Embedding Loss Research Articles

Discrete element method simulations of offshore plate anchor keying behavior in granular soils

The keying mechanism of plate anchors (PLA) embedded into granular sandy soils is investigated in this work using the discrete element method (DEM) modeling to simulate microscale response and to observe the emergent macroscale behavior. Parameters (e.g., padeye eccentricity, loading direction) that influence anchor keying are analyzed in the simulations. The load-displacement response and embedment losses during keying are evaluated and compared to published experimental results from the literature. The keying mechanism of the PLA for different padeye eccentricities and loading directions are investigated. The DEM results are found to have the same trends as published experimental results. The embedment loss has a bilinear response with the padeye eccentricity which is in accordance with the experimental results reported in the literature. Embedment losses increase linearly with increasing loading directions. Microscale observations of mobilized particle friction, particle rotation, contact force network, and steering coefficient during keying are used to provide insights into the keying mechanisms. The potential for particle mobilization is reached more quickly for the larger padeye eccentricities. The particle rotation is the major keying mechanism for all the cases in the simulations. Finally, the granular assembly adjacent to the PLA is steering from horizontal to vertical for all padeye eccentricities.

Transfer Neural Trees: Semi-Supervised Heterogeneous Domain Adaptation and Beyond.

Heterogeneous domain adaptation (HDA) addresses the task of associating data not only across dissimilar domains but also described by different types of features. Inspired by the recent advances of neural networks and deep learning, we propose a deep leaning model of Transfer Neural Trees (TNT), which jointly solves cross-domain feature mapping, adaptation, and classification in a unified architecture. As the prediction layer in TNT, we introduce Transfer Neural Decision Forest (Transfer- NDF), which is able to learn the neurons in TNT for adaptation by stochastic pruning. In order to handle semi-supervised HDA, a unique embedding loss term is introduced to TNT for preserving prediction and structural consistency between labeled and unlabeled target-domain data. We further show that our TNT can be extended to zero shot learning for associating image and attribute data with promising performance. Finally, experiments on different classification tasks across features, datasets, and modalities would verify the effectiveness of our TNT.

Improving Recommendations by Embedding Multi-Entity Relationships With Latent Dual-Metric Learning

Recently, latent vector embedding has become a research hotspot, with its great representative ability to measure the latent relationships among different views. However, most researches utilize the inner product of latent vectors as the representation of relationships, and they develop some embedding models based on this theory. In this paper, we take deep insight into the existing embedding models and find that utilizing the inner product may increase several problems: 1) in latent space, the inner product among three vectors may violate triangle principle; 2) the inner product cannot measure the relationships between vectors in the same category, such as user and user and item and item; and 3) the inner product cannot catch the collaborative relationships (user-user and item-item) for collaborative filtering. Along with this line, we propose a latent vector embedding model for collaborative filtering: latent dual metric embedding (LDME), which utilizes the dual-Euclidean distance in latent space, instead of the inner product, to represent different types of relationships (user-user, item-item, and user-item) with a uniform framework. Specifically, we design an embedding loss function in LDME, which can measure the close and remote relationships between entities, tackle the above problems, and achieve a more clear, well-explained embedding result. Extensive experiments are conducted on several real-world datasets (Amazon, Yelp, Taobao, and Jingdong), where the expiring results demonstrate that LDME can overperform some state-of-the-art user-item embedding models and can benefit the existing collaborative filtering models.

Toward a quick evaluation of the performance of gravity installed anchors in clay: Penetration and keying

