PSI - Issue 54

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

487

6

Networks (NNs) are employed to approximate complex computer simulations with easily assessable functions (D. Cristiani, C. Sbarufatti, F. Cadini, M. Giglio, 2020). 3.4. The Grey-box modelling approach Surrogate models, which rely on numerical models, can effectively leverage experimental data to offer insights and potential warnings regarding the condition of the monitored structure. However, conventional Machine Learning algorithms (i.e., ANNs) often operate as black boxes, generating predictions that may lack physical consistency. On top that, the scarcity of available data covering the complete environmental and operational range of the structure is an additional factor that is frequently overlooked in many cases. To address these limitations, numerous researchers, including (Z. Lai, C. Mylonas, S. Nagarajaiah, E. Chatzi, 2021) and (W. Liu, Z. Lai, K. Bacsa, E. Chatzi, 2022), have taken steps to incorporate physics into the training phase of ML algorithms. This approach is aimed at guiding the training process toward more physically consistent estimations. In the present study, ANNs are constrained by known physical relations (i.e., governing equation of motion) representing the physics-informed component, trained using a smaller amount of numerical data provided by the high-fidelity reduced order model. This physics-informed model introduces physical meaning into the training process, replacing certain parts of training and enabling a better comprehension and interpretation of results, while maintaining a high level of performance and flexibility to various structural monitoring problems. The proposed Physics-Informed Neural Network is utilized as a damage identification tool for the transmission shaft of a military helicopter in the event of a ballistic impact. 4. SHMP system for corrosion monitoring Corrosion is a highly intricate problem, necessitating substantial financial resources and effort for inspections and maintenance activities. In helicopters, corrosion poses a significant threat to flight safety as it can lead to stress raisers and potential hazards for structural components, and consequently to the entire helicopter. Traditionally, maintaining corrosion degradation has relied on frequent inspections, which are costly and reduce the helicopter's operational availability. Efforts have intensified in recent years to establish a reliable framework for online corrosion monitoring, primarily for diagnostic and prognostic purposes. Corrosion monitoring involves assessing and observing structures for signs of corrosion, with the goal of prolonging asset life, enhancing safety, and cutting down on maintenance and inspection expenses. Online corrosion monitoring provides a crucial advantage by detecting early signs of corrosion through trend analysis and monitoring parameters that may contribute to a corrosive environment, such as temperature, pressure, and pH. Additionally, the SHM tool evaluates the effectiveness of corrosion prevention methods, like protective coatings, to determine if alternative inspection and protection techniques are necessary. Within the SAMAS 2 project framework, three types of corrosion damage are investigated and tested. The first group of experimental tests involves artificial corrosion, achieved through electrochemical attacks, and accelerated corrosion, conducted in a Salt Spray Chamber. Additionally, tests on natural corrosion are conducted by exposing representative coupon specimens and full-scale components to marine environments. In all three types of corrosion tests, numerous specimens, including lamina or coupon samples, as well as real components made from various materials like steel alloys and aluminum alloys, are exposed to corrosive conditions. This extensive and diverse database is later utilized by complex algorithms for diagnostic and prognostic purposes. A brief introduction on sensing technologies for corrosion monitoring, categorized into direct and indirect methods, is presented. Direct sensors interact with physical fields affected by structural damage, while indirect sensors measure environmental factors that can influence corrosion rates or properties of reference specimens exposed to corrosive conditions. These indirect methods establish correlations between environmental parameters and corrosion rates, validated by the SHMP tool for accuracy. In the direct corrosion monitoring category, PZT piezoelectric transducers and Eddy Current sensors are highlighted as technologies capable of estimating corrosion damage and detecting fatigue cracks. In contrast, indirect corrosion monitoring employs sensors like the Acuity LUNA system to track environmental factors such as temperature, humidity, and contamination, contributing to the estimation of corrosion growth rates.