Research On The Rolling Bearing Fault Recognition Algorithm Based On Integrated Deep Neural Network

Rolling bearings can reduce the friction between moving parts and make the machine run effectively.It has become an important part of rotating machinery.According to statistics,the failure of rolling bearings is an important factor leading to the failure of rotating machinery.How to accurately and timely identify the faults of rolling bearings has become an important research direction in the field of fault detection.With the rapid development of the Industrial Internet of Things,fault diagnosis has entered the era of big data,and the continuous improvement of artificial intelligence technology has provided a technical approach for bearing fault identification.In this paper,deep neural network is used to study the pattern recognition of bearing faults in a big data environment.Firstly,the characteristics of bearing failures were analyzed.The frequency domain analysis,envelope spectrum analysis and wavelet analysis were used to process the rolling bearing failure signals.The bearing failure information was characterized from multiple angles,and various aspects of bearing failure were extracted feature.The weight initialization method of deep neural network is studied.A sample-based feedforward deep neural network weight initialization method is proposed.The initial weight set by this method is random and related to samples.The proposed algorithm was tested using Case Western Reserve University(CWRU)bearing data and compared with random initialization of weights.The results show that the initial method of weights can accelerate the training of deep feedforward neural networks to a certain extent.The intelligent identification method of bearing faults is studied,and a rolling bearing fault identification algorithm based on integrated deep neural network and correlation coefficient is proposed.The algorithm uses three deep neural networks(DNNs)to identify bearing faults from the frequency domain,wavelet domain,and envelope spectrum perspectives,uses correlation coefficient analysis to evaluate the recognition results of the three DNNs,and uses Fusion processing to improve the reliability of recognition.Using the CWRU bearing data,the proposed fault recognition algorithm was tested with homologous data and non-homologous data,respectively.The test results show that the proposed algorithm has better fault recognition accuracy than a single DNN,whether it is homogeneous data ornon-homogeneous data,which indicates that the method has a strong generalization ability.The research in this paper provides a certain technical support for the intelligent diagnosis of bearing faults.