Bo She

Bio: Bo She is an academic researcher from Naval University of Engineering. The author has contributed to research in topics: Computer science & Wavelet. The author has an hindex of 1, co-authored 6 publications receiving 10 citations.

Rotating Machinery Remaining Useful Life Prediction Scheme Using Deep-Learning-Based Health Indicator and a New RVM

Abstract: Remaining useful life (RUL) prediction plays a significant role in developing the condition-based maintenance and improving the reliability and safety of machines. This paper proposes a remaining useful life prediction scheme combining deep-learning-based health indicator and a new relevance vector machine. First, both one-dimensional time-series information and two-dimensional time-frequency maps are input into a hybrid deep-learning structure network consisting of convolutional neural network (CNN) and long short-term memory network (LSTM) to construct health indicator (HI). Then, the prediction results and confidence interval are calculated by a new RVM enhanced by a polynomial regression model. The proposed method is verified by the public PRONOSTIA bearing datasets. Experimental results demonstrate the effectiveness of the proposed method in improving the prediction accuracy and analyzing the prediction uncertainty.

A hidden feature label propagation method based on deep convolution variational autoencoder for fault diagnosis

Abstract: Vibration signal of mechanical component usually exhibits non-linear and non-stationary characteristics, the key step of fault diagnosis is to extract discriminant features hidden in the vibration signal, in order to improve diagnostic performance and identify new fault class, a novel fault diagnosis method of hidden feature label propagation based on deep convolution variational autoencoder (HFLPDCVA) is proposed. Firstly, the raw vibration signal is transformed into frequency spectrum data by using fast Fourier transform as the input of the model. Secondly, the variational autoencoder is used to construct the convolutional neural network, and the non-fixed dropout parameter is fused to change the network to improve the identification of network hidden layer features. Finally, the label propagation algorithm is applied to process the hidden features of the full connection layer, and the local common features between the unknown label samples and the known label samples are extracted to predict the class of unknown label samples, including the new emerge fault class. The effectiveness of the proposed method is verified on bearing fault data sets under variable working conditions and damage data sets of self-priming centrifugal pump under a single working condition. Experimental results show that the proposed HFLPDCVA method can effectively extract the fault characteristics of the vibration signal, and it also has a significantly higher recognition accuracy rate than other typical deep learning methods and traditional classifiers.

Rolling bearing fault diagnosis method based on dual-tree complex wavelet pack manifold domain noise reduction

Abstract: The invention relates to a rolling bearing fault diagnosis method based on dual-tree complex wavelet pack manifold domain noise reduction The rolling bearing fault diagnosis method based on the dual-tree complex wavelet pack manifold domain noise reduction comprises steps of using an accelerated speed sensor to collect a vibration signal of the rolling bearing, performing dual-tree complex wavelet pack decomposition on the vibration signal, maintaining wavelet pack coefficients of first two nodes, performing threshold noise reduction on wavelet coefficients of the rest nodes, performing single branch reconstruction on the wavelet pack coefficient of each node to perform a high dimensional signal space, using a t distribution random neighbor embedding method to extract low a dimensional manifold, performing inverse reconstruction on the low-dimensional manifold to obtain a high-dimensional space main manifold, obtaining a signal after noise reduction, performing Hilbert envelope demodulation on the signal after noise reduction to obtain an envelope frequency spectrum of the vibration signal, and realizing fault diagnosis of the rolling bearing according to an inner ring fault characteristic frequency and an outer ring fault characteristic frequency of the rolling bearing, a rolling body fault characteristic frequency and a retainer fault characteristic frequency

Nonlinear Model for Condition Monitoring and Fault Detection Based on Nonlocal Kernel Orthogonal Preserving Embedding

Abstract: The dimension reduction methods have been proved powerful and practical to extract latent features in the signal for process monitoring. A linear dimension reduction method called nonlocal orthogonal preserving embedding (NLOPE) and its nonlinear form named nonlocal kernel orthogonal preserving embedding (NLKOPE) are proposed and applied for condition monitoring and fault detection. Different from kernel orthogonal neighborhood preserving embedding (KONPE) and kernel principal component analysis (KPCA), the NLOPE and NLKOPE models aim at preserving global and local data structures simultaneously by constructing a dual-objective optimization function. In order to adjust the trade-off between global and local data structures, a weighted parameter is introduced to balance the objective function. Compared with KONPE and KPCA, NLKOPE combines both the advantages of KONPE and KPCA, and NLKOPE is also more powerful in extracting potential useful features in nonlinear data set than NLOPE. For the purpose of condition monitoring and fault detection, monitoring statistics are constructed in feature space. Finally, three case studies on the gearbox and bearing test rig are carried out to demonstrate the effectiveness of the proposed nonlinear fault detection method.

