PSI - Issue 54

Vasiliki Panagiotopoulou et al. / Procedia Structural Integrity 54 (2024) 482–489 Vasiliki Panagiotopoulou/ Structural Integrity Procedia 00 (2023) 000 – 000

489

8

due to the sheer volume of components vulnerable to corrosion-induced structural integrity issues, gathering data from all of them within the context of SHM becomes impractical, if not impossible. Transfer learning (TL) offers an innovative solution to these challenges by providing a method to align feature and label distributions for different structures, encompassing both labeled source and unlabeled target structures, within a common space. This approach allows classifiers trained on a labeled source structure to effectively generalize to a different, unlabeled target structure. Transfer learning, a subfield of deep learning, seeks to enhance a learner's performance in one domain by transferring knowledge from a related domain. Domain adaptation, a specific subcategory of TL, focuses on minimizing the distance between data distributions originating from source and target domains. Several researchers, including (X. Zhou, C. Sbarufatti, M. Giglio, L. Dong, 2023), have recently applied transfer learning techniques to detect damage within the context of SHM. Within the framework of SAMAS 2, domain adaptation techniques will be applied for exploiting experimental data obtained from a simple structure to leverage the training dataset of the ML algorithm for identifying corrosion damage on real components, described by similar yet distinct geometry, materials, or types of corrosion damage. 5. Conclusions and forthcoming work The results achieved thus far provide strong motivation to advance in this research endeavor. Notably, the Finite Element (FE) model of the TRDL has demonstrated commendable accuracy, while an extensive experimental campaign is currently underway to construct a corrosion database. Nevertheless, there is an urgent need to reduce the computational cost of the FE models to enable real-time applications for damage identification. Furthermore, the development of diagnostic and prognostic algorithms is ongoing, with a particular focus on enhancing their interpretability by integrating physical laws into the training process. SAMAS 2 is expected to conclude by December 2024. Acknowledgements This work has been developed based on the results from SAMAS 2 project (Structural health and ballistic impact monitoring and prognosis on a military helicopter), a Cat.-B project coordinated by the European Defense Agency (EDA) and financed by two nations, Italy and Poland. The project consortium includes the following parties: Italy (Politecnico di Milano, Leonardo S.p.A. - Helicopter Division, Consiglio Nazionale delle Ricerche) and Poland (Instytut Techniczny Wojsk Lotniczych - AFIT, Military Aviation Works No. 1, Institute of Aviation, Military University of Technology). References AL-Shudeifat, M. A. (2010). General harmonic balance solution of a cracked rotor-bearing-disk system for harmonic and sub-harmonic analysis: Analytical and experimantel approach. International Journal of Engineering Science 48 , 921-935. C. Sbarufatti. (2016). Sequential Monte-Carlo sampling based on a committe of artificial neural networks for posterior state estimation and residual lifetime prediction. International Journal of Fatigue 83 , 10-23. C. Zhang, D. Wang, R. Zhu, J. Li, P. Cao. (2023). Dynamic modeling and vibration characteristics analysis for the helicopter horizontal tail drive shaft system with the ballistic impact vertical penetrating damage. Iranian Journal of Science and Technology, Transactions of Mechanical Engineering 47 , 1177-1190. D. Cristiani, C. Sbarufatti, F. Cadini, M. Giglio. (2020). Fatigue damage diagnosis and prognosis of an aeronautical structure based on surrogate modelling and particle filter. Structural Health Monitoring . J.-J. Sinou, A. W. Lees. (2007). A non-linear study of a cracked rotor. European Journal of Mechanics A/Solids , 152-170. J.-J. Sinou, A.W. Lees. (2005). The influence of cracks in rotating shafts. Journal of Sound and Vibration , VOlume 285, Issues 4-5, 1015-1037. S. Pintos, N. Queipo, O. T. Rincon, A. Rincon, M. Morcillo. (2000). Artificial neural netowrk modeling of atmospheric corrosion in the MICAT project. Corrosion Science , Volume 42, Isuue 1, p.35-52. W. Liu, Z. Lai, K. Bacsa, E. Chatzi. (2022). Physics-guided Deep Markov Models for learning nonlineat dynamical systems with uncertainty. Mechanical Systems and Signal Processing 178 , 109276. X. Zhou, C. Sbarufatti, M. Giglio, L. Dong. (2023). A Fuzzy-set-based Joint Distribution Adaptation Method for Regression and its Application to Online Damage Quantification for Structural Digital Twin. Mechanical Systems and Signal Processing . Z. Lai, C. Mylonas, S. Nagarajaiah, E. Chatzi. (2021). Structural identification with physics-informed neural ordinary differential equations. Journal of Sound and Vibration , Volume 508, 116196.