Overview
SoundSentinel is a compact, self-contained device that learns the normal acoustic fingerprint of any running machine and raises an alert when the sound pattern deviates, indicating wear, imbalance, blockage, or imminent failure. It requires no disassembly, no wiring into the machine. Just clip it on and let it listen.
Developed under the working name VigileAcustico for the Italian market, SoundSentinel targets the 2026 Rome Maker Fair as its debut showcase.
How It Works
- Learn Mode: Records a spectral template over a configurable window (default 5 minutes) by computing 1024-point FFT with Hann window and 50% overlap across 512 frequency bins.
- Monitor Mode: Every 3 seconds, captures 200 ms of audio, cross-correlates with the stored template (Pearson rho), and computes per-bin Mean Squared Error. Thresholds above 3.0 sigma trigger alerts.
- Dashboard: Self-hosted ESP32 web server serves a real-time FFT chart (Chart.js), anomaly timeline, and status indicators with REST API polling.
Key Features
- Zero integration cost: Clip it on, press Learn, walk away. No PLC connection, no SCADA integration, no cloud dependency.
- Cross-noise rejection: Template correlation threshold rejects ambient noise from adjacent machines. Piezo vibration disc provides a secondary vibration-only channel.
- Fully offline: All processing runs on-device. Web dashboard is self-hosted. No internet required.
- Portable power: Deep-sleep duty cycling with 2000 mAh LiPo provides 12-18 hours of operation.
Applications
- Industrial: Pumps, compressors, HVAC fans, conveyor motors, cooling towers
- Workshop: 3D printers, CNC mills, laser cutters, vacuum pumps
- Infrastructure: Ventilation systems, wastewater pumps, elevator drives
- Commercial: Server room CRAC units, air handling units, escalators
Technical Stack
- MCU: ESP32-S3 (Xtensa LX7, dual-core, SIMD DSP)
- Audio: INMP441 I2S MEMS microphone (16 kHz, 16-bit)
- Vibration: 35 mm piezo disc (ADC, v2 planned with I2S preamp)
- Display: SSD1306 OLED 0.96" (128x64, I2C)
- Alert: WS2812 RGB LED (NeoPixel)
- Storage: MicroSD card (SPI) for template persistence + anomaly CSV logging
- Framework: PlatformIO with Arduino framework + arduinoFFT
- Connectivity: WiFi STA with AP fallback, optional MQTT
Status
Active (R&D)
- Verified — Phase 1: I2S capture (v0.18, INMP441 MEMS mic, 16 kHz/16-bit streaming)
- Verified — Phase 2: FFT pipeline (v0.19, 1024-pt Floating FFT, Hann window, top-5 peak detection)
- Done — Phase 3: Button input (GPIO 15, press/release events working)
- Coded — Phase 4–5: Template engine (Pearson correlation, running avg + stddev) + web dashboard (REST API), to be deployed on hardware by 28 July 2026
Target: Ready for Maker Faire Rome, end of August 2026.
Roadmap (v2.0): Contact mic hybrid correlation, MQTT multi-node, edge ML.
Learn More
Visit the dedicated project site at soundsentinel.altirratech.eu for the full technical spec, block diagram, and bilingual (EN/IT) information.