Large 3D Models Research Articles

CLAY: A Controllable Large-scale Generative Model for Creating High-quality 3D Assets

In the realm of digital creativity, our potential to craft intricate 3D worlds from imagination is often hampered by the limitations of existing digital tools, which demand extensive expertise and efforts. To narrow this disparity, we introduce CLAY, a 3D geometry and material generator designed to effortlessly transform human imagination into intricate 3D digital structures. CLAY supports classic text or image inputs as well as 3D-aware controls from diverse primitives (multi-view images, voxels, bounding boxes, point clouds, implicit representations, etc). At its core is a large-scale generative model composed of a multi-resolution Variational Autoencoder (VAE) and a minimalistic latent Diffusion Transformer (DiT), to extract rich 3D priors directly from a diverse range of 3D geometries. Specifically, it adopts neural fields to represent continuous and complete surfaces and uses a geometry generative module with pure transformer blocks in latent space. We present a progressive training scheme to train CLAY on an ultra large 3D model dataset obtained through a carefully designed processing pipeline, resulting in a 3D native geometry generator with 1.5 billion parameters. For appearance generation, CLAY sets out to produce physically-based rendering (PBR) textures by employing a multi-view material diffusion model that can generate 2K resolution textures with diffuse, roughness, and metallic modalities. We demonstrate using CLAY for a range of controllable 3D asset creations, from sketchy conceptual designs to production ready assets with intricate details. Even first time users can easily use CLAY to bring their vivid 3D imaginations to life, unleashing unlimited creativity.

Improving the performance of 3D image model compression based on optimized DEFLATE algorithm

This study focuses on optimizing and designing the Delayed-Fix-Later Awaiting Transmission Encoding (DEFLATE) algorithm to enhance its compression performance and reduce the compression time for models, specifically in the context of compressing NX three-dimensional (3D) image models. The DEFLATE algorithm, a dual-compression technique combining the LZ77 algorithm and Huffman coding, is widely employed for compressing multimedia data and 3D models. Three 3D models of varying sizes are selected as subjects for experimentation. The Wavelet algorithm, C-Bone algorithm, and DEFLATE algorithm are utilized for compression, with subsequent analysis of the compression ratio and compression time. The experimental findings demonstrate the DEFLATE algorithm’s exceptional performance in compressing 3D image models. Notably, when compressing small and medium-sized 3D models, the DEFLATE algorithm exhibits significantly higher compression ratios compared to the Wavelet and C-Bone algorithms while also achieving shorter compression times. Compared to the Wavelet algorithm, the DEFLATE algorithm enhances the compression performance of 3D image models by 15% and boosts data throughput by 49%. While the compression ratio of the DEFLATE algorithm for large 3D models is comparable to that of the Wavelet and C-Bone algorithms, it notably reduces the actual compression time. Furthermore, the DEFLATE algorithm enhances data transmission reliability in NX 3D image model compression by 12.1% compared to the Wavelet algorithm. Therefore, the following conclusions are drawn: the DEFLATE algorithm serves as an excellent compression algorithm for 3D image models. It showcases significant advantages in compressing small and medium-sized models while remaining highly practical for compressing large 3D models. This study offers valuable insights for enhancing and optimizing the DEFLATE algorithm, and it serves as a valuable reference for future research on 3D image model compression.

Modeling the dynamic fracture of concrete — A robust, efficient, and accurate mesoscale description

Dynamic loading of concrete often leads to excessive cracking and fragmentation. Since these processes are massively influenced by the underlying heterogeneity of the material, discrete modeling of the lower-scale features strongly improves the simulation results. At the same time, the computational effort is greatly increased. We therefore propose here a mesomechanical simulation approach which is efficient in terms of model generation and computational efforts even for large 3d models. Furthermore, it is robust by limiting the application of computationally less efficient cohesive zone elements to the weak transition boundary between matrix and inclusions. Intermatrix failure is considered by a smeared crack approach and element removal after a certain crack opening has been reached. We apply an improved version of the widely used RHT concrete model, which is enhanced with a principal stress criterion. The resulting approach leads to an accurate replication of experiments in terms of crack propagation and stress wave induced dynamic fragmentation.