Gravity installed anchors (GIAs) are the most recent generation of anchoring solutions to moor floating facilities for deepwater oil and gas developments. Challenges associated with GIAs include predicting the initial embedment depth and evaluating the keying performance of the anchor. The former involves high soil strain rate due to large anchor penetration velocity, while the later influences the subsequent behavior and pullout capacity of the anchor. With the coupled Eulerian–Lagrangian method, three-dimensional large deformation finite element models are established to investigate the penetration and keying of GIAs in non-homogeneous clay. In the penetration model, a modified Tresca soil model is adopted to allow the effects of soil strain rate and strain softening, and user-defined hydrodynamic drag force and frictional resistance are introduced via concentrated forces. In the keying model, the anchor line effects are incorporated through a chain equation, and the keying, diving and pulling out behaviors of the anchor can all be replicated. Parametric studies are undertaken at first to quantify the effects of various factors on the performance of GIAs, especially on the penetration and keying behaviors. Based on the results of parametric studies, fitted formulae are proposed to give a quick evaluation of the anchor embedment depth after the installation, and the shackle horizontal displacement, shackle embedment loss and anchor inclination at the end of the keying. Comparative studies are also performed to verify the effectiveness of the fitted formulae.

Keying behavior of suction embedded plate anchors with flap in clay

Suction embedded plate anchors (SEPLAs) have been applied extensively in deepwater projects due to their high efficiency and low cost. In this paper, a new plasticity model was developed to simulate the keying behavior of SEPLAs. The model was based on three groups of yield envelopes, which considered three different states between the flap and the main plate. Thus, the plasticity model can consider the flap rotation relative to the fluke. The anchor shank was also considered to investigate the three dimensional (3D) effects of the actual SEPLA shape when deriving the yield envelops. A three dimensional large deformation finite element (3D-LDFE) model with flap rotation and shank considered was also set up. Then 3D-LDFE analysis was carried out to verify the plasticity model. The results given by 3D-LDFE analyses agree well with those by plasticity analyses. The effects of the flap on the embedment loss and pullout capacity, and the effects of loading eccentricity and pullout angle on diving behavior are investigated. Optimal padeye offset and pullout angle to drive the anchor diving are proposed.

An efficient approach to incorporate anchor line effects into the coupled Eulerian–Lagrangian analysis of comprehensive anchor behaviors

With the application of innovative anchor concepts and advanced technologies in deepwater moorings, anchor behaviors in the seabed are becoming more complicated and significantly affected by the anchor line. Based on the coupled Eulerian–Lagrangian (CEL) method, a numerical approach incorporating anchor line effects is developed to investigate comprehensive anchor behaviors in the soil, including penetration of drag anchors, keying of suction embedded plate anchors and diving of gravity installed anchors. Compared to the method directly incorporating the anchor line into the CEL analysis, the proposed method is computationally efficient. To examine the robustness and accuracy of the proposed method, numerical probe tests and then comparative studies are carried out. It is found that the penetration (or diving) and keying behaviors of anchors can be well simulated. A parametric study is also undertaken to quantify the effects of various factors on the behavior of OMNI-Max anchors, whose mechanisms are not yet fully understood. The maximum embedment loss of OMNI-Max anchors during keying is not influenced by the initial anchor embedment depth, whereas significantly increases with increasing drag angle at the embedment point. With decreasing initial anchor embedment depth or increasing soil strength gradient, drag angle at the embedment point and diameter of the anchor line, the behavior of OMNI-Max anchors could change from diving to pullout, which is undesirable in offshore engineering practice. If the drag angle increases over a certain limit, the anchor will fail similar to a suction anchor.

Structure Sensitive Hashing With Adaptive Product Quantization.

Hashing has been proved as an attractive solution to approximate nearest neighbor search, owing to its theoretical guarantee and computational efficiency. Though most of prior hashing algorithms can achieve low memory and computation consumption by pursuing compact hash codes, however, they are still far beyond the capability of learning discriminative hash functions from the data with complex inherent structure among them. To address this issue, in this paper, we propose a structure sensitive hashing based on cluster prototypes, which explicitly exploits both global and local structures. An alternating optimization algorithm, respectively, minimizing the quantization loss and spectral embedding loss, is presented to simultaneously discover the cluster prototypes for each hash function, and optimally assign unique binary codes to them satisfying the affinity alignment between them. For hash codes of a desired length, an adaptive bit assignment is further appended to the product quantization of the subspaces, approximating the Hamming distances and meanwhile balancing the variance among hash functions. Experimental results on four large-scale benchmarks CIFAR-10, NUS-WIDE, SIFT1M, and GIST1M demonstrate that our approach significantly outperforms state-of-the-art hashing methods in terms of semantic and metric neighbor search.

