Engineering

Abstract: This study analyses the aerodynamic and acoustic trade-offs of toroidal quadcopter propellers using variants derived from the APC 10×5 design to guide mission-specific adoption in next-generation quadcopters. Three configurations, namely a two-loop elliptical, a two-loop circular, and a three-loop circular design, were evaluated through CFD simulations. Static thrust tests on a thrust stand validated the CFD model, complementing wind-tunnel validation against published APC 10×5 data. Acoustic assessments used the Ffowcs Williams–Hawkings (FW-H) analogy. The results show that toroidal propellers achieve 38.8 to 59.2 % higher static thrust and significant noise reductions relative to the conventional benchmark, although efficiency is compromised by increased induced drag from vortex diffusion. Specifically, the three-loop circular variant reduced the peak tonal sound pressure level by approximately 26.8 dB (a factor of ≈22 in sound pressure) within the human-sensitive range, yet it suffered a ≈41 % lower thrust-to-weight ratio. The two-loop elliptical variant offered a practical balance with 38.8 % higher static thrust, a peak-tone reduction of ≈2.7 dB, and a maximum propulsive efficiency of 0.57 (vs. 0.69 for the benchmark). Propeller choice therefore dictates mission suitability: three-loop designs best serve noise-sensitive tasks such as surveillance, while elliptical designs better serve infrastructure inspection missions requiring a balance of efficiency and thrust.

Abstract: Small unmanned aerial vehicle propellers must generate sufficient thrust while meeting strict noise expectations in civil and industrial drone operations. However, open aerodynamic and geometric data for widely used low-noise propellers remain limited, restricting validation and comparative design studies. This work establishes a public benchmark for the DJI 9455S low-noise propeller and investigates a biomimetic variant inspired by humpback whale tubercles. The commercial propeller was reverse-engineered using high-resolution Faro ScanArm measurements at 30 micrometre resolution to extract its taper and twist distributions. A three-dimensional printed PLA replica and a modified propeller with alternating S1223 high-lift and SD7032 low-drag airfoil sections were manufactured and evaluated using static thrust testing, validated Reynolds-averaged Navier-Stokes simulations, forward-flight calculations and Ffowcs Williams-Hawkings aeroacoustic prediction. The printed replica produced 9-11% less thrust than the commercial propeller above 4000 rpm, highlighting the influence of fused deposition modelling surface roughness, stiffness and geometric deviation. After validation, transient simulations showed that the biomimetic variant increased static thrust by approximately 2%, maintained measurable thrust at advance ratios of 0.4-0.6 and reduced the predicted tonal sound pressure level at the second and third blade-passage harmonics, while leaving the fundamental tone broadly unchanged. These results indicate that full-chord spanwise airfoil alternation can improve UAV propeller tonal acoustic performance while providing reusable benchmark data for future drone propulsion studies.

Abstract: In the present study, three-dimensional steady-state Reynolds-Averaged Navier–Stokes (RANS) simulations were performed to investigate the influence of radial skew angle in axial slot casing treatments (CTs) applied to the first rotor of a transonic three-and-a-half-stage axial compressor. The investigation focuses on three skewed slot configurations with radial skew angles of 35°, 45°, and 60°, while the axially aligned slot configuration (0° skew angle), previously reported and validated, is adopted solely as a reference baseline. All configurations share identical geometric proportions, enabling the isolated assessment of skew-angle effects on compressor aerodynamic behavior. The results demonstrate that the radial skew angle significantly influences the balance between stall margin improvement and efficiency variation, with the magnitude of these effects strongly dependent on operating speed. Under off-design operating conditions, skewed axial slots provide substantial stall margin enhancement relative to the smooth casing configuration, with larger skew angles generally yielding greater improvements in compressor stability. Conversely, at the design speed, increasing skew angle is associated with a progressive efficiency penalty, highlighting the trade-off between aerodynamic stability enhancement and additional aerodynamic losses. Among the skewed configurations investigated, the 45° radial skew angle provides the most favorable compromise, delivering consistent stall margin improvement across the examined operating range while preserving compressor efficiency under nominal operating conditions. Overall, the present study demonstrates that radial skew angle constitutes a critical design parameter for axial slot casing treatments in multistage compressors and should therefore be carefully tailored to the intended operating regime. The findings extend the current understanding of axial-slot flow-control mechanisms and provide practical guidance for the aerodynamic design of casing treatments in transonic axial compressors.

