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.

1. Audio Monitoring

The Sipeed R6+1 microphone array provides PrintSENTINEL with acoustic intelligence through 7 beamforming microphones sampling at 48kHz. MFCC coefficients and frequency spectrum analysis enable detection of distinct audio signatures associated with mechanical failures. This subsystem identifies motor grinding, filament popping, stepper skipping, and belt tension issues—anomalies often missed by visual inspection alone.

2. Bed Impact Monitoring

The ADXL375 high-G accelerometer, mounted directly on the print bed, captures sudden impacts and vibrations that standard sensors miss. With a ±200g range and 3200Hz sampling rate, it extracts 18 features including vibration RMS, impact energy, and peak G-force. This enables early detection of bed warping, layer delamination, and catastrophic head crashes before they destroy hours of print progress.

3. Extruder Vibration Monitoring

The BNO055 IMU mounted on the printhead provides 9 degrees of freedom with on-chip sensor fusion outputting calibrated quaternion data at 100Hz. By analyzing vibration RMS, jerk signatures, and tilt deviations, PrintSENTINEL detects mechanical degradation such as loose belts, bearing wear, and layer shift events. The sensor's ±16g accelerometer and ±2000°/s gyroscope ensure comprehensive motion capture.

5. Motion Tracking Along Axes

Six AS5600 magnetic encoders with 12-bit resolution track motor and driven element positions across all three axes. By comparing motor shaft angles to pulley/leadscrew positions, PrintSENTINEL calculates gear ratios and detects belt slip, step skipping, and layer shifts in real-time. G-Code verification compares expected vs. actual positions, enabling closed-loop correction of motion errors.

8. Subsurface Detection

The 24GHz mmWave FMCW radar represents PrintSENTINEL's most innovative sensing modality—an industry first for desktop 3D printing. Mounted on the Z-axis beam, it scans each layer to detect internal voids, delamination, and density variations invisible to cameras. With ~4mm range resolution and ~10mm penetration in PLA, it extracts layer integrity metrics that identify structural defects before they propagate.

9. Temperature Monitoring

The MLX90640 thermal imaging camera provides a 32×24 pixel thermal map of the entire print area at up to 64Hz. Mounted at 45° for optimal coverage, it simultaneously monitors nozzle temperature, print surface conditions, and bed edge cooling. Feature extraction identifies thermal gradients that predict overheating, cold extrusion, and warping risk—enabling preemptive intervention before defects form.