HAL Portal IMT Nord Europe
Not a member yet
6687 research outputs found
Sort by
Influence of using wood biomass fly ash as filler on the rheological and physical properties of cementitious pastes
No full textInternational audienc
Coupled Modeling of Hydromechanical Behavior of Saturated Concrete Under High Stresses and Dynamic Loading: Influence of Pore Pressure
No full textInternational audienceABSTRACT This paper investigates the dynamic response of water‐saturated concrete under high stress levels, with a particular emphasis on the role of pore pressure. An enhanced elastoplastic damage model, incorporating dual plastic mechanisms, is proposed to capture the coupled hydromechanical behavior of concrete under combined high stress and high strain rate loading. Key improvements include the refinement of the porosity‐volumetric strain relationship, the incorporation of full hydromechanical coupling under dynamic loading, and the integration of strain rate sensitivity into the pore collapse mechanism and material strength. The improved constitutive model and numerical methodology are validated through simulations of uniaxial tensile tests and three sets of compression tests. Parametric studies are conducted to explore the influence of pore pressure on the confined response of concrete under both static and dynamic loading conditions. The results demonstrate that interstitial pore pressure significantly affects both the volumetric and deviatoric behaviors of saturated concrete, with its influence becoming more pronounced under dynamic loading. The findings provide valuable insights into the hydromechanical behavior of concrete structures subjected to extreme loading scenarios
Estimation of the impact of soil water saturation on the significance of erosion events (Hauts-de-France),
No full textInternational audienc
In Situ Monitoring of the Aging of Mineral Particles by Methylglyoxal Using Infrared Spectroscopy. Part I: Unraveling Uptake under Dry Conditions
No full textInternational audienc
Ozonolysis of prenol, a second-generation biofuel, in atmospheric simulation chambers: Temperature dependent kinetics and gas-phase products analysis
No full textInternational audienc
Réduction d’erreur de stéganographie générative par ajustement dans l’espace latent
No full textInternational audienceBy embedding directly in the latent space, generative steganography is a promising way to embed large payloads in an undetectable way. However, the embedding needs to be robust to the error caused by both the compression of the image in the pixel format and the inversion in the latent space. In this paper, we show how to design a potentially error free steganographic scheme by modifying an existing scheme based on high capacity binary spread spectrum [5]. By approximating the generation and inversion channels using a shift hypothesis, we perform error correction by first identifying erroneous components and then altering them in the latent space. In order to minimize detectability w.r.t. steganalysis both in the latent or image spaces, we also refine the norm of the corrected stego latent vector. Results highlight that 37% of PNG images become error free instead 0% without correction. When preserving the norm, we also show that the practical security w.r.t. steganalysis is still important
Ai-driven decision support for spatio-temporal water quality monitoring using aquatic drones
No full textInternational audienc
Développement de systèmes Géothermiques sur stations de Relevage d’eau
No full textInternational audienc
DeepFake Detection based on Noise Residuals
No full textInternational audienceDeepfakes pose major challenges, especially regarding fraud, misinformation, and evidence tampering. As deepfakes become more prevalent, effective detection methods are essential. We introduce DJIN [5], a deepfake detection model that retains noise components by avoiding pooling layers in the initial stages. Pre-trained on ImageNet for steganography using the JIN version, DJIN outperforms CoDE [2] and CLIP [7] and is the best among all detectors mentioned for the In-Distribution dataset. DJIN is highly effective in handling high-quality images and processing images of various sizes. Since deepfake generators typically produce high-quality outputs, an explainability analysis reveals that DJIN leverages image noise by focusing on darker areas in real images and brighter areas in generated one