Session: 01-05-03: Data-Driven Design and Optimization of Aerospace Structures Using AI/ML

Paper Number: 175154

175154 - Vibration-Based Identification of Bolted Joint Conditions in Offshore Wind Turbine Jackets via Deep Learning 

This study presents a deep learning–based approach for the automatic detection and classification of bolted joint conditions in jacket-type structures used in offshore wind turbines. The analysis employs an experimental dataset which includes vibration signals acquired from eight triaxial accelerometers mounted on a scaled model of a wind turbine jacket type foundation. The foundation structure was composed of steel bars connected through bolted joints. The tested conditions comprise a healthy state (bolts tightened to 12 N·m) and three damage levels: partially loosened bolts (9 N·m and 6 N·m) and completely missing bolts. Using these vibration signals, a supervised deep learning model was developed to classify torque conditions and detect missing bolts with high accuracy. The proposed network achieved over 90% accuracy in distinguishing among the four structural classes, demonstrating strong robustness to amplitude variations and white noise interference. These results highlight the potential of deep learning techniques to enhance the reliability of structural integrity assessments in mechanical joints. Moreover, the proposed methodology contributes to the field of Structural Health Monitoring (SHM) by offering a scalable and data-driven framework for early fault detection in bolted connections, paving the way toward integration with digital twin architectures and predictive maintenance strategies in offshore wind energy systems.

Presenting Author: Jersson Xavier Leon Medina Universidad Pedagógica y Tecnológica de Colombia

Presenting Author Biography: PhD in Mechanical and Mechatronics Engineering of National University of Colombia. Particularly, his first PhD was related with pattern recognition and signal processing using machine learning and feature extraction methods in sensor arrays type electronic noses and tongues. PhD in Earthquake Engineering and Structural Dynamics of Polytechnic University of Catalonia in Barcelona Spain, his second PhD was related with the intelligent condition monitoring of structures through data-driven process using machine and deep learning methods, in particular, recurrent neural networks as GRU or LSTM and different dimensionality reduction methods used in structural damage classification. He obtained the Master of science (M.Sc) in mechanical engineering of National University of Colombia and Electromechanical Engineer degree from Pedagogical and Technological University of Colombia-UPTC. Jersson is disciplined, responsible and motivated in research, with personal skills of communication, teamwork, cognitive use of virtual tools and related areas of design engineering. He has English language proficiency. At june of 2025 He has published 20 journal articles belonging to the JCR and SJR. He has also made oral presentations at different leading academic conferences on structural health monitoring such as EWSHM, IWSHM and EURODYN.