Investigation On Intelligent Diagnosis Method Of Critical Transmission Components Based On Data Driven

Transmission parts are the key components of mechanical equipment,and ensuring the reliable operation of transmission parts is of great significance to the safe production of mechanical equipment.Traditional fault diagnosis relies on manual experience,and decreases as the environment deteriorates.With the development of artificial intelligence,intelligent diagnosis has become a research hotspot in the field of fault diagnosis,but the following problems still exist: First,the healthy operation time of the equipment far exceeds the failure time,resulting in imbalance of the signal sample set,the model is easy to overfit,and the diagnostic performance is reduced.Therefore,how to carry out intelligent diagnosis under the condition of sample imbalance is worth studying.Second,intelligent diagnosis relies on a large amount of labeled data to carry out supervised learning.However,the cost of sample labeling is high,and a large number of samples are not labeled.Therefore,how to use unlabeled data sets to carry out fault diagnosis is worth studying.(1)Aiming at the problem of decreased accuracy of intelligent diagnosis when samples are imbalance,a fault diagnosis framework based on Wasserstein GAN(WGAN)is proposed.WGAN is used to learn the feature distribution of samples to expand the unbalanced sample set.The verification was carried out on the open bearing data set of Case Western Reserve University(CWRU)and the Machine Fault Simulator(MFS)bearing data set.A number of models that have excellent performance in processing imbalanced sample data sets are introduced for comparison.The results show that the expanded data set can help improve the performance of the diagnostic model.(2)Aiming at the problem that it is difficult to learn useful representations from large amounts of unlabeled data,a semi-supervised learning fault diagnosis framework based on Mutual Information Neural Estimation(MINE)and Deep Infomax(DIM)is proposed.MINE is used to solve the problem that high-dimensional feature mutual information is difficult to calculate.DIM uses mutual information constraints and structural constraints to unsupervised learning the representation of input samples,and finally uses a small number of labeled samples to achieve fault classification.(3)Aiming at the problem of reduced performance of intelligent diagnosis under complex working conditions,a fault diagnosis framework based on order holospectrum and Convolutional Neural Networks(CNN)is proposed.The accurate amplitude,frequency,and phase information are extracted through order analysis and spectrum correction,and then the holospectrum is used for information fusion to obtain more rotor state information,and finally combined with CNN for automatic pattern classification.The framework can greatly reduce manual dependence and improve the performance of mode diagnosis under complex working conditions.