No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
A Deep Learning method for bearing fault diagnosis based on Cyclic Spectral Coherence and Convolutional Neural Networks
Abstract
Accurate fault diagnosis is critical to ensure the safe and reliable operation of rotating machinery. Data-driven fault diagnosis techniques based on Deep Learning (DL) have recently gained increasing attention due to theirs powerful feature learning capacity. However, one of the critical challenges lies in how to embed domain diagnosis knowledge into DL to obtain suitable features that correlate well with the health conditions and to generate better predictors. In this paper, a novel DL-based fault diagnosis method, based on 2D map representations of Cyclic Spectral Coherence (CSCoh) and Convolutional Neural Networks (CNN), is proposed to improve the recognition performance of rolling element bearing faults. Firstly, the 2D CSCoh maps of vibration signals are estimated by cyclic spectral analysis to provide bearing discriminative patterns for specific type of faults. The motivation for using CSCoh-based preprocessing scheme is that the valuable health condition information can be revealed by exploiting the second-order cyclostationary behavior of bearing vibration signals. Thus, the difficulty of feature learning in deep diagnosis model is reduced by leveraging domain-related diagnosis knowledge. Secondly, a CNN model is constructed to learn high-level feature representations and conduct fault classification. More specifically, Group Normalization (GN) is employed in CNN to normalize the feature maps of network, which can reduce the internal covariant shift induced by data distribution discrepancy. The proposed method is tested and evaluated on two experimental datasets, including data category imbalances and data collected under different operating conditions. Experimental results demonstrate that the proposed method can achieve high diagnosis accuracy under different datasets and present better generalization ability, compared to state of the art fault diagnosis techniques.
... Numerous papers have recently emphasized on diagnosing bearing faults using an AI classifier and new features extraction techniques [16][17][18][21][22][23][24][25][26]. Regardless, the effectiveness of the chosen procedures and features that provide more details about the type of defects determines how well the proposed procedure will perform. ...
... The comparison elements include, the bearing defect types, the defect severity, the motor speed, the extraction method, the used classifier and the accuracy of each work. The works carried out on the bearing defect classification focuses on the common defect types that are outer race, inner race, cage and ball defects; these works can be found in [18,[22][23][24][25][26][32][33][34][35] that proved their higher performances; but these researches are based on complex feature extraction procedures. For example, in [22] Attoui et al. proposed a long algorithm that includes four key steps: data acquisition, feature generation, feature selection, and defect classification. ...
... The researchers in [24] have applied a model, based on mathematical matrixes, called graph modeled singular values besides to Kernel Neural Network for bearing defects classification. Zhuyun Chen et al. [25] and Amrinder Singh Minhas et al. [26] have achieved a good performance of 98.93% and 98% respectively while using more simple feature extraction methodologies. ...
Finding a precise method for improved fault detection and classification when dealing with non-stationary vibration signals is the main goal of this paper. For the detection and classification of induction motor failures, a wavelet packet decomposition (WPD) associated to an artificial neural network (ANN) technique is considered. The effectiveness of this approach depends on the characteristics that have been carefully chosen and prepared to enable the classifier support the healthy conditions of the monitored system with the aid of the measured signal. Different testing data sets of healthy and defective bearings under various rotating speeds are studied to train the ANN classifier in order to demonstrate the effectiveness of the proposed method. The results showed the high performance of this procedure as an efficient method for bearing fault diagnosis.
... Hasan et al. [13] fused the multi-domain information of raw bearing vibration signals into a two-dimensional composite image, and then fed the composite image into a multi-task learning (MTL)-based CNN model for fault diagnostics, which was able to accurately detect faults in the presence of simultaneous changes in speed and health conditions. Chen et al. [14] used cyclic spectral analysis to construct a frequency domain graph as an input to CNN to reveal the hidden periodic behavior of each fault type in bearing vibration signals, which reduces the difficulties with feature learning in deep diagnostic models. Sobie et al. [15] sequentially performed envelope extraction, simultaneous angular domain averaging, and normalization of the bearing vibration signals before feeding them into a CNN, which allows the fault classification of experimental data with different shaft speeds and bearing geometries. ...
... The time-frequency transformed time-frequency map as a two-dimensional image is often used as input to CNN-based diagnostic models. Therefore, this paper compares the diagnostic effectiveness of STFT spectrograms, CWT spectrograms [22], VMD-HT spectrograms [23], MDFVI spectrograms [13], CSCoh spectrograms [14], and the proposed SES spectrograms as inputs. Each task is repeated 10 times, and the average diagnostic accuracy using the proposed SES-CNN method is shown in Table 3, and the results using other methods are also shown in Table 4. Task 6 is run once and its confusion matrix is shown in Figure 11, where the green font represents the number of correctly classified samples. ...
