基于注意力机制的一维卷积神经网络行星齿轮箱故障诊断

针对行星齿轮箱故障信号成分复杂和时变性强的特点,提出了基于注意力机制的一维卷积神经网络(１D-CNN )行星齿轮箱故障诊断方法.首先,将行星齿轮箱各类故障状态的原始振动信号进行分段处理,作为模型的输入;其次,利用一维卷积神经网络对行星齿轮箱的原始振动信号学习齿轮故障特征,结合注意力机制( AM )对特征序列自适应的赋予不同的权重,增强故障特征信息;最后,利用 Softmax 分类器实现行星齿轮箱的故障诊断.通过故障实验验证以及与其他模型的对比,该故障诊断模型具有较强的学习能力,诊断性能优于其他的深度学习模型,有较好的工程实际意义.

Fault Diagnosis of Planetary Gearbox Based on One-dimensional Convolutional Neural Network with Attention Mechanism

Aiming at the characteristics of complex signal components and strong time-variability of planetary gearbox fault , a one-dimensional convolutional neuraln etwork fault diagnosis model for planetary gearbox based on attention mechanism was designed and implemented． Firstly , the originalvibration signals of planetary gearbox in various fault states are processed in sections as the input of the model． Secondly,a one-dimensional convolutional neural network was used to extract the fault features from the original vibration signals of planetary gears , and the Attention Mechanism ( AM ) is used to self-adaptively assign different weights to the feature sequences to enhance the fault feature information． Finally , the fault diagnosis of planetary gearbox is realized by Softmax classifier．Through verification by failure experiments and comparison with other models , it is found that the fault diagnosis model has a strong learning ability , and its diagnostic performance is better than other deep learning models , and has good engineering practical significance．