Boundary Network Research Articles

A Malleable Boundary Network for temporal action detection

Temporal action detection in untrimmed videos is a challenging task aiming to predict the boundary and category of action instances. It can be useful in transportation. In this study, we propose a two-stage framework Malleable Boundary Network (MB-Net) to adaptively regress proposals based on finer scores. In particular, MB-Net consists of a Potential Boundary Generator in the first stage and an Adaptive Proposal Detector in the second stage. First, the Potential Boundary Generator fuses multiple sets of flexible score sequences to obtain tentative proposals through a frame-level feature in an anchor-free way. Then, the Adaptive Proposal Detector employs parallel modules to filter, classify and regress proposals adaptively. Besides, we propose an easy-to-realize feature augmented method Structured Temporal Segment Pooling, which makes full use of the information throughout the whole proposal. Experiments show that MB-Net achieves state-of-the-art performance on popular benchmarks THUMOS-14 and Activity-1.3 with an improvement of 1.9% and 1.2%.

Texture and Grain Boundary Network Evolution of Laser Powder Bed Fusion Processed Pure Ni During Post-printing Annealing

Commercially pure Ni was additively manufactured using laser powder bed fusion. The printed specimens were then subjected to annealing treatments in a range of 700 °C to 1200 °C to investigate the evolution of the microstructure using electron backscatter diffraction mapping and a five-parameter analysis of the grain boundary character distribution. The post-printing annealing treatment resulted in microstructural changes in respect to grain shape and size, the grain boundary character distribution and the overall texture. However, the extent of these changes strongly depended on the annealing temperature. The grain structure and overall texture changed only modestly when annealed below 900 °C. The grains grew by about 10 pct and this led to a small (≈ 15 pct) increase in the areas of boundaries with low energy (111) planes at the expense of higher energy grain boundary planes with the (001) orientation. Static recrystallization was the dominant microstructure evolution mechanism at annealing temperatures greater than 1000 °C, where new equiaxed grains replaced the initial printed microstructure, enhancing the relative areas of low energy grain boundaries (i.e., sum 3 and sum 9) by a factor of more than five. This change increased the population of boundaries with the low energy (111) plane by a factor of five and decreased the strength of the texture by more than 50 pct through twinning. The resultant microstructure is expected to ultimately improve the material properties, where the increase in the relative areas of sum 3 boundaries would enhance the corrosion resistance and fracture toughness of material, and the texture weakening diminishes the anisotropy in mechanical behaviour.

On the propensity of lath martensite intervariant boundaries to hydrogen absorption using molecular statics simulation

In the current study, the influence of boundary character (i.e., tilt, twist and mixed) on hydrogen absorption was, for the first time, investigated for lath martensite using molecular statics simulations. The energies of the martensitic intervariant boundaries varied widely from 306 to 1109 mJ/m2 depending on their character. The symmetric tilt and quasi-tilt boundaries revealed the lowest and highest boundary energies, respectively. This was, to some extent, reflected by distinct boundary unit structure/s and their complexity within a given intervariant boundary type. Interestingly, the intervariant boundary energy was not the main criteria to regulate the propensity of the boundary to hydrogen absorption. Rather, absorption depended on a distinct arrangement of unit structure/s within a given boundary plane, providing different interstitial sites for hydrogen. Indeed, the propensity for hydrogen atom/s was largely determined by the number of inequivalent interstitial sites, which resulted from the interaction of wide varieties of boundary unit structures. The intervariant boundary that consisted of regular arrangements of atoms with fewer boundary unit structure variety exhibited less tendency to H/H2 absorption (e.g., 60°/[111¯] symmetrical tilt boundary). On the contrary, the presence of a substantial number of diverse unit structures within the boundary plane significantly enhanced the tendency for hydrogen segregation (e.g., 51.73°/[116¯11] quasi-tilt and 10.53°/[01¯1¯] twist intervariant boundaries). The current findings provide a fundamental basis for the engineering of the intervariant boundary network within the martensitic microstructure to significantly enhance the hydrogen embrittlement resistance. This is critical for steels used in severe environments, such as hydrogen storage and pipelines for hydrogen transportation.

