Abstract:

Aiming at the problems of high bearing failure rate, high damage and imperceptibility in industrial production, this paper proposes a multi-level neural network fault diagnosis model that is integrated by improved perceptron, dynamic routing algorithm and stochastic optimization algorithm.Then, the fault features are extracted by the improved multi-layer perceptron, and the dynamic routing algorithm is used to predict and classify the fault features. The classification error is calculated from the loss function. In the error back propagation, the learning rate is filtered by the adaptive learning rate algorithm, and the network model is optimized by the random optimization algorithm（Adam） updating the weight. Finally, the numerical simulation experiment of bearing fault classification is carried out. The results show that the fault diagnosis model can diagnose and classify high-precision bearing faults.

Key words: bearing fault diagnosis, percetron;dynamic routing, neural networks

摘要：

针对工业生产中轴承故障发生率高、危害大、不易察觉等问题，提出一种由改进的感知器、动态路由算法和随机优化算法集成的多级神经网络故障诊断模型。通过随机等间隔无重复采样的方式对轴承振动信号数据库进行扩充，并根据故障类型做好对应标签；通过改进的多层感知器提取故障特征，由动态路由算法对所提取特征进行预测分类，进而由损失函数得出分类误差，在误差反向传播中由自适应学习速率算法筛选学习速率，并由随机优化算法（Adam）更新权值以优化网络模型。最后进行轴承故障分类的数值仿真实验，结果表明该故障诊断模型能实现高精度轴承故障诊断与分类。

关键词: 轴承故障诊断, 感知器, 动态路由, 神经网络