Scholarlink

<p>Rolling ball bearings are an essential part of rotating machines that can be affected by various faults, increasing maintenance costs and downtime. The paper presents a comparative study of a knowledge-based approach, including Artificial Neural Network (ANN) and multiscale Convolutional Neural Network (CNN) LeNet-5 architecture, including 1D-CNN and 2D-CNN, aimed to detect and diagnose the failures in components of the bearings at an early stage. The models are trained using nonlinear vibration signals that were recorded from a laboratory experimental setup on Machine Fault Simulator (MFS), focusing on ball bearings with defects in inner race (IR), outer race (OR) and ball (B) itself. The classification accuracy of the models is significantly enhanced through the utilisation of deep features extracted by neural networks employing transfer learning models compared to traditional ANN approaches. A comparative analysis, including neighborhood component analysis, evaluates the performance of ANN 1D CNN and 2D CNN models under various fault conditions. Overall, the result of this paper introduces a promising approach for both efficient and accurate fault detection and severity estimation in bearings of induction motors, potentially reducing the need for extensive manpower and sensor usage.</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 multi-scale deep neural networks for early fault diagnosis in rolling ball bearings

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Rajeev Kumar,
  • R. S. Anand

摘要

Rolling ball bearings are an essential part of rotating machines that can be affected by various faults, increasing maintenance costs and downtime. The paper presents a comparative study of a knowledge-based approach, including Artificial Neural Network (ANN) and multiscale Convolutional Neural Network (CNN) LeNet-5 architecture, including 1D-CNN and 2D-CNN, aimed to detect and diagnose the failures in components of the bearings at an early stage. The models are trained using nonlinear vibration signals that were recorded from a laboratory experimental setup on Machine Fault Simulator (MFS), focusing on ball bearings with defects in inner race (IR), outer race (OR) and ball (B) itself. The classification accuracy of the models is significantly enhanced through the utilisation of deep features extracted by neural networks employing transfer learning models compared to traditional ANN approaches. A comparative analysis, including neighborhood component analysis, evaluates the performance of ANN 1D CNN and 2D CNN models under various fault conditions. Overall, the result of this paper introduces a promising approach for both efficient and accurate fault detection and severity estimation in bearings of induction motors, potentially reducing the need for extensive manpower and sensor usage.