ON THE STRUCTURAL VALUATION IN OBLIQUE WAVES OF A LARGE LIQUEFIED NATURAL GAS CARRIER BY 3D-FEM MODEL APPROACH

This structural valuation in oblique equivalent waves has as a study case a large LNG liquefied natural gas carrier, with an overall length of 289.3 m, structural gas membrane tanks, and a double hull. From the operation load cases, for the numerical study, a selection of the main cases is made, full cargo and ballast. The selected oblique design wave conditions are specific to unrestricted navigation areas, sagging and hogging wave-free surface, with the heading angle covering the oblique relevant cases. The computation of the LNG carrier equilibrium in oblique design waves is done by a three-parameter nonlinear approach, with the own P_QSW software. The 3D-FEM large LNG carrier model is developed by Femap/NX Nastran software. The structural valuation in oblique design waves, by strength and freeboard criteria, of the large LNG carrier involves 3D and 1D models. For practical purposes, the strength valuation results are synthesized as polar diagrams, for the relevant heading angles of the oblique design waves, representing the wave height safety limits for preserving the large LNG carrier structural integrity.

Domain-specific modeling and semantic alignment for image-based 3D model retrieval

Image-based 3D model retrieval aims to search for 3D models according to 2D image queries, which provides a convenient way for the management of large 3D model datasets. Most of the related works put the emphasis on bridging the modality gap between 2D images and 3D models, which faces a lot of challenges due to the huge domain discrepancy. In this paper, we explicitly model and eliminate the domain-specific features of 2D images and 3D models. To alleviate the negative effect of complex background of natural images, we adopt semantic focus loss to constrain networks to learn the most semantically relevant feature representations for both 2D images and 3D models. We conduct extensive experiments on two cross-domain 3D model retrieval datasets, MI3DOR and MI3DOR-2, to show the effectiveness of the proposed method.

VOXELISATION AND VOXEL MANAGEMENT OPTIONS IN UNITY3D

Abstract. Voxels have been used in various application domains successfully for the last several decades. Their main advantage is the underlying discrete data structure allowing to reliably work with surrounding voxels all the time. In this paper, capabilities of the Unity game engine for voxels management and geometry voxelisation are assessed, where 4 native solutions and 7 open-source projects written for Unity are investigated. Although many voxel-based options exist in Unity, they only deal with one part related to voxels. Therefore, the available capabilities to voxelise, visualise, structure and export voxels are combined and extended with the goal of successfully processing large 3D models and geometries. Many voxel visualisation techniques are investigated including mesh, VFX, point clouds and SVO, which have distinctive benefits in various aspects. Possibilities to structure voxels for effective management in simulations and other tasks are shown. Also, it is enabled to export voxels as point clouds and to the Postgres database for further processing, spatial analysis and distribution. One of the main conclusions is the lack of support for state-of-the-art voxel data structures, where the presented platform can easily be extended to support any. This platform can be used by people who deal with 3D discrete data, require voxelising 3D data, visualise voxels in different ways and technologies, as well as manage more efficiently sparsely occupied voxels.

Numerical modelling of lauric acid phase change material using iterative and non-iterative time-advancement schemes

Phase Change Materials are used in different applications as thermal storage solution. Numerical and experimental studies have been performed to investigate and test their properties. But large time consumption of PCM numerical simulation is considered as one of the main limitations that restrict modelling of large dimension 2D and 3D models, in addition to large time scale experiments. Numerical model using ANSYS FLUENT NITA (Non-Iterative Time Advancement) scheme was employed and compared to the default ITA (Iterative Time Advancement) scheme. The comparison included pairs of models at different grid sizes and time-step sizes. It was found that simulation time-saving by NITA scheme exceeded 89 % and reached 96.4 % when compared to the default ITA scheme. The maximum error percentage between them was 0.27 % in the employed model.