Microstructure evolution and failure mechanism of electromigration in Ag-alloy bonding wire

Ag-alloy bonding wire has excellent physical properties and is widely applied in microelectronics, but growing concerns about circuit failure induced by electromigration (EM) have raised doubts about this material. One point of frequent debate is the diffusion mechanism. This study provides a novel perspective by investigating the evolution of the microstructure via electron backscatter diffraction (EBSD) to settle the controversy over the diffusion mechanism of Ag-alloy bonding wire during EM. Particular focuses are the propensity for grain growth, the effects of the coincidence site lattice (CSL) on the grain boundary network, the texture transition, and the influence of thermal energy during EM. A stage-like transition in the dominant diffusion mechanism in different stages of EM is proposed and substantiated from the viewpoint of microstructure evolution.

Comparison of evolving interfaces, triple points, and quadruple points for discrete and diffuse interface methods

The evolution of interfaces is intrinsic to many physical processes ranging from cavitation in fluids to recrystallization in solids. Computational modeling of interface motion entails a number of challenges, many of which are related to the range of topological transitions that can occur over the course of the simulation. Microstructure evolution in a polycrystalline material that involves grain boundary motion is a particularly complex example due to the extreme variety, heterogeneity, and anisotropy of grain boundary properties. Accurately modeling this process is essential to determining processing-structure–property relationships in polycrystalline materials though. Simulations of microstructure evolution in such materials often use diffuse interface methods like the phase field method that are advantageous for their versatility and ease of handling complex geometries but can be prohibitively expensive due to the need for high interface resolution. Discrete interface methods require fewer grid points and can consequently exhibit better performance but have received comparatively little attention, perhaps due to the difficulties of maintaining the mesh and consistently implementing topological transitions on the grain boundary network. This work explicitly compares a recently-developed discrete interface method to a multiphase field method on several classical problems relating to microstructure evolution in polycrystalline materials: a shrinking spherical grain, the steady-state triple junction dihedral angle, and the steady-state quadruple point dihedral angle. In each case, the discrete method is found to meet or outperform the multiphase field method with respect to accuracy for comparable levels of refinement, demonstrating its potential efficacy as a numerical approach for microstructure evolution in polycrystalline materials.

Effect of solution aging treatment on microstructure and properties of Fe‐0.5C‐11Cr corrosion resistant alloy by laser cladding

Fe‐0.5C‐11Cr corrosion resistant alloy coating was prepared on the surface of 3Cr13 stainless steel by powder feeding laser cladding, and the microstructure of the coating layer was controlled by solution aging treatment. The microstructure of the coatings was measured by scanning electron microscopy, and phase composition was determined by X-ray diffraction. The hardness and corrosion resistance of coatings before and after heat treatment were evaluated on microhardness tester and electrochemical workstation, respectively. The results show that the morphology, size and quantity of dendrite carbides in the cladding layer of Fe‐0.5C‐11Cr alloy can be improved by solution and aging treatment. The eutectic carbides in the cladding layer can be changed from block, bone, continuous network of grain boundaries to round granular and discontinuous short rod by solution and aging treatment, which greatly improves the uniformity of the microstructure. There is a dynamic equilibrium and critical time for the melting and growth of carbides at 1050 °C. When the holding time is less than 20 min, the size of carbide decreases with time. After 20 min, fine carbides agglomerate and grow with time. The hardness of the cladding layer after heat treatment is about 9% higher than that of the cladding layer without heat treatment and the uniformity of the hardness is improved. The corrosion resistance of the sample kept for 20 min at 1050 ℃ is the superior because the pitting potential and passivation zone width of the coating kept at 1050 ℃ for 20 min increased by 21.5% and 19.2% compared with the cladding laser without heating, respectively.

Grain growth in thin film under strong temperature gradients

Grain growth in thin films has attracted lots of attention due to the numerous applications of polycrystalline films, on one hand, and modified properties due to changes in the microstructure caused by thermal effects, on the other hand. While the phenomenon of grain growth is well understood in general, effects of e.g. temperature gradients leave questions open still today. In the present study, we investigate the influence of strong temperature gradients on grain growth in thin films. To that aim, a modified three-dimensional Potts model algorithm is employed, where the annealing temperature depends on the position within the sample leading to spatial heterogeneities in the grain boundary network. As a consequence of different mobilities, a drag effect occurs on the boundary network evolution that has serious consequences for the microstructures evolving during grain growth.

