Efficient water leakage detection is crucial for smart building water management. Conventional methods relying on fixed sensors are costly to install and maintain. Micro air vehicles (MAVs) offer a cost-effective solution for confined spaces but have limited processing power and payload capacity. MAVs are generally equipped with lightweight sensors, such as compact thermal cameras, which produce low-resolution images and typically limit the detection range. To enable effective long-range detection, this paper proposes a novel cross-modal self-supervised learning (SSL) setup, where a computer vision algorithm is designed to identify water leakage from thermal images. These thermal-based detections then guide the training of a deep learning model using color input images. Furthermore, a control strategy integrating the color-based deep learning model with thermal imaging is introduced to enable MAVs to provide preliminary long-range leakage detection and precise localization when approaching affected areas. The system was tested in both static setups and flight scenarios. In flight tests, the MAV successfully identified leakage across multiple trials, beginning from various initial positions and targeting different leakage sites. Results from both static and dynamic tests demonstrate the effectiveness of the control strategy and detection algorithm in localizing water leakage in indoor environments.
This paper introduces two novel metrics—failure impact (FI) and uncertainty impact (UI)—to overcome the limitations of traditional technology and integration readiness levels in evaluating hybrid-electric turboprop aircraft. Scored on a 1–9 scale, FI and UI quantify component failure severity and the impact of input uncertainties on system-level metrics like block fuel and maximum takeoff weight. The UI scores are rigorously derived using a physics-based sizing model and variance-based global sensitivity analysis (Sobol indices). Applying these metrics to a 50-passenger aircraft’s critical components (battery, motor, inverter, and thermal management system [TMS]), the study utilizes multi-attribute decision-making (MADM) frameworks across four strategic scenarios. Methods, including multi-attribute utility theory, analytic hierarchy process, best–worst method (BWM), and technique for order preference by similarity to ideal solution, consistently prioritize the battery for testing and resource allocation due to its high FI and UI risks, followed by the TMS. In contrast, the noncompensatory Elimination and Choice Expressing Reality III (ELECTRE III) method ranked the battery last due to extreme risk vetoes. The study concludes that the BWM is the most effective and consistent MADM approach for aerospace applications due to its simplicity.
Energy management (EM) is a critical factor in enhancing the efficiency and extending the lifespan of Nanosatellite missions. This study introduces an advanced Nanosatellite Power Distribution Strategy based on efficient EM using intelligent Fuzzy-based control to maximize task performance. A novel State of Charge (SoC) prediction model is integrated, optimizing charge and discharge cycles to extend battery lifespan. A Fuzzy decision-making is developed to regulate discharge rates and depth of discharge, preserving battery health while maintaining mission tasks. Simulation experiments demonstrate the effectiveness of the proposed EM approach in representative mission scenarios based on Automatic Identification System (AIS) for tracking and fire detection. Additionally, Processor-in-the-Loop (PIL) validation using the Atmel SAM3X8E ARM Cortex-M3 microcontroller confirms the controller’s practical applicability. These advancements provide a robust foundation for offline optimization of Nanosatellite energy management, supporting reliable and sustainable mission operations.
Traditional preflight critical component inspections rely heavily on manual visual checks by maintenance crews, resulting in inefficient detection processes, significant safety risks, and high omission rates. The civil aviation industry urgently requires automated inspection solutions to enhance safety and efficiency. However, existing computer vision approaches are constrained by challenges such as complex lighting variations on aprons and significant scale variations of targets. To address these issues, this paper proposes an autonomous unmanned ground vehicle (UGV)-based detection framework for preflight critical component status detection, enhancing the intelligence of preflight inspections by using a UGV to support maintenance personnel in round-the-plane inspections. Specifically, we propose an aircraft critical component status detection method for preflight inspection via a UGV platform to accurately identify the states of nine key components (e.g., landing gear pins and electronic equipment bay doors) in apron scenarios. The luminance generation branch and multiscale detection branch are designed and introduced into the proposed algorithm to achieve state-of-the-art performance in preflight critical component status detection tasks. To establish an evaluation benchmark, we construct an Aircraft Pre-flight Inspection Safety-critical Detection dataset derived from line maintenance checklists, containing precise mask annotations of nine critical components captured via a UGV. Experimental results on our dataset demonstrate that the proposed method outperforms state-of-the-art object detection algorithms and meets preflight inspection requirements.
This paper investigates the resilience of deep reinforcement learning for autonomous satellite task scheduling under reaction wheel fault conditions. The scheduling problem for an agile Earth-observing satellite is formulated as a partially observable decision-making framework that captures realistic operational constraints, including power management, momentum control, and imaging opportunities. Four representative fault modes are considered: complete loss of control, power limitation, increased friction, and encoder measurement faults. To improve robustness, three training strategies are evaluated: nominal training, curriculum learning, and domain randomization. Policies are trained within a high-fidelity spacecraft simulation environment and evaluated across varying target densities and fault severities. Performance is assessed in terms of cumulative mission reward and mission safety, defined by successful completion without battery depletion or reaction wheel saturation. Results show that policies trained only under nominal conditions exhibit gradual performance degradation as fault severity increases, whereas curriculum learning and domain randomization mitigate this degradation while maintaining comparable performance in the nominal environment. Domain randomization provides the most consistent performance across fault conditions and outperforms a heuristic scheduling approach in most fault scenarios, demonstrating the potential of deep reinforcement learning for resilient autonomous satellite task scheduling.