Abstract: Unmanned Aerial Vehicles (UAVs) have transitioned from localised applications to complex, long-range missions across civil, commercial, and defence domains. Central to this expansion is the capacity to operate Beyond Visual Line of Sight (BVLOS), an operational paradigm entirely reliant on resilient command and control (C2) data links. This perspective traces the architectural evolution of UAV communication systems, tracking the shift from legacy line-of-sight setups to modern Space-Air-Ground Integrated Networks (SAGIN) and 6G Non-Terrestrial Networks (NTNs). While Low Earth Orbit (LEO) satellite constellations resolve the coverage limits of terrestrial infrastructure, they introduce severe physical bottlenecks: dynamic propagation delays and handover-induced stochastic jitter. This paper evaluates how the field has shifted from basic connectivity validation to latency-aware control systems, highlighting multi-layered, risk-aware autonomous safety architectures that dynamically reconcile network volatility with vehicle safety margins.

Abstract: The Seagull Optimization Algorithm (SOA) [1 ]has been successfully applied in various domains, including large-scale industrial engineering optimization, engineering design, UAV (Unmanned Aerial Vehicle) path planning, and wireless sensor network node localization, owing to its advantages such as few parameters and a clear structure. However, when applied to UAV path planning, the original SOA still exhibits several critical limitations: insufficient diversity of the initial population, the lack of a predictive mechanism to guide the search toward the optimal solution, and the tendency to suffer from premature convergence caused by being trapped in local optima during the optimization process. To address these issues, this paper proposes a modified Seagull Optimization Algorithm with Latin Hypercube Sampling and Levy Flight (LLSOA). First, the Latin Hypercube Sampling (LHS) method is adopted to enhance the generation of the initial population, thereby improving the coverage and diversity of the sample space. Second, a Levy Flight perturbation mechanism is introduced into the position updating process to enhance the algorithm’s ability to escape from local optima. To validate the effectiveness of the proposed algorithm, simulation experiments are conducted on UAV path planning in complex three-dimensional environments. The performance of the improved SOA is compared with that of the original SOA, the Dung Beetle Optimization algorithm (DBO) [2], the Grey Wolf Optimization algorithm (GWO) [3], the Pigeon-Inspired Optimization algorithm (PIO) [4], and the Particle Swarm Optimization algorithm (PSO) [5]. The results demonstrate that the proposed improved SOA achieves significant advantages in terms of path quality and computational efficiency. Furthermore, experiments with varying waypoint configurations in different scenarios confirm the strong generalization ability and stability of the proposed method under diverse mission conditions. Finally, potential future application scenarios of this method in the field of UAV path planning are discussed.

Abstract: A graph attention network-enhanced multi-agent proximal policy optimization (GAT-MAPPO) framework is proposed for cooperative guidance in counter-attack/defense scenarios. A dynamic heterogeneous interaction graph is formulated over interceptors and targets at every decision epoch. Through a multi-head graph attention encoder, relational features capturing both inter-interceptor cooperation and target threat dynamics are adaptively aggregated. These graph-enriched observations are processed by a Centralized-Training, Decentralized-Execution (CTDE) MAPPO architecture, guided by a hierarchical reward function that mandates miss distance minimization, simultaneity of arrival consensus, multi-directional encirclement, and smooth control effort. Furthermore, the integration of a three-stage curriculum learning strategy allows for robust cooperative policy derivation across transitions from rectilinear to highly adaptive evasion patterns, eliminating the need for explicit rule engineering. Extensive Monte Carlo simulations confirm GAT-MAPPO’s superior performance: achieving >95% interception success rate in 4-vs-4 scenarios and reducing mean simultaneity error by 41.4% compared to the MAPPO baseline. Comprehensive ablation studies validate the critical roles played by graph attention encoding, reward hierarchy design, and progressive curriculum staging.

