5 Werkstofftechnik
Filtern
Dokumenttyp
- Vortrag (4)
- Zeitschriftenartikel (1)
Sprache
- Englisch (5) (entfernen)
Schlagworte
- Fractography (5) (entfernen)
Organisationseinheit der BAM
- 5.1 Mikrostruktur Design und Degradation (5) (entfernen)
Eingeladener Vortrag
- nein (4)
The aim of a fractographic investigation is the evaluation of macroscopic and microscopic fracture surface characteristics and, as a result, the determination of the fracture mechanism of a component from a failure case. The basis for such evaluations of fracture characteristics comes from actual comparative mechanical testing and from the literature. A fractographic analysis can be very complex and, in any case, requires considerable experience. In the IGF project "iFrakto", software is being developed that quantitatively determines fracture characteristics and fracture mechanisms utilizing digitized expert knowledge, machine learning, and standard 2D and topographical data from SEM imaging. Topographical data are obtained from 4QBSE detector using shape-from-shading technology. In the medium term, a software tool should provide knowledge-based suggestions for the evaluation of fracture surfaces in real time during SEM work or at subsequent evaluation. As a basis for this, round robins were carried out among fractographers in order to create a knowledge base, to query the practice-relevant requirements for such tools and to carry out first practical tests. Actual results are presented and the relevance of the evaluation strategy is evaluated.
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.
The aim of a fractographic investigation is the evaluation of macroscopic and microscopic fracture surface characteristics and, as a result, the determination of the fracture mechanism of a component from a failure case. The basis for such evaluations of fracture characteristics comes from actual comparative mechanical testing and from the literature. A fractographic analysis can be very complex and, in any case, requires considerable experience.
Machine learning methods enables the quantitative determination of fracture characteristics and fracture mechanisms utilizing digitized expert knowledge [1]. Although the application of SE images provides promising results, additional information is required to obtain reliable solutions. As expected, BSE and 3D information helps to improve the classification (Fig. 1). But only a fast, widely integrated, and automated topography measurement can provide the required amount of referenced surface data for the application of machine learning methods.
To fulfil these requirements, topographical data are obtained from a BSE detector with four symmetric segments (4Q-BSE) using shape-from-shading technology [2]. Surface height calculation is performed live during image acquisition and provides immediate feedback in three dimensions. All available signals (SE, BSE and more if applicable) are recorded simultaneously together with the surface topography and stored in a multichannel data file. This guaranties the same geometrical reference for all data, which is required for further analysis (Fig. 2).
When applying machine learning methods to topographic data together with SEM images, topographic information must be provided as depth image. Consequently, a unique height scale is required for all applied data with different magnifications. This requires a calibrated height measurement, which is ensured with the integrated 3D calibration of the topographic acquisition and a dedicated calibration sample. Thus, a large number of data sets from different fracture samples was generated and used as training data for machine learning.
The following work deals with the quantitative fracture surface evaluation in damage analysis. So far, fracture surfaces have almost exclusively been evaluated qualitatively, i.e. the presence of fracture features is documented and their surface proportions are estimated, if necessary. Many years of experience are required, as well as an intensive comparison with defined comparative images from the literature. The aim of this work is the development of classifiers which can recognize fracture mechanisms or fracture features in scanning electron microscope images (SEM). The basis is 46 SEM images, which have been evaluated by fractography experts with regard to fracture features. The existing data set of images is expanded using augmentation methods in order to increase the variability of the data and counteract overfitting. Only convolutional neural networks (CNN) are used to create the classifiers. Various network configurations are tested, with the SegNet achieving the best results.