Filtern
Erscheinungsjahr
- 2023 (4)
Dokumenttyp
Sprache
- Englisch (4)
Schlagworte
- Automatic segmentation (2)
- Dice score (2)
- Multi-phase materials (2)
- Synthetic Training Data (2)
- 3D deep convolutional neural network (3D DCNN) (1)
- 3D imaging (1)
- ACEnet (1)
- Artificial Intelligence (1)
- Automatic Segmentation (1)
- BAM SynthCOND (1)
Organisationseinheit der BAM
- 8 Zerstörungsfreie Prüfung (4) (entfernen)
The greatest challenge when using deep convolutional neural networks (DCNNs) for automatic segmentation of microstructural X-ray computed tomography (XCT) data is the acquisition of sufficient and relevant data to train the working network. Traditionally, these have been attained by manually annotating a few slices for 2D DCNNs. However, complex multiphase microstructures would presumably be better segmented with 3D networks. However, manual segmentation labeling for 3D problems is prohibitive. In this work, we introduce a method for generating synthetic XCT data for a challenging six-phase Al–Si alloy composite reinforced with ceramic fibers and particles. Moreover, we propose certain data augmentations (brightness, contrast, noise, and blur), a special in-house designed deep convolutional neural network (Triple UNet), and a multi-view forwarding strategy to promote generalized learning from synthetic data and therefore achieve successful segmentations. We obtain an overall Dice score of 0.77. Lastly, we prove the detrimental effects of artifacts in the XCT data on achieving accurate segmentations when synthetic data are employed for training the DCNNs. The methods presented in this work are applicable to other materials and imaging techniques as well. Successful segmentation coupled with neural networks trained with synthetic data will accelerate scientific output.
Regardless of the experimental care practiced in acquiring X-ray computed tomography (XCT) data, artifacts might still exist, such as noise and blur. This is typical for fast XCT data acquisitions (e.g., in-situ investigations), or low-dose XCT. Such artifacts can complicate subsequent analysis of the data. Digital filters can moderately cure extensive artifacts. The selection of filter type, intensity, and order of application is not always straight forward. To tackle these problems, a complete sequential multilevel, multi-scale framework: BAM SynthCOND, employing newly designed deep convolutional neural networks (DCNNs), was formulated. Although data conditioning with neural networks is not uncommon, the main complication is that completely artifact-free XCT data for training do not exist. Thus, training data were acquired from an in-house developed library (BAM SynthMAT) capable of generating synthetic XCT material microstructures. Some novel DCNN architectures were introduced (2D/3D ACEnet_Denoise, 2D/3D ACEnet_Deblur) along with the concept of Assertive Contrast Enhancement (ACE) training, which boosts the performance of neural networks trained with continuous loss functions. The proposed methodology accomplished very good generalization from low resemblance synthetic training data. Indeed, denoising, sharpening (deblurring), and even ring artifact removal performance were achieved on experimental post-CT scans of challenging multiphase Al-Si Metal Matrix Composite (MMC) microstructures. The conditioning efficiencies were: 92% for combined denoising/sharpening, 99% for standalone denoising, and 95% for standalone sharpening. The results proved to be independent of the artifact intensity. We believe that the novel concepts and methodology developed in this work can be directly applied on the CT projections prior to reconstruction, or easily be extended to other imaging techniques such as: Microscopy, Neutron Tomography, Ultrasonics, etc.
We propose a novel iterative segmentation algorithm (i.S.Sy.Da.T.A: Iterative Segmentation Synthetic Data Training Algorithm) employing Deep Convolutional Neural Networks and synthetic training data for X-ray tomographic reconstructions of complex microstructures. In our method, we reinforce the synthetic training data with experimental XCT datasets that were automatically segmented in the previous iteration. This strategy produces better segmentations in successive iterations. We test our algorithm with experimental XCT re constructions of a 6-phase Al-Si Matrix Composite reinforced with ceramic fibers and particles. We perform the analysis in 3D with a special network architecture that demonstrates good generalization with synthetic training data. We show that our iterative algorithm returns better segmentations compared to the standard single training approach. More specifically, phases possessing similar attenuation coefficients can be better segmented: for Al2O3 fibers, SiC particles, and Intermetallics, we see an increase of the Dice score with respect to the classic approach: from 0.49 to 0.54, from 0.66 to 0.72, and from 0.55 to 0.66 respectively. Furthermore, the overall Dice score increases from 0.77 to 0.79. The methods presented in this work are also applicable to other materials and imaging techniques.
It is shown that preconditioning of experimental X‐ray computed tomography (XCT) data is critical to achieve high‐precision segmentation scores. The challenging experimental XCT datasets and deep convolutional neural networks (DCNNs) are used that are trained with low‐resemblance synthetic XCT data. The material used is a 6‐phase Al–Si metal matrix composite‐reinforced with ceramic fibers and particles. To achieve generalization, in our past studies, specific data augmentation techniques were proposed for the synthetic XCT training data. In addition, two toolsets are devised: (1) special 3D DCNN architecture (3D Triple_UNet), slicing the experimental XCT data from multiple views (MultiView Forwarding), the i.S.Sy.Da.T.A. iterative segmentation algorithm, and (2) nonlocal means (NLM) conditioning (filtering) for the experimental XCT data. This results in good segmentation Dice scores across all phases compared to more standard approaches (i.e., standard UNet architecture, single view slicing, standard single training, and NLM conditioning). Herein, the NLM filter is replaced with the deep conditioning framework BAM SynthCOND introduced in a previous publication, which can be trained with synthetic XCT data. This leads to a significant segmentation precision increase for all phases. The proposed methods are potentially applicable to other materials and imaging techniques.