Filtern
Dokumenttyp
- Vortrag (34) (entfernen)
Referierte Publikation
- nein (34) (entfernen)
Schlagworte
- Machine learning (34) (entfernen)
Organisationseinheit der BAM
- 6 Materialchemie (16)
- 6.0 Abteilungsleitung und andere (8)
- 8 Zerstörungsfreie Prüfung (8)
- 6.3 Strukturanalytik (6)
- 8.6 Faseroptische Sensorik (5)
- 1 Analytische Chemie; Referenzmaterialien (4)
- 1.6 Anorganische Referenzmaterialien (4)
- 8.4 Akustische und elektromagnetische Verfahren (2)
- VP Vizepräsident (2)
- VP.1 eScience (2)
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.
Artificial intelligence (AI), machine learning, and neural networks have revolutionized fields such as self-driving cars or machine translation. Indeed, AI has progressed so far that scientists such as Stephen Hawking now fear that it may destroy humankind altogether. So, let’s get started and use AI in surface science.
Here we train neural networks to use raw measurement data as input and immediately return the desired output parameters. We discuss the example of X-ray reflectivity measurements of ultrathin films, which contain reciprocal space information and must traditionally be fitted with dynamic scattering theory. Instead, we train the neural network with simulated X-ray data of multilayer structures and then apply it to measurement data. The neural network yields high accuracy results, is robust against noise, and performs significantly faster than fitting algorithms in real-time experiments. The presented neural network data analysis is becoming increasingly attractive, because free software has become very accessible and specialized computer chips (NPU) are currently being rolled out.
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.
We propose a method based on artificial neural networks to extract strain information from wavelength-scanning coherent optical time domain reflectometry (C-OTDR) data. Our neural network algorithm performs more than two orders of magnitude faster than the conventional approach.
This is due to the highly parallel evaluation of the neural networks on a GPU accelerated computer and the fact that conventional correlation and interpolation analysis needs many Iteration steps. This opens the way for real-time C-OTDR strain sensing because the neural Network strain predictions require less time than the measurements themselves. Real-time data Analysis enables long-term sensing e.g. in structural health monitoring, because the large amount of raw data does not have to be stored but can immediately be reduced to the strain data of interest.
In X-ray fluorescence (XRF), a sample is excited with X-rays, and the resulting characteristic radiation is detected to detect elements quantitatively and qualitatively. Quantification is traditionally done in several steps:
1. Normalization of the data
2. Determination of the existing elements
3. Fit of the measured spectrum
4. Calculation of concentrations with fundamental parameters / MC simulations / standard based
The problem with standard based procedures is the availability of corresponding standards. The problem with the calculations is that the measured intensities for XRF measurements are matrix-dependent. Calculations must, therefore, be performed iteratively (= time consuming) in order to determine the chemical composition.
First experiments with gold samples have shown the feasibility of machine learning based quantification in principle. A large number of compositions were simulated (> 10000) and analyzed with a deep learning network. For first experiments, an ANN (Artificial Neural Network) with 3 hidden layers and 33x33x33 neurons was used. This network learned the mapping of spectra to concentrations using supervised learning by multidimensional regression. The input layer was formed by the normalized spectrum, and the output layer directly yielded the searched values. The applicability for real samples was shown by measurements on certified reference materials.
A brief introduction to the efforts we have done in our lab towards AI/ML analysis of SAXS data. For this, we need to extend the data with an extensive, structured hierarchy of metadata and associated data. A practical look into the information stored in our files, and the organization of the files in a data catalog is presented.
Various applications for artificial intelligence in the context of spectroscopy will be presented. in particular, examples from bamline will be featured. After a short introduction to synchrotron radiation, artificial intelligence algorithms for the quantification of X-ray fluorescence measurement are discussed. In the second example, information retrieval by natural language processing is reviewed. As a last example the reconstruction of measurements with the X-ray color camera and coded apertures is presented.
In this talk, the features of X-ray fluorescence analysis with synchrotron radiation will be presented. First, the basics of the origin of synchrotron radiation and X-ray fluorescence analysis and the experimental setup will be discussed. Then, examples of trace element detection, micrometer resolution, and application of the X-ray color camera will be shown.