On the performance of a cointegration-based approach for novelty detection in realistic fatigue crack growth scenarios

Abstract Confounding influences, such as operational and environmental variations, represent a limitation to the implementation of Structural Health Monitoring (SHM) systems in real structures, potentially leading to damage misclassifications. In this framework, this study considers cointegration as a state of the art method for data normalisation in fatigue crack propagation scenarios, where monitoring is performed by a distributed network of strain sensors. Specifically, the work is aimed at demonstrating the effectiveness of cointegration on real engineering data in a new context, where the damage is continuously growing. Cointegration is applied at first in a controlled scenario consisting of a numerical strain simulation by means of a finite element model, modified in order to take realistic temperature fluctuations and sensor noise into account. Afterwards, detrending and anomaly detection performances are verified in two different experimental programmes on realistic aeronautical structures subjected to fatigue crack growth, including a full-scale fatigue test on a helicopter tail boom. Strain measurements are taken from a network of Fibre Bragg Grating (FBG) sensors, known to be extremely sensitive to temperature variations, hence delivering challenging scenarios for cointegration testing. Results are shown to be in good agreement with the experimental evidence, with the cointegration algorithm capable of detecting the onset of damage propagation within a 4 mm increment from a baseline condition.