Scholarlink

Background/Introduction <p>Propeller failures represent one of the most critical risks to unmanned aerial vehicle (UAV)&#xa0;safety. Conventional diagnostic methods often lack robustness, real-time applicability, and scalability, limiting their effectiveness for in-flight monitoring and predictive maintenance.</p> Purpose <p>This study aims to develop and validate a lightweight, real-time diagnostic framework that integrates IoT-based&#xa0;sensing, multi-domain vibration analysis, and optimized machine learning for accurate detection of UAV propeller&#xa0;faults.</p> Methods <p>An embedded ESP32–ADXL335 platform was designed to continuously acquire tri-axial vibration signals at&#xa0;200 Hz and transmit them wirelessly with end-to-end latency under 150 ms and active current consumption of ~12 mA. A&#xa0;multi-domain feature set (time, frequency, and time–frequency) was extracted and evaluated using three classifiers—Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and XGBoost—each optimized through Genetic Algorithm,&#xa0;Markov Chain Monte Carlo, or Grid Search. Additionally, a custom Deep Neural Network (DNN) with residual-style&#xa0;architecture and embedded feature selection was implemented.</p> Results <p>Experimental validation showed that the GA-optimized SVM achieved 98.8% classification accuracy, reducing&#xa0;false alarms by 40% compared to traditional FFT-based approaches. The proposed DNN achieved 100% classification&#xa0;accuracy, surpassing all conventional models and confirming the potential of deep learning for UAV fault diagnostics.</p> Conclusions <p>The proposed framework establishes a scalable, low-cost, and real-time solution for UAV health&#xa0;monitoring. By combining efficient IoT-based sensing with advanced feature analysis and optimized classifiers, it provides&#xa0;a path toward predictive maintenance strategies and improved UAV safety.</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

Real-Time Drone Propeller Fault Detection Using Onboard Vibration Sensors and Optimized Machine Learning

  • 期刊论文

  • 发表日期：

    2025年10月20日

查看全文
  • Mohamed Seif El Islam Lalem,
  • M’hamed Ouadah,
  • Omar Touhami

摘要

Background/Introduction

Propeller failures represent one of the most critical risks to unmanned aerial vehicle (UAV) safety. Conventional diagnostic methods often lack robustness, real-time applicability, and scalability, limiting their effectiveness for in-flight monitoring and predictive maintenance.

Purpose

This study aims to develop and validate a lightweight, real-time diagnostic framework that integrates IoT-based sensing, multi-domain vibration analysis, and optimized machine learning for accurate detection of UAV propeller faults.

Methods

An embedded ESP32–ADXL335 platform was designed to continuously acquire tri-axial vibration signals at 200 Hz and transmit them wirelessly with end-to-end latency under 150 ms and active current consumption of ~12 mA. A multi-domain feature set (time, frequency, and time–frequency) was extracted and evaluated using three classifiers—Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and XGBoost—each optimized through Genetic Algorithm, Markov Chain Monte Carlo, or Grid Search. Additionally, a custom Deep Neural Network (DNN) with residual-style architecture and embedded feature selection was implemented.

Results

Experimental validation showed that the GA-optimized SVM achieved 98.8% classification accuracy, reducing false alarms by 40% compared to traditional FFT-based approaches. The proposed DNN achieved 100% classification accuracy, surpassing all conventional models and confirming the potential of deep learning for UAV fault diagnostics.

Conclusions

The proposed framework establishes a scalable, low-cost, and real-time solution for UAV health monitoring. By combining efficient IoT-based sensing with advanced feature analysis and optimized classifiers, it provides a path toward predictive maintenance strategies and improved UAV safety.