IMAGE-BASED CONTROL POINT DETECTION AND CORRESPONDENCE-FREE GEOREFERENCING

Abstract. In order to appropriately measure properties in 3D models, accurate georeferencing plays a vital role in structural health monitoring. For that purpose, control points are attached to the surface of the structure and measured geodetically. These points can be recovered in the virtual model and associated with the geodetic measurements. Automating the process of detecting and associating control points and geodetic measurements, facilitates the accurate georeferencing of large 3D models. While the number of marker types for control points is steadily increasing, this work claims that – under a plausible assumption – comparatively simple and commonly used marker designs can serve for accurate and robust georeferencing. By assuming that control points are asymmetrically distributed over the surface of the structure, the correspondence of points is determined by their geometric interrelation. In this work, an image-based detector for relatively simple control point types is proposed, applying transfer learning on hierarchical multi-scale attention (HMA) (Tao et al., 2020). For associating detected and geodetically measured points, a RANSAC-based procedure is presented that determines a geometrically consistent transformation between detected and measured points.

Kronecker Neural Networks Overcome Spectral Bias for PINN-Based Wavefield Computation

Wavefield computation constitutes the majority of the computational cost for seismic applications, including reverse-time migration and full waveform inversion (FWI). One of the popular approaches is to solve for the wavefields in the frequency domain by using the Helmholtz equation. However, Helmholtz solvers require inversion of a large stiffness matrix that can become computationally intractable for large 3D models or in the case of modeling high frequencies. Recently, researchers have explored the potential of physics-informed neural networks (PINNs) in solving the Helmholtz equation with limited success. While a number of attractive features have been demonstrated for the PINN-based Helmholtz solvers, their large training cost has been the main impediment in their widespread adoption for wavefield computations. The large training cost is mainly due to the spectral bias of neural networks, which poses difficulty in training the PINN model for high-frequency wavefields. In this work, I employ Kronecker neural networks (KNNs) that form a general framework for neural networks with adaptive activation functions. I, specifically, implement it using a standard feed-forward neural network employing a composite activation function formed by using the inverse tangent (atan), exponential linear unit (elu), locally adaptive sine (l-sin), and locally adaptive cosine (l-cos) activation functions. Thanks to the oscillatory noise added by the sine and cosine terms, the network is able to explore more and learn faster. This allows the network to get rid of saturation regions from the output of each layer and overcome slow convergence. Through numerical tests, the efficacy of the proposed approach in fast and accurate wavefield modeling in the frequency domain is demonstrated.

Multi-Component Thermal Fluid Injection Performance in Recovery of Heavy Oil Reservoirs

Field observations discern that the oil production rate decreases substantially and water cut increases rapidly with the increase of steam injection cycles. Compared with steam drive, the advantage of flue gas (also called multi-component thermal gas) co-injection with steam is that flue gas can increase the reservoir pressure and expand the heating chamber. In this paper, the flue gas generated by fuel burning in the field was injected with steam to improve heavy oil recovery. This technique was investigated in the large laboratory 3D model and implemented in the field as well. The huff-n-puff process efficiency by flue gas, steam, and flue gas–steam co-injection was compared in the experiments. The field practice also demonstrated that the addition of non-condensable gas in the steam huff-n-puff process recovered more oil than steam alone. The temperature profile in the wellbore with flue gas injection is higher than that with steam injection since the low thermal conductivity of N2 reduces the heat loss. With the increase of stimulation cycles, the incremental oil recovery by flue gas injection declines significantly.

On the Implementation and Usability of Crystal Preferred Orientation Evolution in Geodynamic Modeling

AbstractFlow in the Earth's mantle causes the preferred orientation of crystals called lattice/crystal preferred orientation (LPO or CPO). This preferred orientation is one of the main reasons why seismic anisotropy is observed. Seismic anisotropy observations could therefore be used to constrain the mantle flow in geodynamic models through tracking CPO evolution, and computing the resulting elastic tensor and the anisotropy predicted at the surface. Even though there are many types of CPO models, only a few studies include CPO calculations due to the complexity and computational cost. Here, we implemented an extended version of the CPO model D‐Rex into the open‐source community geodynamics code ASPECT. We show that the implementation is correct, how to use it, and that it is feasible and important to use in large 3D models. We also show that it is important to calculate CPO, especially for models focusing on plate boundary of smaller scale flow because the resulting fast axis directions can greatly deviate from the flow direction. The added infrastructure will also allow for future enhancement, testing, and even replacement of the CPO model.

