Distributed Optical Fiber Sensing for Early Debonding Detection in Fatigue-Loaded Bonded Joints: Sensitivity to Training Data Composition and Load-Level Extrapolation

Distributed fiber optic sensing (DFOS) resolves strain along an adhesive bondline at millimetre spacing, but turning such data into a reliable damage indicator under changing and unknown load remains difficult. This paper presents an output-only reference framework for monitoring the debonding of carbon-fiber-reinforced polymer (CFRP) strips bonded to steel plates, with the optical fiber embedded in the adhesive layer, tested in four-point bending fatigue. A reference model trained on the healthy state predicts the strain profile of the bonded region from the strain at its centre, so that it follows the applied load without a force signal, and the residual is evaluated at each position against held-out healthy residuals with a combined absolute--relative threshold. On a specimen under stepped-amplitude loading, a linear regression or a multilayer perceptron reference localized strain anomalies consistent with debonding growing from both ends of the overlap, in the same phase of the test as the co-located strain gauges. An input-selection study showed that the position of the reference inputs, not the model architecture, decides whether damage is localized: once the inputs lie in the debonding zone, linear, neural, convolutional and PCA references alike smear the anomaly over the whole bondline, while a raw-strain Mahalanobis distance without load normalization flagged the load increase itself. The relative threshold coefficient was characterized as a trade-off between false alarms and detection delay under load extrapolation. On a second specimen under constant-amplitude loading, DFOS positions near a strain gauge reached the red zone 30–40 thousand cycles before the gauge departed from its baseline, the first confirmed alarm preceded the change of the global compliance by more than 150 thousand cycles, and no alarm occurred in the undamaged centre of the joint.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-05
DOI
https://doi.org/10.5281/zenodo.23761173
Primary Topic
Structural Health Monitoring Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Distributed Optical Fiber Sensing for Early Debonding Detection in Fatigue-Loaded Bonded Joints: Sensitivity to Training Data Composition and Load-Level Extrapolation

Karel Doubrava, David Bláha, Milan Růžička, Jaroslav Václavík et al.
Zenodo (CERN European Organization for Nuclear Research)
Structural Health Monitoring Techniques
preprint

Distributed Optical Fiber Sensing for Early Debonding Detection in Fatigue-Loaded Bonded Joints: Sensitivity to Training Data Composition and Load-Level Extrapolation

Karel Doubrava, David Bláha, Milan Růžička, Jaroslav Václavík, Milan Dvořák
preprint en

Abstract

Distributed fiber optic sensing (DFOS) resolves strain along an adhesive bondline at millimetre spacing, but turning such data into a reliable damage indicator under changing and unknown load remains difficult. This paper presents an output-only reference framework for monitoring the debonding of carbon-fiber-reinforced polymer (CFRP) strips bonded to steel plates, with the optical fiber embedded in the adhesive layer, tested in four-point bending fatigue. A reference model trained on the healthy state predicts the strain profile of the bonded region from the strain at its centre, so that it follows the applied load without a force signal, and the residual is evaluated at each position against held-out healthy residuals with a combined absolute--relative threshold. On a specimen under stepped-amplitude loading, a linear regression or a multilayer perceptron reference localized strain anomalies consistent with debonding growing from both ends of the overlap, in the same phase of the test as the co-located strain gauges. An input-selection study showed that the position of the reference inputs, not the model architecture, decides whether damage is localized: once the inputs lie in the debonding zone, linear, neural, convolutional and PCA references alike smear the anomaly over the whole bondline, while a raw-strain Mahalanobis distance without load normalization flagged the load increase itself. The relative threshold coefficient was characterized as a trade-off between false alarms and detection delay under load extrapolation. On a second specimen under constant-amplitude loading, DFOS positions near a strain gauge reached the red zone 30–40 thousand cycles before the gauge departed from its baseline, the first confirmed alarm preceded the change of the global compliance by more than 150 thousand cycles, and no alarm occurred in the undamaged centre of the joint.

Zenodo (CERN European Organization for Nuclear Research)
Research Centre Rez (CZ), Czech Technical University in Prague (CZ)
Structural Health Monitoring Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon