Research On Fault Diagnosis Of Rolling Bearing Based On Extreme Learning Machine And Deep Learning Theory

In recent years,the level of science and technology in China has been further improved.The development of information technology has led to the progress of industrialization,making mechanized production increasingly popular.Rolling bearings are the heart of mechanical systems.The failure of the bearing will inevitably lead to the breakdown of the mechanical system.It is of great significance to carry out condition monitoring and fault diagnosis of rolling bearings.The data used by traditional intelligent fault diagnosis methods are features manually extracted from the original data,and these features are used to train the model for fault diagnosis.With the development of new artificial intelligence methods such as deep learning,more and more deep learning methods are used in rolling bearing fault diagnosis.This paper combines deep learning theory with Extreme Learning Machines and proposes a deep learning method called Deep Extreme Learning Machines.At the same time,an experimental platform for rolling bearing fault diagnosis was established to study the effectiveness of deep limit learning machine in rolling bearing fault diagnosis.The main categories of this article are divided into the following three parts:1.This paper introduces a fault diagnosis algorithm of Deep Extreme Learning Machines.Deep Extreme Learning Machines combines the feature extraction ability of Autoencoder with the fast training ability of Extreme Learning Machines,which can greatly reduce the training time while maintaining high classification accuracy.At the same time,it can directly use the original data without feature extraction for training,avoiding the problem of manual selection of sensitive fault features.2.This paper proposes a Sparsity and Neighborhood Preserving Deep Extreme Learning Machines for rolling bearing fault diagnosis.Two manifold learning methods are introduced into the Extreme Learning Machines to improve the classification ability of the Deep Extreme Learning Machines.At the same time,the experiment platform of rolling bearing fault diagnosis is built.Aiming at the problem that vibration signal is difficult to extract in the actual production environment,current signal is used as the diagnosis data to study the effectiveness of current signal in fault diagnosis.The diagnosis results show that the proposed algorithm achieves 97.933% accuracy.3.Aiming at the problem of difficult to obtain data labels in the actual production environment,a Semi-supervised Deep Extreme Learning Machines fault diagnosis algorithm is proposed.The Semi-supervised Deep Extreme Learning Machines combines the layer-by-layer feature extraction capabilities of the Deep Extreme Learning Machines and the semi-supervised learning ability of the Semi-supervised Extreme Learning Machines.When there is less labeled data,the proposed method can make full use of the information contained in unlabeled data to train the network's classifier and improve the diagnostic accuracy.At the same time,the experimental platform of fan bearing fault diagnosis is built.Compared with other supervised and unsupervised methods,the experimental results show that the proposed method achieves 95.60% accuracy in 3.525 s training time,and has higher diagnostic accuracy and stability.