Filtern
Dokumenttyp
- Zeitschriftenartikel (5) (entfernen)
Sprache
- Englisch (4)
- Mehrsprachig (1)
Schlagworte
- Machine Learning (5) (entfernen)
Organisationseinheit der BAM
Die AG Fraktographie im DVM/DGM-Gemeinschaftsgremium „Elektronenmikroskopie in der Materialforschung“ betreibt seit 2013 eine fraktographische online-Datenbank („FractoDB“), die für das interessierte Fachpublikum kostenfrei zur Verfügung steht. Die Analyse und Bewertung von Bruchflächen und der darauf befindlichen Bruchmerkmale ist ein wichtiger Teilaspekt der Schadensanalyse bei der Suche nach Schadensmechanismen und Schadensursachen. Risse und Brüche an realen Bauteilen können nur dann bewertet werden, wenn gut dokumentierte Vergleichsbrüche aus Laborversuchen zur Verfügung stehen, sei es an Proben oder an Vergleichs-Bauteilen. Daher trägt die AG Fraktographie Bildmaterial zusammen, führt systematisch Laborversuche und Ringversuche durch und analysiert Brüche aus Schadensfällen. Die gewonnenen Daten werden zu Datensätzen zusammengestellt und über die Datenbank zugänglich gemacht. Derzeit ist ein Bestand von über 400 Datensätzen mit insgesamt über 4.500 Bildern verfügbar, welcher durchsucht werden kann und in Anlehnung an die VDI 3822 organisiert ist. Weitere Aktivitäten der AG Fraktographie, die in der FractoDB abgebildet werden, sind u.a. die Entwicklung einer fraktographischen Symbolik und die Analyse von Bruchmerkmalen mittels Machine Learning. Über die aktuellen Ergebnisse und Planungen wird berichtet.
Machine learning based multi-sensor fusion for the nondestructive testing of corrosion in concrete
(2020)
Half-cell potential mapping (HP) is the most popular non-destructive testing method (NDT) for locating corrosion damage in concrete. It is generally accepted that HP is susceptible to environmental factors caused by salt-related deterioration, such as different moisture and chloride gradients. Additional NDT methods are able to identify distinctive areas but are not yet used to estimate more accurate test results. We present a Supervised Machine Learning (SML) based approach to data fusion of seven different signal features to obtain higher quality information. SMLs are methods that explore (or learn) relationships between different (sensor) data from predefined data labels. To obtain a representative, labelled data set we conducted a comprehensive experiment simulating the deterioration cycle of a chloride exposed device in the laboratory. Our data set consists of 18 measurement campaigns, each containing HP, Ground Penetrating- Radar, Microwave Moisture and Wenner resistivity data. We compare the performance of different ML approaches. Many outperform the best single method, HP. We describe the intrinsic challenges posed by a data-driven approach in NDT and show how future work can help overcome them.
To date, the destructive extraction and analysis of drilling cores is the main possibility to obtain depth information about damaging water ingress in building floors. The time- and costintensive procedure constitutes an additional burden for building insurances that already list piped water damage as their largest item. With its high sensitivity for water, a ground-penetrating radar (GPR) could provide important support to approach this problem in a non-destructive way. In this research, we study the influence of moisture damage on GPR signals at different floor constructions. For this purpose, a modular specimen with interchangeable layers is developed to vary the screed and insulation material, as well as the respective layer thickness. The obtained data set is then used to investigate suitable signal features to classify three scenarios: dry, damaged insulation, and damaged screed. It was found that analyzing statistical distributions of A-scan features inside one B-scan allows for accurate classification on unknown floor constructions. Combining the features with multivariate data analysis and machine learning was the key to achieve satisfying results. The developed method provides a basis for upcoming validations on real damage cases.
An in-depth insight into the chemistry and nature of the individual chemical bonds is essential for understanding materials. Bonding analysis is thus expected to provide important features for large-scale data analysis and machine learning of material properties. Such chemical bonding information can be computed using the LOBSTER software package, which post-processes modern density functional theory data by projecting the plane wave-based wave functions onto an atomic orbital basis. With the help of a fully automatic workflow, the VASP and LOBSTER software packages are used to generate the data. We then perform bonding analyses on 1520 compounds (insulators and semiconductors) and provide the results as a database. The projected densities of states and bonding indicators are benchmarked on standard density-functional theory computations and available heuristics, respectively. Lastly, we illustrate the predictive power of bonding descriptors by constructing a machine learning model for phononic properties, which shows an increase in prediction accuracies by 27% (mean absolute errors) compared to a benchmark model differing only by not relying on any quantum-chemical bonding features.
Fractography plays a critical role in failure analysis of engineering components and has a considerable importance for safety investigations. Usually, the interpretation of fracture surfaces is done by experts with the help of literature and experimental data, that requires a lot of experience. The use of deep learning (DL) with neural networks in failure analysis becomes more and more relevant with the rapidly developing possibilities. Especially, the modern network architectures can assist fractographers in determining various fracture features on SEM images of the fracture surfaces. The basis for the best possible evaluation is the understanding of the influence of the input data used for training deep neural networks (DNN). Therefore, this study discusses the influence of the selection of the input data used for the prediction quality of these networks in order to take this into account for future data acquisition. Specimens of various metallic materials were subjected to fatigue cracking experiment under laboratory conditions. The fractured surfaces were then imaged using various modes or detectors (such as SE, BSE and topography) in SEM, and those captured images were used to create a training data set. The relevance of the individual data for the quality of the prediction is determined by a specific combination of the different detector data. For the training, the well-established architecture of a UNet-ResNet34 with a fixed set of hyperparameters is used. It has been found in this present study that the combination of all input data significantly increases the prediction accuracy, whereby even the combination of SE and BSE data provides considerable advantages over the exclusive use of SE images.