To construct the micro collision avoidance intelligent unit, efficient solutions have been pursued by modeling a specific collision-sensitive neuron, the lobula giant movement detector (LGMD), identified in the locust's visual neural system. However, the existing LGMD models cooperating with a threshold-based collision avoidance strategy are less robust in the presence of visual variability, including fluctuations in contrast and noise. Intriguingly, biological research underscores the potential of the single-peak modeling through the a function, suggesting a non-threshold collision avoidance strategy by identifying the peak response moment. Adapting the biological finding to the engineering applications, we establish a novel modeling method based on a-form perceptron, making the response curve behave like the a function. The single peak within the a-function-form response occurs before the collision, offering reliable collision warning cues. Numerical experiments substantiate the effectiveness of the proposed model against visual variability. This study extends the limits of the applications of LGMD models in diverse scenarios, and explores the potential of modeling neural computation in constructing micro-intelligent units.
This paper proposes a novel transcritical CO2 refrigeration system integrating a vortex tube and an ejector, aiming to synergistically enhance overall thermodynamic performance by combining the energy separation effect of vortex tube with the energy recovery mechanism of ejector. Firstly, a detailed mathematical model is established to investigate the system behavior. Under the design conditions, the proposed system achieves a cooling capacity of 141.37 kW and a COP of 3.56. Subsequently, a parametric analysis reveals that system efficiency is highly sensitive to variations in key operating variables: increasing the vortex tube inlet pressure, inlet temperature, and cold mass fraction leads to a decline in COP due to dominant compressor power consumption increases, whereas elevating the evaporation temperature enhances efficiency by reducing the compressor pressure ratio. Meanwhile, the vortex tube hot-end pressure exerts a minor influence with an optimal value, and adjusting the ejector back pressure optimizes compressor power consumption distribution. Building upon these findings, a multi-parameter optimization yields a maximum COP of 3.90. Finally, comprehensive performance comparisons across multiple operating conditions demonstrate that the proposed system outperforms conventional and single-component reference configurations, achieving relative COP improvements of approximately 0.1% to 21.1% over varying gas cooler pressures and 1.6% to 21.4% over varying evaporation temperatures. These findings confirm the superior energy-saving potential and robust adaptability of the proposed system for advanced refrigeration applications.
Traditional microwave absorbers with limitations of narrow band absorption, single functionality, and difficulty in conforming to complex surfaces are no longer sufficient to meet the demands of modern equipment for multispectral stealth. This paper proposes a flexible optically transparent metasurface (FOTMS) with simultaneously ultra-broadband millimeter-wave absorption and infrared (IR) stealth. Tri-layer indium tin oxide (ITO) patterned layers were designed and etched onto flexible transparent polyethylene terephthalate (PET) substrate, achieving radar-IR bi-stealth, optical transparency and a flexible configuration. Both simulated and experimental results indicate that the metasurface features absorptivity more than 90% within 15–38.7 GHz, while exhibits 10 dB radar cross section (RCS) reduction for curved configurations. Additionally, the metasurface displays polarization insensitivity and stability at different incidences and polarization angles. Meanwhile, the entire structure maintains a high optical transmittance of 85.09%, and the average infrared emissivity of metasurface is 0.47 for 3–14 μm. This work provides an effective strategy for optical windows and multispectral stealth.
Efficient delivery of humanitarian supplies is crucial for disaster loss reduction. The Truck-Drone Collaborative Routing Problem (TDRP) has mostly been studied under static demand and ideal road networks, neglecting the dynamic revelation of demand points and nonlinear road capacity degradation from disaster damage. This paper introduces the TDRP with Road Damage and Dynamic Demand (TDRP-RD-DD) to minimize global delivery completion time. We build a full-chain model linking disaster intensity, road capacity, and truck travel speed, and develop an event-driven two-layer rolling optimization framework with locally optimal selection criteria for supply and replenishment modes. To solve this NP-hard problem, we design a Rolling Replenishment-aware Adaptive Large Neighborhood Search (RR-ALNS) with three specialized destroy-repair operators, a capacity- and mode-aware Warm-start strategy, and a replenishment-prioritized repair mechanism for infeasible solutions. Experiments on Solomon instances and a real-world disaster-relief case in Hubei Province show that RR-ALNS closely matches CPLEX on CPLEX-proven small-scale instances, with a 0.47% average gap for the default simulated-annealing setting, while scaling effectively to larger dynamic settings. Compared with three baseline algorithms, RR-ALNS achieves stronger feasibility and overall lower makespan in both benchmark and real-world tests. Ablation and sensitivity analyses further support the value of truck-drone collaboration, collaborative replenishment, and explicit road-damage-to-speed modeling. This work extends dynamic truck-drone routing to post-disaster settings with road damage and demand revelation, and provides quantitative support for emergency logistics decision-making.
High-fidelity wind tunnel data are costly and sparse across the full flight envelope, while abundant CFD simulations carry inherent systematic bias, making multi-fidelity data fusion a core challenge in aerodynamic modeling. To reduce high-fidelity expenses and improve the multi-source information value, this study applies a multi-fidelity aerodynamic data fusion framework based on physics-informed autoencoder transfer learning (AE-TL). The latent-space dimension is chosen to reflect the distinct physical attributes of the system and is set to two for the specified Mach number, Reynolds number, and configuration, thereby reducing model complexity from the outset. A two-stage pretraining–fine-tuning strategy is adopted: the shallowest encoder layers are frozen to preserve the low-dimensional latent variability shared across fidelity levels, while only the deeper layers and the decoder are fine-tuned to achieve high-fidelity accuracy and mitigate overfitting under limited-data conditions. Using only 62 wind-tunnel samples from the CHN-F1 flying-wing benchmark model, AE-TL reduced the lift-coefficient RMSE by 59.3%, 93.0%, and 92.8% compared with XGBoost, Co-Kriging, and a comparable standard neural-network transfer-learning baseline (Standard NN-TL), respectively; the corresponding R² values for lift, drag, and pitching moment were 0.9996, 0.9995, and 0.9921. AE-TL achieves a throughput of 62 predictions in 2.1 ms on a single CPU, comparable to the millisecond-scale inference speeds of the other neural-network and Gaussian-process models; it is also approximately 32 times faster than XGBoost (66.9 ms) while attaining the highest predictive accuracy. These results demonstrate the effectiveness of physics-informed autoencoder transfer learning for multi-fidelity aerodynamic data synthesis.
