Scholarlink

<p>In this paper, a multi-fault detection framework is suggested for linear induction motors (LIMs) that depend on deep learning (DL) and high-fidelity finite element modelling (FEM). Four fault conditions are utilized through static eccentricity, broken secondary conductor, inter-turn fault in the stator, and mechanical irregularities in combination, which were simulated through a FEM in a two-dimensional time domain. In order to assure fidelity in the simulation, electromagnetic parameters including flux density, field intensity, and eddy current distribution were rigorously calculated and checked using published parameters of LIMs. The FEM outputs were taken as multi-domain diagnostic features that consist of wavelet coefficients, thrust responses, stator currents, line profiles, and fast Fourier transform (FFT) spectra that were then fed into a spatial–temporal feature tensor. It has been trained and validated on the simulated dataset and the hybrid convolutional neural network (CNN) and long short-term memory (LSTM) network provided 97.8% classification accuracy and demonstrated 25–40% higher sensitivity in detecting early-stage faults compared to FFT, wavelet, and traditional machine learning-based methods. In addition, the proposed DL plus FEM framework identified fault-induced variations of up to 25% in flux density, 25% in field intensity, and 40% in eddy current density, indicating its effectiveness as simulation-validated pathway for future real-time implementation and predictive maintenance applications.</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

A deep learning-assisted FEM approach for comprehensive fault detection in linear induction motors

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Chilukoti Varaha Narasimha Raja,
  • S Sandeep Kumar Reddy Vaka,
  • Baseem Khan,
  • Rajkumar Sivanraju,
  • Khaled S. Alatawi,
  • Fahad M. Almasoudi

摘要

In this paper, a multi-fault detection framework is suggested for linear induction motors (LIMs) that depend on deep learning (DL) and high-fidelity finite element modelling (FEM). Four fault conditions are utilized through static eccentricity, broken secondary conductor, inter-turn fault in the stator, and mechanical irregularities in combination, which were simulated through a FEM in a two-dimensional time domain. In order to assure fidelity in the simulation, electromagnetic parameters including flux density, field intensity, and eddy current distribution were rigorously calculated and checked using published parameters of LIMs. The FEM outputs were taken as multi-domain diagnostic features that consist of wavelet coefficients, thrust responses, stator currents, line profiles, and fast Fourier transform (FFT) spectra that were then fed into a spatial–temporal feature tensor. It has been trained and validated on the simulated dataset and the hybrid convolutional neural network (CNN) and long short-term memory (LSTM) network provided 97.8% classification accuracy and demonstrated 25–40% higher sensitivity in detecting early-stage faults compared to FFT, wavelet, and traditional machine learning-based methods. In addition, the proposed DL plus FEM framework identified fault-induced variations of up to 25% in flux density, 25% in field intensity, and 40% in eddy current density, indicating its effectiveness as simulation-validated pathway for future real-time implementation and predictive maintenance applications.