arXiv subjects

Peter Kenesei

Publications and source records attributed to Peter Kenesei.

14 recordsLinked to original sources

arXiv2026

Unsupervised Semantic Segmentation in Synchrotron Computed Tomography with Self-Correcting Pseudo Labels

X-ray computed tomography (CT) is a widely used imaging technique that provides detailed examinations into the internal structure of an object with synchrotron CT (SR-CT) enabling improved data quality by using higher energy, monochromatic X-rays. While SR-CT allows for improved resolution, time-resolved experimentation, and reduced imaging artifacts, it also produces significantly larger datasets than conventional CT. Accurate and efficient evaluation of these datasets is a critical component of these workflows; yet is often done manually representing a major bottleneck in the analysis phase. While deep learning has emerged as a powerful tool capable of providing a wide range of purely data-driven solutions, it requires a substantial amount of labeled data for training and manual annotation of SR-CT datasets is impractical in practice. In this paper, we introduce a novel framework that enables automatic segmentation of large, high-resolution SR-CT datasets by eliminating the need to hand label images for deep learning training. First, we generate pseudo labels by clustering on the voxel values identifying regions in the volume with similar attenuation coefficients producing an initial semantic map. Afterwards, we train a segmentation model on the pseudo labels before utilizing the Unbiased Teacher approach to self-correct them ensuring accurate final segmentations. We find our approach improves pixel-wise accuracy and mIoU by 13.31% and 15.94%, respectively, over the baseline pseudo labels when using a magnesium crystal SR-CT sample. Additionally, we extensively evaluate the different components of our workflow including segmentation model, loss function, pseudo labeling strategy, and input type. Finally, we evaluate our approach on to two additional samples highlighting our frameworks ability to produce segmentations that are considerably better than the original pseudo labels.

cs.CV↗

arXiv2023

Rapid detection of rare events from in situ X-ray diffraction data using machine learning

High-energy X-ray diffraction methods can non-destructively map the 3D microstructure and associated attributes of metallic polycrystalline engineering materials in their bulk form. These methods are often combined with external stimuli such as thermo-mechanical loading to take snapshots over time of the evolving microstructure and attributes. However, the extreme data volumes and the high costs of traditional data acquisition and reduction approaches pose a barrier to quickly extracting actionable insights and improving the temporal resolution of these snapshots. Here we present a fully automated technique capable of rapidly detecting the onset of plasticity in high-energy X-ray microscopy data. Our technique is computationally faster by at least 50 times than the traditional approaches and works for data sets that are up to 9 times sparser than a full data set. This new technique leverages self-supervised image representation learning and clustering to transform massive data into compact, semantic-rich representations of visually salient characteristics (e.g., peak shapes). These characteristics can be a rapid indicator of anomalous events such as changes in diffraction peak shapes. We anticipate that this technique will provide just-in-time actionable information to drive smarter experiments that effectively deploy multi-modal X-ray diffraction methods that span many decades of length scales.

cs.LG↗

arXiv2023

Full-Field Nanoscale X-ray Diffraction-Contrast Imaging using Direct Detection

Recent developments in x-ray science provide methods to probe deeply embedded mesoscale grain structures and spatially resolve them using dark field x-ray microscopy (DFXM). Extending this technique to investigate weak diffraction signals such as magnetic systems, quantum materials and thin films proves challenging due to available detection methods and incident x-ray flux at the sample. We present a direct detection method focusing on DFXM studies in the hard x-ray range of 10s of keV and above capable of approaching nanoscale resolution. Additionally, we compare this direct detection scheme with routinely used scintillator based optical detection and achieve an order of magnitude improvement in exposure times allowing for imaging of weakly diffracting ordered systems.