Intelligent fault diagnostics based on deep learning provides a favorable guarantee for the reliable operation of equipment, but a trained deep learning model generally has low prediction accuracy in cross-domain diagnostics. To solve this problem, a deep learning fault diagnosis method based on the reconstructed envelope spectrum is proposed to improve the ability of rolling bearing cross-domain fault diagnostics in this paper. First, based on the envelope spectrum morphology of rolling bearing failures, a standard envelope spectrum is constructed that reveals the unique characteristics of different bearing health states and eliminates the differences between domains due to different bearing speeds and bearing models. Then, a fault diagnosis model was constructed using a convolutional neural network to learn features and complete fault classification. Finally, using two publicly available bearing data sets and one bearing data set obtained by self-experimentation, the proposed method is applied to the data of the fault diagnostics of rolling bearings under different rotational speeds and different bearing types. The experimental results show that, compared with some popular feature extraction methods, the proposed method can achieve high diagnostic accuracy with data at different rotational speeds and different bearing types, and it is an effective method for solving the problem with cross-domain fault diagnostics for rolling bearings.
... They can adaptively extract the general feature representation from complex data and have been widely adopted in engineering practices [11]. For instance, Chen et al. [12] presented a cyclic spectral coherence and convolutional neural network (CSCohCNN) for bearing fault diagnosis, where the cyclic spectral coherence maps of vibration signals are calculated as a new type of input to improve the recognition performance. Kumar et al. [13] proposed a novel conventional neural network (NCNN) to diagnose bearing defects using transfer learning. ...
... To validate the effectiveness and superiority of the proposed MCRNets, five state-of-the-art intelligent diagnosis methods, including modified CNN (MCNN) [27]、CSCohCNN [12]、NCNN [13]、Bi-LSTM [1] and ResNet18 [28], are employed as comparison models. Besides, one self-made gearbox dataset and three open-source datasets are used to test the model performance, and all models are randomly executed 10 times on a computer platform consisting of an Intel Core i9-9900 K, NVIDIA GeForce RTX 2080Ti and 128G RAM. ...
... They can adaptively extract the general feature representation from complex data and have been widely adopted in engineering practices [11]. For instance, Chen et al. [12] presented a cyclic spectral coherence and convolutional neural network (CSCohCNN) for bearing fault diagnosis, where the cyclic spectral coherence maps of vibration signals are calculated as a new type of input to improve the recognition performance. Kumar et al. [13] proposed a novel conventional neural network (NCNN) to diagnose bearing defects using transfer learning. ...
... To validate the effectiveness and superiority of the proposed MCRNets, five state-of-the-art intelligent diagnosis methods, including modified CNN (MCNN) [27]、CSCohCNN [12]、NCNN [13]、Bi-LSTM [1] and ResNet18 [28], are employed as comparison models. Besides, one self-made gearbox dataset and three open-source datasets are used to test the model performance, and all models are randomly executed 10 times on a computer platform consisting of an Intel Core i9-9900 K, NVIDIA GeForce RTX 2080Ti and 128G RAM. ...
With the excellent capacity of feature representation and nonlinear mapping, deep learning with stacking deeply has aroused goer research interest in the field of intelligent fault diagnosis. However, under the case that mechanical failure signals, including gears, bearings, etc., essentially follow the excitation mechanism and modulation principle, an interpretable expression of deep learning architecture for intelligent diagnosis has been rarely discussed. Motivated by this issue, this study presents a novel interpretable multiplication convolution network (MCN), where three designed layers, including a feature separator, a feature extractor, and a classifier, are operated on spectrum samples input. Different from the conventional models, a series of multiplication filtering kernels (MFKs) are analytically designed to extract the differential modes from spectrum samples in an ex-ante interpretable way. The separated modes are stacked into a filtered mode map. A convolution layer is later used as the feature extractor to further abstract high-level feature representations. Finally, a dense decision layer is taken as the classifier for fault identification. Specially, to strengthen the sensing ability of MFKs, an anti-aliasing constraint is introduced to improve the information diversity of the separator. In essence, MCN operates in a novel framework collaborating signal processing with deep learning. Experimental results validate the effectiveness of the proposed MCN. Besides, feature map visualizations are further implemented to verify that the desired fault-sensitive modes in spectrum samples can be precisely mined, which provides the MCN with higher recognition accuracy and good ex-post interpretability. Benefiting from analytic kernel design, MCN has fewer model parameters as a lightweight efficient architecture, which shows enormous potential in the application of edge intelligent fault diagnosis. Related source codes can be available at: https://github.com/CQU-BITS/MCN-main.
... Therefore, it has been promptly studied in fault diagnosis [19][20][21][22][23][24]. Among these methods, CNN is prominent [25]; for example, a cyclic spectral-coherence-based CNN can achieve high diagnosis accuracy and better generalization ability by applying domain-related techniques to reduce the difficulty of feature learning and obtain high-level feature representations [26]. A common CNN with Gramian noise reduction can improve the performance of denoising and classification for bearing fault diagnosis [27]. ...
... Processes 2024, 12, x FOR PEER REVIEW 3 of 28 spectral-coherence-based CNN can achieve high diagnosis accuracy and better generalization ability by applying domain-related techniques to reduce the difficulty of feature learning and obtain high-level feature representations [26]. A common CNN with Gramian noise reduction can improve the performance of denoising and classification for bearing fault diagnosis [27]. ...
A time–frequency residual convolution neural network (TFRCNN) was proposed to identify various rolling bearing fault types more efficiently. Three novel points about TFRCNN are presented as follows: First, by constructing a double-branch convolution network in the time domain and the frequency domain, the respective features in the time domain and the frequency domain were extracted to ensure the rich and complete feature representation of raw data sources. Second, specific residual structures were designed to prevent learning degradation of the deep network, and global average pooling was adopted to improve the network’s sparsity. Third, TFRCNN was better than the other models in terms of prediction accuracy, robustness, generalization ability, and convergence. The experimental results demonstrate that the prediction accuracy rate of TFRCNN, trained using mixing load data, reached 98.88 to 99.92% after optimizing the initial learning rate and choosing the optimizer and loss function. It was verified that TFRCNN can adaptively learn to extract deep fault features, accurately identify bearing fault conditions, and overcome the limitations of classical shallow feature extraction and classification methods, as well as common convolution neural networks. Hence, this investigation revealed TFRCNN’s potential for bearing fault diagnosis in practical engineering applications.
... Yan et al [11] have elicited a new method for multi-sensor fault diagnosis of machinery based on Adaptive Multiple Feature Modal Decomposition (AMFMD) and Multi-Attention Fused Residual Convolutional Neural Networks (MAFResCNN). Chen et al [12] who elicited a deep learning method based on cyclic spectral coherence and convolutional neural networks for bearing fault diagnosis. Considering the physical properties of the bearing's health status and the defects, a cyclic spectral analysis was proposed to obtain 2D CSCoh images for fault diagnosis. ...
Aiming at the problem that the data distribution of bearings across operating conditions generates offset resulting in insufficient diagnostic accuracy of the original model for new data, a cross-condition bearing fault detection method based on online drift detection and domain adaptation is proposed. First, the original one-dimensional vibration signals collected are transformed by a two-dimensional wavelet transform to convert the time-frequency image dataset. Second, the drift detection of the data across operating conditions is carried out using Random Forest (RF), and the 3σ criterion as well as the drift detection judgment criteria are set. Next, the source domain model based on Googlenet is used to extract features from the target domain data, and the Whale Optimization Algorithm to Improve Local Preserving Projection Algorithm (WOA-LPP) algorithm is combined to construct a brand-new projection space to align the features of the source and target domains. Then, the source and target domain features are reconstructed by combining the LPP optimal projection matrix to construct a fully connected network trained by the source domain features. Finally, probabilistic label-based decision fusion is proposed to integrate multiple classifiers to reduce the effects of model training randomness and strong noise interference. Validated by the publicly available Western Reserve University bearing data, the method proposed in this paper has good detection accuracy as well as robustness across operating conditions, which can effectively improve the defects of shifting data distribution and degradation of model accuracy under variable speed.
... Liu et al. 14 applied Matrix profile to mine the signal segments containing fault information and then applied CNN to analyze these segments for bearing fault diagnosis. Chen et al. 15 combined CNN and cyclic spectrum coherence (CSCoh) for rolling bearing fault diagnosis, in which CNN was utilized to extract the features from the 2-D representations by CSCoh. ...