Abstract: Metallic wing covers are defined here as wing skins plus mechanically attached or integrally machined stringers; an integrally stiffened panel is the monolithic case in which the skin and stiffeners are machined from one plate. This structured narrative review examines upper and lower metallic wing covers as manufacturing objects in civil transport, business-aircraft, and selected military fixed-wing programmes. Aircraft-level evidence is concentrated from 1950 onward, with earlier peen-forming origins included only where they explain later industrial adoption. The evidence base combines peer-reviewed papers, SAE Technical Papers, patents, trade literature, government reports and supplier disclosures, so the review uses explicit source weighting rather than statistical aggregation. Evidence is graded by source strength, and patents are treated as capability evidence rather than proof of production use unless independently corroborated. The synthesis shows that route selection is governed by structural scale, cover role, curvature class, alloy and temper, inherited stock state, panel architecture, and compensation or validation capability. The upper/lower cover divergence that emerges is directly evidenced for selected Airbus, Gulfstream and B-1B cases and is treated, for Boeing and other lineages where cover-separated routes are not publicly disclosed, as a mechanistically supported inference rather than a demonstrated production rule. The strongest public evidence supports peen and particle-impact routes for selected directional or inflected lower-cover cases, Creep Age Forming (CAF) for large smooth heat-treatable covers, and specialised laser or hybrid routes where corroborated by supplier, patent, or named-programme evidence. Modern, non-Western, business-jet, and many military programmes are frequently supported only by patent, supplier, lineage, or contextual evidence and are therefore interpreted cautiously.

Abstract: Fiber-reinforced polysiloxane composites (FRPCs) with glass, silica, carbon, graphite, car-bon/polybenzimidazole, alumina, and quartz fibers in different architectures, infiltrated with a high char yield polysiloxane resin, were developed for hypersonic applications. Techneglas manufactures the polysiloxane resin. FRPCs were manufactured by the Koo Research Group. Material characterization of thermal stability, flammability, ablation, thermophysical, and mechanical properties of these high-performance FRPCs was per-formed. Thermal stability properties were characterized using thermogravimetric analysis, and flammability properties using microscale combustion calorimetry for the FRPCs. Ablation properties using oxy-acetylene test bed with advanced diagnostics were performed at sev-eral heat flux test conditions, exposure times for recession rate, mass loss rate, front surface temperature, and back-face heat-soaked temperature to compare material performance. The microstructures of these FRPCs before and after ablation and mechanical testing were in-vestigated using scanning electron microscopy and micro-computed tomography. Thermophysical properties of the virgin and char FRPC were measured at elevated tem-peratures. Using these material properties and surface thermochemistry analysis, material response modeling was performed and validated with aerothermal test data. Mechanical properties, such as tensile, compression, and flexural were conducted via ASTM stand-ards. The above material properties of these FRPCs compared favorably with several commercial ablatives under similar extreme aerothermal environments.

Abstract: Anomaly signals in the Attitude Determination and Control System (ADCS) of nanosatellites can significantly degrade mission performance, especially in the absence of robust fault detection, isolation, and recovery (FDIR) mechanisms. Thus, traditional threshold-based approaches, while portable and compact, may overlook subtle faults, whereby abnormal sensor signals or current spikes within the threshold may compromise the operation of the entire ADCS as a subsystem. Furthermore, the lack of interpretable detection methods further limits the development of reliable machine learning (ML) FDIR solutions. To address these limitations, this work presents a wavelet-based anomaly detection framework that introduces a two-stage hybrid architecture combining a lightweight Convolutional Neural Network (CNN) for fault detection with logistic regression for fault classification, both based on Discrete Wavelet Transform (DWT) detail coefficients extracted from sensor and actuator data. The framework was validated using a statistics-based anomaly dataset for a 1U CubeSat ADCS simulated in MATLAB, in which anomalies are introduced at the component level with controlled variations in magnitude, frequency, and waveform, ensuring 99% statistical significance. Additionally, to demonstrate operational feasibility, constraints for onboard implementation were considered by executing the proposed framework in a Processor-in-the-Loop (PIL) environment. For benchmarking, lightweight detection and classification algorithms were compared, including Out-Of-Limits (OOL) and compact machine learning approaches. Finally, to identify the framework’s limitations and trace faulty events to physical phenomena, Grad-CAM, SHAP, and impurity analysis were performed on the proposed algorithms as primary interpretability tools. Consequently, the results demonstrate accurate fault detection and identification to support both autonomous FDIR actions and ground operator decision-making. The proposed validation framework and dataset provide a reproducible basis for advancing anomaly detection onboard nanosatellites.

