Scholarlink

<p>Unmanned aerial vehicles (UAVs) are becoming prevalent in human environments, and their utilization is expanding across many applications. Therefore, even a minor breakdown in a drone’s executive systems can provide a genuine threat to individuals. As a result, there is a need to address the issue of guaranteeing the reliability and safety of their flight missions and achieving efficient diagnostics of potential drone malfunctions. A well-known issue often occurring with multi-rotors is propeller damage. Hence, its timely identification is crucial for ensuring the secure functioning of drones. This study examines these frequently encountered faults using the PADRE repository dataset. Despite its potential, machine learning’s dependence on traits not derived from the time domain presents a challenge. So, starting from this challenge, we constructed a lightweight Bidirectional Long Short-Term Memory (BiLSTM) model using only two bidirectional layers combined with time-domain statistical features extracted from the inertial measurement unit sensors’ data. Experimental findings demonstrate the efficiency of the proposed fault detection and diagnosis model with an accuracy of over <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(99\%\)</EquationSource> </InlineEquation> across 20 distinct propeller fault scenarios.</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

Lightweight deep neural network based fault detection and diagnosis for quadrotor propeller faults

  • 期刊论文

  • 发表日期：

    2025年10月13日

查看全文
  • Zineb Adaika,
  • Mohamed Boumehraz

摘要

Unmanned aerial vehicles (UAVs) are becoming prevalent in human environments, and their utilization is expanding across many applications. Therefore, even a minor breakdown in a drone’s executive systems can provide a genuine threat to individuals. As a result, there is a need to address the issue of guaranteeing the reliability and safety of their flight missions and achieving efficient diagnostics of potential drone malfunctions. A well-known issue often occurring with multi-rotors is propeller damage. Hence, its timely identification is crucial for ensuring the secure functioning of drones. This study examines these frequently encountered faults using the PADRE repository dataset. Despite its potential, machine learning’s dependence on traits not derived from the time domain presents a challenge. So, starting from this challenge, we constructed a lightweight Bidirectional Long Short-Term Memory (BiLSTM) model using only two bidirectional layers combined with time-domain statistical features extracted from the inertial measurement unit sensors’ data. Experimental findings demonstrate the efficiency of the proposed fault detection and diagnosis model with an accuracy of over \(99\%\) across 20 distinct propeller fault scenarios.