Anomaly Detection Overview

Anomaly detection identifies unexpected behavior in telemetry data — patterns that differ significantly from what is considered normal. These anomalies can indicate system malfunctions, equipment degradation, security breaches, or operational inefficiencies.

Trendz anomaly detection is unsupervised — it requires no manual thresholds or labeled training data. Instead, it learns what “normal” looks like from historical telemetry and then scores deviations in new data automatically.

The typical workflow:

1. Build Anomaly Detection Model — configure input scope, choose a detection strategy, and train the model on historical telemetry.
2. Explore results — review detected anomalies per device using the activity heatmap, ranked item table, and cluster analysis.
3. Set up continuous monitoring — enable periodic scanning so the model keeps detecting anomalies as new data arrives.
4. Enable alarms — automatically trigger ThingsBoard alarms when anomaly scores exceed defined thresholds.

The default setup takes under 2 minutes and requires no ML expertise. Advanced users can also fine-tune clustering algorithms, segmentation, distance functions, and scoring thresholds directly.

Trendz anomaly detection is based on two core metrics:

Metric	Description
Anomaly Score	A numeric value representing how far a data segment deviates from expected (normal) behavior. Measures the intensity of the anomaly.
Anomaly Score Index	A composite metric that combines anomaly score with the duration of the anomaly. Helps prioritize anomalies that have a greater cumulative impact over time.

Example — pump vibration analysis:

• Anomaly A — a sharp vibration spike lasting ~5 seconds. High anomaly score, but short duration keeps the score index low.
• Anomaly B — a longer-lasting deviation without large spikes. Lower anomaly score, but the score index is higher due to extended duration — suggesting greater long-term impact on pump health.

Use Anomaly Score to detect sharp, short-term deviations. Use Anomaly Score Index to surface sustained anomalies that may cause greater damage over time.

Trendz covers the full anomaly detection lifecycle — from building and training a model to analyzing results, setting up continuous monitoring, and embedding insights into ThingsBoard dashboards. Explore the capabilities below.

• 

  Create Anomaly Model

  Choose a detection intent, configure input scope and sensitivity, then build your first model in minutes.

• 

  Analyze Anomalies

  Review detected anomalies per device using the activity heatmap, ranked item table, and cluster analysis to assess model quality and prioritize issues.

• 

  Anomaly Detection Interface

  Explore the Summary, Models, and Anomalies tabs — fleet-wide KPIs, full model list, and a unified view of all detected anomaly events.

• 

  Set Up Monitoring

  Enable periodic scanning, configure anomaly score saving, and automatically generate ThingsBoard alarms when anomaly scores exceed defined thresholds.

• 

  Visualize Anomalies

  Embed anomaly data in ThingsBoard dashboards using Anomaly View Reports, Anomaly Fields in custom views, or Business Entity Fields for saved telemetry.

• 

  Advanced Properties

  Fine-tune clustering algorithms, segmentation window settings, score thresholds, and anomaly detection parameters for full control over model behavior.