HAL-INSA Toulouse
Not a member yet
34325 research outputs found
Sort by
Static vs. dynamic characterization of p-GaN HEMTs: Discrepancies in electrical characteristics and their dependence on bias history
No full textInternational audienceQuasi-static electrical characteristics of p-GaN HEMTs fluctuate with bias history. This study evidences that dynamic operation is fortunately highly reproducible without pre-conditioning. The original experimental setup highlights that quasi-static data alone is insufficient for modeling dynamic behavior, while allowing precise detection of discrepancies, enabling improved transient modeling
Artificial intelligence in photovoltaic fault diagnosis: A Natural Language-Based Topic-tSNE Fusion analysis
No full textInternational audienceTimely fault detection in photovoltaic systems is critical for ensuring energy efficiency, reliability, and cost-effectiveness. However, the nonlinear and weather-dependent behavior of photovoltaic systems poses challenges for accurate diagnosis. This study presents a large-scale review of 983 scientific publications on artificial intelligence-based photovoltaic fault detection, using a novel methodology called Topic-tSNE Fusion. This approach integrates topic modeling, dimensionality reduction, and expert analysis to extract and visualize dominant research themes. Four key machine learning paradigms are identified: supervised, unsupervised, semi-supervised, and reinforcement learning. Among them, supervised methods, particularly neural networks and support vector machines, are the most frequently applied, showing accuracies above 95% in controlled conditions. The analysis also reveals growing use of semi-supervised and hybrid approaches to overcome data scarcity. Commonly monitored variables include irradiance, voltage, and current, while the most studied faults are shading, open-circuit, and degradation. Several open-access datasets supporting fault diagnosis research are catalogued. Overall, the proposed method enables a more objective and scalable review process and uncovers emerging trends, such as the shift toward lightweight artificial intelligence for edge deployment and frugal diagnostic architectures. The methodology is scalable and adaptable to other domains facing similar challenges in knowledge synthesis and system monitoring
Towards a fictitious magnetic field trap for both ground and Rydberg state 87 Rb atoms via the evanescent field of an optical nanofiber
No full textInternational audienceCold Rydberg atoms, known for their long lifetimes and strong dipole-dipole interactions that lead to the Rydberg blockade phenomenon, are among the most promising platforms for quantum simulations, quantum computation and quantum networks. However, a major limitation to the performance of Rydberg atom-based platforms is dephasing, which can be caused by atomic motion within the trap. Here, we propose a trap for 87 Rb cold atoms that confines both the electronic ground state and a Rydberg state, engineered to minimize the differential light shifts between the two states. This is achieved by combining a fictitious magnetic field induced by optical nanofiber (ONF) guided light and an external bias magnetic field. We calculate trap potentials for the cases of one- and two-guided modes with quasi-linear and quasi-circular polarizations, and calculate trap depths and trap frequencies for different values of laser power and bias fields. Moreover, we discuss the impact of the quadrupole polarisability of the Rydberg atoms on the trap potential and demonstrate how the size of a Rydberg atom influences the ponderomotive potential generated by the nanofiber-guided light field. This work expands on the idea of light-induced fictitious magnetic field traps and presents a practical approach for creating quantum networks using Rydberg atoms integrated with ONFs to generate 1D atom arrays
Latent Conditioned Loco-Manipulation Using Motion Priors
No full textInternational audienceAlthough humanoid and quadruped robots provide a wide range of capabilities, current control methods, such as Deep Reinforcement Learning, focus mainly on single skills. This approach is inefficient for solving more complicated tasks where high-level goals, physical robot limitations and desired motion style might all need to be taken into account. A more effective approach is to first train a multipurpose motion policy that acquires low-level skills through imitation, while providing latent space control over skill execution. Then, this policy can be used to efficiently solve downstream tasks. This method has already been successful for controlling characters in computer graphics. In this work, we apply the approach to humanoid and quadrupedal loco-manipulation by imitating either simple synthetic motions or kinematically retargeted dog motions. We extend the original formulation to handle constraints, ensuring deployment safety, and use a diffusion discriminator for better imitation quality. We verify our methods by performing loco-manipulation in simulation for the H1 humanoid and Solo12 quadruped, as well as deploying policies on Solo12 hardware. Videos and code are available at https://gepetto.github.io/LaCoLoco
Current fluctuations for the second class particle: joint statistics
No full textInternational audienceWe consider totally asymmetric simple exclusion process (TASEP) with a single second class particle and periodic boundary conditions. Using Bethe ansatz, we compute stationary large deviations for the joint statistics of the current of first and second class particles. At large scales, the generating function of the joint cumulants shows an unexpected connection to current fluctuations of TASEP with open boundaries
Gradient-based optimization of core-shell particles with discrete materials for directional scattering