Influence of padeye offset on bearing capacity of three-dimensional plate anchors

This paper reports advanced three-dimensional large-deformation finite element analyses investigating the effect of the padeye offset on the behaviour of plate anchors. The analyses varied the normalized padeye offset from 0 to 0.5 and identified two counterbalancing effects on the anchor bearing capacity induced by the padeye offset. These are the change in anchor inclination with respect to the direction of loading (detrimental due to the reduction of bearing area) and the reduction of loss of embedment (beneficial, as the anchor mobilizes stronger soil). The results provide insights to optimize the design of plate anchors. The underlying concepts also have applications for other types of anchors.

Capacity of dynamically embedded plate anchors as assessed through field tests

A dynamically embedded plate anchor (DEPLA) is a rocket-shaped anchor that penetrates to a target depth in the seabed by the kinetic energy obtained through free-fall and by the anchor’s self-weight. After embedment, the central shaft is retrieved leaving the anchor flukes vertically embedded in the seabed. The flukes constitute the load bearing element as a plate anchor. This paper presents and considers field data on the embedment depth loss due to the plate anchor keying process and the subsequent bearing capacity factor of the plate anchor element. The loss in plate anchor embedment was significantly higher than that reported from corresponding centrifuge tests and is reflected in the larger padeye displacements required to mobilize peak capacity in the field tests. Measured plate capacities and plate rotations during keying indicate that the end of keying coincides with the peak anchor capacity. Experimental bearing capacity factors are in the range Nc = 14.3–14.6, which is appreciably higher than existing solutions for vanishingly thin circular plates. The higher Nc for the DEPLA is considered to be due to a combination of the cruciform fluke arrangement and the fluke (or plate) thickness.

Large-Scale Unsupervised Hashing with Shared Structure Learning.

Hashing methods are effective in generating compact binary signatures for images and videos. This paper addresses an important open issue in the literature, i.e., how to learn compact hash codes by enhancing the complementarity among different hash functions. Most of prior studies solve this problem either by adopting time-consuming sequential learning algorithms or by generating the hash functions which are subject to some deliberately-designed constraints (e.g., enforcing hash functions orthogonal to one another). We analyze the drawbacks of past works and propose a new solution to this problem. Our idea is to decompose the feature space into a subspace shared by all hash functions and its complementary subspace. On one hand, the shared subspace, corresponding to the common structure across different hash functions, conveys most relevant information for the hashing task. Similar to data de-noising, irrelevant information is explicitly suppressed during hash function generation. On the other hand, in case that the complementary subspace also contains useful information for specific hash functions, the final form of our proposed hashing scheme is a compromise between these two kinds of subspaces. To make hash functions not only preserve the local neighborhood structure but also capture the global cluster distribution of the whole data, an objective function incorporating spectral embedding loss, binary quantization loss, and shared subspace contribution is introduced to guide the hash function learning. We propose an efficient alternating optimization method to simultaneously learn both the shared structure and the hash functions. Experimental results on three well-known benchmarks CIFAR-10, NUS-WIDE, and a-TRECVID demonstrate that our approach significantly outperforms state-of-the-art hashing methods.

Installation and capacity of dynamically embedded plate anchors as assessed through centrifuge tests

