Scholarlink

This paper presents a fault diagnosis technique in three-phase induction motors (IM) utilizing stator current signature as feature input to four different machine learning algorithms. The performance of k-Nearest Neighbor (kNN), Support Vector Machine (SVM), Logistic Regression (LR), and Random Forest (RF) classifiers has been analyzed for classifying external faults and abnormalities in three-phase IM. A comparative evaluation of instantaneous and RMS currents has been carried as feature input for classifying Normal Load (NL), Overload (OL), Overvoltage (OV), Undervoltage (UV), Single-Phasing (SP), and Voltage Unbalanced (VUB) conditions. Stator currents from both the experimental setup and simulation have been used to evaluate performance. The analysis suggests that the RMS current-based SVM classifier performed consistently and reliably with simulation and experimental datasets demonstrating better-generalized capabilities.

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

Data-Driven Fault Diagnosis in Three-Phase IMs: Harnessing Current Signature Data with Machine Learning Algorithms

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Uvesh Sipai,
  • Hagos L. Shifare,
  • Nishant Kothari

摘要

This paper presents a fault diagnosis technique in three-phase induction motors (IM) utilizing stator current signature as feature input to four different machine learning algorithms. The performance of k-Nearest Neighbor (kNN), Support Vector Machine (SVM), Logistic Regression (LR), and Random Forest (RF) classifiers has been analyzed for classifying external faults and abnormalities in three-phase IM. A comparative evaluation of instantaneous and RMS currents has been carried as feature input for classifying Normal Load (NL), Overload (OL), Overvoltage (OV), Undervoltage (UV), Single-Phasing (SP), and Voltage Unbalanced (VUB) conditions. Stator currents from both the experimental setup and simulation have been used to evaluate performance. The analysis suggests that the RMS current-based SVM classifier performed consistently and reliably with simulation and experimental datasets demonstrating better-generalized capabilities.