Bridging the gap: Enhancing structural damage localization accuracy using multi-modal sensor fusion

Electromagnetic interference and signal attenuation make traditional piezoelectric sensors (PZT) more susceptible to unreliable structural health monitoring outcomes. This study proposes a novel, multi-modal sensing framework that fuses electrical PZT actuation with optical fiber Bragg grating sensing to achieve robust damage localization. An end-to-end 1D-convolutional neural network is designed to process raw temporal wave signals across a broadband frequency range of 50–250 kHz. Unlike conventional methods that rely on manual, physics-heavy feature extraction, the proposed convolutional neural network architecture learns temporal and spatial wave–damage interactions directly from standardized 1D signals. The model is validated using a five-fold cross-validation scheme and an extensive ablation study to facilitate scientific rigor. The results demonstrate a significant improvement in performance. While a PZT-only configuration achieved an accuracy of only 27% (near-random), the hybrid fusion model achieved an aggregated accuracy of 66% and a Macro-F1 score of 0.66. The model shows good sensitivity in distinguishing symmetric damage locations and identifying remote damage sites. This research provides a data-driven justification for hybrid sensing architectures, proving that the integration of optical strain data with electrical excitation is essential for overcoming the signal-to-noise limitations of monomodal structural health monitoring systems. This framework offers a good scalable, interference-immune solution for real-time aerospace and civil infrastructure monitoring.