Domain adaptation-based transfer learning methods have been widely investigated in fault diagnosis of rotating machinery, but their basic convolution or recurrent structure is subject to poor global feature representation ability, which hinders the learning of domain-irrelevant modulation information. In addition, the “black box” nature of deep learning models limits their applications in high risk-sensitive scenarios. In this paper, an interpretable domain adaptation transformer (IDAT) is proposed for the transferable fault diagnosis of rotating machinery. First, a multi-layer domain adaptation transformer framework is proposed, which can capture the global information that is crucial for learning the modulation information of different domains, and meanwhile reduce the feature distribution discrepancy. Second, an ensemble attention weight is applied to enable the transfer learning framework to be interpretable, which is implemented by averaging the integral values of the multi-head attention maps along the key direction. In addition, the raw vibration signals are embedded as the input of the model, which provides an end-to-end fault diagnosis. The proposed IDAT is tested by various cross-condition and cross-machine bearing fault diagnosis tasks, and results confirm the advantages of the method.
... 47 Generally, in the optimization design process of color routing, deep learning is typically applied by first collecting a large amount of spectral data for processing and analysis, aiming to extract valuable mapping relationships through training. 48 Subsequently, the corresponding optimization parameters need to be formulated, typically reflecting the optical efficiency of spectral splitting for each pixel channel in the color routing problem. Finally, through a combination of strategies, such as gradient descent optimization using neural networks or multi-objective optimization algorithms, the color routing structure tends to be iteratively optimized towards the desired outcome. ...
Nano-color routing has emerged as an immensely popular and widely discussed subject in the realms of light field manipulation, image sensing, and the integration of deep learning. The conventional dye filters employed in commercial applications have long been hampered by several limitations, including subpar signal-to-noise ratio, restricted upper bounds on optical efficiency, and challenges associated with minia-turization. Nonetheless, the advent of bandpass-free color routing has opened up unprecedented avenues for achieving remarkable optical spectral efficiency and operation at sub-wavelength scales within the area of image sensing applications. This has brought about a paradigm shift, fundamentally transforming the field by offering a promising solution to surmount the constraints encountered with traditional dye filters. This review presents a comprehensive exploration of representative deep learning-driven nano-color routing structure designs, encompassing forward simulation algorithms, photonic neural networks, and various global and local topology optimization methods. A thorough comparison is drawn between the exceptional light-splitting capabilities exhibited by these methods and those of traditional design approaches. Additionally, the existing research on color routing is summarized, highlighting a promising direction for forthcoming development, delivering valuable insights to advance the field of color routing and serving as a powerful reference for future endeavors.
... This method efficiently extracts domain-invariant features, enhancing the performance of cross-domain testing significantly. Chen et al [43] developed a fault diagnosis method based on cyclic spectral coherence (CSCoh) and convolutional neural networks (CNNs) with two-dimensional mapping representation. Compared to existing techniques, their DL-based approach improves rolling body bearing fault identification performance and demonstrates higher diagnosis accuracy across different datasets, with superior generalization ability. ...
As a key component of mechanical equipment, real-time monitoring and diagnosis of rolling bearings play a critical role in ensuring the stable operation of equipment and the safety of operators. In order to present the current status and trends of fault diagnosis research on rolling bearings more intuitively, the scientific knowledge mapping was used to visualize and analyze the relevant literature in the article. The results show that the number of publications in this area of research has grown significantly in recent years, with China, India, the United States, and England having contributed significantly. The journals such as MECHANICAL SYSTEMS AND SIGNAL PROCESSING, MEASUREMENT, and JOURNAL OF SOUND AND VIBRATION have played an important role in disseminating cutting-edge technologies in this field. In addition, the exploration of modern methods based on data-driven and artificial intelligence, as well as their application to real-world problems, are gradually becoming the focus of research. Through summarising and analysing, the application of modern data processing techniques, the development of more efficient and practical intelligent fault diagnosis techniques, and the close integration of laboratory research and practical applications will become future research trends.
... The most typical network model of deep learning is CNN and some variants based on CNN. [19][20][21] Zhao et al. 22 proposed a normalized CNN, which shows strong performance in sample training and classification, and can be effectively applied to rolling bearing fault diagnosis. Wang et al. 23 studied the adaptive normalized CNN, which solved the great challenge of fault detection of planetary gear boxes caused by variable speed and variable load, and finally significantly improved the fault accuracy caused by the change of operating modes. ...
