流离失所(心理学)
算法
计算机科学
故障检测与隔离
方位(导航)
声学
噪音(视频)
断层(地质)
度量(数据仓库)
人工智能
控制理论(社会学)
探测理论
信号处理
计算机视觉
作者
Weiyu Zhang,Wenjie Zhou
出处
期刊:IEEE-ASME Transactions on Mechatronics
[Institute of Electrical and Electronics Engineers]
日期:2026-01-01
卷期号:: 1-13
标识
DOI:10.1109/tmech.2026.3694121
摘要
The traditional offline calibration method cannot detect the loosening fault of the displacement sensor in the magnetic bearing system under the mobile platform. This situation will lead to the reduction of the system control accuracy, and then cause the risk of suspension instability. To solve this problem, an online self-diagnosis method based on the resonance sparse signal decomposition (RSSD) algorithm optimized by deep reinforcement learning is proposed in this article. This method adopts a two-level decision-making architecture combining nondominated sorting genetic algorithm III (NSGA-III) and dueling deep Q-network. Moreover, multipoint optimal minimum entropy deconvolution adjusted algorithm using multipoint kurtosis spectrum is combined to accurately separate the characteristic frequency of loosening fault. To validate the method, the flywheel battery system is used as an example. First, the nonlinear dynamic model of displacement sensor loosening fault is established. Then, the specific architecture of RSSD-RL is detailed, and the verification experiment is carried out through the simulated loosening fault signal. Finally, the reliability and real-time performance of the proposed method are verified by fault detection performance experiments.
科研通智能强力驱动
Strongly Powered by AbleSci AI