| CNN extracted the signal characteristics layer by layer through the local perception of convolution kernel,but the rotation speed and sampling frequency of the vibration signal of rotating equipment are not the same.Extracting different signal features with fixed convolution kernel will affected the local feature perception and ultimately affected the learning effect and recognition accuracy.In order to solve this problem,the matching between the size of convolution kernel and the signal(rotation speed,sampling frequency)was optimized,and the matching relation was obtained.Through the study of this paper,the ability of extracting vibration features of CNN was improved,and the accuracy of vibration state recognition was finally improved.Firstly,a fault diagnosis model based on one-dimensional Convolution Neural Network(CNN)was established for one-dimensional bearing vibration signals.In this model,one-dimensional bearing vibration signal--Bently RK4 rotor test bench vibration signal as input.After that,the strong local perception and feature extraction ability of CNN is used to realize the fault feature extraction of the original one-dimensional vibration signal,and finally the state recognition and fault classification of the signal is realized.Secondly,For the nonlinear,complex components and high noise of wind turbine bearing signals affect fault features and state recognition.The vibration signals of different sampling frequencies and different speeds collected by the wind turbine gear power transmission fault simulation experimental platform were taken as the input of one-dimensional CNN,and the parameters of CNN were optimized according to the signals.Then feature extraction and state recognitio n are carried out for the signals with different sampling frequencies and different rotational speeds.Then,optimization experiments of convolution kernel matching are carried out for signals with different sampling frequencies and different rotating speeds.For the feature extraction and state recognition of CNN,the fundamental problem lies in whether the local perception of CNN can effectively reflect the specific characteristics of the signal.Therefore,this paper based on the vibration signal of the main two parameters: sampling frequency and equipment running speed.The convolution kernel matching optimization experiment based on the CNN model was carried out.By changing the size of the convolution kernel,the optimal percept ual field most suitable for the signal was found,and through the calculation of the optimal convolution kernel calculation formula.The fault identification accuracy of CNN model is improved to realize accurate,rapid and efficient fault diagnosis.Finally,this paper proposed the pre-fully connected deep CNN(PFC-CNN)method and based on this,established the bearing vibration state recognition model.the CNN’s front fully connected structure was design and the front fully connected layer was used to reduce the complexity of the signal and extracted the global characteristics of the signal.On this basis,the fusion feature learning of global features and local features(which got by CNN’s local perceptual field)was studied.and the back-propagation training of the whole network was studied,the discrimination degree and recognition accuracy of the corresponding features of different state categories were finally improved.Compared with other fault diagnosis methods,the results show that this method has higher fault identification r ate and certain advantages. |