The master cylinder of most pump trucks is equipped with a waterproof valve, whose purpose is to prevent water from the tank from entering the master cylinder. Once waterproof valve fails to failure, the waterproof valve at the main cylinder can only be supported by a BS seal (this seal is very easy to fail), which results in oil emulsification and pollution of the hydraulic system. Therefore, a fault diagnosis method combining a multi-sensor high-dimensional time-domain feature expansion map (MHTFEM) with an attentional convolutional capsule network (ACCN) is proposed. In this method, the raw vibration signals acquired by all sensors are first preprocessed to generate a high-dimensional feature matrix. Then the different high-dimensional feature matrices are stitched, expanded and generated into grayscale images, followed by randomly dividing the training set and the testing set. Finally, the training set is brought into the ACCN for training and the testing set is brought into the network model for fault type identification. A test bench was built to confirm the effectiveness of the method for waterproof valve fault diagnosis. This provides a method to achieve intelligent fault diagnosis of construction machinery to ensure its reliability.
... 47 Generally, in the optimization design process of color routing, deep learning is typically applied by first collecting a large amount of spectral data for processing and analysis, aiming to extract valuable mapping relationships through training. 48 Subsequently, the corresponding optimization parameters need to be formulated, typically reflecting the optical efficiency of spectral splitting for each pixel channel in the color routing problem. Finally, through a combination of strategies, such as gradient descent optimization using neural networks or multi-objective optimization algorithms, the color routing structure tends to be iteratively optimized towards the desired outcome. ...
Nano-color routing has emerged as an immensely popular and widely discussed subject in the realms of light field manipulation, image sensing, and the integration of deep learning. The conventional dye filters employed in commercial applications have long been hampered by several limitations, including subpar signal-to-noise ratio, restricted upper bounds on optical efficiency, and challenges associated with miniaturization. Nonetheless, the advent of bandpass-free color routing has opened up unprecedented avenues for achieving remarkable optical spectral efficiency and operation at sub-wavelength scales within the area of image sensing applications. This has brought about a paradigm shift, fundamentally transforming the field by offering a promising solution to surmount the constraints encountered with traditional dye filters. This review presents a comprehensive exploration of representative deep learning-driven nano-color routing structure designs, encompassing forward simulation algorithms, photonic neural networks, and various global and local topology optimization methods. A thorough comparison is drawn between the exceptional light-splitting capabilities exhibited by these methods and those of traditional design approaches. Additionally, the existing research on color routing is summarized, highlighting a promising direction for forthcoming development, delivering valuable insights to advance the field of color routing and serving as a powerful reference for future endeavors.
... In view of the existence of noise and non-smoothness of vibration signals and the greater advantage of CNNs in processing 2D data, Xiao et al [15] transformed bearing vibration signals into grayscale maps and combined CNNs to achieve fault diagnosis of rotating machinery. Chen et al [16] transformed diagnostic signals into 2D images as network inputs for fault diagnosis using cyclic spectra. Wang et al [17] proposed a method to apply the Markov migration field to the images that carries diagnostic time series data. ...
To address the problems of low diagnostic accuracy and slow diagnostic speed of the convolutional neural network fault diagnosis method in rolling bearing diagnosis, a new rolling bearing fault diagnosis method based on Fast Fourier Transform (FFT) image coding and Lightweight-Convolutional Neural Network (LCNN) is proposed. The method is mainly divided into three stages: firstly, the original signal is reconstructed by noise reduction using a joint noise reduction method of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Permutation Entropy(PE), and Wavelet Threshold Denoise(WTD); then, the frequency spectra and phase spectra feature fusion data of the noise-reduced and reconstructed bearing vibration signals are obtained by FFT, the feature fusion data are encoded into a heat map, and the image coding data-set is fed into an improved L-CNN for fault diagnosis. Experiments were carried out using the Guangdong University of Petrochemical Technology bearing fault data-set and the Case Western Reserve University bearing fault data-set with diagnostic accuracies of 98.75% and 99%, respectively. The results demonstrate that the method can effectively classify bearing fault vibration signals with the advantages of a fast diagnosis, high accuracy, and good generalization ability.
... To address this issue, alternative methods focusing on the relationship between system input and output have been proposed. Examples include Generalized Canonical Correlation Analysis (GCCA) [22], Canonical Variate Analysis (CVA) [23], and Canonical Variate Dissimilarity Analysis (CVDA) [24]. Although existing approaches demonstrate notable performances, fault detection in scenarios characterized by non-Gaussian distributed data and non-linear systems, with minimal false alarms and missed detections, remains a pertinent challenge. ...