Partial adversarial domain adaptation by dual-domain alignment for fault diagnosis of rotating machines.

Abstract: Domain adaptation (DA) techniques have succeeded in solving domain shift problem for fault diagnosis (FD), where the research assumption is that the target domain (TD) and source domain (SD) share identical label spaces. However, when the SD label spaces subsume the TD, heterogeneity occurs, which is a partial domain adaptation (PDA) problem. In this paper, we propose a dual-domain alignment approach for partial adversarial DA (DDA-PADA) for FD, including (1) traditional domain-adversarial neural network (DANN) modules (feature extractors, feature classifiers and a domain discriminator); (2) a SD alignment (SDA) module designed based on the feature alignment of SD extracted in two stages; and (3) a cross-domain alignment (CDA) module designed based on the feature alignment of SD and TD extracted in the second stage. Specifically, SDA and CDA are implemented by a unilateral feature alignment approach, which maintains the feature consistency of the SD and attempts to mitigate cross-domain variation by correcting the feature distribution of TD, achieving feature alignment from a dual-domain perspective. Thus, DDA-PADA can effectively align the SD and TD without affecting the feature distribution of SD. Experimental results obtained on two rotating mechanical datasets show that DDA-PADA exhibits satisfactory performance in handling PDA problems. The various analysis results validate the advantages of DDA-PADA.

Experimental Study on Multi-Domain Fault Features of AUV with Weak Thruster Fault

Abstract: As the most important device of an Autonomous Underwater Vehicle (AUV), thrusters are one of the main sources of fault. If the thruster fault can be diagnosed in the early stage, it would give more time to guarantee the safety of an AUV. Fault feature extraction is the premise of fault diagnosis. The traditional feature calculation methods extract fault features from one domain. These methods work well in the case of high fault severity, but poorly in the case of weak fault severity. In addition, for weak faults, the fault features extracted by the traditional methods may not meet the monotonic relationship with fault severity and cannot be used in fault severity identification. Aiming at these problems, through experimental data analysis, this paper excludes the features that do not meet the law from the 52 selectable fault features in the time domain, frequency domain and time-frequency domain. Aiming at the problem that there is no useful feature in the frequency domain, a new feature calculation method is proposed, and the order of magnitude of the available feature is given, which provides concise and accurate information for subsequent fault feature fusion and fault severity identification.

RenyiBS: Renyi entropy basis selection from wavelet packet decomposition tree for phonocardiogram classification

Abstract: Wavelet packet transform (WPT) is a powerful mathematical tool for analyzing nonlinear biomedical signals, such as phonocardiogram (PCG). WPT decomposes a PCG signal into a full binary tree of details and approximation coefficients. Appropriate nodes of the tree could be selected as a basis for generating features. Motivated by this, we propose the Renyi entropy basis selection (RenyiBS) method. In RenyiBS method, we use the Renyi entropy as an information measure to choose the best basis of the wavelet packet tree of PCG signals for feature selection and classification. The Renyi entropy estimates the spectral complexity of a signal, which is vital for characterizing nonlinear signals such as PCGs. After selecting the best basis, we define features on the coefficients of the selected nodes. Then, we classify PCGs using the support vector machine (SVM) classifier. In the simulation, we examine a set of 820 heart sound cycles, including normal heart sounds and three types of heart murmurs. The three murmurs examined include aortic regurgitation, mitral regurgitation, and aortic stenosis. We achieved the promising result of 99.74% accuracy, confirming the ability of Renyi entropy to select an appropriate basis of the wavelet packet tree and extracting the nonlinear behavior of particular heart sounds. Besides, the superiority of our proposed information measure in comparison with other information measures reported before is shown.