Abstract:Aiming at the engineering pain points of insufficient health-evaluation accuracy, delayed fault early-warning and difficulty in dynamic quantification of single-UAV task weights for UAV swarms under complex operating environments, research is carried out on UAV-swarm health-assessment technology integrated with multiple machine-learning algorithms. Analyses are conducted on five key links including distributed preprocessing of multi-source sensor data, intelligent fault identification for single UAV via ensemble learning, percentage-scale health quantification of single UAV by binomial logistic regression with log-linear kernel, adaptive weight solving with genetic-fuzzy comprehensive evaluation, and weighted fusion of swarm health. Combined with FMECA failure-mode analysis theory and full-life-cycle monitoring data of UAVs, an end-to-end fully-automatic integrated health-assessment system is constructed. It realizes 0-100 continuous percentage-based health quantification and working-condition-driven adaptive solving of single-UAV weights, and remedies the defects of conventional assessment schemes such as fragmented processes, discrete binary judgment of health status, expert-dependent fixed weights and poor coordination among multiple algorithms. Simulation tests obtain a stable faul-detection rate above 93.7% and a false-alarm rate below 2.4%. Only comprehensive-condition simulation verification is carried out for electromagnetic disturbance, low temperature and data-packet-loss scenarios, and quantitative performance tests under multi-gradient interference have not been completed. The technology supports online real-time operation-maintenance and advance fault early-warning for large-scale heterogeneous UAV swarms to facilitate dynamic task allocation, and is also applicable to civil scenarios such as infrastructure inspection and emergency rescue.
