154092 research outputs found
Sort by
Discontinuous Galerkin Finite Element Methods for Linear Port-Hamiltonian Dynamical Systems
No full textIn this paper, we present discontinuous Galerkin (DG) finite element discretizations for a class of linear hyperbolic port-Hamiltonian dynamical systems. The key point in constructing a port-Hamiltonian system is a Stokes-Dirac structure. Instead of following the traditional approach of defining the strong form of the Dirac structure, we define a Dirac structure in weak form, specifically in the input-state-output form. This is implemented within broken Sobolev spaces on a tessellation with polyhedral elements. After that, we state the weak port-Hamiltonian formulation and prove that it relates to a Poisson bracket. In our work, a crucial aspect of constructing the above-mentioned Dirac structure is that we provide a conservative relation between the boundary ports. Next, we state DG discretizations of the port-Hamiltonian system by using the weak form of the Dirac structure and broken polynomial spaces of differential forms, and we provide a priori error estimates for the structure-preserving port-Hamiltonian discontinuous Galerkin (PHDG) discretizations. The accuracy and capability of the methods developed in this paper are demonstrated by presenting several numerical experiments.</p
Author Correction: Explaining the adaptation gap through consistency in adaptation planning
No full textA reproducible framework for synthetic data generation and instance segmentation in robotic suturing
Get PDFPurpose: Automating suturing in robotic-assisted surgery offers significant benefits including enhanced precision, reduced operative time, and alleviated surgeon fatigue. Achieving this requires robust computer vision (CV) models. Still, their development is hindered by the scarcity of task-specific datasets and the complexity of acquiring and annotating real surgical data. This work addresses these challenges using a sim-to-real approach to create synthetic datasets and a data-driven methodology for model training and evaluation. Methods: Existing 3D models of Da Vinci tools were modified and new models–needle and tissue cuts–were created to account for diverse data scenarios, enabling the generation of three synthetic datasets with increasing realism using Unity and the Perception package. These datasets were then employed to train several YOLOv8-m models for object detection to evaluate the generalizability of synthetic-trained models in real scenarios and the impact of dataset realism on model performance. Additionally, a real-time instance segmentation model was developed through a hybrid training strategy combining synthetic and a minimal set of real images. Results: Synthetic-trained models showed improved performance on real test sets as training dataset realism increased, but realism levels remained insufficient for complete generalization. Instead, the hybrid approach significantly increased performance in real scenarios. Indeed, the hybrid instance segmentation model exhibited real-time capabilities and robust accuracy, achieving the best Dice coefficient (0.92) with minimal dependence on real training data (30–50 images). Conclusions: This study demonstrates the potential of sim-to-real synthetic datasets to advance robotic suturing automation through a simple and reproducible framework. By sharing 3D models, Unity environments and annotated datasets, this work provides resources for creating additional images, expanding datasets, and enabling fine-tuning or semi-supervised learning. By facilitating further exploration, this work lays a foundation for advancing suturing automation and addressing task-specific dataset scarcity.</p
A simulation environment for robot-assisted endovascular interventions
Get PDFPurpose: Cardiovascular diseases are the leading cause of mortality globally. Advances in interventional radiology and endovascular devices have made endovascular procedures effective alternatives to traditional open surgery, leading to their routine application in clinical practice. Within this framework, novel technologies, including robotic platforms and navigation software, have been developed to assist clinicians in executing endovascular interventions with improved dexterity, enhanced guidance, and superior clinical training, ultimately yielding better patient outcomes. Methods: This study aims to develop a model-based simulation environment within the SOFA framework, to enable shape and force sensing for endovascular robotic procedures. The vascular catheter was modeled using beam theory, and realistic interactions between the catheter and vascular models were established using the finite element method (FEM) with both linear elastic and nonlinear hyper-elastic models. Experiments measured contact forces and positional changes during catheter insertion, comparing anatomical deformations with simulation results. Results: Experimental tests validated the simulated force and displacement measurements. The catheter contact force showed an absolute error of 0.0371 N (30.45%). Catheter tip displacement averaged 3.1 mm, and the proximal segment’s Fréchet distance averaged 3.6 mm. For the anatomical model, the elastic FEM model performed best, with deformation measurement errors of 34%, 19%, and 59% across three different force scenarios. Conclusion: The results indicate that the integration of advanced physical modeling, realistic human–robot interactions, and enhanced computational capabilities will facilitate the development of innovative solutions, enabling clinicians to achieve greater accuracy and reliability in minimally invasive surgical (MIS) applications, particularly in endovascular interventions.</p
