What is the difference between anomaly detection and classification?

Classification requires labeled examples of each condition (normal, fault-type-A, fault-type-B). Anomaly detection learns only from normal data and flags anything different. Use anomaly detection when you cannot collect failure examples — which is the common case for equipment that rarely fails.

Can the Arduino Nano 33 BLE detect anomalies in audio?

Yes. The Sense variant has a built-in MP34DT05 MEMS microphone. Edge Impulse's spectral analysis processes audio features the same way as vibration data. This enables acoustic anomaly detection — detecting unusual machine sounds like grinding, clicking, or humming changes.

How sensitive is the anomaly detection on Arduino Nano 33 BLE?

Sensitivity depends on the anomaly threshold and training data quality. Edge Impulse's K-means block reports a distance score — higher means more anomalous. With 30 minutes of normal data and a threshold at the 95th percentile, expect to detect significant operational changes while tolerating normal variation.

Hardware Guide

Arduino Nano 33 BLE for Anomaly Detection with Edge Impulse

The Arduino Nano 33 BLE Sense runs Edge Impulse anomaly detection using its built-in sensors — accelerometer, temperature, and microphone. Edge Impulse's K-means anomaly detection block trains on normal data and flags deviations, requiring no labeled failure examples.

Published 2026-04-01

Hardware Specs

Spec	Arduino Nano 33 BLE
Processor	ARM Cortex-M4F @ 64 MHz
SRAM	256 KB
Flash	1 MB
Key Features	Built-in 9-axis IMU (LSM9DS1) on Arduino Nano 33 BLE, Arduino ecosystem, Ultra-low-power BLE, Built-in microphone (Sense variant)
Connectivity	Bluetooth 5.0 LE, 802.15.4 (Thread/Zigbee), NFC, USB 2.0
Price Range	$5 - $8 (chip), $20 - $35 (dev board)

Compatibility: Good

The Arduino Nano 33 BLE Sense includes an LSM9DS1 IMU, HTS221 temperature/humidity sensor, and MP34DT05 microphone — all usable for anomaly detection without external hardware. The 256 KB SRAM provides 8x the 32 KB minimum for anomaly models. Edge Impulse's anomaly detection block uses K-means clustering or autoencoders, both lightweight (<15 KB models). The nRF52840's 64 MHz Cortex-M4F handles inference quickly. Edge Impulse's data collection daemon works over USB serial with the Nano 33 BLE — no special firmware needed. BLE 5.0 enables wireless anomaly alerts to phones or BLE gateways. The main advantage is accessibility: Arduino IDE + Edge Impulse Studio is the lowest-barrier path from concept to working anomaly detector. The limitation is the 64 MHz clock and 1 MB flash — sufficient for anomaly detection, but leaves no headroom for complex multi-model pipelines.

Getting Started

1

Connect the Nano 33 BLE Sense to Edge Impulse

Flash the Edge Impulse firmware via Arduino IDE or use the edge-impulse-daemon CLI. The built-in sensors appear automatically in Edge Impulse Studio's data collection interface — no wiring required.
2

Collect normal operating data

Attach the board to the target equipment (motor, pump, conveyor). Stream accelerometer data during normal operation for 30-60 minutes. Edge Impulse needs only normal data — no labeled failures required for anomaly detection.
3

Configure anomaly detection in Edge Impulse

Add a Spectral Analysis processing block for vibration features. Then add an Anomaly Detection (K-means) learning block instead of Classification. This learns the normal feature space and flags anything outside it.
4

Deploy and set up BLE alerts

Export as Arduino library from Edge Impulse. In your Arduino sketch, run inference in loop() and check the anomaly score. When the score exceeds your threshold, advertise the anomaly via BLE or trigger a visual indicator (LED, buzzer).

Alternatives

ESP32 with TFLite Micro

Built-in Wi-Fi for cloud reporting. 520 KB SRAM and dual-core for concurrent monitoring. More capable but requires external sensors and TFLite expertise.

ESP32-C3 with TFLite Micro

Cheapest Wi-Fi option ($1-3 chip). 400 KB SRAM. Good for high-volume deployments. Requires external sensors and more development effort than the Arduino + Edge Impulse path.

Explore More

More nRF52840 guides More Anomaly Detection guides All resources Find the right MCU

FAQ

What is the difference between anomaly detection and classification?: Classification requires labeled examples of each condition (normal, fault-type-A, fault-type-B). Anomaly detection learns only from normal data and flags anything different. Use anomaly detection when you cannot collect failure examples — which is the common case for equipment that rarely fails.
Can the Arduino Nano 33 BLE detect anomalies in audio?: Yes. The Sense variant has a built-in MP34DT05 MEMS microphone. Edge Impulse's spectral analysis processes audio features the same way as vibration data. This enables acoustic anomaly detection — detecting unusual machine sounds like grinding, clicking, or humming changes.
How sensitive is the anomaly detection on Arduino Nano 33 BLE?: Sensitivity depends on the anomaly threshold and training data quality. Edge Impulse's K-means block reports a distance score — higher means more anomalous. With 30 minutes of normal data and a threshold at the 95th percentile, expect to detect significant operational changes while tolerating normal variation.

Orchestrate Anomaly Detection with ForestHub

Sensors and devices flag anomalies; ForestHub on the Linux edge gateway correlates them over MQTT/Modbus/OPC-UA and acts on the line as a deterministic, auditable graph.

Get Started Free