262 research outputs found
Data Set of PLOS Computational Paper PCOMPBIOL-D-18-02181R1
No full textFigures Data of PLOS Computational paper:Modeling of the axon plasma membrane structure and its effects on protein diffusionAuthors: Yihao Zhang, Anastasios V. Tzingounis, and George LykotrafitisCorresponding Author: George Lykotrafitis, Ph.D.University of ConnecticutStorss, CT UNITED STATES</div
The state of modern Greek language as spoken in Victoria
No full textDeposited with permission of the author. © 1986 Dr. Anastasios TamisThis thesis reports a sociolinguistic study, carried out between 1981 and 1984, of the state of the Modern Greek (MG) language in Australia, as spoken by native-speaking first-generation Greek immigrants in Victoria. Particular emphasis is given to the analysis of those characteristics of the linguistic behaviour of these Greek Australians which can be attributed to the contact with English and to other environmental, social and linguistic influence. (For complete abstract open document
New historical evidence for Anastasios Emm. Papas
No full textNo AbstractThe author’s attention has been drawn to the existence of this historicalevidence in the National Archives of Vienna, by his friend the writer EteoclesGregoriadis together with the numbers of the relevant files. Most of the documents were written in the old German script. Thus the author asked for the help of his friend and former colleague at the University of Thessaloniki and director of the Goethe Institute, Graf Kurt v. Posadowsky, for reading andstudying those documents. Without his help this study would have been impossible. This new evidence concerns the sojourn of Anastasios Papas·—son of Emmanuel Papas, leading figure of the Greek Revolution—in Austria andGermany between the 3rd January and 11th March 1822. There is informationabout his short imprisonment in Trieste, after his arival from Vienna. He then visits various towns in Germany and after negotiations with the Philhellene professor Fr. Thiersch in Munich, he purchases large quantities of ammunition to be despatched to Greece. He finally arrives in Greece early in 1824, and takes part—together with his three brothers who were already fighting—in the struggle for the liberation of the common great fartheland
Ευρωστία σε εκτός κατανομής δεδομένα για κρίσιμες εφαρμογές της όρασης υπολογιστών
No full textArtificial intelligence (AI) has progressed explosively in recent years. Driven by the advent of deep learning, AI is being used in a variety of applications, across multiple scientific fields, in industry as well as in medicine. Out-of-distribution (OOD) robustness is crucial in mission-critical computer vision applications because these scenarios often involve encountering unforeseen or novel situations that may differ significantly from the training data. In mission-critical contexts, such as autonomous vehicles, medical diagnosis, or security systems, the models need to make reliable and safe decisions. If the model encounters situations or inputs that fall outside the distribution it was trained on, it may provide inaccurate or unreliable predictions, leading to potentially dangerous consequences. Ensuring OOD robustness is essential to enhance the generalization capabilities of computer vision models, enabling them to handle diverse and unexpected scenarios in real-world applications. It helps prevent the system from making critical errors when faced with novel inputs, thereby improving safety, reliability, and performance in mission-critical tasks. The emergence of Out-of-Distribution (OOD) robustness or Domain Generalization research has become a crucial tool for achieving reliable performance in medical imaging and autonomous driving. In the context of medical imaging, OOD robustness is vital because medical datasets can vary significantly due to differences in patient demographics, imaging equipment, and conditions. Researchers and practitioners recognize the need for models that can generalize well to diverse and previously unseen medical scenarios to ensure accurate diagnoses and treatment plans. Similarly, in autonomous