多尺度时空特征驱动的无人机编队故障诊断OA
Multi-scale spatio-temporal feature-driven fault diagnosis for quadrotor formations
风扰环境下无人机编队的执行器故障特征易与全局气流扰动耦合,导致故障定位、类型识别和严重程度量化精度下降.为提高复杂环境中的编队诊断鲁棒性,构建了包含稳态风、阵风、湍流及多类执行器故障的四旋翼无人机编队仿真平台,并生成多风速、多轨迹、多规模的故障数据集.在此基础上,提出一种变量级注意力多尺度故障诊断模型.该模型利用变量级注意力自适应强化故障敏感观测量,通过多尺度卷积提取邻机状态和飞行参数之间的空间耦合特征,并采用门控循环单元(GRU)刻画故障随时间演化的动态特性;进一步结合级联式多任务网络与动态权重平衡机制,实现故障无人机定位、故障执行器定位、故障类型识别和故障程度量化的协同诊断.实验结果表明,在0∼9 m/s风扰测试条件下,所提模型的故障执行器定位和故障类型识别准确率均保持在99%以上,故障程度量化平均绝对误差低于0.01.结果验证了所提模型在风扰环境下对编队执行器故障的诊断精度和泛化能力.
Actuator fault features in quadrotor formations are easily coupled with global airflow disturbances,which degrades fault localization,type identification,and severity estimation.To improve diagnostic robustness in complex en-vironments,a quadrotor formation simulation platform is constructed by incorporating steady wind,gust,turbulence,and representative actuator faults,and a dataset covering multiple wind speeds,trajectories,and formation scales is generated.A multi-scale fault diagnosis model with variable-level attention is then developed.The model adaptively enhances fault-sensitive observables through variable-level attention,extracts spatial coupling features among neighboring quadrotors and flight parameters using multi-scale convolutions,and captures temporal fault evolution with a gated recurrent unit(GRU).A cascaded multi-task network with dynamic task-weight balancing is further used to jointly perform faulty quadrotor lo-calization,faulty actuator localization,fault type identification,and fault severity quantification.Experimental results show that,under wind speeds from 0 to 9 m/s,the proposed model maintains over 99%accuracy in faulty actuator localiza-tion and fault type identification,while keeping the mean absolute error of severity estimation below 0.01.These results demonstrate the diagnostic accuracy and generalization capability of the proposed model for actuator faults in quadrotor formations under wind disturbance.
马亚杰;曹旭;姜斌;李映兮
南京航空航天大学 自动化学院,江苏 南京 211106南京航空航天大学 自动化学院,江苏 南京 211106南京航空航天大学 自动化学院,江苏 南京 211106南京航空航天大学 自动化学院,江苏 南京 211106
无人机编队风扰执行器故障故障诊断多尺度特征提取多任务学习
quadrotor formationwind disturbanceactuator faultfault diagnosismulti-scale feature extractionmulti-task learning
《控制理论与应用》 2026 (9)
1901-1916,16
国家自然科学基金项目(62273177,62020106003,62233009),江苏省基础研究计划项目(BK20250081),高校学科创新引智基地项目(B20007),中央高校基本科研业务费项目(NI2024001)资助. Supported by the National Natural Science Foundation of China(62273177,62020106003,62233009),the Basic Research Program of Jiangsu(BK20250081),the Programme of Introducing Talents of Discipline to Universities of China(B20007)and the Fundamental Research Funds for the Central Universities(NI2024001).
评论