Deep learning segmentation approach for forest height retrieval with multichannel SAR
No full textDeep learning has emerged as a transformative approach in big data analysis, gaining traction across various domains, including remote sensing, due to advancements in data availability and computational power. Neural networks, with their ability to hierarchically abstract complex patterns, serve as robust nonlinear models capable of representing intricate mathematical relationships [1]. Consequently, deep learning (DL) methods have become integral to remote sensing tasks [2], particularly Synthetic Aperture Radar (SAR) image processing [3]. Researchers have further explored DL’s potential to advance 3D Tomographic SAR (TomoSAR) reconstruction techniques. Early DL-based methodologies for TomoSAR applications, especially in urban settings [4], approached scatterer position estimation as a classification problem. In this paradigm, discrete elevation positions within a scene corresponded to classification labels, enabling scatterer location determination in range-azimuth cells. In polarimetric SAR tomography, the PolGAN method [5] reframed forest height estimation as an image pan-sharpening problem. By integrating PolInSAR and LiDAR inputs with adversarial feedback from PolGAN discriminators, this method achieved high-resolution forest height estimates with enhanced vertical accuracy. Inspired by these advancements, recent efforts have shifted toward leveraging DL for more robust height reconstruction. For instance, a pixel-wise Tomographic SAR Neural Network (TSNN) [6] demonstrated the adaptability of DL in height estimation tasks. Building on TSNN’s foundational concepts, this study introduces a more flexible and spatially adaptive framework: multiChannel SAR foresT height reconStruction Neural nETwork (CATSNET). Designed to process multi-channel data, CATSNET accommodates both multi-baseline (MB) and multi-polarization (MP) inputs. Unlike conventional pixel-wise methods that rely solely on local correlations between heights and covariance matrices, CATSNET employs a patch-based convolutional neural network (CNN) model. This enables the integration of spatial context and neighboring pixel information, significantly improving height estimation stability and reducing susceptibility to noise. CATSNET is tailored to address two primary objectives: 1. Leveraging patch-based features to enhance robustness over pixel-wise methods. 2. Ensuring generalization to diverse geographic regions. The inclusion of neighboring spatial information allows the model to better capture the inherent spatial correlation in ground and canopy height profiles, improving the accuracy and adaptability of height estimation across various regions. To achieve this, training patches are annotated with quantized LiDAR-derived heights, such as canopy height models (CHM) and digital terrain models (DTM), reframing the height estimation problem as a segmentation task. The CATSNET architecture is based on a U-Net structure, with a contracting encoder path to extract hierarchical features and an expansive decoder path to restore spatial resolution. The final output layer employs a 1×1 convolution to map 32-dimensional feature vectors into class predictions representing forest or ground heights. The proposed model was evaluated using data from the Paracou site, surveyed during the TropiSAR campaign over French Guiana’s forests in 2009 [7]. Small-footprint LiDAR data was used to construct the training dataset by extracting polarimetric and interferometric elements from the covariance matrix for each range-azimuth pixel in the multi-channel SAR stack. Experimental results demonstrated that CATSNET outperformed existing methods in height reconstruction, closely aligning with LiDAR reference data while minimizing outliers. Its patch-based approach proved particularly effective, achieving superior accuracy in estimating both forest and ground heights. These findings establish CATSNET as a reliable, adaptable, and generalizable framework for SAR-based height reconstruction tasks. To date, experiments have been conducted using airborne SAR data, as spaceborne SAR systems typically have long repeat acquisition intervals, which can result in significant forest decorrelation. However, upcoming missions such as NASA’s NISAR, ESA’s Biomass, and DLR’s Tandem-L are expected to provide low-frequency data and improved acquisition timelines, mitigating temporal decorrelation issues. The proposed method presents a promising approach for retrieving forest information with high accuracy and efficiency, making it highly suitable for applications on a global scale