physics.ins-det↗

arXiv2022

fairDMS: Rapid Model Training by Data and Model Reuse

Extracting actionable information rapidly from data produced by instruments such as the Linac Coherent Light Source (LCLS-II) and Advanced Photon Source Upgrade (APS-U) is becoming ever more challenging due to high (up to TB/s) data rates. Conventional physics-based information retrieval methods are hard-pressed to detect interesting events fast enough to enable timely focusing on a rare event or correction of an error. Machine learning~(ML) methods that learn cheap surrogate classifiers present a promising alternative, but can fail catastrophically when changes in instrument or sample result in degradation in ML performance. To overcome such difficulties, we present a new data storage and ML model training architecture designed to organize large volumes of data and models so that when model degradation is detected, prior models and/or data can be queried rapidly and a more suitable model retrieved and fine-tuned for new conditions. We show that our approach can achieve up to 100x data labelling speedup compared to the current state-of-the-art, 200x improvement in training speed, and 92x speedup in-terms of end-to-end model updating time.

cs.LG↗

arXiv2022

Temperature mapping of stacked silicon dies from x-ray diffraction intensities

Increasing power densities in integrated circuits has led to an increased prevalence of thermal hotspots in integrated circuits. Tracking these thermal hotspots is imperative to prevent circuit failures. In 3D integrated circuits, conventional surface techniques like infrared thermometry are unable to measure 3D temperature distribution and optical and magnetic resonance techniques are difficult to apply due to the presence of metals and large current densities. X-rays offer high penetration depth and can be used to probe 3D structures. We report a method utilizing the temperature dependence of x-rays diffraction intensity via the Debye-Waller factor to simultaneously map the temperature of an individual silicon die that is a part of a stack of dies. Utilizing beamline 1-ID-E at the Advanced Photon Source (Argonne), we demonstrate for each individual silicon die, a temperature resolution of 3 K, a spatial resolution of 100 um x 400 um and a temporal resolution of 20 s. Utilizing a sufficiently high intensity laboratory source, e.g., from a liquid anode source, this method can be scaled down to laboratories for non-invasive temperature mapping of 3D integrated circuits.

physics.app-ph↗

arXiv2022

Bridging Data Center AI Systems with Edge Computing for Actionable Information Retrieval

Extremely high data rates at modern synchrotron and X-ray free-electron laser light source beamlines motivate the use of machine learning methods for data reduction, feature detection, and other purposes. Regardless of the application, the basic concept is the same: data collected in early stages of an experiment, data from past similar experiments, and/or data simulated for the upcoming experiment are used to train machine learning models that, in effect, learn specific characteristics of those data; these models are then used to process subsequent data more efficiently than would general-purpose models that lack knowledge of the specific dataset or data class. Thus, a key challenge is to be able to train models with sufficient rapidity that they can be deployed and used within useful timescales. We describe here how specialized data center AI (DCAI) systems can be used for this purpose through a geographically distributed workflow. Experiments show that although there are data movement cost and service overhead to use remote DCAI systems for DNN training, the turnaround time is still less than 1/30 of using a locally deploy-able GPU.

cs.LG↗

arXiv2021

BraggNN: Fast X-ray Bragg Peak Analysis Using Deep Learning

X-ray diffraction based microscopy techniques such as High Energy Diffraction Microscopy rely on knowledge of the position of diffraction peaks with high precision. These positions are typically computed by fitting the observed intensities in area detector data to a theoretical peak shape such as pseudo-Voigt. As experiments become more complex and detector technologies evolve, the computational cost of such peak detection and shape fitting becomes the biggest hurdle to the rapid analysis required for real-time feedback during in-situ experiments. To this end, we propose BraggNN, a deep learning-based method that can determine peak positions much more rapidly than conventional pseudo-Voigt peak fitting. When applied to a test dataset, BraggNN gives errors of less than 0.29 and 0.57 pixels, relative to the conventional method, for 75% and 95% of the peaks, respectively. When applied to a real experimental dataset, a 3D reconstruction that used peak positions computed by BraggNN yields 15% better results on average as compared to a reconstruction obtained using peak positions determined using conventional 2D pseudo-Voigt fitting. Recent advances in deep learning method implementations and special-purpose model inference accelerators allow BraggNN to deliver enormous performance improvements relative to the conventional method, running, for example, more than 200 times faster than a conventional method on a consumer-class GPU card with out-of-the-box software.

eess.IV↗

arXiv2020

Non-destructive determination of phase, size, and strain of individual grains in polycrystalline photovoltaic materials

