FIGURE 25 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Confusion matrices of different feature extraction approaches: (a) EMD-TSMRE; (b) WPT-TSMRE; (c) DTCWT-TSMRE; (d) DTCWPT-RE; (e) TSMRE applied to raw faulty data.
Source publication
Most existing fault diagnosis methods for rolling bearings are single-stage; these methods can only judge the fault type but cannot detect the existence of a fault. Moreover, the uncertainty in pattern recognition may lead to misclassification of healthy bearings as faulty ones. This paper proposes a multistage fault detection scheme for rolling be...
Citations
... One should consider the usability scenarios of the runtime optimization techniques presented in this paper when using the last-hour practical applications of wavelets , as those presented in [9] and [10]. ...
... Maximum values of the ratio between the energy of the 5-th node and the sum between the energies of the 5-th and 13th nodes when the 15-th and 17-th harmonics act jointly in a T7 topology.These figures represent the extreme values of the ratio E5/(E5+E13) where Ex represents the energy of the node x, in the case when the associated pair of harmonics (15-th and 17th) act jointly and their magnitudes are variable within the range[1,10]% of the polluted sine wave's Root Mean Square (RMS) value. Actually, the WPT transform should be used quite for the cases when certain harmonic orders change such as they cannot be considered ideal sine waves and decompositions like Fast Fourier Transform are unable to yield accurate results. ...
Wavelet packet trees represent a topic which grows in popularity when it comes to analysis of electrical waveforms. It allows for time-frequency analysis providing information on narrower ranges of frequency (as compared to the faster Discrete Wavelet Decomposition), but the computational resources are significantly greater than that involved in other types of wavelet-based analysis. In order to allow for this type of analysis to be usable in real-time applications, that is – to reduce the runtime, original algorithms were conceived and tested. In the first part of this work, previously implemented algorithms are briefly described, along with their pros and cons. Afterward, a new runtime optimization algorithm is proposed. Details on data structures, workflow, tests and study of errors are provided. This algorithm diminishes with up to 59% the runtime required by the application of the superposition theorem in order to evaluate the contribution of clustered harmonics using WPT.
Ensuring safe machine operation in industrial environments requires accurate bearing fault diagnosis. However, maintaining consistent data distribution between input training and test sets in practical online engineering tests based on depth models can be challenging. Inconsistencies in data distribution can hinder internal feature extraction and significantly affect the accuracy of online diagnosis. Additionally, different types of fault data are strongly nonlinearly coupled in space, requiring a deep model with powerful feature learning capabilities to reduce coupling effects between data. However, in practical engineering, deep models face difficulties in learning-rich fault features when few labeled training samples are available, leading to overfitting and poor generalization issues. To overcome these challenges, this study proposes an intelligent diagnosis method based on the Mel-scale frequency cepstrum coefficient (MFCC) and deep convolutional neural network with spatial adjacent region dropout (DCNN-SARD). Firstly, MFCC is utilized to extract sign fault features from vibration signals in different frequency bands, thus reducing the effect of inconsistent data distribution on fault diagnosis accuracy. Secondly, according to the different fault types, the obtained signature fault features are organized into one-dimensional vectors to construct block sample datasets. Finally, the fault features and input to the next convolution are enriched by creating a spatial descent region on the output feature map of the DCNN and randomly dropping the activation of neighboring kernels. This improves the generalizability of the model without increasing its structural complexity. The effectiveness of the proposed algorithm is experimentally validated on two common bearing datasets, and its reliability is verified on a large gas turbine bearing dataset.
Manufacturing systems are becoming more sophisticated and expensive, particularly with the development of the intelligent
industry. The complexity of the architecture and concept of Smart Manufacturing (SM) makes it vulnerable to several faults
and failures that impact the entire behavior of themanufacturing system. It is crucial to find and detect any potential anomalies
and faults as soon as possible because of the low tolerance for performance deterioration, productivity decline, and safety
issues. To overcome these issues, a variety of approaches exist in the literature. However, the multitude of techniques make
it difficult to choose the appropriate method in relation to a given context. This paper proposes a new architecture for a
conceptual model of intelligent fault diagnosis and self-healing for smart manufacturing systems. Based on this architecture,
a review method for the different approaches, sub-approaches and methods used to develop a Fault Detection and Diagnosis
(FDD) and Self-Healing-Fault-Tolerant (SH-FT) strategy dedicated to smart manufacturing is defined. Moreover, this paper
reviews and analyzes more than 256 scientific articles on fault diagnosis and self-healing approaches and their applications
in SM in the last decade. Finally, promising research directions in the field of resilient smart manufacturing are highlighted.