Effects of Grain Boundary Engineering on the Microstructure and Corrosion Fatigue Properties of 316L Austenitic Stainless Steel

Grain boundary engineering (GBE) treatment was performed through thermomechanical processing (TMP) to optimize the grain boundary character distribution (GBCD) of 316L austenitic stainless steel. The effects of TMP on the GBCD and corrosion fatigue properties in high-temperature and high-pressure water were investigated. The results indicated that a high fraction (about 74%) of special boundaries as well as the interrupted network of random high-angle grain boundaries were obtained through 5% strain followed by annealing at 1,273 K for 90 min. The Σ9 and Σ27 boundaries were generated by the reaction of special boundaries. The highest corrosion fatigue life for 3,187 cycles was obtained when the TMP parameters of the 316L ASS were of 5% strain, annealing temperature of 1,273 K, and annealing time of 45 min. The low-energy special boundaries had strong intergranular corrosion resistance, but the strength of these boundaries was not enough to resist the propagation of transgranular fatigue cracks.

Engineering grain boundary anisotropy to elucidate grain growth behavior in alumina

Current grain growth models have evolved to account for the relationship between grain boundary energy/mobility anisotropy and the five degrees of grain boundary character. However, the role of grain boundary networks on overall growth kinetics remains poorly understood. To experimentally investigate this problem, a highly textured Al2O3 was fabricated by colloidal casting in a strong magnetic field to engineer a unique spatial distribution of grain boundary character. Microstructural evolution was quantified and compared to an untextured sample. From this comparison, a prevalence of (0001)/(0001) terminated grain boundaries with anisotropic networks were identified in the textured sample. These boundaries and their networks were found to be driving grain growth at a faster rate than predicted by models. These findings will allow better modelling of grain growth in real systems by experimentally exploring the impact thereon of grain boundary plane anisotropy and relative energy/mobility differences between neighboring boundaries.

On the critical strain of thermal-mechanical processing to tailor grain boundary character distribution in INCOLOY 925 alloy

Through a series of thermal-mechanical processing (TMP) routes with various ratios of cold rolling and subsequent high-temperature annealing, the samples with different critical strain states were prepared to explore the evolution of grain boundary character distribution (GBCD) in INCOLOY 925 during grain boundary engineering (GBE). The GBE microstructure was characterized using the electron backscatter diffraction to study the influence of TMP parameters on the evolution of GBCD and the penetration of random high angle grain boundaries. The results show that the different states of pre-strain correspond to the variational formation mechanism of grain boundary network (GBN). The strain induced boundary migration is the predominant mechanism with the rolling reduction lower than critical strain. With the slight increase of strain, the recrystallization nucleation can be initiated to generate the strain-free grains at a certain annealing temperature. Moderate strain is needed to ensure adequate growth for twin related domains (TRDs). Besides, with the increase of strain and temperature, the dominant mechanism of GBCD gradually transfers from the new twinning mechanism to the Σ3 regeneration mechanism. Specifically, the TMP sample with 10% rolling reduction followed by annealing at 1075 °C for 10min shows relatively higher grain size difference ratio and average grains numbers in the TRDs, indicating that a larger initial grain size can increase critical strain and delay the nucleation of strain-free grains during TMP treatment. The moderately higher grain size shows an opposite effect on the formation of GBN. Finally, an estimation model of critical strain, as a function of initial grain size and annealing temperature, was constructed based on the artificial neural network, and the predicted results show a high consistency with the actual microstructure evolution.

Spreading monoclinic boundary network between hexagonal primary grains for high performance Ni-rich cathode materials

Knowledge of structure-performance relationship is a fundamental issue in the field of material design and engineering. Functional-directed fine tune of the crystal structure has always been inspiring but rarely implemented in energy storage materials. Here we develop an approach to improve the performance of LiNi0.8Co0.1Mn0.1O2 (NCM811), a typical Ni-rich layered cathode material, through building monoclinic surfaces onto hexagonal primary grains, simply accomplished by oxidizing the flake-like primary precursors with KMnO4. In this way, the local octahedral ligand field has been engineered by inducing Jahn-Teller distortion of low spin Ni3+ state, resulting in a three-dimensional monoclinic functional network spreading over a secondary particle. Such an elaborate monoclinic architecture stabilizes the hexagonal structure of primary grains from phase transitions, and also offers an interconnected highway for both ionic and electronic transportations. Accordingly, an enhanced cycling stability and an outstanding rate capability have been achieved in our designed NCM811 material. Our approach starts a prospective way of designing Ni-rich cathode material with local electronic and structural engineering, which could be expanded to widespread battery researches.

