PINN approach for detecting anomalous base excitations using merged signals from multilocation piezoelectric harvesters
Part Of
Proceedings of SPIE - The International Society for Optical Engineering
Journal Volume
13946
Start Page
139460W
ISSN
0277786X
ISBN (of the container)
9781510698338
ISBN
9781510698338
Date Issued
2026-04-16
Author(s)
Wang, K. W.
Abstract
Reliable detection of anomalous base excitations is vital for structural health monitoring, particularly when excitation conditions differ across locations within a structure. This study introduces a self-powered sensing approach using two piezoelectric energy harvesters deployed at separate locations, each subjected to distinct excitation frequencies. When connected in parallel, the harvesters yield a merged voltage signal that compactly encodes excitation information. To decode this signal, a physics-informed neural network (PINN) is developed for inverse estimation of excitation magnitudes from limited sampled data. The PINN training integrates governing differential equations with measured time-domain voltage responses to enhance robustness. Results show that the method accurately reconstructs excitation magnitudes and effectively identifies anomalous local excitations, validating its potential as a structural health monitoring sensor. Beyond anomaly detection, the device simultaneously harvests energy from ambient excitations across multiple frequencies, offering a compact, self-sustaining, and intelligent sensing solution for structural health monitoring applications.
Event(s)
20th Active and Passive Smart Structures and Integrated Systems, 16 March 2026 - 18 March 2026, Vancouver
Subjects
anomalous excitation detection
inverse vibration problem
multi-location monitoring
Physics-Informed Neural Network (PINN)
piezoelectric harvesters connected in parallel
Publisher
SPIE
Type
conference paper