Yazar "Mowla, Md Najmul" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • [ X ]
    Öğe
    Aerodynamic coefficient prediction of bio-inspired camber morphing wings with flexible surfaces using an explainable transformer
    (Elsevier France-Editions Scientifiques Medicales Elsevier, 2026) Mowla, Md Najmul; Durhasan, Tahir; Asadi, Davood; Kesilmi, Zehan; Jafari, Javad Rashid
    Bio-inspired morphing wings with flexible surfaces can enhance aerodynamic efficiency at low Reynolds numbers (Re), yet predicting their fluid-structure interaction remains challenging. We present PhysAero-MHANet, a physics-aware, interpretable deep learning framework coupled with controlled wind tunnel experiments for aerodynamic prediction of camber-morphing finite wings. The campaign yielded 911 samples spanning Re is an element of [3 x 10(4), 1 x 10(5)], camber deflections up to 10(degrees), and angles of attack from-18(degrees )to 18(degrees). Experiments showed up to 34% drag reduction at small angles of attack, a stall delay of approximate to 6(degrees), a maximum lift coefficient C-L,C-max approximate to 1.44, and a peak lift-to-drag ratio C-L/C-D approximate to 8.84. The proposed model is a transformer-based multi-task surrogate with physics-informed attention, hierarchical cross-feature fusion, and shapley additive explanations (SHAP) for interpretability. Against 11 machine-learning, deep-learning, and attention baselines, PhysAero-MHANet achieved R-2 approximate to 0.985 and MAPE < 12% across lift (C-L), drag (C-D), and rolling moment (C-M,C-R) predictions. These results provide new insight into morphing-wing aerodynamics and support real-time control, performance optimization, and integration into unmanned aerial vehicles (UAVs) and micro aerial vehicles (MAVs).
  • [ X ]
    Öğe
    Safe multi-agent uav flight planning for 6G-enabled internet of things (IoT) networks using deep reinforcement learning
    (Adana Alparslan Türkeş Bilim ve Teknoloji Üniversitesi, 2026) Mowla, Md Najmul; Asadi, Davood
    This thesis investigates deployment-oriented autonomy for unmanned aerial vehicles (UAVs) in 6G-enabled Green Internet of Things (IoT) environments. Mobility decisions must jointly address collision avoidance, kinematic executability, energy sustainability, and connectivity under dynamic and partially observable conditions. The tasks are formulated as Markov decision processes and solved using four deep reinforcement learning frameworks: (1) proximal policy optimization (PPO) with kinematic optimization (KinOpt) for smooth, curvature-bounded, flight-feasible trajectories; (2) decentralized multi-agent PPO for smart agriculture with moving hazards and Simultaneous Wireless Information and Power Transfer (SWIPT)-inspired replenishment; (3) Reconfigurable Intelligent Surface (RIS)-supported multi-agent soft actor–critic (MASAC) jointly optimizing mobility, relay/recharge behavior, and connectivity; and (4) an Ensemble Distributional Dueling Double Deep Q-Network (ED3QN) with risk-aware action selection and a safety shield, evaluated in 2D benchmarks and a 3D Light Detection and Ranging (LiDAR) setting. The proposed multi-agent PPO achieves 100% success with an average reward of 1026.33 (baseline: 710.00) and a computation time of 34.84 ms. In RIS-assisted networking, MASAC attains 1.00 ± 0.00 success, 341.67 ± 8.32 final battery, 80.00 ± 4.21 harvested energy, 0.6291 ± 0.013 connectivity ratio, and 650.33 ± 11.6 total reward, outperforming MADDPG (p < 0.05). ED3QN achieves 100% success with zero collisions and path-efficiency 1.010–1.067, while PPO+KinOpt reduces trajectory length from 54.000 m to 34.463 m and smoothness cost from 87.967 to 2.107 rad. Overall, explicit feasibility and sustainability modeling yield more deployable UAV behavior for 6G aerial networking and energy-constrained IoT missions.