Abstract: The paper examines the impact of gas turbine engine component manufacturing quality on the efficiency criteria of its life test. Known methods for selecting test parameters apply maximum damageability equivalence and minimum test time as test efficiency criteria. This study also proposes taking into account the maximization of engine lifecycle profits through the proper selection of test parameters. Engine components that determine its life were selected: the turbine blade, rotor bearing, reducer driving gear, fan bearing, and DC and AC generators. Both the mathematical expectation and variance of the quality parameters were varied during the study. The manufacturing quality of engine components and assemblies is characterized by geometric, mechanical, and physical parameters. These parameters include bearing fit diameters, initial radial clearance, turbine blade geometry, mechanical properties and gear shape, generator insulation quality, and others. Parameters selection was based on the life cycle simulation model. The following results were obtained in the course of the study: (1) Manufacturing accuracy has a more significant impact on test results than deviations from mean values of initial state parameters; (2) Despite the fact that variation in production parameters from the standard values do not affect the comparability of test results, they lead to an acceleration of the testing process. At the same time, this entails a decrease in overall economic efficiency throughout the entire life cycle of the product; (3) The overall profitability of a production run of engines is primarily determined by the quality characteristics of the turbine blades, and least on fan bearing quality parameters; (4) in the case where there is a full guarantee of engine component manufacturing quality, only short-term (acceptance, control) tests can be carried out.

Abstract: The rapid expansion of unmanned aerial vehicles (UAVs) applications in logistics, surveillance, and defense highlights the need for scalable and reliable energy delivery solutions. Conventional charging approaches constrain operational endurance and scalability, requiring frequent returns to base. This paper presents a laser-based wireless power transmission system designed to enable safe, contactless and efficient power transfer from ground to air. The proposed architecture integrates a high-power optical source, a hierarchical beam-pointing system combining coarse and fine steering, and a receiver-side sensing and energy-conversion module. The system is designed to be adaptable across different UAV classes, from lightweight platforms to larger aerial systems. An experimental campaign is conducted to validate the main system functions under representative operating conditions. Beam propagation, pointing accuracy, and control response are characterized through laboratory and outdoor tests, including long-range spot measurements and closed-loop steering validation. Overall, the study demonstrates the feasibility of laser-based wireless energy transfer for UAV applications and provides an experimental foundation for the development of persistent aerial operations in both civil and defense scenarios.

Abstract: Small penetrator probes have been proposed regularly as low(er) cost landing elements particularly for network science. This study reviews such a concept with regard to its soil penetration capability with numerical and experimental investigations. A reference Micro Mars Lander (MML) is considered to deliver a 10 kg payload to the Martian surface. It decelerates the penetrator probe with a mechanical decelerator to 40–60 m/s before impact. The remaining kinetic energy is distributed to the soil and an internal load limiter on impact, which restricts the maximum g-load acting on the payload. A semi-empirical force-displacement law and a discrete element method simulation were used to describe penetration behavior. Major impact force constituents are the velocity-dependent, drag-like displacement of the soil particles and the compaction of soil along the penetration path. Based on the simulation results, a test penetrator was designed and tested by impacting it on different soil conditions with a maximum impact velocity of 10 m/s. Cohesive Mars soil simulant, non-cohesive quartz sand, and stones of various sizes were used to assess different impact conditions. The comparison of test and simulation data identifies their respective capabilities and limitations. Recommendations for use and findings for further research are deduced therefrom.

