PrintSENTINEL

The Inspiration: Beyond the Visible Failures

Every 3D printing enthusiast knows the heartbreak of a 20-hour print failing in the final minutes. Most monitoring systems today rely on simple cameras that only see what is on the surface, often detecting a failure long after the damage is done. We realized that to truly safeguard the manufacturing process, a machine needs more than just "eyes"—it needs a central nervous system.

We set out to build PrintSENTINEL, an advanced multi-modal monitoring system that gives industrial-grade intelligence to the desktop printer. Our goal was to move beyond simple visual checks and create a system that could "hear" a motor grinding, "feel" a bed impact, and even "see" beneath the surface of the plastic to detect internal voids.

By fusing 152 distinct features from seven different sensing domains—including 24GHz mmWave Radar, 7-mic Beamforming Arrays, and 9-DOF IMUs—we have created a digital guardian that understands the print state better than a human operator. Leveraging the dual-core power of the Arduino UNO Q (RP2350) and the machine learning capabilities of Edge Impulse, we have achieved a system that doesn't just watch for failures; it predicts and classifies them with 98.98% accuracy in real-time.

This is not just a sensor mount—it is the future of autonomous, error-free manufacturing.

152 features, 7 domains, one unified intelligence

This diagram illustrates PrintSENTINEL's multi-level fusion pipeline that transforms raw sensor data into actionable intelligence. A 1kHz master clock synchronizes all inputs, which flow through early fusion (feature concatenation), attention fusion (intra/inter-modal weighting), and late fusion (ensemble voting). Temporal modeling via Bi-LSTM captures sequential patterns, while contextual G-Code integration grounds predictions in expected printer behavior.

The complete PrintSENTINEL nervous system

This architecture diagram shows how 13 sensors across 4 functional domains connect to the Arduino UNO Q via TCA9548A I2C multiplexer and TXS0108E level shifters. Motion tracking (IMU, accelerometer, encoders), environmental sensing (thermal camera, microphone array), and advanced subsurface detection (radar, load cell) combine to generate 152 features. UART integration with the printer MCU enables real-time G-Code verification and closed-loop control.

10 failure modes, one intelligent response

PrintSENTINEL classifies printer states into 10 distinct failure modes using multi-sensor fusion from 8 input sources. Each class is mapped to a severity-based response: LOW (continue monitoring), MEDIUM (alert user), HIGH (pause print), or CRITICAL (immediate stop). This tiered approach balances print preservation with safety, enabling autonomous intervention proportional to the detected threat level.