No full textInternational audienceDesigning nanophotonic structures traditionally grapples with the complexities of discrete parameters, such as real materials, often resorting to costly global optimization methods. This paper introduces an approach that leverages generative deep learning to map discrete parameter sets into a continuous latent space, enabling direct gradient-based optimization. For scenarios with non-differentiable physics evaluation functions, a neural network is employed as a differentiable surrogate model. The efficacy of this methodology is demonstrated by optimizing the directional scattering properties of core-shell nanoparticles composed of a selection of realistic materials. We derive suggestions for core-shell geometries with strong forward scattering and minimized backscattering. Our findings reveal significant improvements in computational efficiency and performance when compared to global optimization techniques. Beyond nanophotonics design problems, this framework holds promise for broad applications across all types of inverse problems constrained by discrete variables
Positively not SOS: pseudo-moments and extreme rays in exact arithmetic
No full textA polynomial that is a sum of squares (SOS) of other polynomials is evidently positive. The converse is not true, there are positive polynomials which are not SOS. This note focuses on the problem of certifying, in exact arithmetic, that a given positive polynomial is not SOS. Using convex duality, this can be achieved by constructing a separating linear functional called a pseudo-moment certificate. We present constructive procedures to compute such certificates with rational coefficients for several famous forms (homogeneous polynomials) that are known to be positive but not SOS. Our method leverages polynomial symmetries to reduce the problem size and provides explicit integer-based formulas for generating these rational certificates. As a by-product, we can also generate extreme rays of the pseudomoment cone in exact arithmetic
Polynomial slowdown in space-inhomogeneous branching Brownian motion
No full textWe consider a branching Brownian motion in in which particles independently diffuse as standard Brownian motions and branch at an inhomogeneous rate which depends only on the angle of the particle. We assume that is maximal when , which is the preferred direction for breeding. Furthermore we assume that b(\theta) = 1 - \beta \abs{\theta}^\alpha + O(\theta^2), as , for and \beta>0. We show that if is the maximum distance to the origin at time , then is tight whereand is explicit in terms of the first eigenvalue of a certain operator
Understanding the Impact of Value Selection Heuristics in Scheduling Problems
No full textInternational audienceIt has been observed that value selection heuristics have less impact than other heuristic choices when solving hard combinatorial optimization (CO) problems. It is often thought that this is because more time is spent on unsatisfiable sub-problems where the value ordering is irrelevant. In this paper we investigate this belief in the scheduling domain and come up with a more detailed explanation. We find that, even though there are less relevant choices to be made on hard instances, each mistake tends to have a bigger impact, to a point where the potential gain from a value heuristic predominates. Moreover, we observe two interesting and relatively surprising phenomena when solving scheduling problems. First, the accuracy of a given value selection heuristic decreases with the optimality gap. Second, the computational penalty of a mistake increases with the accuracy of the heuristic. For the first observation, we argue that on hard problems, constraint propagation removes a large portion of choices that align with the intuition behind the heuristic. This means that the heuristic faces mostly difficult choices. For the second observation, we argue that simple heuristics tend to make more mistakes on intuitive choice points, and the computational cost for refuting these mistakes is smaller than for those made by a more accurate heuristic
Parametric experimental-numerical investigation of the impact behavior of Carbon/Epoxy composites
No full textInternational audienceThis article provides test/model dialogue of impact on composite laminates at low and medium impact velocity. An experimental setup featuring a compressed air cannon launches projectiles at composite laminates, while high-speed imaging, load measurement, and infrared thermography enable in-situ damage monitoring. Numerical simulations employ the Discrete Ply Model (DPM) to analyze the impact events and damage evolution. The study examines various impact parameters, such as projectile energy, velocity, and impact location, to assess their effects on structural response and damage mechanisms. Experimental results highlight the effectiveness of these techniques in capturing the dynamic behavior and quantifying damage from impacts. Numerical simulations offer insights into damage mechanisms, including delamination, matrix cracking, and fiber breakage, as well as their interactions. The findings elucidate damage formation during single impact and emphasize the significance of spatial impact distribution and local bending stiffness in understanding damage patterns. Validation of the numerical model against experimental results confirms its predictive capabilities, showcasing its potential for virtual testing and parametric studies. By integrating the strengths of both approaches, this article enhances understanding of the single impact behavior of composite structures, providing a valuable experimental and numerical database for researchers to test and validate their models, ultimately paving the way for improved design and performance in engineering applications