Abstract: This research is a survey of published studies in unmanned aerial vehicles utilizing machine learning and a systematic review of what these studies have accomplished in the past ten years. It focuses on the application of supervised learning, unsupervised learning, and reinforcement learning in four types of unmanned aerial vehicles: fixed-wing, hybrid VTOL, single-rotor, and multi-rotor.It is found, according to this survey, that the application of all three types of machine learning (supervised, unsupervised, and reinforcement) have increased over the past 12 months, 24 months, five years, and ten years, with reinforcement learning application getting the highest increasing trend, followed by supervised learning. Unsupervised learning application is also increasing but the lowest. It is also found that among all four periods, the past ten years showed a significant increase in the machine learning application. Then per geographic region, China gets the highest count of published papers, followed by the North America and Europe. However, per category of unmanned aerial vehicles, it is found that the multi-rotor UAVs has the highest count in application of machine learning, followed by the fixed-wing and single-rotor UAVs. Though hybrid VTOL UAVs have important application, however, their use of machine learning was the least in terms of published papers count.

Abstract: The spacecraft reachable domain has become increasingly important for orbital game analysis due to growing on-orbit activities such as servicing, debris removal, and space situational awareness. This paper provides a comprehensive review of reachable domain theory and its applications in orbital games. A unified mathematical framework is established through three complementary classification dimensions, including spatial attributes that distinguish absolute from relative reachable domains, temporal attributes that differentiate free-time from fixed-time reachable domains, and informational attributes that contrast deterministic with predictive reachable domains. Solution methods are systematically reviewed according to this taxonomy, covering analytical and semi-analytical methods, numerical optimization approaches, and geometric and sampling methods for spatial scale reachable domains, as well as linearized ellipsoidal approximation, exact envelope determination, and fast analytical approximation for time scale reachable domains. Applications are examined through three representative scenarios: one-on-one pursuit-evasion games, multi-agent cooperative games, and threat avoidance and defense games. Key limitations of existing approaches are identified, including modeling fidelity, computational efficiency, and scalability under uncertainty. Future research directions are outlined to address these challenges.

Abstract: Background: Multirotor unmanned aerial vehicles (UAVs) are critically constrained by battery endurance, achieving only 15–25 minutes per charge in operational configurations — a fundamental limit driven by the electrochemical ceiling of lithium polymer chemistry. Piezoelectric energy harvesting (PEH), which converts structural vibration injected by onboard BLDC motors into usable electrical power, offers a mechanically passive, structurally integrated supplement requiring no additional rotating machinery. Methods: This systematic review synthesises 38 peer-reviewed studies and technical reports identified via PRISMA-ScR methodology across Web of Science, Scopus, and Google Scholar, spanning 2008–2026. Each study was critically assessed against five criteria: method type, experimental validation status, UAV platform, key quantitative result, and reliability designation (Highest/High/Medium/Low) according to an explicit evidence hierarchy. An original normalized power density analysis (mW/cm²) enables cross-study comparison on a common dimensional basis. Results: Three convergent design conclusions are established, two at Highest reliability. The central finding — arm-root patch placement outperforms motor-mount placement by 12.7–75× in harvested power — is replicated by experimental laser Doppler vibrometry (12.7:1, Perez et al. [3]) and analytically verified Euler–Bernoulli FEA (75:1, Omar [5]). An energy budget analysis reveals that the highest reported output (5.35 mW, four-arm flight experiment) represents approximately 0.002% of nominal hover power draw, confirming PEH as a sensor-power supplement rather than a propulsion augmentation strategy. Four critical research gaps are identified with explicit priority assignments. A costed four-stage roadmap charts the path from bench validation (Stage 1, ≤USD 320, 0–12 months) to multi-platform DRL-integrated flight systems (Stage 4, ≤USD 15,000, 3–5 years). Conclusions: Root-zone placement (first 20% of arm span from hub) and broadband conditioning are now sufficiently evidence-grounded to be treated as settled engineering principles. Magnetic plucking is identified as the highest-priority unexplored broadband strategy. The end-to-end in-flight demonstration gap is a resource and integration challenge, not a knowledge gap, making it immediately tractable for a well-equipped university research group.

