Changqing Shen

Research on the Sparse Representation for Gearbox Compound Fault Features Using Wavelet Bases

The research on gearbox fault diagnosis has been gaining increasing attention in recent years, especially on single fault diagnosis. In engineering practices, there is always more than one fault in the gearbox, which is demonstrated as compound fault. Hence, it is equally important for gearbox compound fault diagnosis. Both bearing and gear faults in the gearbox tend to result in different kinds of transient impulse responses in the captured signal and thus it is necessary to propose a potential approach for compound fault diagnosis. Sparse representation is one of the effective methods for feature extraction from strong background noise. Therefore, sparse representation under wavelet bases for compound fault features extraction is developed in this paper. With the proposed method, the different transient features of both bearing and gear can be separated and extracted. Both the simulated study and the practical application in the gearbox with compound fault verify the effectiveness of the proposed method.

A hybrid technique based on convolutional neural network and support vector regression for intelligent diagnosis of rotating machinery

Rolling element bearings and gears are the most common machine elements. As they are extensively used in rotating machinery, their health conditions are crucial to the safe operation. The signals measured from rotating machines are usually affected by the working conditions and background noises. Thus, identifying faults from the mixed signals is a challenging and important task. Deep learning is initially developed for image recognition. Recently, it has attracted increasing attention in machinery fault diagnosis research. However, the generalization ability of the default classifier of it is not very satisfying. Thus, combining the feature learning ability of deep learning and the existing classifiers with satisfactory generalization ability is necessary. In this article, a hybrid technique based on convolutional neural network and support vector regression is proposed. The former part is used to promote feature extraction capability, and the latter part is used for multi-class classification. The efficiency of the proposed scheme is validated using the real acoustic signals measured from locomotive bearings and vibration signals measured from the automobile transmission gearbox. Results confirm that the method proposed is able to capture fault characteristics from the raw data, and both bearing faults and gear faults can be detected successfully.

A parameterized Doppler distorted matching model for periodic fault identification in locomotive bearing

The locomotive bearings support the whole weight of the train under the high speed and its continuous running is a key factor of the train safety. The collected Doppler distorted signal greatly increases the difficulties in detecting the whelmed fault information. To overcome this disadvantage, a novel Doppler distorted correlation matching model using the single side Laplace wavelet and acoustic theories is constructed to recognize the bearing fault-related impulse intervals. The parameterized Doppler distorted model is assessed by the correlation coefficient with the bearing fault signal in waveform. The optimal Doppler distorted model, that is the Doppler distorted correlation matching model which matches the fault component in the acoustic signal could be used for identifying the fault impulse interval. Then, the bearing fault can be successfully detected from the parameters of the Doppler distorted correlation matching model due to its maximum similarity to the real signal. Except for the simulation study, the proposed model is also employed to match the fault components in some real acoustic bearing fault signals. The recognized fault impulse period in the Doppler distorted correlation matching model’s initial model keep in good accordance with the correct fault impacts, which represent the locomotive bearing fault characteristics.

Adaptive deep feature learning network with Nesterov momentum and its application to rotating machinery fault diagnosis

Abstract The effective fault diagnosis of rotating machinery is critical to ensure the continuous operation of equipment and is more economical than scheduled maintenance. Traditional signal processing-based and artificial intelligence-based methods, such as wavelet packet transform and support vector machine, have been proved effective in fault diagnosis of rotating machinery, which prevents unexpected machine breakdowns due to the failure of significant components. However, these methods have several disadvantages that make them unable to automatically and effectively extract valid fault features for the effective fault diagnosis of rotating machinery. A novel adaptive learning rate deep belief network combined with Nesterov momentum is developed in this study for rotating machinery fault diagnosis. Nesterov momentum is adopted to replace traditional momentum to enable declining in advance and to improve training performance. Then, an individual adaptive learning rate method is used to select a suitable step length for accelerating descent. To confirm the utility of the proposed deep learning network architecture, two examinations are implemented on datasets from gearbox and locomotive bearing test rigs. Results indicate that the method achieves impressive performance in fault pattern recognition. Comparisons with existing methods are also conducted to demonstrate that the proposed method is more accurate and robust.