Effect of thermomechanical processing via rotary swaging on grain boundary character distribution and intergranular corrosion in 304 austenitic stainless steel

In the present work, the thermomechanical processing consisting of rotary swaging deformation and annealing treatment was performed to optimize the grain boundary character distribution (GBCD) of 304 austenitic stainless steel. To systematically investigate the effect of GBCD evolution on intergranular corrosion (IGC), double loop electrochemical potentiokinetic reactivation and electrolytic oxalic acid etch tests were employed. The experimental results show that the fraction of low-Σ coincident site lattice (CSL) boundaries increased from 58.1% to 74.0% for the specimen swaged to 0.06 true strain and then annealed at 1050 °C for 5 min duration. By characterizing the evolution of GBCD as a function of strain level in terms of low-Σ CSL boundaries fraction, average twin-related domain (TRD) size, average number of grain per TRD and fractal dimension of the maximum random boundary connectivity, the grain boundary engineering (GBE) microstructure was realized by the occurrence of prolific multiple twinning events during strain-induced boundary migration while static recrystallization has a detrimental effect on optimizing GBCD for the prolific of new strain-free grains with random boundaries. The IGC resistance of the GBE-treated 304 austenitic stainless steel is enhanced by inhibiting the nucleation and propagation of IGC cracks, resulting from the increase in the fraction of low-Σ CSL boundaries, especially Σ3 boundaries and the disruption of random boundary network connectivity.

Three-dimensional imaging of grain boundaries via quantitative fluorescence X-ray tomography analysis

Three-dimensional visualization of material composition within multiple grains and across complex networks of grain boundaries at nanoscales can provide new insight into the structure evolution and emerging functional properties of the material for diverse applications. Here, using nanoscale scanning X-ray fluorescence tomography, coupled with an advanced self-absorption correction algorithm developed in this work, we analyze the three-dimensional gain distributions and compositions in a Ce0.8Gd0.2O2-δ-CoFe2O4 mixed ionic-electronic conductor system with high accuracy and statistical significance. Our systematic investigation reveals an additional emergent phase and uncovers highly intriguing composition stability ranges for the multiple material phases within this system. The presented visualization of composition variations across complex interfaces, supported by our quantitative composition analysis, discloses mechanistic pathways of the diverse phase transformations occurring in the material synthesis, providing insights for the optimization of transport properties in the mixed ionic-electronic conductor system.

Semantic segmentation of building extraction in very high resolution imagery via optimal segmentation guided by deep seeds

The automatic segmentation of buildings from satellite images is extremely challenging due to the complex shapes of buildings. Deep convolutional neural networks (DCNNs) have recently enabled accurate pixel-level labeling tasks, which ensures precise boundaries. However, the large number of pixel-level labels required for DCNN-based semantic segmentation of buildings represents a considerable obstacle to the process. We propose a framework that relies on deep seeds and optimal segmentation to extract buildings from very high resolution imagery to solve the aforementioned issue. Input images are cropped rather than resized to a low-resolution image and passed directly to DCNN for segmentation. An image resize transformation will lead to a significant loss of image detail, especially for images with high resolution. For both image patches and resized images, a classification network is utilized to locate deep seeds within building and nonbuilding classes. Then, a boundary map can be predicted by passing the resized image through a convolutional oriented boundary network. A hierarchical segmentation tree is built to determine the optimal segmentation by seeking the optimal tree cut. The final segmentation is achieved by developing a graphic model representing a set of nodes that distribute information from deep seeds to unmarked regions. According to the experiments conducted on the ISPRS two-dimensional semantic labeling contest (Potsdam) and the WHU building datasets, the proposed framework is able to significantly enhance the building segmentation accuracy. We have achieved such improvements through a cost-effective computing method without training a segmentation network.

