Research On Fault Identification Method Of Rolling Bearing Based On Deep Belief Network

With the development of technology,industrial equipment is moving in the direction of increasing speed,precision and efficiency.In order to ensure the normal operation of these devices,the fault monitoring system needs to collect equipment data to reflect the health status of the devices and the traditional artificial fault diagnosis technology gradually unable to adapt to changes in equipment.Bearing as an important component of industrial equipment,about 30%of industrial equipment failure is caused by bearing failure.And bearing information with diverse working conditions,complex conditions,non-linear and non-stationary characteristics,research and use of advanced theories and methods to mine information from the bearing data to efficiently and accurately identify the operation of equipment has become a new problem in the field of mechanical equipment fault diagnosis and monitoring.As a new field in machine learning research,deep learning shows great performance in handling images,sounds and texts with powerful non-linear expressions.Deep belief network,as a model of deep learning,has powerful characterization ability in feature recognition,classification and non-linear mapping.Therefore,in this paper,based on the complexity,nonlinearity and non-stationarity of the vibration signal of the rolling bearing,and the complex and difficult training of the deep neural network in the existing research,the paper studies the fault diagnosis of the rolling bearing based on the deep belief network.Based on the deep learning theory,a method of fault diagnosis for rolling bearing based on deep belief network is proposed by adopting the method of extracting features and building deep belief network.First of all,in order to reduce the impact of noise on the diagnostic results,simplify the structure of the deep belief network and improve the diagnostic efficiency,the signal characteristics are used to replace the original signal as the input of the deep belief network.In order to extract more rolling bearing fault characteristics and describe the real-time situation of bearings in a comprehensive and accurate way,the characteristics of time domain,frequency domain and time-frequency domain are extracted from vibration signals,where the time domain features include maximum,average,peak-to-peak,root mean square,kurtosis,waveform factor,peak factor,kurtosis factor,pulse factor and margin factor,frequency domain features include center frequency and frequency variance,and the features of time-frequency domain is the relative value of the energy of each frequency band of the vibration signal after wavelet packet decomposition.Then elaborated the principle of deep belief network and the classification ability based on deep belief network is studied.Mainly studies the influence of different neural network layers and the number of iterations on the classification ability of DBN,and obtains the ideal layer number and the number of iterations,which provides the basis for the follow-up study.Then the fault diagnosis upper monitor is designed based on MFC.Finally,in order to verify the fault diagnosis ability of deep belief network,the comparison between single hidden layer BP neural network and multiple hidden layer BP neural networks is made.Furthermore,the training error of deep belief network is further analyzed,and it is verified that deep belief network has better fault diagnosis ability and efficiency.Experiments on the Case Western Reserve University bearing dataset show that the method proposed in this paper has a fault recognition rate of 100%,indicate that the deep belief network has better classification performance and can diagnose faults from many fault types of signals and has higher fault recognition accuracy,and is suitable for dealing with the complexity,non-linearity and non-stationary signal,can offer the convenience for trouble detection.