We demonstrate a non-destructive approach to provide structural properties on the grain level for the absorber layer of kesterite solar cells. Kesterite solar cells are notoriously difficult to characterize structurally due to the co-existence of several phases with very similar lattice parameters. Specifically, we present a comprehensive study of 597 grains in the absorber layer of a 1.64% efficient Cu2ZnSnS4 (CZTS) thin-film solar cell, from which 15 grains correspond to the secondary phase ZnS. By means of three dimensional X-ray diffraction (3DXRD), we obtained statistics for the phase, size, orientation, and strain tensors of the grains, as well as their twin relations. We observe an average tensile stress in the plane of the film of ~ 70 MPa and a compressive stress along the normal to the film of ~ 145 MPa. At the grain level, we derive a 3D stress tensor that deviates from the biaxial model usually assumed for thin films. 41% of the grains are twins. We calculate the frequency of the six types of $Σ$3 boundaries, revealing that 180° rotations along axis <221> is the most frequent. This technique can be applied to polycrystalline thin film solar cells in general, where strain can influence the bandgap of the absorber layer material, and twin boundaries play a role in the charge transport mechanisms.

cond-mat.mtrl-sci↗

arXiv2019

High-energy coherent X-ray diffraction microscopy of polycrystal grains: first steps towards a multi-scale approach

We present proof-of-concept imaging measurements of a polycrystalline material that integrate the elements of conventional high-energy X-ray diffraction microscopy with coherent diffraction imaging techniques, and that can enable in-situ strain-sensitive imaging of lattice structure in ensembles of deeply embedded crystals over five decades of length scale upon full realization. Such multi-scale imaging capabilities are critical to addressing important questions in a variety of research areas such as materials science and engineering, chemistry, and solid state physics. Towards this eventual goal, the following key aspects are demonstrated: 1) high-energy Bragg coherent diffraction imaging (HE-BCDI) of sub-micron-scale crystallites at 52 keV at current third-generation synchrotron light sources, 2) HE-BCDI performed in conjunction with far-field high-energy diffraction microscopy (ff-HEDM) on the grains of a polycrystalline sample in an smoothly integrated manner, and 3) the orientation information of an ensemble of grains obtained via ff-HEDM used to perform complementary HE-BCDI on multiple Bragg reflections of a single targeted grain. These steps lay the foundation for integration of HE-BCDI, which typically provides a spatial resolution tens of nanometers, into a broad suite of well-established HEDM methods, extending HEDM beyond the few-micrometer resolution bound and into the nanoscale, and positioning the approach to take full advantage of the orders-of-magnitude improvement of X-ray coherence expected at fourth generation light sources presently being built and commissioned worldwide.

cond-mat.mtrl-sci↗

arXiv2018

Phase retrieval for Bragg coherent diffraction imaging at high X-ray energies

Coherent X-ray beams with energies $\geq 50$ keV can potentially enable three-dimensional imaging of atomic lattice distortion fields within individual crystallites in bulk polycrystalline materials through Bragg coherent diffraction imaging (BCDI). However, the undersampling of the diffraction signal due to Fourier space compression at high X-ray energies renders conventional phase retrieval algorithms unsuitable for three-dimensional reconstruction. To address this problem we utilize a phase retrieval method with a Fourier constraint specifically tailored for undersampled diffraction data measured with coarse-pitched detector pixels that bin the underlying signal. With our approach, we show that it is possible to reconstruct three-dimensional strained crystallites from an undersampled Bragg diffraction data set subject to pixel-area integration without having to physically upsample the diffraction signal. Using simulations and experimental results, we demonstrate that explicit modeling of Fourier space compression during phase retrieval provides a viable means by which to invert high-energy BCDI data, which is otherwise intractable.

cond-mat.mtrl-sci↗

arXiv2017

Sparse recovery of undersampled intensity patterns for coherent diffraction imaging at high X-ray energies