A path generation method for wire and arc additive remanufacturing of complex hot forging dies

Wire and arc additive remanufacturing (WAAR) technology has become a new solution for hot forging dies repair and remanufacturing. In this study, a path generation method is proposed for WAAR of hot forging dies. At first, a WAAR process of the hot forging die is presented, and considering the characteristics of large welding heat input and complex 3D digital model, the hybrid path planning strategy is confirmed as an appropriate strategy for WAAR. The developed hybrid path generation method for WAAR consists of three main steps: determine the direction of the scan line; divide and fill the internal area; and connect the sub-paths. The relatively optimal scanning direction is determined by calculating the length and inclination angle of each line segment in the contour lines, which reduces the possibility of sharp angles. The internal region is divided according to the position of the selected extreme points, and the path space is adjusted to avoid the occurrence of the unfilled phenomenon. At the stage of sub-path connection, some criteria are proposed to reduce the number of sub-paths. At last, the effectiveness and robustness of the proposed method are validated through the planar deposition experiment and the WAAR process of four damaged hot forging dies.

PINNeik: Eikonal solution using physics-informed neural networks

The eikonal equation is utilized across a wide spectrum of science and engineering disciplines. In seismology, it regulates seismic wave traveltimes needed for applications like source localization, imaging, and inversion. Several numerical algorithms have been developed over the years to solve the eikonal equation. However, these methods require considerable modifications to incorporate additional physics, such as anisotropy, and may even breakdown for certain complex forms of the eikonal equation, requiring approximation methods. Moreover, they suffer from computational bottleneck when repeated computations are needed for perturbations in the velocity model and/or the source location, particularly in large 3D models. Here, we propose an algorithm to solve the eikonal equation based on the emerging paradigm of physics-informed neural networks (PINNs). By minimizing a loss function formed by imposing the eikonal equation, we train a neural network to output traveltimes that are consistent with the underlying partial differential equation. We observe sufficiently high traveltime accuracy for most applications of interest. We also demonstrate how the proposed algorithm harnesses machine learning techniques like transfer learning and surrogate modeling to speed up traveltime computations for updated velocity models and source locations. Furthermore, we use a locally adaptive activation function and adaptive weighting of the terms in the loss function to improve convergence rate and solution accuracy. We also show the flexibility of the method in incorporating medium anisotropy and free-surface topography compared to conventional methods that require significant algorithmic modifications. These properties of the proposed PINN eikonal solver are highly desirable in obtaining a flexible and efficient forward modeling engine for seismological applications.

Visualizing 3D imagery by mouth using candy-like models.

Handheld models help students visualize three-dimensional (3D) objects, especially students with blindness who use large 3D models to visualize imagery by hand. The mouth has finer tactile sensors than hand, which could improve visualization using microscopic models that are portable, inexpensive, and disposable. The mouth remains unused in tactile learning. Here, we created bite-size 3D models of protein molecules from "gummy bear" gelatin or nontoxic resin. Models were made as small as rice grain and could be coded with flavor and packaged like candy. Mouth, hands, and eyesight were tested at identifying specific structures. Students recognized structures by mouth at 85.59% accuracy, similar to recognition by eyesight using computer animation. Recall accuracy of structures was higher by mouth than hand for 40.91% of students, equal for 31.82%, and lower for 27.27%. The convenient use of entire packs of tiny, cheap, portable models can make 3D imagery more accessible to students.

A Coupled Flow–Geomechanical Modeling of Out-of-Sequence Fracturing Using a Dual-Lattice Implementation of Synthetic-Rock-Mass Approach

