Mechanism Design And Control Of Flexible Bionic Tensioned Ankle Exoskeleton

Rolling element bearing is one of the most widely used mechanical parts in rotating equipment,and is also a high-risk part with frequent failures.Once the defect exists,it is easy to lead to abnormal operations of the whole mechanical equipment,even irreversible loss.Deep learning provides useful tools for intelligent fault diagnosis of bearings.However,in practical applications,the diagnostic performance of the model is influenced by some problems,such as lack of samples and large differences between training and test samples,which results in poor diagnosis accuracy and low generalization ability.Therefore,this thesis makes full use of deep learning and aims to study the qualitative and quantitative analysis methods for rolling bearings.The main research contents and innovations are summarized as follows:(1)The deep autoencoder(DAE)has excellent extraction capability of dimensional features,and use the unsupervised learning that can avoid the influence caused by the difficulty of acquiring enough labelled data,but it ignores the function of labels.Moreover,few fault samples also make the feature extraction of the DAE difficult,which is critical for accuract fault diagnosis.To solve these problems,an improved autoencoder based on deep clustering is proposed.The graph convolutional neural network is used to assist the DAE to obtain the class information,and the prototype network is used as a classifier with few samples to form an intelligent fault diagnosis model for the case of less faulty samples.Experimental results indicate that the proposed method has better diagnosis ability and anti-noise performance under varying working conditions(small domain shifts).(2)Considering that bearing data acquisition conditions vary greatly and fault knowledge of different equipment is difficult to transfer,a deep clustering graph convolutional network based on multi-adversarial learning is proposed to improve the generalization ability and diagnostic accuracy of the diagnosis model.In the deep clustering framework,the generative adversarial network(GAN)is introduced.On the one hand,it is used for stable model training to avoid model collapse caused by data differentiation.On the other hand,it promotes the diagnostic model to standardize the distribution of data in the same category while distinguishing different data.Furthermore,the domain adaptive network is used to construct the learning framework and shorten the distribution difference between training and test samples.Multiple experimental and actual bearing datasets are used to demonstrate the superiority of the proposed model.The results show that the proposed network can be used for accurate fault diagnosis of bearings with large domain shifts,such as cross-condition and crossequipment.(3)In order to ensure the safe operation of bearings and reduce the loss caused by failures,it is necessary to accurately assess the bearing performance degradation,and thus a multiscale stacked auto-encoder(MSAE)is proposed.In this method,the multidimensional information of bearing vibration time series is obtained by using a generative adversarial framework.The nested scatter plot(NSP)is designed to collect and characterize the bearing degradation.Then,the convolutional neural network is combined with the long and short-term memory network to transform the bearing information to some images and then generate the health indicator of bearings.After that,the degradation state analysis and remaining useful lifetime(RUL)prediction of bearings are realized.The entire life-cycle datasets are used to verify the effectiveness of the proposed method.The results show that it can use a small amount of bearing data to determine bearing health and accurately estimate the degradation and RUL of bearings,which provides a guarantee for the safe operation of bearings.