Filtern
Dokumenttyp
- Vortrag (19) (entfernen)
Sprache
- Englisch (12)
- Deutsch (6)
- Mehrsprachig (1)
Referierte Publikation
- nein (19)
Schlagworte
- Concrete (5)
- Data fusion (5)
- Machine Learning (5)
- NDT (5)
- Machine learning (4)
- Materials Informatics (3)
- Small Data (3)
- Beton (2)
- Corrosion (2)
- Datenanalyse (2)
Organisationseinheit der BAM
Eingeladener Vortrag
- nein (19)
Das maschinelle Lernen (ML) wurde erfolgreich zur Lösung vieler Aufgaben in der zerstörungsfreien Prüfung im Bauwesen (ZfPBau) eingesetzt. Allerdings ist die Erstellung von Referenzdaten in den meisten Fällen extrem teuer und daher viel knapper als in anderen Forschungsbereichen. Auch decken die verfügbaren Daten mitunter nur ein einziges Szenario ab, so dass die Leistungsindikatoren oft nicht die tatsächliche Leistung des ML-Modells in der praktischen Anwendung widerspiegeln. Schätzungen, die die Übertragbarkeit von einem Szenario auf ein anderes quantifizieren, sind erforderlich, um dieser Herausforderung gerecht zu werden und den Weg für Anwendungen in der Praxis zu ebnen.
In diesem Beitrag stellen wir Werkzeuge zur Beschreibung der Unsicherheit von ML in neuen ZfPBau-Szenarien vor. Zu diesem Zweck haben wir einen bestehenden Trainingsdatensatz zur Klassifizierung von Korrosionsschäden der Bewehrung in Beton um eine neue Fallstudie erweitert. Die Messungen wurden an großflächigen Betonproben mit eingebauter chloridinduzierter Korrosion des Bewehrungsstahls durchgeführt. Das Experiment simulierte den gesamten Lebenszyklus von chloridinduzierten Sichtbetonbauteilen im Labor. Unser Datensatz umfasst Potenzialfeld- und Radarmessungen. Die einzigartige Fähigkeit, die Schädigung zu überwachen und eine gezielte Korrosion einzuleiten, ermöglichte es, die Daten zu labeln - was für die Konstruktion von ML-Modellen entscheidend ist. Um die Übertragbarkeit zu untersuchen, erweitern wir unser Modell um Metadaten - wie etwa Konstruktionsmerkmale des Prüfkörpers und Umweltbedingungen. Dies erlaubt es, die Veränderung dieser Merkmale in neuen Szenarien mit statistischen Methoden als Unsicherheiten auszudrücken. Wir vergleichen verschiedene auf Stichproben und statistischer Verteilung basierende Ansätze und zeigen, wie diese Methoden eingesetzt werden können, um Wissenslücken von ML-Modellen in der ZfP zu schließen
ML has been successfully applied to solve many NDT-CE tasks. This is usually demonstrated with performance metrics that evaluate the model as a whole based on a given set of data. However, since in most cases the creation of reference data is extremely expensive, the data used is generally much sparser than in other areas, such as e-commerce. As a result, performance indicators often do not reflect the practical applicability of the ML model. Estimates that quantify transferability from one case to another are necessary to meet this challenge and pave the way for real world applications.
In this contribution we invetigate the uncertainty of ML in new NDT-CE scenarios. For this purpose, we have extended an existing training data set for the classification of corrosion damage by a new case study. Our data set includes half-cell potential mapping and ground-penetrating radar measurements. The measurements were performed on large-area concrete samples with built-in chloride-induced corrosion of reinforcement. The experiment simulated the entire life cycle of chloride induced exposed concrete components in the laboratory. The unique ability to monitor deterioration and initiate targeted corrosion initiation allowed the data to be labelled - which is crucial to ML. To investigate transferability, we extend our data by including new design features of the test specimen and environmental conditions. This allows to express the change of these features in new scenarios as uncertainties using statistical methods. We compare different sampling and statistical distribution-based approaches and show how these methods can be used to close knowledge gaps of ML models in NDT.
With 8% of man-made CO2 emissions, cement production is an important driver of the climate crisis. By using alkali-activated binders, part of the energy-intensive clinker production process can be dispensed. However, as numerous raw materials are involved in the manufacturing process here, the complexity of the materials increases by orders of magnitude. Finding a properly balanced binder formulation is like looking for a needle in a haystack. We have shown for the first time that artificial intelligence (AI)-based optimization of alkali-activated binder formulations can significantly accelerate research.
The "Sequential Learning App for Materials Discovery" (SLAMD) aims to accelerate practice transfer. With SLAMD, materials scientists have low-threshold access to AI through interactive and intuitive user interfaces. The value added by AI can be determined directly. For example, the CO2 emissions saved per ton of cement can be determined for each development cycle: the more efficient the AI optimization, the greater the savings.
Our material database already includes more than 120,000 data points of alternative binders and is constantly being expanded with new parameters. We are currently driving the enrichment of the data with a life cycle analysis of the building materials.
Based on a case study we show how intuitive access to AI can drive the adoption of techniques that make a real contribution to the development of resource-efficient and sustainable building materials of the future and make it easy to identify when classical experiments are more efficient.
High greenhouse gas emissions from the production of building materials are a major contributor to the current climate crisis. However, developing alternative building materials is complex. Traditional laboratory methods are reaching their limits. Artificial intelligence, on the other hand, can give research a new dynamic.