Summary In out-of-sequence (OOS) pinpoint fracturing, Stage 1 is fractured, followed by Stage 3, after which Stage 2 (center fracture) is placed between Stages 1 and 3 (outside fractures). The center fracture can exploit the reduced stress anisotropy to activate planes of weakness (e.g., fissures) and create branch fractures that can connect hydraulic fractures to stress-relief fractures, ultimately enhancing fracture connectivity and complexity. It has been trialed in western Siberia (2014) and western Canada (2017 to 2019) with overall operational and production performance success. Previous fracture-modeling works calibrated by OOS fracturing trials have either used shear-decoupled planar-fracture models (in which slippage along the shear planes restricts the displacement to a limited area because of displacement damping)—which are unable to reproduce out-of-plane fracture complexity, and to dynamically track the change in stress anisotropy and orientation—or discrete-fracture-network (DFN) models, which often exaggerate the fracture-network connectivity, and reproduce unrealistically high fracture-network-extension pressures in the stimulated reservoir volume (SRV). This work attempts to resolve the issues in planar-fracture and DFN models by more realistically addressing the dominant mechanisms of OOS fracturing, dynamic changes in the stress anisotropy and orientation, activation of pre-existing planes of weaknesses, and poroelasticity using an iteratively coupled flow–geomechanical model that uses the dual-lattice implementation of the synthetic-rock-mass (SRM) model with a robust, fully coupled, iterative flow/stress solution to capture the following: Nonlinear deformations caused by induced tensile- and shear-fracture-complexity propagation Induced stress shadowing in and around the SRV Sliding of opened, pre-existing joints, fractures, and fissures using the smooth-joint model (SJM) Propagation of the hydraulic fracture as an aggregate of intact matrix fracturing and opening and slip of pre-existing fluid-filled planes of weakness (e.g., joints, fractures, fissures) Permeability enhancement in the main tensile and complex fractures following the updated deformation aperture from the coupled solution The results (fracture geometries and treatment pressures) of the three models (planar-fracture, DFN, and SRM with lattice models) are compared after using each model for treatment-pressure history matching of an OOS-fracturing trial. The calibrated, coupled SRM with lattice model more reasonably reproduces the measured fracture-extension pressures and end-of-job pressures from OOS pinpoint fracturing treatments, and it reveals the following: The dynamic change in the stress-field orientation and magnitude during OOS fracturing leads to a reduction in stress anisotropy and complex out-of-plane fracturing in the SRV for center fractures. Center fractures tend to be narrower and shorter if sufficient out-of-zone growth is attained in the absence of strong vertical containment, making OOS fracturing an option for penetrating multistacked zones in one treatment. Where center fractures are shorter or near-well fracture complexity is generated, OOS fracturing can be considered in treating the child wells to reduce fracture hits. Compared with planar-fracture and DFN models, this coupling technique achieves the following: Accounts for dominant mechanisms of complex shear and tensile fracturing Renders fast computation in simulating large 3D models with dual-lattice implementation of SRM with SJM Reproduces fracture surface area and SRV permeability more realistically Leads to a more reasonable history match of the measured OOS-fracturing pressures

3D inversion of magnetotelluric data by using a hybrid forward-modeling approach and mesh decoupling

We have developed a new algorithm for the inversion of magnetotelluric (MT) data. The developed algorithm is built to be fast, versatile, and accurate. A fast inversion algorithm has to include a fast forward-modeling routine. To achieve that, a hybrid approach consisting of finite-difference (FD) and finite-element (FE) methods is used to benefit from the speed of the FD method and the flexibility to add topographic features of the FE method. To reduce the number of cells, and thus reducing the size of the system to be solved in the forward and pseudoforward solutions, different meshes for various groups of frequencies are used. Then, these are mapped onto the inversion mesh by a mesh-decoupling technique to further accelerate the inversion. To build a versatile inversion algorithm, the capability of using different data types is implemented. In addition to the impedance tensor and the magnetic transfer function, the algorithm also computes the phase tensor and phase vector, which are distortion-free forms of MT data. It is also possible to invert intersite data and their respective phase tensors using the developed code. Furthermore, the distortion matrix can also be estimated as a parameter. The new code is tested with different noisy and distorted synthetic data measured on a surface with topography to evaluate the inversion accuracy and computational efficiency. The results indicate that the code is accurate and that the runtimes are reasonable for the large 3D models considered. Using four graphics processing units, the hybrid forward-modeling approach and the mesh-decoupling technique together result in a 12 times speedup for the examples presented in this study.

