Research On Self-learning Fault Diagnosis Of Rotating Machinery

In the process of product production,there are often sudden failures of equipment,which may lead to some economic losses or serious casualties.However,these sudden faults are generally caused by the failure of early detection and long-term accumulation.Therefore,it is particularly important to establish the equipment fault diagnosis model with scientific methods so as to diagnose the running state of the equipment in time.In this paper,firstly,the development of mechanical fault diagnosis technology and the common methods of establishing fault diagnosis models at the present stage are thoroughly understood,and the common fault forms of rolling bearings and the characteristics of different fault forms in their vibration signals are analyzed.In addition,in order to obtain the actual vibration data of the rolling bearing,this project designed and built a set of experimental platform for fault diagnosis of rotating machinery.For feature extraction of bearing vibration signal,this paper introduces classification separability index as the criterion to evaluate the feature extraction effect.When the feature vectors extracted by the previous feature extraction methods are analyzed by category separability index,it is found that the classification separability of the feature vectors extracted by the previous methods is poor.Based on this,a vibration signal feature extraction method based on EEMD and improved SVD differential spectrum is proposed in this paper.By studying the relationship between the change of the singular value and the signal itself,the method of selecting the effective singular value is determined.The simulation experiment shows that,compared with the previous feature extraction method,this method reduces the average distance between feature classes extracted from vibration signals of different fault types by 44.87%,and increases the average distance between classes by 105.55%,which greatly improves the feature extraction effect of bearing vibration signals.For the establishment of fault diagnosis model,the previous fault diagnosis methods are limited to the known fault type,unable to realize the discrimination of the unknown fault type,and lack of self-learning function.Therefore,this paper proposes a self-learning fault diagnosis method based on Affinity Propagation clustering algorithm and Back propagation network.Experimental results show that this method can correctly identify the known type of fault and the unknown type of fault,and the recognition accuracy is up to 100%.Finally,the self-learning fault diagnosis method proposed in this paper is experimentally verified by using the bearing vibration data of the independently built experimental platform for fault diagnosis of rotating machinery.The experimental results show that the self-learning fault diagnosis method proposed in this paper can correctly identify known and unknown faults,and the model has self-learning function.