A DATA DRIVEN BASED METHODOLOGY FOR STURCTURAL HEALTH MONITORING WITH DISTRIBUTED OPTICAL FIBRE SENSORS

Abstract

Structural health monitoring (SHM) is a means for maintaining structural integrity, safety and reliability by analysing various structural responses (i.e., mechanical signals) to pinpoint the anomalies of the structures due to damage. It is not an easy task to filter the noise and fluctuation of mechanical signals to successful find the damage-induced anomalies, but it might be achieved by machine learning algorithms. However, the successful implementation of a machine learning requires a large amount of training data, which is always available. In this work, a novel machine learning (ML) model, combining k-nearest neighbors kernel (KNN) and deep neural network (DNN), was proposed that can be trained by insufficient/incomplete SHM data. In addition, the damage states can be identified by Kernel Principle Component Analysis (KPCA). To demonstrate the accuracy of this model, training and validation data were taken from the strains of the braided composite beam under progressive three-point bending. The strain signals were measured by embedded distributed optical fibre sensors (DOFS). The prediction of the proposed novel ML model demonstrates a good agreement with the experimental observations for validation, which provides a novel approach for sufficient/incomplete training data. © 2023 International Committee on Composite Materials. All rights reserved.