An Impact Localization Solution Using Embedded Intelligence-Methodology and Experimental Verification via a...

Area of Science:

• Engineering
• Materials Science
• Computer Science

Background:

• Embedded intelligence (EI) integrates machine learning into resource-scarce IoT devices.
• Wireless Sensor Networks (WSNs) are crucial for IoT applications.
• Current structural health monitoring (SHM) often uses expensive equipment and complex methods.

Purpose of the Study:

• To develop an extreme-edge system for real-time structural health monitoring (SHM).
• To enable on-device impact detection and localization on thin plates using low-cost hardware.
• To create a material and sensor location-agnostic SHM solution.

Main Methods:

• A novel methodology for creating an experimental time-series dataset using ceramic piezoelectric transducers (PZTs).
• Implementation of TinyML technology for on-device intelligence.
• Validation of Random Forest and shallow neural network models for impact localization.

Main Results:

• Real-time impact localization with less than 400 ms latency.
• Achieved higher than 90% accuracy in impact detection and localization.
• Demonstrated an agnostic approach, independent of material properties and sensor placement.

Conclusions:

• The developed extreme-edge system offers an efficient and accurate solution for SHM.
• TinyML enables sophisticated machine learning tasks on resource-constrained IoT devices.
• This work advances the integration of EI in IoT for practical structural monitoring applications.