Abstract: Graphite nozzles remain the dominant choice for small hybrid and solid rocket motors operating on laboratory and university budgets, owing to their low cost, ease of machining, and rapid turnaround during iterative design campaigns. These same programs, however, must contend with the fact that graphite erodes through coupled thermochemical and mechanical mechanisms when exposed to the oxidizing species generated by high-energy propellant combustion, and the resulting throat-area growth fundamentally alters the time histories of chamber pressure, thrust, and delivered specific impulse. This paper presents a nozzle-erosion reconstruction model that extracts the time-resolved throat area from coupled thrust and chamber-pressure measurements using the thrust coefficient relationship, scales the reconstructed area history against pre- and post-test throat measurements, identifies the onset and rate of erosion, and accounts for variable sensor lag between the thrust-stand and pressure-transducer signal chains. The model is exercised on two complementary sets of laboratory-scale GOX/ABS hybrid hot-fire data that together span roughly two orders of magnitude in total throat-area change and peak chamber pressures from 0.5 to 3.4~MPa: a controlled three-operating-point campaign conducted in support of the NASA Plume-Surface Interaction (PSI) program, and a set of higher-pressure firings from the laboratory development series in which the technique was matured. Reconstructed erosion-onset times, erosion rates, and total throat-diameter change are reported for each firing, the reconstruction accuracy is characterized as a function of erosion magnitude, and the chamber-pressure dependence of graphite erosion is examined across the combined envelope. The results demonstrate the robustness of the reconstruction technique and provide a reusable framework for post-test reconstruction of transient nozzle geometry in rocket-engine ground testing.

Abstract: Urban UAV logistics planning optimizes depot locations for delivery distance while treating airspace capacity as a fixed constraint. We show that this decoupling is not merely a modeling simplification — it produces a structural error that cost monitoring cannot detect in time. When depot configurations begin diverging under airspace capacity pressure, the window for low-cost infrastructure correction closes before any cost signal appears.The central finding is a lead–lag relationship between structural and cost responses: at demand levels where total cost differences remain negligible, depot layouts already diverge substantially under airspace capacity pressure. Cost is a lagging indicator, not a planning trigger.To close this gap, we derive the Ω index, a closed-form demand threshold with three city-level parameters — capacity gap ΔC (from building-footprint data via a FAR volumetric method), depot distance differential Δd (from road-network geometry), and mean OD delivery distance d (from weighted building-centroid sampling) — separating a regime where distance-based planning is analytically sufficient from one where airspace capacity must be internalized into the location decision. Applied to Dongli District, Tianjin, the threshold lies above current UAV density but within a foreseeable planning horizon given observed growth rates in Chinese pilot cities. The depot relocation distance Δreloc provides a leading structural indicator, detectable well before any cost signal emerges. The Ω index gives planners a computable early-warning trigger for airspace-integrated infrastructure review — before congestion, not after.

Abstract: In window tracking mode, stray light and detector readout noise can submerge star-spot signals in star sensor images. The resulting degradation reduces centroid extraction accuracy and may even cause extraction failure, thereby preventing precise attitude determination. This study uses the self-supervised spatiotemporal denoising model ASTERIS as the baseline. ASTERIS integrates 3D spatiotemporal inputs with a global attention mechanism for joint noise modeling, thereby providing stronger denoising and restoration capability than conventional methods such as multi-frame stacking. However, ASTERIS lacks adaptive compensation for subpixel jitter in on-orbit star images and has difficulty preserving the high-frequency morphology of star spots, affecting denoising performance and centroiding accuracy. To address these limitations, this study introduces two improvements. First, frame-by-frame spatial deformable convolution is incorporated into the decoder upsampling stage to adaptively compensate for subpixel offsets, actively suppress background noise, and lower the parameter count. Second, a complex-valued frequency-domain loss with a high-frequency weighted mask is designed to jointly constrain the amplitude and phase spectra, thereby preserving high-frequency star-spot details. Experimental results show that, for star images with extremely low signal-to-noise ratios, the proposed method improves the peak signal-to-noise ratio by approximately 60-fold and reduces the centroid localization error to approximately 0.1 pixels. This performance is substantially better than that of the original ASTERIS model, which improves the peak signal-to-noise ratio by approximately 9-fold and yields an error of approximately 0.4 pixels, and the multi-frame stacking method, which improves the peak signal-to-noise ratio by approximately 4-fold and yields an error of approximately 0.5 pixels. The deep learning method presented in this paper provides a novel solution for centroid extraction of star sensors under strong noise interference in orbit and achieves satisfactory results. Future work will focus on lightweight network design to enable on-orbit engineering applications.

