Refine
Document Type
Language
- English (14)
Keywords
- Artificial intelligence (1)
- Data augmentation (1)
- Deep learning (1)
- Flaw detection (1)
- Machine learning (1)
- Microfluidic devices (1)
- NDE (1)
- NDT (1)
- Non-destructive evaluation (1)
- Non-destructive testing (1)
- Synthetic training data (1)
- Thin fluid films (1)
- Ultrasonic testing (1)
- Ultrasonics (1)
- micropump; electrostatic actuation; nano e-drive; in-plane reciprocating displacement micropump; MEMS (1)
Institute
BTU
We present new results on femtosecond LIPSS on silicon, fostering the dynamic model of self-organized structure formation. The first set of experiments demonstrates LIPSS formation by irradiation with a femtosecond white light continuum. The ripples are, as usual, perpendicular to the light polarization with a fluence-dependent wavelength between 500 and 700 nm. At higher dose (fluence × number of shots), the LIPSS turn to much coarser structures. The second set of experiments displays the dose dependence of pattern evolution at about threshold fluence. In contrast to the general case of multi-pulse LIPSS, where a strong dependence of the structures on the laser polarization is observed, single-shot exposition of silicon at about the ablation threshold results in a concentric pattern of very regular sub-wavelength ripples following the oval shape of the irradiated spot, without any reference to the laser polarization. When increasing the number of pulses, the usual, typical ripples develop and then coalesce into broader perpendicular structures, interlaced with remnants of the first, finer ripples.
Ultrasonic Testing (UT) has seen increasing application of machine learning (ML) in recent years, promoting higher-level automation and decision-making in flaw detection and classification. Building a generalized training dataset to apply ML in non-destructive evaluation (NDE), and thus UT, is exceptionally difficult since data on pristine and representative flawed specimens are needed. Yet, in most UT test cases flawed specimen data is inherently rare making data coverage the leading problem when applying ML. Common data augmentation (DA) strategies offer limited solutions as they don’t increase the dataset variance, which can lead to overfitting of the training data. The virtual defect method and the recent application of generative adversarial neural networks (GANs) in UT are sophisticated DA methods targeting to solve this problem. On the other hand, well-established research in modeling ultrasonic wave propagations allows for the generation of synthetic UT training data. In this context, we present a first thematic review to summarize the progress of the last decades on synthetic and augmented UT training data in NDE. Additionally, an overview of methods for synthetic UT data generation and augmentation is presented. Among numerical methods such as finite element, finite difference, and elastodynamic finite integration methods, semi-analytical methods such as general point source synthesis, superposition of Gaussian beams, and the pencil method as well as other UT modeling software are presented and discussed. Likewise, existing DA methods for one- and multidimensional UT data, feature space augmentation, and GANs for augmentation are presented and discussed. The paper closes with an in-detail discussion of the advantages and limitations of existing methods for both synthetic UT training data generation and DA of UT data to aid the decision-making of the reader for the application to specific test cases.