A dynamically embedded plate anchor (DEPLA) is a rocket or dart shaped anchor that comprises a removable central shaft and a set of four flukes. Similar to other dynamically installed anchors, the DEPLA penetrates to a target depth in soft seabed sediments by the kinetic energy obtained through free-fall in water and the self-weight of the anchor. In this paper DEPLA performance was assessed through a series of beam centrifuge tests conducted at 200 times earth׳s gravity. The results show that the DEPLA exhibits similar behaviour to other dynamically installed anchors during installation, with tip embedments of 1.6–2.8 times the anchor length. After anchor installation the central shaft of the DEPLA, termed a follower, is retrieved and reused for the next installation, leaving the DEPLA flukes vertically embedded in the soil. The load–displacement response during follower retrieval is of interest, with mobilisation of frictional and bearing resistance occurring at different rates. The load required to extract the DEPLA follower is typically less than three times its dry weight. The vertically embedded DEPLA flukes constitute the load bearing element as a circular or square plate. The keying and pullout response of this anchor plate is similar to other vertically embedded plate anchors, with an initial stiff response as the anchor begins to rotate, followed by a softer response as the rotation angle increases, and a final stiff response as the effective eccentricity of the padeye reduces and anchor capacity is fully mobilised. For the padeye eccentricity ratios considered (0.38–0.63 times the plate breadth or diameter), the loss in plate anchor embedment is between 0.50 and 0.66 times the corresponding plate breadth or diameter. Finally, the bearing capacity factors determined experimentally are typically in the range 14.2–15.8 and are higher than numerical solutions for flat circular and square plates. This is considered to be due to the cruciform fluke arrangement which ensures that the failure surface extends to the edge of the orthogonal flukes and mobilises more soil in the failure mechanism.

Improving Plate Anchor Design with a Keying Flap

Suction embedded plate anchors have been increasingly applied in deepwater projects because of their efficiency and limited cost. Current industry practice equips the fluke of the anchor with an outwardly rotating keying flap aiming at a reduction of the loss of embedment during keying. However, recent experimental and numerical research has questioned the efficiency of the keying flap currently used in practice because it has demonstrated that the flap does not open during keying as was intended. This paper presents a two-dimensional plain strain large deformation finite-element analysis by assuming the soil as homogenous to investigate the performance of plate anchors equipped with various keying flap designs. The results of the paper agree with previous centrifuge and finite-element analysis that the outwardly keying flap has no effect on the loss of embedment and results in a reduction of the bearing capacity postkeying. The performance of an alternative keying flap design where the flap is rotating inward is also discussed. The new design effectively reduces the loss of embedment during keying and exhibits a bearing capacity higher than the current design.

Large deformation finite element analysis investigating the performance of anchor keying flap

Suction embedded plate anchors, as used in practice, include a hinged keying flap aligned with the main anchor plate (fluke). The flap was designed to rotate away from the shank during keying of the anchor, to minimise loss of anchor embedment and the associated reduction in pull-out resistance. However, it has been observed experimentally that the flap is activated only once keying is nearly completed, and thus does not assist in reducing embedment loss. In this paper, the performance of the keying flap is investigated using a large deformation finite element approach based on frequent mesh regeneration. The moment acting at the flap hinges, determined from the soil pressures and shear forces on the keying flap, is quantified. The moment controlling the flap activation depends mainly on the net soil pressure on the front or back face of the flap, and this prevents activation of the flap while the anchor is rotating significantly. The flap is only activated once the anchor motion becomes dominated by forward translation, normal to the plate. Various factors affecting flap activation and embedment loss are explored, including flap height, eccentricity of the padeye, soil strength profile, initial anchor depth and load inclination.

A plasticity model to assess the keying of plate anchors

Suction-embedded plate anchors (SEPLAs) have been developed to answer the growing need for anchors to withstand significant vertical loading. The concept combines the advantage of suction caissons (known penetration depth and location) and ‘drag-embedded' plate anchors (efficiency and low cost). The main issue associated with SEPLAs relates to the keying process, as the anchor is first loaded, and the associated loss of embedment and reduction in capacity. The paper presents a plasticity model developed to predict the trajectory and load development during anchor keying, and up to peak load. Rigid plasticity is assumed, allowing the kinematics of the anchor to be determined from a yield surface and associated plastic potential. The trajectory and performance of a typical SEPLA are predicted using the model, and are compared with results from centrifuge tests and large-deformation finite-element analysis. The anchor loss of embedment ranged from ∼0·2 to 1·5 times the anchor height for loading inclinations between 40° and 90° from the horizontal. The model was used further to calculate the anchor loss of embedment and capacity for varying padeye offsets. Results indicated that the loss of embedment could be reduced significantly by increasing the offset, but at the detriment of the ultimate anchor capacity.