Large-Amplitude Vibrations of Spider Web Structures

Spider silk, as a natural material, shows exceptional performance in its properties. The combination of the superior properties of spider silk and the geometry of spider structures make the spider web very resilient. A spider web structure can be considered as a cable-like structure with inappreciable torsional, bending and shear rigidities. An investigation emphasising on natural frequencies and corresponding mode shapes with and without the consideration of geometric nonlinearity is presented in this paper. This study is the world’s first discovery of large amplitude free vibration behaviours of spider web structures. Large deformable finite element 3D models of spider web structures have been developed and validated. By using the energy method, the variational model of the spider web structures have been established to further extend the finite element model, consisting of the strain energy due to axial deformation, kinetic energy due to the spider web movement and the virtual work caused by the self-weight per unit unstretched length. The emphasis of this study is placed on the linear and geometric nonlinear behaviour of the spider web structures considering different structural patterns and material properties. To determine the large-amplitude free vibrational behaviours, a series of pretension load is applied to the first step in Abaqus to initiate the nonlinear strain-displacement relationships enabling a precursor to free vibrations. The parametric studies stemming from structural patterns (the number of radial and capture threads), elastic modulus, density, and inertia moment have been highlighted. The insight will help engineers and scientists to adapt the concept of spider webs, their geometric properties, and damage patterns for the design of any structural membranes, preventing any failure from dynamic resonances and nonlinear phenomena.

Efficient implementation of critical plane for 3D stress histories using triangular elements

Multiaxial fatigue and the application of critical plane criteria are briefly reviewed. Methods to increase the efficiency of such criteria are discussed. A simple methodology is also presented where the critical plane search space is discretised using triangular elements and adaptively refined. It is shown to increase accuracy in the damage search with fewer candidate plane evaluations, useful for large 3D models. Parallel computations are also leveraged.

Effective Collaborative Design of Large Virtual 3D Model using Multiple AR Markers

In the new era of industry, collaboration amongst partners will be the most challenging part of design and manufacturing. There are many ways for companies to collaborate with remote experts or partners in designing, manufacturing, or distributing the new product, depending on how complicated the product is. With the progress in information technology, many techniques can be used for supporting collaborative work. For the design that needs specific or complicates visual appearance, Augmented Reality (AR) is the most promising one. The virtual 3D model can be distributed to the remote partners for evaluation and providing comments. In this research, the technique in generating a stable large 3D model for remote design collaboration using multiple markers was developed. Users from different locations can work on the same virtual model. The platform for the mobile application was also developed. The results from the experiment with this setup confirmed that the system could be used for real collaboration.

Automatic Shadow Detection in Urban Very-High-Resolution Images Using Existing 3D Models for Free Training

Up-to-date 3D city models are needed for many applications. Very-high-resolution (VHR) images with rich geometric and spectral information and a high update rate are increasingly applied for the purpose of updating 3D models. Shadow detection is the primary step for image interpretation, as shadow causes radiometric distortions. In addition, shadow itself is valuable geometric information. However, shadows are often complicated and environment-dependent. Supervised learning is considered to perform well in detecting shadows when training samples selected from these images are available. Unfortunately, manual labeling of images is expensive. Existing 3D models have been used to reconstruct shadows to provide free, computer-generated training samples, i.e., samples free from intensive manual labeling. However, accurate shadow reconstruction for large 3D models consisting of millions of triangles is either difficult or time-consuming. In addition, due to inaccuracy and incompleteness of the model, and different acquisition time between 3D models and images, mislabeling refers to training samples that are shadows but labeled as non-shadows and vice versa. We propose a ray-tracing approach with an effective KD tree construction to feasibly reconstruct accurate shadows for a large 3D model. An adaptive erosion approach is first provided to remove mislabeling effects near shadow boundaries. Next, a comparative study considering four classification methods, quadratic discriminant analysis (QDA) fusion, support vector machine (SVM), K nearest neighbors (KNN) and Random forest (RF), is performed to select the best classification method with respect to capturing the complicated properties of shadows and robustness to mislabeling. The experiments are performed on Dutch Amersfoort data with around 20% mislabels and the Toronto benchmark by simulating mislabels from inverting shadows to non-shadows. RF is tested to give robust and best results with 95.38% overall accuracy (OA) and a value of 0.9 for kappa coefficient (KC) for Amersfoort and around 96% OA and 0.92 KC for Toronto benchmarks when no more than 50% of shadows are inverted. QDA fusion and KNN are tested to be robust to mislabels but their capability to capture complicated properties of shadows is worse than RF. SVM is tested to have a good capability to separate shadow and non-shadows but is largely affected by mislabeled samples. It is shown that RF with free-training samples from existing 3D models is an automatic, effective, and robust approach for shadow detection from VHR images.