Bearing condition monitoring in the field of industrial machinery has increasingly relied on the incorporation of artificial intelligence techniques. This paper introduces a fault detection and diagnosis methodology for bearing condition monitoring processes, utilizing the Mahalanobis Squared Distance (MSD). In the initial phase, a health index, namely MSD, is proposed to accurately indicate the health condition of the spherical bearing in an induction motor based on vibration signals. The MSD serves as a pre-classification stage, effectively addressing the issue of data overlap and facilitating the identification of distinct data classes, particularly in cases where non-linear and non-Gaussian data are prevalent. In the subsequent phase, a DL-based approach utilizing transfer learning is employed for the classification of the labeled dataset by MSD. Three established models, namely AlexNet, VGG19, and ResNet50, pre-trained on the ImageNet dataset, are considered. These models are further fine-tuned using scalogram images generated through the application of continuous wavelet transform (CWT) on the vibration signals obtained from spherical roller bearings. This integrated approach for fault detection and diagnosis is presented and validated using the Intelligent Maintenance Systems (IMS) bearing dataset. The results obtained demonstrate the reliability and efficacy of the proposed approach in accurately detecting and diagnosing bearing faults. Furthermore, the experimental findings indicate that the proposed approach surpasses existing state-of-the-art methods documented in the relevant literature.
... To address this issue, alternative methods focusing on the relationship between system input and output have been proposed. Examples include Generalized Canonical Correlation Analysis (GCCA) [22], Canonical Variate Analysis (CVA) [23], and Canonical Variate Dissimilarity Analysis (CVDA) [24]. Although existing approaches demonstrate notable performances, fault detection in scenarios characterized by non-Gaussian distributed data and non-linear systems, with minimal false alarms and missed detections, remains a pertinent challenge. ...
Bearing condition monitoring in the field of industrial machinery has increasingly relied on the incorporation of artificial intelligence techniques. This paper introduces a fault detection and diagnosis methodology for bearing condition monitoring processes, utilizing the Mahalanobis Squared Distance (MSD). In the initial phase, a health index, namely MSD, is proposed to accurately indicate the health condition of the spherical bearing in an induction motor based on vibration signals. The MSD serves as a pre-classification stage, effectively addressing the issue of data overlap and facilitating the identification of distinct data classes, particularly in cases where non-linear and non-Gaussian data are prevalent. In the subsequent phase, a DL-based approach utilizing transfer learning is employed for the classification of the labeled dataset by MSD. Three established models, namely AlexNet, VGG19, and ResNet50, pre-trained on the ImageNet dataset, are considered. These models are further fine-tuned using scalogram images generated through the application of continuous wavelet transform (CWT) on the vibration signals obtained from spherical roller bearings. This integrated approach for fault detection and diagnosis is presented and validated using the Intelligent Maintenance Systems (IMS) bearing dataset. The results obtained demonstrate the reliability and efficacy of the proposed approach in accurately detecting and diagnosing bearing faults. Furthermore, the experimental findings indicate that the proposed approach surpasses existing state-of-the-art methods documented in the relevant literature.
... Wang et al. (Wang, Zhuang, Duan, & Cheng, 2016) enhanced CNNs' generalization for fault detection by applying Morlet wavelet decomposition, bilinear interpolation, and rectified linear units to grayscale images derived from vibration signals. Another approach, combining Cyclic Spectral Coherence (CSCoh) with CNNs, has been proposed for diagnosing rolling bearing faults, showing improved detection performance (Chen, Mauricio, Li, & Gryllias, 2020). However, these methods pose challenges in real-world engineering due to the high demand for training samples. ...
Bearings are critical components in rotating machinery, and their failure can lead to costly repairs and downtime. To prevent such failures, it is important to detect and diagnose bearing faults early. In recent years, deep-learning techniques have shown promise for detecting and diagnosing bearing faults automatically. While these algorithms can all achieve diagnostic accuracy of over 90%, their generalizability and robustness in complex, extreme variable loading conditions have not been thoroughly validated. In this paper, a feature extraction method based on Synchro-squeezing Wavelet Transform (SSWT), Random projection (RP), and deep learning (DL) is presented. To fulfill the data requirements of neural networks, data augmentation is initially utilized to augment the size of the original data. Subsequently, the SSWT technique is employed to convert the signals from the Time domain to the Time-Frequency domain, resulting in the conversion of the 1-D signal to a 2-D feature image. To decrease the complexity of deep learning computation, data preprocessing involves utilizing Random projection to reduce feature dimensionality. The final step involves constructing a Convolutional Neural Network (CNN) model that can identify fault features from the obtained Time-Frequency images and perform accurate fault classification. By utilizing the CWRU and IMS datasets to evaluate the method, the study demonstrates that the suggested approach outperforms advanced techniques in terms of both diagnostic accuracy and robustness.
