Under in-plane compressive load conditions, the growth of a delamination initially induced by an impact can be followed by a fast growth after a threshold level, which leads to a catastrophic failure in composite structures. To avoid reaching this critical level, it is essential to uncover the delamination size and growth pattern in real time. Ultrasonic Guided Waves (UGW) have a strong capability to interrogate and monitor the structure in real-time and thus track the growth of damage, which may occur during the flight cycles. Although various types of damage affect the monitored UGW signals, it is challenging to determine from the UGW signals what types of damage and at what rate of growth are occurring within the structure. UGW signals can be acquired at defined intervals and then analysed to possibly detect different types of damages, such as delamination, and to quantify the rate of damage growth over fatigue cycles. However, correlating the UGW-based Damage Indicators (DIs) with the specific type of damage, such as delamination, and damage growth is a challenging task as the relation between these DIs and the actual damage state is very complex. Therefore, in this study, a supervised Deep Neural Network-based (DNN) prediction model is proposed aiming to diagnose the delamination size of the composite structure by correlating the UGW-based DIs with the quantified time-varying delamination size. UGW data is collected through a network of permanently installed piezoelectric transducers (PZTs). The delamination size is obtained through ultrasonic C-Scan technique at defined cycles. DIs are extracted in time, frequency, and time-frequency domains and used as the input for the DNN-based regression model. Each sensor-actuator path is considered as an independent set of indicators, which are separated for training, validation, and testing purposes. The effect of the different paths on the delamination size prediction is presented along with the model performance on measured delamination growth in woven type composite sample. @en

A health indicator (HI) serves as an intermediary link between structural health monitoring (SHM) data and prognostic models, and an efficient HI should meet prognostic criteria, i.e., monotonicity, trendability, and prognosability. However, designing a proper HI for composite structures is a challenging task due to the complex damage accumulation process during operational conditions. Additionally, designing a HI that is independent of historical SHM data (i.e., from the healthy state until the current state) is even more challenging as HI and remaining useful life prediction are time-dependent phenomena. A reliable SHM technique is required to extract informative time-independent data, and a powerful model is necessary to construct a proper HI from that data. The lamb wave (LW) technique is a useful SHM method that can extract such time-independent data. However, translating the LW data at each time step to the appropriate HI value is a challenge. AI—deep learning in this case—offers significant mathematical potential to meet this difficulty. A semi-supervised learning approach is developed to train the model using the simulated ideal HIs as the targets. The model uses the current LW data, without prior or subsequent data, to output the current HI value. Prognostic criteria are then calculated using the entire HI trajectory until the end-of-life. To validate the proposed approach, aging experiments from NASA’s prognostics data repository are used, which include composite specimens subjected to a tension-tension fatigue loading and monitored using the LW technique. The LW data is first processed using the Hilbert transform, and then, upper and lower signal envelopes in two states – baseline and current time – are used to feed the deep learning model. The results confirm the effectiveness of the proposed approach according to the prognostic criteria. The effect of different triggering frequencies of the LW system on the results is also discussed in terms of the prognostic criteria. @en

Developing comprehensive health indicators (HIs) for composite structures encompassing various damage types is challenging due to the stochastic nature of damage accumulation and uncertain events (like impact) during operation. This complexity is amplified when striving for HIs independent of historical data. This paper introduces an AI-driven approach, the Hilbert transform-convolutional neural network under a semi-supervised learning paradigm, to designing reliable HIs (fulfilling requirements, referred to as 'fitness'). It exclusively utilizes current guided wave data, eliminating the need for historical information. Ensemble learning techniques were also used to enhance HI quality while reducing deep learning randomness. The fitness equation is refined for dependable comparisons and practicality. The methodology is validated through investigations on T-single stiffener CFRP panels under compression-fatigue and dogbone CFRP specimens under tension-fatigue loadings, showing high performance of up to 93% and 81%, respectively, in prognostic criteria.@en