An automatic and robust features learning method for rotating machinery fault diagnosis based on contractive autoencoder

Abstract Fault diagnosis of rotating machinery is vital to improve the security and reliability as well as avoid serious accidents. For instance, robust fault features are crucial to achieve a high diagnosis precision. However, traditional feature extraction methods rely on an abundant amount of expertise and human interference. As a breakthrough in fault diagnosis, deep learning holds the potential to automatically extract discriminative features without much prior knowledge and human interference. However, only a few deep learning models are designed to deal with noise and extract robust features. Contractive autoencoder (CAE) is a potential tool to grasp the internal factors and directly obtain the hidden robust features by penalizing the Frobenius norm of the Jacobian matrix of the hidden features with respect to the inputs. Thus, this paper proposes a method based on stacked CAE for automatic robust features extraction and fault diagnosis of rotating machinery. Gearbox and bearing fault diagnosis experiments are conducted, and the testing accuracy of the proposed method is approximately 100% for both two cases and higher than that of other methods, which fully validates the effectiveness and superiority of the proposed method. In addition, experiments and correlation analysis under different signal-to-noise ratios (SNRs) are conducted. Results show that the diagnosis accuracies of the proposed method are higher than those of the stacked autoencoder (AE) network under each SNR, especially when under 0 dB, the testing accuracies of the proposed method are 4.14% and 5.88% higher than those of the stacked AE network in two case studies, and the correlation coefficients of the CAE are higher than those of the AE, which demonstrate the capability of CAE in mining more robust features compared to the regular AE automatically and the superiority of the proposed method in fault diagnosis.

Hierarchical diagnosis network based on sparse deep neural networks and its application in bearing fault diagnosis

As a breakthrough in artificial intelligence, deep learning allows for the automatic extraction of features without considerable prior knowledge and the determination of the complex non-linear relationship of the input parameters. Owing to these advantages, deep neural networks (DNNs) are superior to traditional artificial neural networks with shallow architectures, and are thus becoming widely used in the fault diagnosis field. To overcome the difficulties in bearing fault type classification and severity diagnosis, a two-layer hierarchical fault diagnosis network based on sparse DNNs is presented in this paper. In the proposed hierarchical fault diagnosis network, the first layer is responsible for fault type identification, and the second layer is responsible for fault severity diagnosis. An autoregressive (AR) model is established using empirical mode decomposition (EMD) to obtain the AR parameters as the input vectors of the proposed diagnosis network. The AR parameters are regarded as the low-level features which are helpful for the diagnosis network mine more useful high-level features. Experiments and comparison analyses are conducted to verify the performance of the proposed hierarchical fault diagnosis network. Results fully demonstrate the superiority of the proposed hierarchical fault diagnosis network in fault type classification or fault severity diagnosis.

A self-adaptive deep belief network with Nesterov momentum for the fault diagnosis of rolling element bearings

Effective fault diagnosis of rotating machinery helps prevent unexpected machine breakdowns resulting from the failure of essential components. Traditional artificial intelligence methods, such as artificial neural networks and support vector machine, have been proved to be effective in fault identification. However, extracting features manually requires a high degree of expertise in signal processing. Deep belief network (DBN) has gained popularity as a new method for machine learning because of its potential merits such as its capability to extract effective features automatically in fault diagnosis. Therefore, a novel adaptive learning rate DBN with Nesterov momentum is proposed in this study for the fault diagnosis of rolling element bearings. An experiment is conducted using a dataset of bearing health states obtained from a test rig to substantiate the utility of the proposed DBN architecture. Results show that the proposed method demonstrates impressive performance in fault pattern recognition. Comparison analyses are further conducted to demonstrate that the advanced method performs better than current methods.

Sparse representation of gearbox compound fault features by combining Majorization-Minimization algorithm and wavelet bases

Bearing and gear are essential components in gearbox, which is easily damaged and breaks down. Thus gearbox compound fault diagnosis has become a challenging topic in recent decades. Both bearing and gear faults in the gearbox tend to result in different transient impulse responses in the vibration signal and therefore it is necessary to present a method whose main task is to extract the different fault features. In this paper, a sparse representation method combining majorization minimization (MM) algorithm and different wavelet bases is proposed to resolve the problem. Through the proposed method, different transients buried in the noisy signal can be converted into sparse coefficients. Both the simulation study and the practical application show the proposed method is effective in extracting gearbox compound fault features.

Extraction of transients via the combination of wavelet basis and majorization minimization algorithm

Ill-posed linear inverse problems (IPLIP), such as restoration and reconstruction, are core topics of transient feature extraction in cyclostationary signal processing. This paper presents a novel method for extracting transients through exploiting both the wavelet basis and the majorization minimization algorithm. To solve the objective function, majorization minimization (MM) algorithm is applied with a designed strictly convex quadratic function. To implement the sparsity of the results, an optimal basis with its atom selected by correlation filtering is proposed. Through an iterative optimization procedure, transients hidden in the noisy signal can be converted into sparse coefficients. Simulated study concerning cyclic transient signal shows the effectiveness of this method. Applications on a rotating machine test rig with bearing fault also demonstrate the validity of the technique. Both the simulated study and the applications show that the proposed method is powerful in the representation of transients and is an effective tool to extract the transient features of bearings.