基于KD-DenseNet的旋转机械故障诊断模型

针对传统旋转机械故障诊断算法在复杂多变的工况下,缺乏良好的自适应与泛化性的问题。提出了基于DenseNet的卷积核dropout(KD)智能故障诊断模型KD-DenseNet。将各类故障状态的原始振动信号进行重叠分段预处理,并将预处理得到的数据作为KD-DenseNet的输入进行训练,使用训练得到的模型对不同工况下的振动信号进行自适应特征提取与分类,并将dropout应用于卷积核中以提高模型对振动信号的处理速度及抗干扰性,最终得到故障类型判定结果。KD-DenseNet的应用避免了梯度弥散现象,提高了有效特征的提取效率,解决了传统特征提取方法中无法有效挖掘特征、无法自适应于任务进行调整等问题。

A fault diagnosis model of rotating machinery based on KD-DenseNet

The traditional fault diagnosis algorithm lacks good adaptability and generalization due to the complex and changeable working conditions of rotating machinery. To solve the problems, an intelligent fault diagnosis model KD-DenseNet based on kernel dropout (KD) is proposed. Firstly, preprocessing the original vibration signals of various fault states by overlapping and segmenting. Then, the pre-processed data are trained as input of KD-DenseNet, at the same time, dropout is applied to kernel to improve the processing speed and anti-interference ability of the model for vibration signals. Finally, the result of fault type determination is obtained. The application of KD-DenseNet avoids the gradient dispersion phenomenon, improves the extraction efficiency of effective features, and solves the problems that traditional feature extraction methods can not mine features and adapt to task adjustment effectively.