[신간의 별자리x] 우리/미술, 그리고 ‘슬픔의 박물관’

Applications of machine learning to machine fault diagnosis: A review and roadmap

Deep learning for smart manufacturing: Methods and applications

A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox

Currently, vibration analysis has been widely considered as an effective way to fulfill the fault diagnosis task of gearboxes in wind turbines (WTs) However, vibration signals are usually with abundant noise and characterized as nonlinearity and nonstationarity Therefore, it is quite challenging to extract robust and useful fault features from complex vibration signals to achieve an accurate and reliable diagnosis This paper proposes a novel feature representation learning approach, named stacked multilevel-denoising autoencoders (SMLDAEs), with the aim to learn robust and discriminative fault feature representations through a deep network architecture for diagnosis accuracy improvement In our proposed approach, we design an MLD training scheme, which uses multiple noise levels to train AEs It enables to learn more general and detailed fault feature patterns simultaneously at different scales from the complex frequency spectra of the raw vibration data, and therefore helps enhance the feature learning and fault diagnosis capability Furthermore, SMLDAE-based fault diagnosis is performed with an unsupervised representation learning procedure followed by a supervised fine-tuning process with label information for classification Our approach is evaluated by using the field vibration data collected from a self-designed WT gearbox test rig The results show that our proposed approach learned more robust and discriminative fault feature representations and achieved the best diagnosis accuracy compared with the traditional approaches

Stacked Multilevel-Denoising Autoencoders: A New Representation Learning Approach for Wind Turbine Gearbox Fault Diagnosis

The vibration signals of faulty rotating machinery are typically nonstationary, nonlinear, and mixed with abundant compounded background noise. To extract the potential excitations from the observed rotating machinery, signal demodulation and time–frequency analysis are indispensable. This work proposes a novel particle swarm optimization-based variational mode decomposition method, which adopts the minimum mean envelope entropy to optimize the parameters ( $\alpha$ and $K$ ) in the existing variational mode decomposition. The proposed fault-detection framework separated the observed vibration signals into a series of intrinsic modes. A certain number of the intrinsic modes are then selected by means of the Hilbert transform-based square envelope spectral kurtosis. Subsequently, in this study, the feature representations were reconstructed via the selected intrinsic modes; then, the envelope spectra of the real faulty conditions were generated in the rotating machinery. To verify the performance of the proposed method, a testbed platform of a gearbox with a combination of different faults was implemented. The experimental results demonstrated that the proposed method represented the patterns of the fault frequency more explicitly than the available empirical mode decomposition, the local mean decomposition, and the wavelet package transform method.

Novel Particle Swarm Optimization-Based Variational Mode Decomposition Method for the Fault Diagnosis of Complex Rotating Machinery

Reliable and quick response fault diagnosis is crucial for the wind turbine generator system (WTGS) to avoid unplanned interruption and to reduce the maintenance cost. However, the conditional data generated from WTGS operating in a tough environment is always dynamical and high-dimensional. To address these challenges, we propose a new fault diagnosis scheme which is composed of multiple extreme learning machines (ELM) in a hierarchical structure, where a forwarding list of ELM layers is concatenated and each of them is processed independently for its corresponding role. The framework enables both representational feature learning and fault classification. The multi-layered ELM based representational learning covers functions including data preprocessing, feature extraction and dimension reduction. An ELM based autoencoder is trained to generate a hidden layer output weight matrix, which is then used to transform the input dataset into a new feature representation. Compared with the traditional feature extraction methods which may empirically wipe off some "insignificant' feature information that in fact conveys certain undiscovered important knowledge, the introduced representational learning method could overcome the loss of information content. The computed output weight matrix projects the high dimensional input vector into a compressed and orthogonally weighted distribution. The last single layer of ELM is applied for fault classification. Unlike the greedy layer wise learning method adopted in back propagation based deep learning (DL), the proposed framework does not need iterative fine-tuning of parameters. To evaluate its experimental performance, comparison tests are carried out on a wind turbine generator simulator. The results show that the proposed diagnostic framework achieves the best performance among the compared approaches in terms of accuracy and efficiency in multiple faults detection of wind turbines.

Representational Learning for Fault Diagnosis of Wind Turbine Equipment: A Multi-Layered Extreme Learning Machines Approach

High-precision fault diagnosis is vital for widely used multistage gearbox systems. Intelligent monitoring is difficult due to the fuzzy boundaries and a variety of unseen single or simultaneous faults of such complex machinery. To solve this problem, local mean decomposition is applied to extract features effectively from the original nonstationary and nonlinear vibration signals. By exploiting the diverse functionalities of extreme learning machines (ELM) in both regression and classification, a novel dual-ELM network is proposed, in which one ELM is employed to count the number of faults and the other is used to identify the specific single- or simultaneous-fault scenarios. The proposed dual-ELM-based multilabel classifier does not rely on an empirically specified threshold. Thus, it is more self-adaptive than the existing probabilistic-based classifiers. In addition, by inheriting the advantages of the original ELM, the dual-ELMs do not require iterative fine-tuning of parameters. Finally, the training speed of the dual-ELMs is much faster than other combinations of the existing classifiers. Experimental results under various loading conditions show that the proposed dual-ELM-based fault diagnostic framework is versatile at detecting single and simultaneous faults accurately and quickly.

Single and Simultaneous Fault Diagnosis With Application to a Multistage Gearbox: A Versatile Dual-ELM Network Approach

The data-driven fault indicator for rotating machinery is designed to reveal the possible fault scenarios from the observed statistical vibration signals. This study develops a novel ensemble extreme learning machine (EELM) network to replace the conventional layout by combining binary classifiers (e.g., binary relevance) for compound-fault diagnosis of rotating machinery. The proposed EELMs consist of two sub-networks, namely, the first extreme learning machine (ELM) for clustering, and the second for multi-label classification. The first network generates the Euclidean distance representations from each point to every centroid with unsupervised clustering, and the second identifies potential output tags through multiple-output-node multi-label learning. Compared to the existing multi-label classifiers (e.g., multi-label radial basis function, rank support vector machine, back-propagation multi-label learning, and binary classifiers with binary relevance), the theoretical verification reveals EELMs perform the best in hamming loss, one-error, training time, and achieves the best overall evaluation for the two real-world databases (e.g., Yeast and Image). Regarding the real test for the compound-fault diagnosis of rotating machinery, this paper applies the particle swarm optimization-based variational mode decomposition to decompose the raw vibration signals into a series of intrinsic modes, and selects ten time-domain indicators and five frequency-domain statistical characteristics for feature extraction. The experimental results illustrate that the EELM-based fault diagnosis method achieves the best overall performance.

Xian-Bo Wang

Papers

Novel Particle Swarm Optimization-Based Variational Mode Decomposition Method for the Fault Diagnosis of Complex Rotating Machinery

Representational Learning for Fault Diagnosis of Wind Turbine Equipment: A Multi-Layered Extreme Learning Machines Approach

Single and Simultaneous Fault Diagnosis With Application to a Multistage Gearbox: A Versatile Dual-ELM Network Approach

Ensemble extreme learning machines for compound-fault diagnosis of rotating machinery

A novel fault diagnosis procedure based on improved symplectic geometry mode decomposition and optimized SVM