A Functionalized Pyrrole Coating to Improve Fiber–Matrix Adhesion in Carbon Fiber Reinforced Polyphenylene Sulfide Composites
Get PDFThe bonding between fiber and matrix has a substantial impact on the mechanical performance of fiber reinforced polymer composites. This paper presents a newly developed pyrrole-based coating to improve the fiber–matrix adhesion within a carbon fiber (CF) polyphenylene sulfide (PPS) composite. Pyrrole is known to bond to carbon, while bonding to the polymer can be achieved by functionalizing the derivative. Two different pyrrole derivatives were tested, namely one containing a functional group toward the matrix and one pyrrole derivative lacking this functional group. The results show that this functional group on the pyrrole-based coating improves the fiber–matrix adhesion significantly, resulting in an increase in short beam strength (SBS) and in-plane shear (IPS) strength of approximately 25% for continuous CF PPS laminates.</p
Editorial:Forward and inverse solvers in multi-modal electric and magnetic brain imaging: theory, implementation, and application
Get PDFExploring the Feasibility of Friction Screw Extrusion Additive Manufacturing for AA7075 Aluminum Alloy
Get PDFThis study explores the feasibility of Friction Screw Extrusion Additive Manufacturing (FSEAM) for fabricating an AA7075 alloy build. A 54-layer build was manufactured on an AA7075 substrate at a controlled print velocity of 50 mm/min. Macroscopic analysis displayed consistent quality with minimal defects, while microstructural analysis revealed a fine, equiaxed microstructure with crack-like defects between some layers, indicating insufficient temperature and nozzle force during the process. Vickers microhardness testing showed significant hardness reduction compared to original feedstock, attributed to frictional heating during FSEAM. The results underscore the challenges of maintaining mechanical properties and temperature control when processing high-strength alloys like AA7075
Tracing Ion Migration in Halide Perovskites with Machine Learned Force Fields
Get PDFHalide perovskite optoelectronic devices suffer from chemical degradation and current–voltage hysteresis induced by migration of highly mobile charged defects. Atomic scale molecular dynamics simulations can capture the motion of these ionic defects, but classical force fields are too inflexible to describe their dynamical charge states. Using CsPbI3 as a case study, we train machine learned force fields from density functional theory calculations and study the diffusion of charged halide interstitial and vacancy defects in bulk CsPbI3. We find that negative iodide interstitials and positive iodide vacancies, the most stable charge states for their respective defect type, migrate at similar rates at room temperature. Neutral interstitials are faster, but neutral vacancies are 1 order of magnitude slower. Oppositely charged interstitials and vacancies, as they can occur in device operation or reverse bias conditions, are significantly slower and can be considered relatively immobile
Versatile Tasks on Integrated Aerial Platforms Using Only Onboard Sensors:Control, Estimation, and Validation
No full textConnecting multiple aerial vehicles to a rigid central platform through passive spherical joints holds the potential to construct a fully actuated aerial platform. The integration of multiple vehicles enhances efficiency in tasks like mapping and object reconnaissance. This article proposes a control and state estimation framework for the integrated aerial platform (IAP), enabling it to perform versatile tasks like object reconnaissance and physical interactive tasks with only onboard sensors. In the framework, the 6-D motion control serves as the low-level controller, while the high-level controller comprises a 6-D admittance filter and a perception-aware attitude correction module. The 6-D admittance filter, serving as the interaction controller, is adaptable for aerial interaction tasks. The perception-aware attitude correction algorithm is carefully designed by adopting a geometric model predictive controller (MPC). This algorithm, incorporating both offline and online calculations, proves to be well-suited for the intricate dynamics of an IAP. A 6-D direct wrench controller is also developed for the IAP. Notably, both the interaction controller and the direct wrench controller operate without reliance on force/torque sensors. Instead, a wrench observer algorithm is devised, considering external disturbances. In addition, based on the kinematics constraints of the multiple aerials in the platform, a fusion algorithm for multiple visual-inertial odometry and kinematics constraints is developed, providing more accurate localization. A prototype of the IAP is constructed, and its capabilities are demonstrated through experiments including perception-aware object reconnaissance, aerial mapping, aerial peg-in-hole task, and 6-D contact wrench generation. All experiments are conducted exclusively with onboard sensors. These tasks exemplify the merits of the proposed IAP and validate the effectiveness of the proposed control framework and fusion algorithm.</p