Direction Générale de l'Armement

Because of their interaction with the surrounding plasma, spacecraft in operation experience high absolute and differential charging between surfaces made of different materials. Differential charging may ultimately lead to the formation of an electrostatic discharge (ESD) that equalizes the potential of the satellite and the plasma environment and, when triggered on a solar array, generates a plasma bubble that neutralizes, partially or totally, the cover glasses. Charging simulations are commonly used to evaluate the risk of ESD formation on satellites. However, as simulations do not consider the effect of ESDs, computed differential potentials on satellite can reach thousands of volts, while discharges are known to occur at lower values (typically around 500 V). This article presents the coupling between a charging model and a flashover propagation model, applied on a satellite with an 8- m(2) solar array. Simulations are presented for several charge-discharge cycles, where the flashover almost completely neutralizes the whole surface of the solar array. A parametric study is carried out to investigate the effect of the ESD ignition voltage, the environment, and the cathode spot material on the ESD occurrence frequency, the ESD duration, and the flashover current amplitude. This type of simulation can be used to further understand parameters strengthening the ESD risks.

When a hypersonic vehicle travels through a planetary atmosphere, it can face a severe degradation of its wireless communication systems due to the creation of a plasma around it, a problem commonly known as the "communication blackout." One solution proposed in the literature is to apply an external static magnetic field within the reentry plasma layer to create a "magnetic window" that allows the propagation of electromagnetic waves. However, this static magnetic field leads to complex phenomena within the plasma, such as nonpurely transverse waves, mode conversions, and resonances. In order to better understand the physics involved, we propose an analytical model that takes into account the inhomogeneity of the electron density and collision frequency in the plasma layer, the nonuniform static magnetic field produced by permanent magnets, and the oblique incidence of the electromagnetic waves. This model demonstrates a transmission enhancement of 85 dB at an altitude of 50 km for the Radio Attenuation Measurement C vehicle, considering a right-hand circular polarized electromagnetic wave at 1575.42 MHz, while more than half of the incident power is transmitted through the plasma slab over a field of view of 74 degrees, still preserving the polarization purity of the incident EM wave.

We develop a Wigner-based phase-space framework for mean paraxial wave propagation in random media. Starting from the random parabolic wave equation, we derive the exact evolution of the realization-dependent Wigner distribution and identify the ensemble-averaged Wigner function as the natural second-order state variable. The averaged equation contains a closure defect, given by a mixed field–medium correlation, which prevents a closed transport equation from being obtained without additional assumptions. We therefore organize the modelling as a hierarchy from the random wave equation to an exact Wigner formulation, then to a nonlocal kinetic closure, and finally to a local Fokker–Planck reduction in the small-angle regime. For the minimal homogeneous isotropic Fokker–Planck model, we derive closed evolution laws for the quadratic moments, exhibit the cubic-in-distance contribution to beam spreading, and obtain explicit Gaussian and Gauss–Schell propagation formulas. These analytical results are used to validate a phase-space splitting solver in one-dimensional transverse benchmarks. Comparisons with nonlocal kinetic models show that the diffusive approximation is accurate for narrow momentum-transfer kernels and loses validity in a controlled way as finite-jump effects become significant. Finally, we introduce a first atmospheric specialization based on a regularized turbulence spectrum, yielding an effective diffusion coefficient expressed in terms of standard atmospheric parameters.

This paper proposes a framework for constructing interval-valued coverage probability maps under stochastic vehicle dynamics and geometric set-evaluation indeterminacy. Classical Monte Carlo coverage estimation relies on binary inclusion tests that assume exact geometric evaluation: each spatial cell is declared either covered or uncovered for every trajectory realization. In practice, however, covered regions are computed numerically and may only be available through finite-resolution set approximations, leading to partial coverage or unresolved inclusion cases. To address this issue, we introduce a three-valued inclusion test that distinguishes certified full coverage, certified non-coverage, and indeterminate outcomes. This logic is embedded into an interval-valued Monte Carlo estimator by defining lower and upper interpretations of indeterminate cases. The resulting empirical interval-valued estimator converges almost surely, as the number of samples increases, to a deterministic probability interval enclosing the true cell coverage probability. For each trajectory realization, the covered region is evaluated using certified set-based computations based on separators and a branch-and-contract paving algorithm, which provide inner and outer approximations required for the three-valued cell evaluation. The resulting interval estimates are assembled into coverage probability maps over spatial cells, and an adaptive probability-driven refinement strategy refines the final map according to an interval-width criterion. Numerical simulations with a stochastic Dubins-type underwater vehicle illustrate the approach and highlight the trade-off between geometric resolution, computational effort, and interval-valued uncertainty representation.

We compare lightweight automata-based models (n-grams) with neural architectures (LSTM, Transformer) for next-activity prediction in streaming event logs. Experiments on synthetic patterns and five real-world process mining datasets show that n-grams with appropriate context windows achieve comparable accuracy to neural models while requiring substantially fewer resources. Unlike windowed neural architectures, which show unstable performance patterns, n-grams provide stable and consistent accuracy. While we demonstrate that classical ensemble methods like voting improve n-gram performance, they require running many agents in parallel during inference, increasing memory consumption and latency. We propose an ensemble method, the promotion algorithm, that dynamically selects between two active models during inference, reducing overhead compared to classical voting schemes. On real-world datasets, these ensembles match or exceed the accuracy of non-windowed neural models with lower computational cost.