Abstract: Commercial quadrotor UAVs are fundamentally constrained to 15–25 minutes of flight per charge by the energy density limitations of lithium-polymer batteries, motivating concurrent advances in structural energy harvesting and energy-aware navigation. This paper presents a comprehensive, physics-coupled framework that extends single-arm to full four-arm PZT-5A piezoelectric energy harvesting on the DJI F450 platform and rigorously evaluates three deep reinforcement learning (DRL) algorithms — DQN, PPO, and SAC — for energy-aware autonomous navigation. Six PZT-5A patch locations per arm are characterised via Euler–Bernoulli finite element analysis (FEA), establishing the arm-root position (P3, 15% span) as universally optimal across all four arms, yielding 0.0600 mW average power and outperforming the motor-mount location by a factor of 75. Symmetric deployment of P3 patches on all four arms produces a combined 0.2400 mW average power and 144 mJ per standard 10-minute mission. When the SAC navigation policy preferentially allocates flight time toward maximum-throttle climb phases, energy recovery increases to 444 mJ per mission. SAC achieves 82.2 ± 2.7% navigation success with 24.2 ± 1.8% battery consumption — Pareto-optimal over PPO (71.7 ± 3.1%, 29.2 ± 1.8%) and DQN (57.8 ± 2.6%, 36.1 ± 2.2%) — with all pairwise differences statistically significant (ANOVA: F = 93.96, p < 0.001; Cohen's d >= 3.6). The harvested energy offsets 14.8% of total eight-VL53L1X proximity sensor energy demand per mission (444 mJ harvested vs. 3,000 mJ sensor consumption over 600 s); during climb phases (168 s), the 93.6 mJ harvest covers 11.1% of the 840 mJ sensor demand — a net deficit of 746 mJ per climb phase that is honestly quantified. This partial offset yields 44.4 J of primary battery relief over 100 operational missions. All results are independently validated by 43/43 unit tests and bench experiments within ±18%. While direct propulsion endurance extension is negligible (0.0049% of LiPo capacity), the physics-derived reward signal improves navigation success by 10.5 percentage points over energy-blind baselines, establishing a reproducible methodology for coupling structural mechanics to DRL. A complete sim-to-real deployment roadmap and open-source codebase are provided.

Abstract: The present work investigates fluid–-structure instabilities and stall flutter of a pitching NACA0012 airfoil through numerical simulations. The flow is modeled using the compressible Navier–-Stokes equations in a non-inertial rotating reference frame, while the structural dynamics are represented by a torsional spring–-mass–-damper system. The analysis focuses on the effects of reduced velocity, equilibrium angle of attack, and elastic axis position on the aeroelastic behavior. The results show a transition from steady flow to vortex-shedding regimes and, at higher reduced velocities, to limit-cycle oscillations. Increasing the equilibrium angle of attack leads to an earlier onset of instability and stronger aerodynamic forcing, while moving the elastic axis downstream has a similar destabilizing effect due to the larger aerodynamic moment arm. Frequency analysis highlights the progressive coupling between fluid and structural dynamics: vortex shedding dominates at low reduced velocity, whereas the structural frequency governs the response in the limit-cycle regime. The study provides a consistent description of the mechanisms driving stall flutter and of the parameters influencing aeroelastic stability.