Scholarlink

<p>Fault detection of rotary machinery is paramount in preventing unforeseen breakdowns and providing high operational efficiency- a factor that is of great significance in any industrial setting where reliability is a crucial consideration. In this study, vibration signals were recorded at various loads and speeds combinations aiming to compare the performance of various Machine Learning (ML) models- Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Artificial Neural Networks (ANN). The core aim of the current study is to identify the robustness of SVM, kNN, and ANN models in the diagnosis of bearing defects under varying loads and speeds. At different speeds (1000&#xa0;rpm, 1200&#xa0;rpm, and 1400&#xa0;rpm) and load conditions (4&#xa0;kg, 8&#xa0;kg, and 12&#xa0;kg), four bearing conditions (i.e. healthy, ball defect, inner race defect, and outer race defect) have been investigated. Time-domain and time-frequency domain methods, such as Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT) were used to analyze the acquired vibration signals. Among the models, ANN achieved the best accuracy of 98.75%, outperforming SVM (96.94%) and kNN (94.44%) in fault classification under varied operating conditions. These findings highlight the superior generalization and fault sensitivity capabilities of ANN. The results indicate how ML models, in particular ANN, can be effectively used to achieve robust and real-time condition-based monitoring and predictive maintenance of rotary systems.</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

Vibration-based Condition Monitoring for Intelligent Anomaly Detection in Rotary Machinery Under Dynamic Environments

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Chirag Mongia,
  • Shankar Sehgal,
  • Deepam Goyal

摘要

Fault detection of rotary machinery is paramount in preventing unforeseen breakdowns and providing high operational efficiency- a factor that is of great significance in any industrial setting where reliability is a crucial consideration. In this study, vibration signals were recorded at various loads and speeds combinations aiming to compare the performance of various Machine Learning (ML) models- Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Artificial Neural Networks (ANN). The core aim of the current study is to identify the robustness of SVM, kNN, and ANN models in the diagnosis of bearing defects under varying loads and speeds. At different speeds (1000 rpm, 1200 rpm, and 1400 rpm) and load conditions (4 kg, 8 kg, and 12 kg), four bearing conditions (i.e. healthy, ball defect, inner race defect, and outer race defect) have been investigated. Time-domain and time-frequency domain methods, such as Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT) were used to analyze the acquired vibration signals. Among the models, ANN achieved the best accuracy of 98.75%, outperforming SVM (96.94%) and kNN (94.44%) in fault classification under varied operating conditions. These findings highlight the superior generalization and fault sensitivity capabilities of ANN. The results indicate how ML models, in particular ANN, can be effectively used to achieve robust and real-time condition-based monitoring and predictive maintenance of rotary systems.