Considerations on the Design of Keying Flap of Plate Anchors

AbstractOne of the critical issues associated with plate anchor performance and design relates to the reduction of the loss of embedment during the keying process. As deep water offshore sediments typically exhibit an increasing soil strength with depth, the loss of embedment results in a reduction in anchor bearing capacity. A keying flap hinged to the main plate has been developed and adopted by industry with the aim to reduce the loss of embedment by limiting the vertical motion of the anchor. However, uncertainties remain regarding the behavior and the performance of the keying flap. This paper presents a series of numerical analyses performed to investigate the flap rotation mechanism and the condition of activation of the flap. They are compared with existing centrifuge modeling. The numerical results validate the centrifuge observations and demonstrate the nonactivation of the keying flap for typical anchor pull-out conditions.

Keying of Rectangular Plate Anchors in Normally Consolidated Clays

The loss in anchor embedment during keying, as it rotates to become normal to the cable load, reduces the uplift capacity of anchors in normally consolidated clay. The keying behavior of plate anchors has been studied previously by using centrifuge and field model tests. In this paper, a large deformation finite-element approach incorporating frequent mesh regeneration and allowing for evolution of the anchor-chain profile, was developed to simulate the keying process of rectangular and strip plate anchors. A parametric study was undertaken to quantify the loss in anchor embedment during keying in terms of the anchor geometry, soil properties, loading eccentricity, and inclination. The embedment loss decreased dramatically with increasing loading eccentricity and decreasing chain angle at the mudline to the horizontal. The loss in anchor embedment during keying increased as the local soil strength increased relative to the weight of the anchor, up to a limit determined by the eccentricity of loading. In c...

Loss in Anchor Embedment during Plate Anchor Keying in Clay

Vertically installed plate anchors have been investigated in this paper by numerical analysis and centrifuge modeling. In the numerical analysis, the large deformation finite-element method (remesh...

Three-dimensional numerical analysis of the keying of vertically installed plate anchors in clay

Plate anchors, such as suction embedded plate anchors and vertically driven plate anchors, offer economically attractive anchoring solutions for deep/ultra-deep water offshore developments. The rotation/keying processes of plate anchors will cause embedment losses, which lead to decreases of the uplift resistances of the anchors in normally consolidated soil. In the present paper, the keying processes of vertically installed strip and square plate anchors are simulated using the 3-D large deformation finite element method. The effects of loading eccentricity and pullout angle on the embedment loss during keying are investigated. Both the development of the uplift resistance and the soil flow mechanisms are presented. The numerical results show that the loading eccentricity e/ B has a much larger effect on the embedment loss than the pullout angle does. The anchor shape has a minimal effect on the loss in anchor embedment. The shape factors (square/strip) are 1.05–1.09 for loss of embedment and 1.10–1.19 for capacity.

Influence of the installation process on the performance of suction embedded plate anchors

This paper describes a series of centrifuge tests performed in order to investigate the influence of the installation process on the capacity of a suction embedded plate anchor. A 1/145th reduced scale model suction caisson and plate anchor were used for this purpose, embedded in a normally consolidated clay sample. The full installation and retrieval process was simulated as well as loading of the anchor to failure. Results are compared with similar pullout tests where the anchor was pre-embedded manually in the sample. The keying process and the anchor capacity were also investigated. Results show a loss of performance of the suction embedded anchor immediately following the retrieval of the caisson due to weakening of the clay in the vicinity of the anchor. As the clay regains strength with time, the anchor capacity increases to match those of the jacked anchors. The normalised anchor capacity following consolidation agreed well with theoretical solutions. The loss of anchor embedment during the keying process was observed to be lower for the suction embedded anchors than for the jacked anchors, but both sets of data correlated closely with the load inclination at the anchor padeye.