousticDevRecords,
    "temperatureRecords": temperatureRecords,
    "vibrationRecords": vibrationRecords
};

Rule chain configuration

Purpose: Analyze the rolling window data, classify anomalies, and generate a plain-language summary. Works with OpenAI or other LLM providers.

Steps:

• Download the json file with the “Equipment Health Analysis” rule chain configuration.

• Navigate to the "Rule chains" page. Click the "plus" icon button, and select "Import rule chain" from the dropdown menu.
• In the opened window, upload the JSON file with the rule chain configuration and click "Import".
• Locate "AI request" node and enter its edit mode.
• Click "Create new" AI model.
• Enter a name for the AI model, choose the AI provider, and paste your API key. This example uses o4-mini from OpenAI. Test connectivity before saving. Then, click "Save".
• Apply changes to "AI request" node.
• Save rule chain.

5. Update the EquipmentSensor profile to reference the “Equipment Health Analysis” rule chain.

Important details:

• System and user prompts can reference incoming message data:
  • $[*] — entire message body
  • ${*} — entire message metadata
  • $[key] — a specific message body field
  • ${key} — a specific metadata value
• Supported response formats: TEXT, JSON, and JSON Schema (we use JSON Schema here).
• A deduplication node with a 5-second interval is used to reduce AI token usage.

Testing

Step 1. Create a test device Equipment Sensor 1 with the EquipmentSensor profile.

Step 2. Copy the “Check connectivity” command from the device details.

Your command will look something like this:

1
2
3
curl -v -X POST http://localhost:8080/api/v1/6sED1ALqyJg0P6ezIODH/telemetry \
--header Content-Type:application/json \
--data "{temperature:25}"

  * Where localhost:8080 is the host of your ThingsBoard instance, and 6sED1ALqyJg0P6ezIODH is the device access token.

Step 3. Modify and send test data.

Step 3.1 No alarm case:

Send the following test data to ThingsBoard. Be sure to replace:

• $THINGSBOARD_HOST_NAME with the host of your ThingsBoard instance,
• $YOUR_DEVICE_ACCESS_TOKEN with your device's access token.

1
2
3
curl -v -X POST http://$THINGSBOARD_HOST_NAME/api/v1/$YOUR_DEVICE_ACCESS_TOKEN/telemetry \
--header Content-Type:application/json \
--data '{"vibration":4.2,"temperature":70,"acousticDev":5}'

There won't be an alarm created.

Step 3.2 Bearing wear detection:

Simulate a bearing wear event by sending the following test data to ThingsBoard. Be sure to replace:

• $THINGSBOARD_HOST_NAME with the host of your ThingsBoard instance,
• $YOUR_DEVICE_ACCESS_TOKEN with your device's access token.

1
2
3
curl -v -X POST http://$THINGSBOARD_HOST_NAME/api/v1/$YOUR_DEVICE_ACCESS_TOKEN/telemetry \
--header Content-Type:application/json \
--data '{"vibration":8.2,"temperature":88,"acousticDev":5}'

An alarm will be created. Below is an example of the AI output:

1
2
3
4
{
  "anomaly": "Bearing Wear",
  "summary": "Vibration has reached 7.4 mm/s and temperature is at 86 °C accompanied by irregular acoustic patterns, indicating bearing wear. Recommend immediate bearing inspection and replacement to avoid catastrophic failure."
}

Step 3.3 Misalignment detection:

Simulate a misalignment event by sending the following test data to ThingsBoard. Be sure to replace:

• $THINGSBOARD_HOST_NAME with the host of your ThingsBoard instance,
• $YOUR_DEVICE_ACCESS_TOKEN with your device's access token.

1
2
3
curl -v -X POST http://$THINGSBOARD_HOST_NAME/api/v1/$YOUR_DEVICE_ACCESS_TOKEN/telemetry \
--header Content-Type:application/json \
--data '{"vibration":32.2,"temperature":38,"acousticDev":5}'

An alarm will be created. Below is an example of the AI output:

1
2
3
4
{
  "anomaly": "Misalignment",
  "summary": "A sudden vibration spike to 32.2 mm/s without a corresponding temperature rise or acoustic deviation indicates likely misalignment in the drive train. Please perform an immediate shaft-and-coupling alignment check to prevent further mechanical damage."
}

Performance & Cost Controls

• Batching & debounce: Send to AI only after N points or every T seconds.
• Early filters: Skip AI call if all metrics are comfortably within normal bands.
• Payload compacting: Send statistics (min/mean/max/std, trend) instead of full arrays if prompt allows.

Troubleshooting

• No alarms:
  1. Verify raw device data is available in ‘Latest telemetry’ tab.
  2. Enable and browse the calculated field debug events.
  3. Enable and browse the corresponding rule nodes debug events.
• High costs: Increase the de-duplication period, tune prompt or switch AI model.

Next steps

Experiment with AI prompts and share your feedback with community!