This paper proposes a novel intelligent fault diagnosis method to automatically identify different health conditions of wind turbine (WT) gearbox. Unlike traditional approaches, where feature extraction and classification are separately designed and performed, this paper aims to automatically learn effective fault features directly from raw vibration signals while classify the type of faults in a single framework, thus providing an end-to-end learning-based fault diagnosis system for WT gearbox without additional signal processing and diagnostic expertise. Considering the multiscale characteristics inherent in vibration signals of a gearbox, a new multiscale convolutional neural network (MSCNN) architecture is proposed to perform multiscale feature extraction and classification simultaneously. The proposed MSCNN incorporates multiscale learning into the traditional CNN architecture, which has two merits: 1) high-level fault features can be effectively learned by the hierarchical learning structure with multiple pairs of convolutional and pooling layers; and 2) multiscale learning scheme can capture complementary and rich diagnosis information at different scales. This greatly improves the feature learning ability and enables better diagnosis performance. The proposed MSCNN approach is evaluated through experiments on a WT gearbox test rig. Experimental results and comprehensive comparison analysis with respect to the traditional CNN and traditional multiscale feature extractors have demonstrated the superiority of the proposed method.

Multiscale Convolutional Neural Networks for Fault Diagnosis of Wind Turbine Gearbox

Failure diagnosis using deep belief learning based health state classification

A review on data-driven fault severity assessment in rolling bearings

A novel bearing fault diagnosis model integrated permutation entropy, ensemble empirical mode decomposition and optimized SVM

One of the most powerful tools to assess the dynamical characteristics of time series is entropy. Sample entropy (SE), though powerful, is not fast enough, especially for long signals. Permutation entropy (PE), as a broadly used irregularity indicator, considers only the order of the amplitude values and hence some information regarding the amplitudes may be discarded. To tackle these problems, we introduce a new method, termed dispersion entropy (DE), to quantify the regularity of time series. We gain insight into the dependency of DE on several straightforward signal-processing concepts via a set of synthetic time series. The results show that DE, unlike PE, can detect the noise bandwidth and simultaneous frequency and amplitude change. We also employ DE to three publicly available real datasets. The simulations on real-valued signals show that the DE method considerably outperforms PE to discriminate different groups of each dataset. In addition, the computation time of DE is significantly less than that of SE and PE.

Dispersion Entropy: A Measure for Time-Series Analysis

A bearing fault diagnosis method has been proposed based on multi-scale entropy (MSE) and adaptive neuro-fuzzy inference system (ANFIS), in order to tackle the nonlinearity existing in bearing vibration as well as the uncertainty inherent in the diagnostic information. MSE refers to the calculation of entropies (e.g. appropriate entropy, sample entropy) across a sequence of scales, which takes into account not only the dynamic nonlinearity but also the interaction and coupling effects between mechanical components, thus providing much more information regarding machinery operating condition in comparison with traditional single scale-based entropy. ANFIS can benefit from the decision-making under uncertainty enabled by fuzzy logic as well as from learning and adaptation that neural networks provide. In this study, MSE and ANFIS are employed for feature extraction and fault recognition, respectively. Experiments were conducted on electrical motor bearings with three different fault categories and several levels of fault severity. The experimental results indicate that the proposed approach cannot only reliably discriminate among different fault categories, but identify the level of fault severity. Thus, the proposed approach has possibility for bearing incipient fault diagnosis.

Bearing fault diagnosis using multi-scale entropy and adaptive neuro-fuzzy inference

Feature extraction from vibration signals has been investigated extensively over the past decades as a key issue in machine condition monitoring and fault diagnosis. Most existing methods, however, assume a linear model of the underlying dynamics. In this study, the feasibility of devoting nonlinear dynamic parameters to characterizing bearing vibrations is studied. Firstly, fuzzy sample entropy (FSampEn) is formulated by defining a fuzzy membership function with clear physical meaning. Secondly, inspired by the multiscale sample entropy (multiscale SampEn) which is originally proposed to quantify the complexity of physiological time series, we placed approximate entropy (ApEn), fuzzy approximate entropy (FApEn) and the proposed FSampEn into the same multiscale framework. This led to the developments of multiscale ApEn, multiscale FApEn and multiscale FSampEn. Finally, all four multiscale entropies along with their single-scale counterparts were employed to extract discriminating features from bearing vibration signals, and their classification performance was evaluated using support vector machines (SVMs). Experimental results demonstrated that all four multiscale entropies outperformed single-scale ones, whilst multiscale FSampEn was superior to other multiscale methods, especially when analyzed signals were contaminated by heavy noise. Comparisons with statistical features in time domain also support the use of multiscale FSampEn.

A comparative study on ApEn, SampEn and their fuzzy counterparts in a multiscale framework for feature extraction *

Sample entropy (SampEn) has been applied in many literatures as a statistical feature to describe the regularity of a time series. However, as components of mechanical system usually interact and couple with each other, SampEn may cause inaccurate or incomplete description of a mechanical vibration signal due to the fact that SampEn is calculated at only one single scale. In this paper, a new method, named multiscale entropy (MSE), taking into account multiple time scales, was introduced for feature extraction from fault vibration signal. MSE in tandem with support vector machines (SVMs) constitutes the proposed intelligent fault diagnosis method. Details on the parameter selection of SVMs were discussed. In addition, performances between SVMs and artificial neural networks (ANNs) were compared. Experiment results verified the proposed model.

An Intelligent Fault Diagnosis Method Based on Multiscale Entropy and SVMs

A parametric time-frequency representation is presented based on time-varying autoregressive model (TVAR), followed by applications to non-stationary vibration signal processing. The identification of time-varying model coefficients and the determination of model order, are addressed by means of neural networks and genetic algorithms, respectively. Firstly, a simulated signal which mimic the rotor vibration during run-up stages was processed for a comparative study on TVAR and other non-parametric time-frequency representations such as Short Time Fourier Transform, Continuous Wavelet Transform, Empirical Mode Decomposition, Wigner-Ville Distribution and Choi-Williams Distribution, in terms of their resolutions, accuracy, cross term suppression as well as noise resistance. Secondly, TVAR was applied to analyse non-stationary vibration signals collected from a rotor test rig during run-up stages, with an aim to extract fault symptoms under non-stationary operating conditions. Simulation and experimental results demonstrate that TVAR is an effective solution to non-stationary signal analysis and has strong capability in signal time-frequency feature extraction.

/pdf/time-frequency-representation-based-on-time-varying-1549erdxhr.pdf

Hesheng Liu

Papers

Bearing fault diagnosis using multi-scale entropy and adaptive neuro-fuzzy inference

A comparative study on ApEn, SampEn and their fuzzy counterparts in a multiscale framework for feature extraction *

An Intelligent Fault Diagnosis Method Based on Multiscale Entropy and SVMs

Time-frequency representation based on time-varying autoregressive model with applications to non-stationary rotor vibration analysis

Cooperative Positioning Enhancement for HDVs and CAVs Coexisting Environment Using Deep Neural Networks