Capsule Boundary Network With 3D Convolutional Dynamic Routing for Temporal Action Detection

Temporal action detection is a challenging task in video understanding, due to the complexity of the background and rich action content impacting high-quality temporal proposals generation in untrimmed videos. Capsule networks can avoid some limitations of the invariance caused by pooling and inability from convolutional neural networks, which can better understand the temporal relations for temporal action detection. However, because of the extremely computationally expensive procedure, capsule network is difficult to be applied to the task of temporal action detection. To address this issue, this paper proposes a novel U-shaped capsule network framework with a k-Nearest Neighbor (k-NN) mechanism of 3D convolutional dynamic routing, which we named U-BlockConvCaps. Furthermore, we build a Capsules Boundary Network (CapsBoundNet) based on U-BlockConvCaps for dense temporal action proposal generation. Specifically, the first module is one 1D convolutional layer for fusing the two-stream with RGB and optical flow video features. The sampling module further processes the fused features to generate the 2D start-end action proposal feature maps. Then, the multi-scale U-Block convolutional capsule module with 3D convolutional dynamic routing is used to process the proposal feature map. Finally, the feature maps generated from the CapsBoundNet are used to predict starting, ending, action classification, and action regression score maps, which help to capture the boundary and intersection over union features. Our work innovatively improves the dynamic routing algorithm of capsule networks and extends the use of capsule networks to the temporal action detection task for the first time in the literature. The experimental results on benchmarks THUMOS14 show that the performance of CapsBoundNet is obviously beyond the state-of-the-art methods, e.g., the mAP@tIoU = 0.3, 0.4, 0.5 on THUMOS14 are improved from 63.6% to 70.0%, 57.8% to 63.1%, 51.3% to 52.9%, respectively. We also got competitive results on the action detection dataset of ActivityNet1.3.

Tuning the microstructure and mechanical properties of TiAl-based alloy through grain boundary engineering

In order to modify the room-temperature plasticity of TiAl-based alloy, grain boundary character distribution (GBCD) in TiAl-based alloy was tuned through multidirectional isothermal forging (MDIF) combined with annealing. The experimental results showed that MDIF provided appropriate driving force for the regeneration of the coincidence site lattice (CSL) boundaries through recovery and recrystallization process. Combination of MDIF and annealing at 1100 °C for 90 min promoted the formation of a large number of annealing twins and increased the fraction of low-ΣCSL boundaries to 65.88% and the ratio of (Σ9+Σ27)/Σ3 to 15.12%, which contributed to the disruption of random boundary networks and enhanced the room-temperature plasticity of the TiAl-based alloy.

The influence of parent austenite characteristics on the intervariant boundary network in a lath martensitic steel

The influence of the parent austenite deformation state on the intervariant boundary network (i.e., population, plane orientation, and connectivity) of a lath martensitic microstructure was investigated using conventional EBSD mapping and five-parameter boundary analysis approach along with quantification of boundary connectivity using homology metrics. The lath martensite largely revealed a bimodal misorientation angle distribution, closely matched with the Kurdjumov–Sachs (K–S) orientation relationship. The application of deformation significantly changed the distribution, gradually reducing the intensity of the 60° misorientation angle peak. This was largely ascribed to substructure development within the parent austenite upon deformation, which stimulates particular variant/s having a habit plane (011)α′ closely parallel to the primary (111)γ and/or secondary (1overline{1 }1)γ slip plane. The interaction of these variants eventually promoted specific intervariant boundaries (e.g., 60°/[111], 10.5°/[011], and 49.5°/[110]) at the expense of 60°/[011]. The application of deformation in the parent austenite did not change the intervariant boundary plane character distribution, which mostly exhibited an anisotropic character terminated on {110} planes because of the displacive nature of the martensitic transformation. However, the extent of anisotropy progressively decreased with increasing strain in the austenite prior to transformation. The grain boundary network connectivity was markedly altered due to the local variant selection induced by the deformation. Deformation in the austenite regime generally decreased the connectivity of boundaries having a {110} plane orientation. The intervariant boundaries with the {110} twist character also displayed a similar trend, though the connectivity of {110} tilt boundaries progressively enhanced with increasing strain. The former was closely matched with a decrease in the population of 60°/[110] intervariant boundaries with the strain. The current findings suggest that the intervariant boundary network of lath martensite can be manipulated through changes in the parent austenite deformation state which, in turn, can be used to enhance key final product properties such as toughness.Graphical abstract

