Research On Fault Diagnosis Technology Of Rotating Machine Based On CNN

Rotating machine has a wide range of applications in the national economy.Rotating machine in high-end mechanical equipment will often cause great economic losses if there are some failures,and may even cause casualties.The research on fault diagnosis and online monitoring technology for rotating machine can quickly find faults and locate faults in advance,which is of great significance in engineering.With the rapid development of technology,especially automation and information technology,the degree of intelligence of aerospace,energy,manufacturing and other related mechanical equipment is changing with each passing day,and the complexity is also getting higher and higher.The diagnosis method is gradually no longer applicable to the fault diagnosis of current complex systems,so if you want to improve the reliability of complex systems,then under the current background of data-driven development of digitalization and intelligence,how to intelligently complex systems The study of fault diagnosis methods is of great significance."Feature extraction" and "pattern recognition" are two important complementary problems in fault diagnosis.Poor solution can affect the effect of fault diagnosis.Looking at the current method of feature extraction by staff through relevant prior knowledge greatly limits the accuracy and automation of fault diagnosis.In contrast,the deep learning method of the deep convolution model can better perform feature extraction,and can directly identify the signal,so it has good performance in "feature extraction" and "pattern recognition" of rotating machine fault signals.Application scenarios.The paper takes a series of Research on mechanical noise reduction methods and fault diagnosis of machinery,and a new type of noise reduction methods and fault diagnosis methods for rotating machinery are proposed.The main research contents of this article are summarized as follows:1.First,it introduces in detail some of the research carried out around the fault diagnosis technology of rotating machinery at home and abroad,and describes the current application of convolutional neural networks in mechanical fault diagnosis,and summarizes the fault diagnosis methods of convolutional neural networks to sum up.Based on this,the structural framework of this paper is proposed.2.Aiming at the problem that the noise in the original vibration signal of rotating machinery will reduce the fault diagnosis result,a new adaptive threshold de-noising method is designed.Taking the rolling bearing fault signal as an example,the EMD method,the improved EEMD method and the CEEMDAN method based on it are compared and analyzed using indicators such as completeness and root mean square error.Finally,based on the CEEMDAN method,the method for establishing the optimal noise reduction correlation model is proposed: first,the signal is decomposed into multiple IMFs by the CEEMDAN algorithm,and then the permutation entropy value of each IMF and the average permutation entropy value of all IMFs are calculated separately,and through wavelet Threshold denoising method is used to reduce the noise of IMFs whose signal is higher than the average permutation entropy.Finally,all IMFs are reconstructed to complete the adaptive denoising of the signal.3.As the convolutional neural network has the characteristics of automatic feature extraction and fault recognition,a new design structure LiCNN is adopted for the convolutional neural network in this paper,and the vibration signal in the time domain is correspondingly faulted based on this structure diagnosis.The diagnosis method is applied to the bearing data of CWRU.The results show that the convolutional neural network can not only automatically extract the signal features,but also the pattern recognition rate of the 16 failure modes on the CWRU bearing database has reached 100%.4.The electro-spindle is an important functional part of the CNC machine tool.Studying the fault diagnosis of the electro-spindle can find the potential failure of the electro-spindle to quickly locate and repair the fault.This is also for the improvement of the reliability of the electro-spindle and even the CNC machine tool.Quick repair provides technical support.Using the adaptive denoising method proposed in this paper and the convolutional neural network to carry out a fault diagnosis case for the vibration signal of the electric spindle,and a series of comparisons with traditional fault diagnosis methods,the results show that the diagnostic accuracy of the method proposed in this paper better than other methods.