... As an alternative to shallow machine learning methods, deep learning models with strong feature learning ability and end-to-end diagnosis characteristics are applicable for fault diagnosis of USMs, such as convolutional neural networks (CNNs) [18,19], deep belief networks (DBNs) [20], and long short-term memory (LSTM). CNN can automatically extract local spatial correlation in multivariate time series data and shows advantages in handling strong nonlinearities between the input and output [21]. ...
The ultrasonic motor is peculiarly prone to failure due to continuous high-frequency friction-related power transfer, whose failure mechanisms are remarkably different from traditional induction motors. Intelligent fault diagnosis provides a way to alarm and avoid catastrophic losses proactively. However, previous studies using deep learning usually ignore the inherent geometric structure of the signal distribution. This paper proposes an intelligent multi-signal fault diagnosis framework for ultrasonic motors to restore the linear or nonlinear manifold structure by preserving the internal structure by integrating graph regularization with deep neural networks. Firstly, the one-dimensional CNN to learn spatial correlations and BiLSTM to exploit temporal dependencies are coalesced to build the deep neural network. Then, an improved k-nearest neighbor graph is proposed to protect the geometric structure information and force the latent features to be more concentrated within their classes. Moreover, the layer in the deep architecture to integrate graph regularization is designed to reduce computation cost, and an adaptive decay strategy is considered to adjust the coefficient of graph regularized automatically. A two-stage training algorithm is developed by considering the time to calculate the graph regularization term. Finally, the proposed multi-signal fault diagnosis framework is validated using datasets from the fault injection experiment of ultrasonic motors in China’s Yutu rover of Chang’e lunar probe. Experimental results show that the proposed method can effectively discriminate different fault types.
... Among them, deep learning models have been widely studied and applied because of their ability to automatically extract useful features from complex data. Chen et al. proposed a deep learning fault diagnosis method based on cyclic spectral coherence diagrams and convolutional neural networks, which improved the identification of rolling bearing faults, but all fault types appeared in the training set [12]. Arellano-Espitia et al. demonstrated the advantages of applying deep learning technology for fault diagnosis in different bearing technologies [13]. ...
Confronting the challenge of identifying unknown fault types in rolling bearing fault diagnosis, this study introduces a multi-scale bearing fault diagnosis method based on transfer learning. Initially, a multi-scale feature extraction network, MBDCNet, is constructed. This network, by integrating the features of vibration signals at multiple scales, is dedicated to capturing key information within bearing vibration signals. Innovatively, this study replaces traditional convolution with dynamic convolution in MBDCNet, aiming to enhance the model’s flexibility and adaptability. Furthermore, the study implements pre-training and transfer learning strategies to maximally extract latent knowledge from source domain data. By optimizing the loss function and fine-tuning the learning rate, the robustness and generalization ability of the model in the target domain are significantly improved. The proposed method is validated on bearing datasets provided by Case Western Reserve University and Jiangnan University. The experimental results demonstrate high accuracy in most diagnostic tasks, achieving optimal average accuracy on both datasets, thus verifying the stability and robustness of our approach in various diagnostic tasks. This offers a reliable research direction in terms of enhancing the reliability of industrial equipment, especially in the field of bearing fault diagnosis.
... Convolutional Neural Network (CNN) is a supervised deep learning algorithm developed in recent years [30], which has been applied in the field of fault diagnosis by scholars for its powerful capability in automatic feature extraction. In the literature [31] CNN was used to achieve bearing fault diagnosis and lubrication performance degradation in rotating machinery. ...
Since bearing fault signal in complex running status is usually characterized as nonlinear and non-stationary, it is difficult to extract accurate affluent features and achieve effective fault identification via conventional signal processing tools. In this article, a rolling bearing fault diagnosis technique based on variational mode decomposition and weighted multidimensional feature entropy fusion is proposed to address this issue, which is mainly composed of three procedures. First, the original signal undergoes the variational model decomposition. Next, the signal features are extracted by weighted multidimensional feature entropy as the input of the diagnosis model. Finally, the classification is performed by a convolutional neural network. The method is applied in simulation and experimental analysis. The experimental results show that the proposed method, which demonstrates strong immunity to noise and robustness, can more effectively and adaptively extract the fault features of rolling bearings and achieve the goal of identifying the rolling bearing fault category and damage degree under variable operating conditions. Meanwhile, this approach exhibits superior accuracy and identification performance to some similar entropy-based hybrid approaches referred to in this article, with a promising prospect in industrial application.
