Scholarlink

<p>The structural integrity of the propellers is a prime factor in UAV flight safety, but reliable real-time fault detection is problematic on resource-constrained hardware. In this effort, a lightweight diagnostic framework is developed that is capable of identifying edge cuts, cracks, and surface cuts using only standard onboard IMU data. The vibration signature from five flight trajectories in the DronePropA dataset was employed to carry out the analysis. Since onboard processing latency is a key issue, computationally intensive deep learning architectures are avoided. A 180-dimensional feature vector is created by extracting 30 statistical time-domain descriptors from the dual-IMU system. Validation using a Leave-One-Trajectory-Out protocol show that the proposed stacked ensemble classifier results in an accuracy of 97.9% and a macro-F1 score of 0.978. Interestingly, the whole system has an average inference time of 2.1&#xa0;ms, which proves that we can use it for real-time edge computing on common CF flight controllers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Intelligent fault detection of UAV propellers through time-domain vibration analysis and ensemble learning

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Karam J. Mohammed,
  • Ahmed Ali Farhan Ogaili,
  • Abdul-Rasool Kareem Jweri,
  • Alaa Abdulhady Jaber,
  • Luttfi A. Al-Haddad

摘要

The structural integrity of the propellers is a prime factor in UAV flight safety, but reliable real-time fault detection is problematic on resource-constrained hardware. In this effort, a lightweight diagnostic framework is developed that is capable of identifying edge cuts, cracks, and surface cuts using only standard onboard IMU data. The vibration signature from five flight trajectories in the DronePropA dataset was employed to carry out the analysis. Since onboard processing latency is a key issue, computationally intensive deep learning architectures are avoided. A 180-dimensional feature vector is created by extracting 30 statistical time-domain descriptors from the dual-IMU system. Validation using a Leave-One-Trajectory-Out protocol show that the proposed stacked ensemble classifier results in an accuracy of 97.9% and a macro-F1 score of 0.978. Interestingly, the whole system has an average inference time of 2.1 ms, which proves that we can use it for real-time edge computing on common CF flight controllers.