driving, OOD robustness is essential as driving conditions can be highly dynamic and unpredictable. Ensuring that self-driving vehicles can handle unforeseen scenarios, such as adverse weather conditions, unusual environment configurations, or unexpected obstacles, is critical for their safe deployment in the wild. OOD robustness research in both medical imaging and autonomous driving aims to enhance the generalization capabilities of machine learning models, enabling them to perform reliably in real-world scenarios beyond the training distribution. This research contributes to the development of more trustworthy and resilient systems in these mission-critical domains. This study proposes methodologies and advancements aimed at enhancing OOD robustness in mission-critical applications. From transfer learning techniques tailored for medical imaging to novel sensor configurations for UAV perception systems and state-of-the-art deep learning architectures for image recognition, significant progress has been made in addressing the challenges posed by OOD data. In the domain of medical imaging, we explored methodologies for enhancing the generalization capabilities of diagnostic models, considering factors such as data heterogeneity, limited sample sizes, and domain shifts across different healthcare facilities. For UAV sense and avoid systems, we investigated techniques for perceptual robustness to ensure safe operation in dynamic environments. In image recognition, we examined approaches for mitigating the impact of OOD data, such as adversarial training, domain generalisation, and uncertainty estimation, to enhance model reliability across diverse datasets and environmental conditions. In summary, this PhD thesis highlights the critical importance of OOD robustness in mission-critical applications and underscores the need for continued research and innovation in this area. By synthesizing insights from diverse studies and identifying key challenges and advancements, this PhD thesis aims to contribute to the ongoing discourse on enhancing the reliability and safety of AI-driven systems in real-world scenarios. Through interdisciplinary collaboration and rigorous experimentation, we strive to develop effective solutions that ensure the resilience and efficacy of AI technologies across medical imaging, UAV sense and avoid systems, and image recognition domains.Η τεχνητή νοημοσύνη AI έχει προχωρήσει εκρηκτικά τα τελευταία χρόνια. Η τεχνητή νοημοσύνη και η Βαθιά Μάθηση χρησιμοποιείται σε ποικίλες εφαρμογές, σε πολλά επιστημονικά πεδία, τόσο στη βιομηχανία όσο και στην ιατρική. Η ευρωστία σε εκτός κατανομής δεδομένα OOD είναι ζωτικής σημασίας σε εφαρμογές που είναι κρίσιμες για την αποστολή, επειδή αυτά τα σενάρια συχνά συνεπάγονται αντιμετώπιση μη-προβλεπόμενων ή νέων καταστάσεων που μπορεί να διαφέρουν σημαντικά από τα δεδομένα εκπαίδευσης. Σε κρίσιμα για την αποστολή πλαίσια, όπως αυτόνομα οχήματα, ιατρικά διάγνωση, ή συστήματα ασφαλείας, τα μοντέλα πρέπει να είναι αξιόπιστα και με ασφαλείς αποφάσεις. Εάν το μοντέλο συναντήσει καταστάσεις ή εισόδους που δεν εμπίπτουν στηη κατανομή στην οποία εκπαιδεύτηκε, μπορεί να προκύψουν ανακριβείς ή αναξιόπιστες προβλέψεις, οδηγώντας σε δυνητικά επικίνδυνες συνέπειες. Εξασφάλιση OOD ευρωστίας είναι απαραίτητη για τη βελτίωση των δυνατοτήτων γενίκευσης του αλγορίθμου σε μοντέλα όρασης, που τους επιτρέπουν να χειρίζονται διαφορετικά και απροσδόκητα σενάρια σε πραγματικές εφαρμογές. Βοηθά να αποτρέψει το σύστημα από το να γίνουν κρίσιμα σφάλματα όταν αντιμετωπίζονται νέες εισροές, βελτιώνοντας έτσι την ασφάλεια, την αξιοπιστία, και απόδοση σε κρίσιμα περιβάλλοντα. Η εμφάνιση της ευρωστίας εκτός κατανομής OOD ή του Τομέα Γενίκευσης έχει γίνει ένα σημαντικό εργαλείο για την επίτευξη αξιόπιστης επιδόσης στην ιατρική απεικόνιση και την αυτόνομη οδήγηση. Στο πλαίσιο της ιατρικής απεικόνισης, η ευρωστία OOD είναι ζωτικής σημασίας επειδή τα ιατρικά σύνολα δεδομένων μπορεί να ποικίλλουν σημαντικά λόγω των διαφορών στα δημογραφικά στοιχεία των ασθενών και τον εξοπλισμό απεικόνισης. Οι ερευνητές και οι επαγγελματίες αναγνωρίζουν την ανάγκη για μοντέλα που