Coherent X-ray photons with energies higher than 50 keV offer new possibilities for imaging nanoscale lattice distortions in bulk crystalline materials using Bragg peak phase retrieval methods. However, the compression of reciprocal space at high energies typically results in poorly resolved fringes on an area detector, rendering the diffraction data unsuitable for the three-dimensional reconstruction of compact crystals. To address this problem, we propose a method by which to recover fine fringe detail in the scattered intensity. This recovery is achieved in two steps: multiple undersampled measurements are made by in-plane sub-pixel motion of the area detector, then this data set is passed to a sparsity-based numerical solver that recovers fringe detail suitable for standard Bragg coherent diffraction imaging (BCDI) reconstruction methods of compact single crystals. The key insight of this paper is that sparsity in a BCDI data set can be enforced by recognising that the signal in the detector, though poorly resolved, is band-limited. This requires fewer in-plane detector translations for complete signal recovery, while adhering to information theory limits. We use simulated BCDI data sets to demonstrate the approach, outline our sparse recovery strategy, and comment on future opportunities.

eess.SP↗

arXiv2017

Microstructure analysis of bismuth absorbers for transition-edge sensor X-ray microcalorimeters

Transition-edge sensors (TESs) as microcalorimeters offer high resolving power, owning to their sharp response and low operating temperature. In the hard X-ray regime and above, the demand for high quantum-efficiency requires the use of absorbers. Bismuth (Bi), owing to its low heat carrier density and high X-ray stopping power, has been widely used as an absorber material for TESs. However, distinct spectral responses have been observed for Bi absorbers deposited via evaporation versus electroplating. Evaporated Bi absorbers are widely observed to have a non-Gaussian tail on the low energy side of measured spectra. In this study, we fabricated Bi absorbers via these two methods, and performed microstructure analysis using scanning electron microscopy (SEM) and X-ray diffraction microscopy. The two types of material showed the same crystallographic structure, but the grain size of the evaporated Bi was about 40 times smaller than that of the electroplated Bi. This distinction in grain size is likely to be the cause of their different spectral responses.

physics.ins-det↗

arXiv2017

Eliminating the non-Gaussian spectral response of X-ray absorbers for transition-edge sensors

Transition-edge sensors (TES) as microcalorimeters for high-energy-resolution X-ray spectroscopy are often fabricated with an absorber made of materials with high Z (for X-ray stopping power) and low heat capacity (for high resolving power). Bismuth represents one of the most compelling options. TESs with evaporated bismuth absorbers have shown spectra with undesirable and unexplained low-energy tails. We have developed TESs with electroplated bismuth absorbers over a gold layer that are not afflicted by this problem and that retain the other positive aspects of this material. To better understand these phenomena, we have studied a series of TESs with gold, gold/evaporated bismuth, and gold/electroplated bismuth absorbers, fabricated on the same die with identical thermal coupling. We show that bismuth morphology is linked to the spectral response of X-ray TES microcalorimeters.

physics.ins-det↗

arXiv2017

Influences of Granular Constraints and Surface Effects on the Heterogeneity of Elastic, Superelastic, and Plastic Responses of Polycrystalline Shape Memory Alloys

Deformation heterogeneities within microstructures of polycrystalline shape memory alloys (SMAs) during superelastic stressing are studied using both experiments and simulations. In situ X-ray diffraction, specifically the far-field high energy diffraction microscopy (ff-HEDM) technique was used to non-destructively measure the grain-averaged statistics of position, crystal orientation, elastic strain tensor and volume for hundreds of austenite grains in a superelastically loaded nickel-titanium (NiTi) SMA. This experimental data were also used to create a synthetic microstructure within a finite element model. The development of intragranular stresses were then simulated during tensile loading of the model using anisotropic elasticity. Driving forces for phase transformation and slip were calculated from these stresses. The grain-average responses of individual austenite crystals examined before and after multiple stress-induced transformation events showed that grains at the specimen interior carry more stress and plastically deform less than the surface grains as the superelastic response shakes down. Examination of the heterogeneity within individual grains showed that regions near grain boundaries exhibit larger stress variation compared to the grain interiors. This intragranular heterogeneity is more strongly driven by the constraints of neighboring grains than the initial stress state and orientation of the individual grains.

cond-mat.mtrl-sci↗