基于加权密集连接网络和注意力机制的滚动轴承故障诊断

Rolling bearing fault diagnosis based on weighted dense connection network and attention mechanism

摘要:

针对当前诊断方法对滚动轴承故障特征表征困难以及在噪声干扰大的环境中检测性能下降的问题，提出了一种基于加权密集连接网络和注意力机制的滚动轴承故障诊断的方法，该方法由特征提取和故障分类两部分组成。在特征提取部分，首先采用加权密集连接网络从轴承振动信号中提取特征，并将不同空间级别的特征进行组合以增强信息的多样性，然后利用注意力机制突出重要信息，获得准确的表征故障特征。故障分类模型以表征的特征信息作为输入，经过Softmax函数输出每种故障类型的诊断结果。实验结果表明，所提模型在加性噪声干扰的情况下具有良好的诊断性能，比其他方法更具优势。

Abstract:

Aiming at the difficulties in characterizing rolling bearing fault features with current diagnosis methods and the degradation of detection performance in a strong noisy environment, a method of rolling bearing fault diagnosis based on weighted dense connection network and attention mechanism is proposed. The method consists of feature extraction and fault classification. In the feature extraction part, firstly, the weighted dense connection network is used to extract features from the bearing vibration signal, and the features of different spatial levels are combined to enhance the diversity of information. Then, attention mechanism is used to highlight important information to obtain accurate fault features. The fault classification model takes the characteristic information as the input and outputs the diagnosis results of each fault type through softmax function. Experimental results show that the proposed model has good diagnostic performance in the case of additive noise interference, and has more advantages than other methods.