μπορούν να γενικεύουν καλά σε ποικίλα και άγνωστα προηγουμένως ιατρικά σενάρια για την εξασφάλιση ακριβών διαγνώσεων και σχεδίων θεραπείας. Ομοίως, στην αυτόνομη οδήγηση, η ευρωστία OOD είναι απαραίτητη, καθώς οι συνθήκες οδήγησης μπορεί να είναι εξαιρετικά δυναμικές και απρόβλεπτες. Η διασφάλιση ότι τα αυτόνομα οχήματα μπορούν να χειριστούν απρόβλεπτα σενάρια, όπως αντίξοες καιρικές συνθήκες, ασυνήθιστες διαμορφώσεις περιβάλλοντος ή απροσδόκητα εμπόδια, είναι κρίσιμη για την ασφαλή ανάπτυξή τους. Η έρευνα ευρωστίας OOD τόσο στην ιατρική απεικόνιση όσο και στην αυτόνομη οδήγηση στοχεύει να ενισχύσει τις δυνατότητες γενίκευσης των μοντέλων μηχανικής μάθησης, επιτρέποντάς τους να αποδίδουν αξιόπιστα σε πραγματικά σενάρια πέρα από τη κατανομή εκπαίδευσης. Αυτή η έρευνα συμβάλλει στην ανάπτυξη πιο αξιόπιστων και ανθεκτικών συστημάτων σε αυτούς τους κρίσιμους για την αποστολή τομείς. Στα πλαίσια της Διαδακτορικής Διατριβής αναπτύξαμε καινοτόμες μεθόδους που σχετίζονται με την ευρωστία OOD στην ιατρική απεικόνιση, τα συστήματα αίσθησης και αποφυγής UAV και την αναγνώριση εικόνας. Στον τομέα της ιατρικής απεικόνισης, αναπτύξαμε μεθοδολογίες για τη βελτίωση των δυνατοτήτων γενίκευσης των διαγνωστικών μοντέλων, λαμβάνοντας υπόψη παράγοντες όπως η ετερογένεια των δεδομένων, τα περιορισμένα μεγέθη δειγμάτων και οι μετατοπίσεις τομέα σε διαφορετικές εγκαταστάσεις υγειονομικής περίθαλψης. Για συστήματα αίσθησης και αποφυγής UAV, δημιουργήσαμε τεχνικές αντιληπτικής ευρωστίας για να διασφαλίσουμε την ασφαλή λειτουργία σε δυναμικά περιβάλλοντα. Στην αναγνώριση εικόνων, εξετάσαμε προσεγγίσεις για τον μετριασμό του αντίκτυπου των δεδομένων OOD, όπως η εκπαίδευση σε αντίθεση, η γενίκευση τομέα και η εκτίμηση αβεβαιότητας, για να ενισχύσουμε την αξιοπιστία του μοντέλου σε διάφορα σύνολα δεδομένων. Μέσω αυτής της Διδακτορικής Διατριβής ανακαλύψαμε πολλές μεθοδολογίες και προόδους που στοχεύουν στην ενίσχυση της ευρωστίας OOD σε κρίσιμες εφαρμογές, από τεχνικές μεταφοράς μάθησης προσαρμοσμένες για ιατρική απεικόνιση έως καινοτόμες διαμορφώσεις δεδομένων για συστήματα αντίληψης UAV και αρχιτεκτονικές βαθιάς μάθησης για την αναγνώριση εικόνων. Έχει σημειωθεί σημαντική πρόοδος στην αντιμετώπιση των προκλήσεων που τίθενται από τα δεδομένα OOD. Επίσης, αυτή η Διδακτορική Διατριβή υπογραμμίζει την σημασία της ευρωστίας OOD σε κρίσιμες εφαρμογές και υπογραμμίζει την ανάγκη για συνεχή έρευνα και καινοτομία σε αυτόν τον τομέα. Με τη δημιουργία καινοτόμων αλγορίθμων, τη σύνθεση γνώσεων από διάφορες μελέτες και τον εντοπισμό βασικών προκλήσεων και προόδων, αυτή η Διδακτορική Διατριβή στοχεύει να συμβάλει στη συνεχή συζήτηση για την ενίσχυση της αξιοπιστίας και της ασφάλειας των συστημάτων που βασίζονται στην τεχνητή νοημοσύνη σε σενάρια πραγματικού κόσμου. Μέσω διεπιστημονικής συνεργασίας και πειραματισμού, αναπτύξαμε αποτελεσματικές λύσεις που διασφαλίζουν την ανθεκτικότητα και την αποτελεσματικότητα των τεχνολογιών τεχνητής νοημοσύνης σε ιατρικές απεικονίσεις, συστήματα αίσθησης και αποφυγής UAV και τομείς αναγνώρισης εικόνων
Measurement of damage growth in ultrasonic spot welded joint
No full textUltrasonic spot welding is a joining technique for thermoplastic composites with great potential regarding processing speed and cost. To investigate the damage tolerance and possible inherent damage arresting behavior of multi-spot welded joints, a technique is necessary to measure damage growth in the joints under cyclic loading. Visual inspection is not possible because the damage is not located on the outside surface and conventional techniques such as C-scan are not practical during a fatigue test because the specimen would have to be removed from the setup. This paper details a methodology for quantifying damage growth rates in singlespot welded joints using surface strain measurements made by Digital Image Correlation. This represents the first step towards developing a methodology for quantifying damage progression behavior in complex multi-spot welded joints.Structural Integrity & CompositesAerospace Structures & Computational Mechanic
Development and characterization of hybrid thin-ply composite materials