Novel materials are usually developed manually in the laboratory rather than on a computer. This makes the processes time-consuming, difficult and expensive. With the app SLAMD (Sequential Learning App for Materials Discovery), materials researchers can explore the potential of artificial intelligence to speed up materials research and easily apply AI in the lab. The app was developed by our team at the Federal Institute for Materials Research and Testing (BAM) led by Prof. Sabine Kruschwitz together with a team in the Department of Building Materials and Construction Chemistry at TU Berlin led by Prof. Dietmar Stephan.
It uses material composition and characterization data to predict ideal material candidates. It can be used to optimize many material properties simultaneously and even incorporates database information such as carbon footprint, material cost or resource availability. Unlike the usual data-intensive AI methods, SLAMD optimally integrates existing knowledge and human feedback, and provides numerous decision support tools to precisely navigate complex scientific knowledge processes towards success.
In this talk, we will present some case studies where we were able to find suitable advanced materials in a few months instead of several years. We will talk about the challenges we overcame and the future potential we see for this approach to developing the green materials of the future.
We present the results of a machine learning (ML)- inspired data fusion approach, applied to multi-sensory nondestructive testing (NDT) data. Our dataset consists of Impact-Echo (IE), Ultrasonic Pulse Echo (US) and Ground Penetrating Radar (GPR) measurements collected on large-scale concrete specimens with built–in simulated honeycombing defects. In a previous study we were able to improve the detectability of honeycombs by fusing the information from the three different sensors with the density based clustering algorithm DBSCAN. We demonstrated the advantage of data fusion in reducing the false positives up to 10% compared to the best single sensor, thus, improving the detectability of the defects. The main objective of this contribution is to investigate the generality, i.e. whether the conclusions from one specimen can be adapted to the other. The effectiveness of the proposed approach on a separate full-scale concrete specimen was evaluated.
This talk demonstrates the results of the IGSTC-project entitled "NDT-Data Fusion".
Project approach:
Nondestructive testing (NDT) of concrete buildings allows to coordinate efficient repair measures. Multi-sensor platforms collect large data sets. Nevertheless, data analysis is typically performed manually. Data Fusion uses the full potential of a multi-sensory data set in order to:
improve information quality (reliability, robustness, accuracy, clarity, completeness) and enables automated algorithm based data analysis.
We present the project achievements, namely:
- Development of building scanner system for multisensory NDT
- Laboratory multi sensor investigations
- Development of data fusion concept for honeycombing and pitting corrosion
- Field testing
Half-cell potential mapping (HP) is the most popular nondestructive test (NDT)-method for the localization of corrosion damage in concrete. It is generally recognized, that HP is prone to the environmental factors that arise from salt induced deterioration, such as varying moisture and chloride gradients. Additional NDT-methods are capable to determine distinctive areas, but cannot yet be used to estimate more accurate testing results. We introduce a supervised machine learning (SML) based approach for data fusion to make use of the additional sensor information. SML are methods that explore relations between different (sensor) data from predefined data labels. We use a simple linear classifier named logistic regression to distinguish defect and intact areas. The test performance improves drastically compared to the best single method, HP. In order to generate representative, labeled data we conducted a comprehensive experiment that simulates the deterioration-cycle of a chloride-exposed building part in the lab. Our data set consist of 18 measurement campaigns, each containing HP-, ground-penetrating-radar-, microwave-moisture-, and Wenner-resistivity-data. We detail the challenges that arise with a data driven approach in NDT and how we addressed them.
Angetrieben durch Digitalisierung und die sogenannte Industrie 4.0 steigt die Erwartungshaltung gegenüber der Nutzung von Sensorik. Die Unsicherheit, Widersprüchlichkeit und Redundanz separat verarbeiteter einzelner Quellen soll durch die synergetische Zusammenführung heterogener Datensätze überwunden werden. Dabei steht der zunehmenden Aufgabenkomplexität eine ebenso zunehmende Verfügbarkeit an Sensoren, Daten und Rechenleistung gegenüber.
Der Begriff Datenfusion fasst Ansätze zusammen, die Daten zu abstrakteren, aber besser verständlichen Informationen verarbeiten. Die Fusion ist Kernbestandteil effektiver Assistenzsysteme und beweist in vielzähligen Aufgaben - von militärischen Anwendungen über Flug- und Fahrassistenzsysteme, bis in den Heimbereich – ihr großes Potenzial. Durch die wachsende Automatisierung bei der Messdatenerfassung wird Datenfusion auch in der industriellen Qualitätsprüfung und –sicherung zunehmend attraktiver.
Der Vortrag gibt einen Überblick über den breiten Themenkomplex und widmet sich dabei im Theorieteil speziell der Fragen, welche Informationen in multivariaten Datensätzen stecken und wie sie extrahiert werden können. Anschließend wird ein Beispiel für die erfolgreiche Anwendung zur zerstörungsfreien Prüfung von Betonbauteilen vorgestellt. Der dargestellte Datensatz ist klein, heterogen, hochdimensional und unausgeglichen. Anhand von Algorithmen mit unterschiedlicher Leistungsfähigkeiten hinsichtlich Anpassungsfähigkeit und Invarianz gegenüber Höherdimensionalität wird erläutert welche Prozesse zur Verbesserung der Informationsqualität nötig sind.