Monitoring Fatigue of Grouted Connections Using Autoencoders and In-Situ Fiber Optical Sensors

Abstract

This paper investigates a traditional autoencoder trained with time series data for unsupervised damage detection and localization. The results are validated using innovative fiber optical sensory inside the grouted connection of a monopile sub-structure. Sensors are installed due to a limited understanding of the degradation mechanisms in grouted connections under cyclic loading. In most experiments, the grouted connection has only been assessed by destructively examining the grouted connection after the appearance of a complete crack pattern. To address these limitations, the signals of the innovative fiber optical sensory are used to gain insight into the crack development inside the grout. Further, the dynamic response of the structure is evaluated using an autoencoder to monitor the changes linked to the degradation process. The fiber sensors are placed within the grout between the diagonal shear keys, which has been identified as the compression strut failure from previous experiments. The cyclic load was repetitively increased to simulate the degradation process and induce structural change inside the grouted connection. After each testing phase, the structure was excited using a shaker, and the acceleration sensors were evaluated using the autoencoder. The results of the study show a high sensitivity of the residuals to minor structural changes that can be explained using the sensors within the grout. Additionally, the local stiffness reduction can be localized evaluating the autoencoder residuals. The study presents the innovative sensory to follow the crack opening of the grout due to cyclic loading. Further, these local changes are detected and localized using an autoencoder with time series data in vibration-based SHM.