No full textThin-ply composites are recognized as a key solution for the manufacturing of high-performance composite structures due to the unique mechanical properties and the increased design versatility that they offer. They are obtained with state-of-the-art fiber spreading methods where high-count (6-24K filaments) tows of technical fibers (carbon, glass) are thinned by spreading into flat unidirectional tapes which are then combined with a polymer matrix to create pre-impregnated (prepregs) tapes of reduced thickness. In recent years, the industrialization of fiber spreading and impregnation processes enabled the large-scale production of homogenous thin-ply prepregs with thicknesses down to about 15μm per ply, which attracted the interest of the research community. However, the high production cost due to the complexity of the manufacturing methods and the inherent brittleness of thin-ply composites limit their wider adoption by the composites industry[1]. Fiber hybridization (i.e combining at least two types of fibers in a common matrix) is emerging as a promising approach for alleviating these drawbacks towards laminates with balanced characteristics in terms of mechanical properties and cost-efficiency. Currently, most studies on thin-ply hybrids employ simple interlayer (ply-by-ply) configurations mainly due to difficulties in manufacturing of more complex hybrid architectures[2]. However, simulation tools predict that notable improvements can be obtained from more complex intralayer (tow-by-tow) and intrayarn (fiber-by-fiber) hybrid architectures[3]. This work focuses on the study of existing fiber spreading methodologies, the development of equipment, and the optimization of composite processing at North Thin Ply Technology (NTPT) Renens, Switzerland, that allowed the manufacturing of hybrid composites with a high degree of fiber dispersion and controlled microstructure. Hybrid prepregs were produced by combining various ratios of dissimilar fibers following different processing routes. Composite laminates were manufactured and a versatile microstructural analysis tool was developed that enabled correlations between the manufacturing route, the resulting microstructural features describing the degree of co-dispersion, and the mechanical performance of the final part. Acknowledgments The research leading to these results has been performed within the framework of the HyFiSyn project and has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 765881. Delamination growth in fibre reinforced polymer composites is generally evaluated with experiments that have been standardized for quasi-static load conditions. These tests characterize unidirectional delamination growth in mode I (DCB), mode II (ELS or ENF) of mixed mode conditions (MMB). However, little attention is paid in literature to the applicability of these tests to in-service delamination problems that are generally characterized by planar delamination growth. In this study, the relation between planar delamination growth, induced by transverse quasi-static indentation loading, and these unidirectional delamination tests was investigated. To that aim, prior planar delamination growth tests reported in literature, performed at EPFL, were analysed to identify up to what extent this planar growth could be correlated to the concepts of strain energy release and strain energy density. Once this appeared to successful, an experimental setup was designed to measure the delamination boundary during the transverse indentation loading of planar delamination specimens made of nontransparent carbon fibre reinforced polymer composites. With that set-up, quasi-static and fatigue planar delamination growth experiments were performed, and delamination contours could be successfully captured. While the quasi-static tests revealed limited growth, evaluation with numerical simulations revealed that the indentation force required to extend the delamination quasi-statically would cause damage to the specimen. This is attributed to the increasing length of the delamination contour when delaminations expand, which is not the