模糊粒化非监督学习结合随机森林融合的旋转机械故障诊断

Fault Diagnosis of Rotating Machinery in Combination with Unsupervised Learning of Fuzzy Granulation and Random Forest Fusion

在旋转机械的智能故障诊断中,复杂网络结构的非监督学习方法调节参数多,训练时间长,而结构简单的网络诊断准确率不够理想。针对以上问题,采用模糊信息粒化和稀疏自编码器搭建并行结构的学习网络,并行结构的稀疏自编码器同时对粒化后重新构成的多个有效参量信息自适应的进行特征提取,随后使用随机森林方法对提取的特征进行融合分类。实验结果表明该方法可以有效实现高精度故障诊断;且与常用的串行多网络处理结构相比,降低了网络参数调节的复杂度和多层网络的前后影响,并且提高了诊断精度,减少了训练时间。

In the intelligent fault diagnosis of rotating machinery,the unsupervised learning with the complex network structure has some problems,such as too large parameters,long training time and not satisfactory diagnosis accuracy with the simple learning structure.Aiming at the above problems,this paper uses fuzzy information granulation and sparse auto-encoder to construct a parallel learning network.At the same time,the sparse auto-encoder of the parallel structure adaptively extracts the features of multiple parameters reconstructed after granulation,and then the extracted features are fused and classified with random forest.Experimental results show that this method can identify faults with high accuracy.Compared with the commonly used methods with the serial network structure,the method not only reduces the network parameter adjustment complexity and the multi-layer network influence but also reduces the training time,while improving the diagnostic accuracy.