A practical guide to characterizing irradiated nuclear fuels using FIB tomography

Focused ion beam (FIB) tomography with combined electron backscatter diffraction (EBSD) and energy dispersive x-ray spectroscopy (EDS) is a technique capable of statistically characterizing the microstructure and spatial compositional variation of nuclear fuel in three-dimensions (3D). The 3D visualization from FIB tomography provides a comprehensive picture of the interconnected microstructural and compositional features that can impact fuel performance. While these features are often characterized with surface examination, the complexity and relationship of fission products and grain boundary networks may not fully be captured by these 2D methods. This work presents a practical guide to FIB tomography that is tailored to nuclear fuel characterization. The steps used to collect and process the data are provided along with the scripts used to process the data. Additionally, suggestions for future characterization efforts utilizing this approach are given.

Investigation of short-term creep properties of a coarse-grained Inconel 718 fabricated by directed energy deposition compared to traditional Inconel 718

Short-term creep performance of a coarse-grained Inconel 718 alloy fabricated by Directed energy deposition (DED) technique has been investigated at 650 °C in comparison to a conventional wrought material. In order to study the effect of heat treatment and sample orientation on the creep behaviour, as-deposited specimens oriented parallel and perpendicular to the building direction have been treated by double aging (DA), solution annealing + double aging (SA) and homogenization + solution annealing + double aging (HSA) conditions. Microstructural changes along with their impacts on high temperature mechanical properties have been examined with respect to grain size, crystallographic orientation, precipitates, and dislocation substructure. Results show that, superior creep strength and creep ductility of DED-manufactured DA specimen can be achieved compared to wrought specimen. The process-induced dislocation substructure and the irregular columnar grains with the random network of grain boundaries primarily contribute to the superior creep behaviour. However, when the unique columnar grain features formed during DED process are replaced by recrystallized, equiaxed grain structure in the HSA case, the worst creep performance is observed due to the coarse grains with regular-shaped grain boundaries facilitating the crack propagation. In addition, anisotropy of creep performance with respect to sample orientation observed for the heat-treated DED specimens can be significantly alleviated by a higher temperature of solution treatment.

Texture-aware dual domain mapping model for low-dose CT reconstruction.

Remarkable progress has been made for low-dose computed tomography (CT) reconstruction tasks by applying deep learning techniques. However, establishing an intrinsic link between deep learning techniques and CT texture preservation is still one of the significant challenges for researchers to further improve the effect of low-dose CT (LDCT) reconstruction. Most of the existing deep learning-based LDCT reconstruction methods are derived from popular frameworks, and most models focus on the image domain. Even few existing methods start with dual domains (sinogram and image) by considering the processing of the data itself, the final performances are limited due to the lack of perception of textures. With this in mind, we propose a method for texture perception on dual domains, so that the reconstruction process can be uniformly driven by visualeffects. The proposed method involves the processing of two domains: the sinogram domain and the image domain. For the sinogram domain, we have designed a novel dilated residual network (S-DRN) which aims to increase the receptive field to obtain multiscale information. For the image domain, we propose a self-attention (SA) residual encoder & decoder network (SRED-Net) as the denoising network for obtaining much acceptable edges and textures. In addition, the composite loss function composed of the feature loss constructed by the proposed boundary and texture feature-aware network (BTFAN) and the mean square error (MSE) can obtain a higher image quality while retaining more details and fewer artifacts, thereby obtaining better visual image quality. The proposed method was validated using both the American association of physicists in medicine (AAPM)-Mayo clinic LDCT data sets and a real clinic data. Experimental results demonstrated that the new method has achieved the state-of-the-art performance on objective indicators and visual metrics in terms of denoising and texture restoration. Compared with single-domain or existing dual-domain processing strategies, the proposed texture-aware dual domain mapping network (TADDM-Net) can much better improve the visual effect of reconstructed CT images. Meantime, we also provide much intuitive evidence in terms of modelinterpretability.