基于深度置信网络的旋转机械在线故障诊断

On-line fault diagnosis of rotating machinery based on deep confidence network

摘要:

针对现有旋转机械在线故障诊断算法所存在的数据遍历耗时长,检测准确率低,故障分类准确率低等不足,提出一种基于深度置信网络的故障诊断算法。先基于受限的玻尔兹曼机搭建深度置信网络框架,利用数据标签在输入层和后端的受限玻尔兹曼机之间建立联系；然后利用k-means算法压缩聚类处理数据集降低数据集的规模和复杂度；最后在不同故障特征的分类诊断方面,引入加入核函数的SVM分类算法,提升对不同机械故障类型的分类精度。实验结果显示,提出的旋转机械故障在线诊断方案的迭代效率高,数据遍历耗时少,训练集和测试集 F1指标的分别为97.9%和97.4%,训练集和测试集结果相对于三种传统诊断算法分别提升了3.70%、4.37%、4.82,和3.95%、4.50%和4.50%。

Abstract:

Aiming at the shortcomings of existing online fault diagnosis algorithms for rotating machinery, such as long data traversal time, low detection accuracy and low fault classification accuracy, a fault diagnosis algorithm based on deep confidence network is proposed. Firstly, a deep confidence network framework is built based on the constrained Boltzmann machine, and the connection between the input layer and the back end of the constrained Boltzmann machine is established by using data labels. Then k-means algorithm is used to compress cluster processing data set to reduce the size and complexity of data set. Finally, in the classification and diagnosis of different fault characteristics, SVM classification algorithm with kernel function is introduced to improve the classification accuracy of different mechanical fault types. The experimental results show that the proposed online diagnosis scheme for rotating machinery faults has high iteration efficiency and less time consuming for data traversal. The F1 index of training set and test set is 97.9% and 97.4% respectively, and the results of training set and test set are improved by 3.70%, 4.37% and 4.82 respectively compared with the three traditional diagnosis algorithms. And 3.95%, 4.50% and 4.50%.