Abstract
With the development of modern industries, the working environment of rotating machinery has become increasingly complicated. Therefore, it is very meaningful to accurately identify the type of equipment failure under variable operating conditions. This paper presents a rotating machinery fault diagnosis method based on dynamic learning rate deep belief network (DBN) with adaptive structure (PSO-DDBN). Firstly, the wavelet packet energy entropy principle was used to obtain the characteristic matrix of the original data, and then the characteristics of the data under variable conditions were distinguished. Secondly, in order to adjust the structure of DBN, the loss function of DBN was used to construct the convergence function in particle swarm optimization (PSO) adaptive process. The dynamic learning rate strategy was applied to the training process of the network. The network gradient value in each iteration was recorded and the dynamic learning rate function was constructed to achieve the purpose of dynamically adjusting the network learning rate and making the network convergence faster and more stable. Then, the performance of PSO-DDBN was verified by the data of bearing and gearbox under variable conditions. Finally, other intelligent diagnosis algorithms were compared with this method, and the results showed that this method had better universality and fault classification ability.
Highlights
A S a real-time and efficient advanced monitoring and diagnosis technology, fault diagnosis technology [1] has been widely concerned by scholars in the industry, and a variety of fault diagnosis technologies have been proposed
Until 2006, Hinton et al First proposed the concept of deep belief network (DBN) and proposed unsupervised layer by layer training algorithm with the help of Restricted Boltzmann machines (RBM) to realize the effective expression of data features [18, 19]
In order to verify the effectiveness of the proposed method, this paper relied on the public data platform of Western Reserve University in the United States as shown in Figure 6 [34], and the QPZZ-II rotating machinery failure test platform as shown in Figure 7, to verify the proposed method.The test platform shown in Figure 7 is the data acquisition platform of our laboratory
Summary
A S a real-time and efficient advanced monitoring and diagnosis technology, fault diagnosis technology [1] has been widely concerned by scholars in the industry, and a variety of fault diagnosis technologies have been proposed. For nonlinear feature extraction methods, such as kernel principal component [9], kernel PLS [10], artificial neural network (ANN) [11, 12] , support vector machine (SVM) [13, 14] and stochastic resonance [15,16,17] are used to describe more complex data features. Kuremoto et al [26] used PSO to optimize the number of RBM neurons and network learning rate in the training process, and achieved good results in time series prediction. How to extract useful features from a large number of original data, establish a robust and stable model, and improve the classification performance is the key to fault diagnosis
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
