← Back to Gallery

Integration of MEMS Accelerometers in Regional Seismological Networks: Piloting A Low-cost Edge Computing Approach

We deployed low-cost MEMS accelerometer nodes integrated with the SeisComP monitoring system to evaluate whether distributed, cost-effective sensors can augment traditional regional seismic networks. The system architecture consists of WiFi-enabled ESP32-S3 microcontrollers equipped with high-resolution ADXL345 accelerometers, sampling 3-axis acceleration at 100 Hz with adaptive gravity subtraction for seismic signal isolation. Each node implements TensorFlow Lite Micro inference (25ms latency, 45KB model size) trained to classify noise, P-waves, and S-waves from 4-second sliding windows. We combined deep learning classification with classical STA/LTA algorithms to provide robust fallback capabilities.

Edge processing enables intelligent event filtering before network transmission—only seismic phases exceeding confidence thresholds (>0.5) trigger MQTT message publication to a server running the SeisComP infrastructure. Published picks include NTP-synchronized UTC timestamps, signal-to-noise ratios, classification confidence, and optionally the full waveform window (400 samples × 3 channels) for analyst review. A local storage mechanism captures events during network outages with automatic synchronization upon reconnection, ensuring data integrity. The complete system including utilizing off-the-shelf developer boards, power management, and weatherproof enclosure costs about $50 per node, enabling dense spatial deployment previously infeasible with traditional instrumentation. Custom PCBs would lower these per-unit costs.

Initial beta testing focuses on evaluating MEMS sensor performance against traditional seismometers in detecting local and regional events (M>2.0). Key metrics under evaluation include detection latency compared to network stations, false alarm rates in urban noise environments, P-wave pick accuracy relative to analyst picks, and network bandwidth efficiency achieved through edge filtering. This deployment represents an important step toward understanding whether distributed, intelligent sensor networks can effectively complement traditional seismological infrastructure.

Session: Network Seismology: Recent Developments, Challenges and Lessons Learned [Poster]

Type: Poster

Room: Exhibit Hall A+B

Date: 4/15/2026

Presentation Time: 08:00 AM (local time)

Presenting Author: Jacob Walter

Student Presenter: No

Invited Presentation: 

Poster Number: 32

Additional Authors