Filtern
Dokumenttyp
- Zeitschriftenartikel (3)
- Vortrag (2)
- Beitrag zu einem Tagungsband (1)
Sprache
- Englisch (6)
Schlagworte
Organisationseinheit der BAM
Eingeladener Vortrag (wissenschaftliche Konferenzen)
- nein (2)
Composite overwrapped pressure vessels (COPV) for hydrogen will play an important role in emission free mobility and energy storage. The DELFIN project (2018 – 2022) was set up with the goal of developing improved COPVs that are lighter, cheaper and more durable. A key aspect of this project was to conduct full-scale impact tests with different gas pressure levels that were used to evaluate the vessels’ crash performance and to verify numerical impact simulations. These tests were followed by controlled pressure experiments until the burst pressure was reached. The procedure was monitored with acoustic emission testing (AT) and subsequent computer tomography (CT). The residual COPV stability and the AT results are compared to different scenarios regarding the impact energies and impact angles. The event localization from AT shows a clear correlation of the emitted energy with the damaged areas. We classify the recorded signals into groups that can be related to the failure mechanism and compare those to the pressure evolution before failure of the COPV. All tests indicate that large impact energies lead to a significant reduction of the burst pressure of COPVs, whereas a higher internal pressure can have a stabilizing effect that reduces the damaging effect.
A good understanding of the structural stability of hydrogen composite overwrapped pressure vessels (COPV) is important for the cost-effective design and safe operation of hydrogen storage systems. Acoustic emission (AE) monitoring is a non-destructive method sensitive to microstructural damages such as e.g. fiber breakage, and matrix cracking in COPVs. This study proposes a novel approach for damage monitoring by integrating acoustic emission techniques with machine learning (ML) algorithms to classify and predict damage types in COPVs. However, training accurate classification models requires extensive labeled datasets, which are very challenging to generate due to the nature of AE signal data and the lack of in-situ observations of microscopic failures in COPVs. Our research overcomes this limitation by automating the labeling process of AE signal data for different COPVs using unsupervised ML methods. The most representative features were extracted and then selected from recorded AE signals. Different unsupervised clustering algorithms were utilized based on various extracted feature combinations. The most stable clustering result was achieved and later used as appropriate labels for training classification algorithms. A deep neural network-based deep learning (DL) architecture was used to train discriminative models on AE data, identify patterns, and classify damage types into different classes with improved accuracy and speed for each COPV. Results demonstrate the potential of the proposed combined deep learning approach to train predictive models in identifying failure patterns. The trained models based on individual COPVs show high training, validation, and test accuracy for unseen datasets and offer enhanced predictive capabilities by following advanced DL techniques compared to traditional monitoring methods. The proposed method highlights its potential to improve the efficiency and safety of hydrogen storage systems.
A total of nine heavily damaged carbon-fibre overwrapped pressure vessels are tested. The burst pressure of the damaged vessels correlates with the impact energy and the internal pressure during impact. No internal pressure leads to larger deformation and lower burst pressure. The vessels are tested with acoustic emissions prior to bursting. It can be shown that major instabilities give rise to a significantly increased number of acoustic signals. The location of occurrence corresponds to the impact regions. The signals are studied in terms of their frequency content and it can be shown that high frequency signals experience strong damping so that there are only few recorded high-frequency events and even fewer that are locatable. The data is further used in a cluster analysis to assign the signals to different failure types. With the given experimental design, no clear patterns emerge so that the failure association
remains ambiguous.
Two carbon fiber reinforced type IV pressure vessels are subjected to step-wise pressurization until burst, while monitored using acoustic emissions (AE). Unlike most prior studies, AE data is collected throughout the entire damage progression. The vessels, manufactured with differing parameters, failed in distinct composite layers – A-type in the hoop layers and B-type in the helical layers. The AE signals are evaluated to study material degradation and identify fiber breaks as signs of critical damage accumulation. The signals are distributed randomly across the surface, with localized accumulation only minutes before rupture, close to the rupture plane. The difference in manufacturing parameters did not result in any clear difference in the AE activity. Felicity and Shelby ratios show consistent decline with increasing pressure, suggesting potential for damage assessment and burst prediction. It is discussed how these ratios are affected by coupling quality of the AE sensors, the shape of the pressurization profile and prior loadings. Different signal features based on the amplitude and the frequency content are extracted for a classification into failure mechanisms. Based on previous studies, AE signals corresponding to fiber breaks have a characteristic high-frequency spectrum and show a delay in occurrence, with an increase in the number of breaking fibers towards the end of the experiment. Indeed, high frequency signals tend to occur later and signals in specific peak-frequency ranges (350 – 400 kHz, 500 kHz) somewhat resemble the expected behavior. However, the dataset is too variable and too incongruent for any clear interpretations. Likely reasons are signal propagation effects, the complex composite structure, simultaneous occurrence of signals and measurement uncertainties. A review of relevant studies is provided to show that similar issues affect also previous works. Successfully identifying fiber breaks in large-scale, complex composite structures based on AE data, and turning this into an applicable health-monitoring technique, therefore remains a challenge.
Accurate damage classification of Composite Pressure Vessels (CPVs) is crucial for understanding failure behaviour of hydrogen storage systems. Acoustic Emission (AE) monitoring is a non-destructive testing technique capable of detecting signals from different failure mechanisms such as fiber breakage and matrix cracking, supporting durability assessment of CPVs. Therefore, the main objective of this study is to combine AE and advanced deep learning techniques to develop a robust framework for automatic and accurate identification and classification of damage mechanisms across various CPVs.
The evolutionary Genetic Algorithms (GA) was used for feature selection, followed by unsupervised clustering to generate automatic labels for model training. Two different FCNN and CNN-LSTM architectures were used to train individual models based on different AE datasets. Later, Adaptive Transfer Learning (ATL) and Meta Ensemble Learning (MEL) techniques were applied to handle data variability and train predictive generalized model over varied AE datasets. The ATL fine-tunes a pre-trained models to leverage their knowledge, while MEL uses pre-trained models’ predictions as meta features to train a meta model.
Experimental results demonstrate that while both generalized ATL and MEL trained models perform well across different AE datasets, the MEL framework outperforms ATL method in terms of evaluation metrics. The Mean-Accuracy score reaches 0.9026, and 0.9900 for ATL, and MEL, respectively. The most accurate multi-class classification results was achieved using MEL method in terms of the Mean-Accuracy and Recall metrics. The proposed framework provides a scalable, adaptive approach for automated damage classification using AE signals across diverse CPVs in real-world settings.
