Research On Intelligent Diagnosis Of Compound Bearing Fault Driven By Multi-label Decoupling

Rolling bearing is a vital part of mechanical equipment,and its operating condition is closely related to the normal operation of mechanical equipment.Fault diagnosis study of rolling bearings,early detection of fault information and maintenance measures are of great significance to ensure the safe production of mechanical equipment.However,in the actual industrial environments,different parts of the bearing are often damaged at the same time and influence each other,resulting in compound faults.The features of compound faults are easily affected by more serious faults,and weak faults are easily ignored,making diagnosis much more difficult.Traditional compound fault diagnosis methods rely on a great deal of professional knowledge and experience to design complex feature extraction processes.Most of them consider the compound fault as a new fault type,and it is difficult to identify the specific fault categories among them.The identification and decoupling of compound fault are a major challenge at present.At the same time,the data that can be collected during bearing operation is limited,and it is difficult to obtain massive data for experiments.Therefore,the problem of small sample compound fault diagnosis needs to be solved urgently.In addition,most studies use signals acquired by a single sensor,which contains incomplete feature information,leading to strong uncertainty in the diagnosis results.In this paper,the vibration signal of rolling bearing is studied based on deep learning method as the theoretical basis,and three compound fault intelligent diagnosis methods are proposed.First,combining convolutional neural network and multi-label learning,an end-to-end multi-label convolutional neural network model is proposed,which effectively achieves intelligent identification and decoupling of bearing compound faults without relying on any professional knowledge.Second,the capsule network is used for small sample compound fault diagnosis to solve the problem that convolutional neural network requires a large number of samples for training.And an improved capsule network is proposed by combining the squeeze-andexcitation block and the self-attention routing mechanism.The results show that this method can balance diagnostic accuracy and efficiency,with significant improvement in diagnostic accuracy under small sample conditions and strong robustness under strong noise environment and on different data sets.Third,a dual-channel capsule network is proposed on the basis of improved capsule network,which extracts the features collected by sensors in different directions through the residual module separately and achieves feature fusion through the squeeze-and-excitation block,effectively solving the problem of limited information collected by a single sensor and further improving the performance of compound fault diagnosis under small sample conditions.