case with standard unidirectional specimen. With the fatigue tests, however, delamination growth was achieved, but interestingly enough two phases were observed; first the delamination propagated in a planar fashion, while at some point in time work did not exceed an apparent threshold. Instead of no growth, however, the delamination still increased but then in a transverse manner. What makes this study of particular interest, is that the strain energy density as criterion could capture the strain energy offered (work) along the entire delamination contour, while the strain energy release rate described the resistance to delamination growth. This latter observation is in agreement with the original concept employed by Griffith when he formulated the basis of linear elastic fracture mechanics. This presentation present the experiments performed, the analysis of results, and will conclude with a proposal how to relate standard unidirectional tests to planar growth, considering that these standard tests contain little to no information on transverse phenomena with respect to strain energy density (work) and strain energy release (dissipation).Aerospace Manufacturing Technologie
The role of matrix boundary in the microstructure of unidirectional composites
No full textFinding new ways to evaluate the variability of microstructures, and its effect on macroscopic properties such as permeability and mechanical performance [1,2] is of increasing interest in the composite field. The variability of microstructural features at a three-dimensional level is not fully understood and its effect on macroscale properties is not well established, and mostly analyzed at a phenomenological level [3]. We introduced in recent work a method based on X-ray Computed Tomography for the threedimensional reconstruction of the fibrous microstructure of unidirectional tapes at a single fibre resolution [4]. A schematic of the workflow is represented in Figure 1. Three descriptors are introduced in the work to describe increasing level of complexity in the microstructural organization, from a single fiber path level with differential tortuosity, to group behavior with collective motion, to fibre network connectivity with length of contact. These descriptors and their interdependence highlight local effects like edge-core segregation in microstructural characteristics. However, in order to achieve a more complete definition of the unidirectional tape domain, understanding of matrix-based features and its interrelation with fiber architecture descriptors is needed. In this work, we expand the methodology of Gomarasca et al. [4], to account for matrix-based phenomena such as tape boundary variability, and void formation and morphology. This will be showcased on a unidirectional composite tape including both fiber-based and matrix-based analysis. These methods enable advanced characterization and modelling of microstructural formation and evolution during composite manufacturing.Aerospace Manufacturing TechnologiesAerospace Structures & Computational Mechanic
Deep learning based prediction of fibrous microstructure permeability
No full textKnowledge of permeability of fibrous microstructures is crucial for predicting the mold fill times and resin flow path in composite manufacturing. Herein we report a method to rapidly predict the permeability of 3D fibrous microstructures. Our method relies on predicting the permeability of 2D cross-sections via deep neural networks and extending this capability to 3D microstructures via circuit analogy as a means of reduced order modeling. Approximately 50% of the permeability predictions of 2D cross-sections have 10% or less deviation from the permeability results obtained via flow simulations in Geodict. Computational time required for predicting the permeability of 3D microstructures is reduced from hours to less than 10 seconds. This framework enables fast and accurate prediction of micro-permeability and serves as the first building block towards prediction of fabric mesostructures’ permeability via deep learning based methods.Aerospace Manufacturing Technologie
- …
