We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
基于多种卷积核特征提取自适应融合的 滚动轴承故障诊断方法.
- Authors
尚东方; 申浩; 王正
- Abstract
The traditional convolutional neural network model uses a single type of convolutional kernel, which faces problems such as insufficient feature extraction and low fault recognition rate when dealing with actual data under complex working conditions. Therefore, a fault diagnosis method for rolling bearings based on adaptive fusion of multiple convolutional kernel feature extraction (MCK-CNN) is proposed. Firstly, the one-dimensional vibration signal of the bearings is transformed into a two-dimensional time-frequency map through wavelet transform. The feature of time frequency map is initially extracted through a common feature extraction networkand parallelly processed through a conventional Convolution and Involutional network. Then, two networks composed of different convolutional kernels are used for feature extraction in different ways, and the two types of features are adaptively fused through CBAM attention module. Finally, the fused features are input into fully connected layer and the classification results are output through Softmax function. The experimental results from CWRU and laboratory bearing datasets show that the MCK-CNN model has high training efficiency and fault recognition rate.
- Publication
Bearing, 2023, Issue 11, p81
- ISSN
1000-3762
- Publication type
Article
- DOI
10.19533/j.issn1000-3762.2023.11.016