科研之友
微信
新浪微博
Facebook
分享链接
摘要
Currently, the volume of data from rotating machinery is undergoing explosive, exponential, big data-based growth, which also brings about the problem that the amount of labeled data is insufficient. Therefore, how to use large amounts of unlabeled data effectively in unsupervised learning to improve the accuracy of fault recognition is a challenging task in the field of fault diagnosis. In order to implement deep unsupervised self-adaptive recognition of fault conditions in rotating machinery, a brand new method of rolling bearing fault recognition based on deep fuzzy clustering neural networks (DFCNNs) is proposed. The proposed novel DFCNN model incorporates the high-level unsupervised feature extraction layer of deep belief networks (DBNs) into an improved adaptive non-parametric weighted-feature Gath-Geva (ANWGG) fuzzy clustering algorithm. The vibration spectrum signals from a rolling bearing are directly input into the DFCNN model to use DBN to extract multi-layer and unsupervised representative features of data, and ANWGG in DFCNN is then employed for unsupervised clustering. The validity of the proposed fault recognition method is verified in three cases of measured vibration signals from faulty bearings. Experimental results demonstrate that, compared to the existing fault diagnosis models, the new DFCNN model can not only overcome the restrictions of the traditional deep learning model (which is too dependent on labeled information for fault recognition), but is also beneficial in the implementation of unsupervised recognition of machinery fault conditions.
- 出版日期2018-12
- 单位东南大学