... The tool may find application to generate time-frequency maps to be input in Convolutional Neural Network (CNN) architectures, as observed in works such as [38,39]. The accuracy of automatic detection algorithms is, to some extent, contingent on the careful selection of an appropriate signal processing tool, as highlighted in [40]. The computation efficiency while generating time-frequency maps of a large amount of data are critical and, thus, the present method presents an interesting trade-off between computational efficiency and physical representation of the failure phenomena. ...
The analysis of electrical machine faults during start-up, and variable speed and load conditions offers numerous advantages for fault detection and diagnosis. In this context, diagnosing localized bearing faults through vibration signals remains challenging, particularly in developing physically meaningful, simple, and resampling-free techniques to monitor fault characteristic components throughout machine start-up. This study introduces a straightforward method for qualitatively identifying the time-frequency evolutions of localized bearing faults during the start-up of an inverter-fed machine. The proposed technique utilizes the time-frequency representation of the envelope spectrum, effectively highlighting characteristic fault frequencies during transient operation. The method is tested in an open-source dataset including transient vibration signals. In addition, the work studies the method limitations induced by the mechanical transfer path, when the bearing surroundings are not directly accessible for vibration acquisition. The proposed methodology efficiently identifies incipient localized bearing faults during inverter-fed machine start-up when the fault signature is not highly attenuated.
... Meanwhile, Deep Learning (DL) [28] is gradually becoming a trend in the field of intelligent fault diagnosis due to its powerful automatic feature learning capability of extracting and utilizing features from a large amount of fault information. Chen et al. [29] proposed a model based on cyclic spectral coherence and deep convolutional neural networks (CNNs) to improve model performance for accurate fault identification. Fu et al. [30] proposed a multiscale CNN composite model based on self-attention and Inception residual connectivity module. ...
To address the problem that uneven sample distribution can affect the accuracy and stability of fault diagnosis outcomes, we propose a deep transfer learning-Res2Net-convolutional block attention mechanism model. Firstly, the deep migration technique is used to transfer weights of the imbalanced source domain samples to the balanced target domain, expanding the data samples. Secondly, in the feature extraction and detection phase, All eight residual blocks are embedded in convolution block attention to ensure that interference signals are suppressed and that the key fault features are retained. Multilayer feature fusion extracts faulty sample features from multiple residual blocks of the network, which are combined into the feature fusion layer by parallel splicing. Finally, this model is experimentally validated using two different bearing datasets, and the combined evaluation indexes of the new model under the severe imbalance condition are 95.80% and 95.85%, respectively, demonstrating the feasibility and excellent performance of the model.
... Islam and Kim [31] employed wavelet packet transform for preprocessing the raw acoustic emission signal and crafted an adaptive CNN for the classification of multi-fault bearings. Chen et al [32] proposed a method based on cyclic spectral coherence and CNN, using the cyclic spectral good feature extraction ability of coherence and the excellent classification ability of CNN to achieve bearing fault diagnosis. The above study shows that the deep learning method improves the performance of bearing fault diagnosis. ...
Bearings are crucial components in rotating machinery equipment. Bearing fault diagnosis plays a significant role in the maintenance of mechanical equipment. In practical industrial settings, equipment conditions often vary continuously, making it challenging to collect data for all operating conditions for bearing fault diagnosis. This paper proposes a transfer learning approach for bearing fault diagnosis based on Adaptive Batch Normalization (AdaBN) and a combined optimization algorithm. Initially, a ResNet neural network is trained using source domain data. Subsequently, the trained model is transferred to the target domain, where AdaBN is applied to mitigate domain shift issues. Furthermore, a combined optimization algorithm is employed during model training to enhance fault diagnosis accuracy. Experimental validation is conducted using bearing data from the CWRU dataset and NEFU dataset. Comparison shows that AdaBN and the combined optimization algorithm improve bearing fault diagnosis accuracy effectively. On the NEFU dataset, the diagnostic accuracy exceeds 95%.
... However, dealing with diverse features and detecting faults using 1D data proved to be a challenge. Recently, research has increasingly shifted towards converting 1D time-series data into 2D images for feature extraction and fault detection, and advancements in deep learning have already demonstrated excellent performance in image classification [3,[9][10][11]. ...
This paper proposes fault diagnosis methods aimed at proactively preventing potential safety issues in robot systems, particularly human coexistence robots (HCRs) used in industrial environments. The data were collected from durability tests of the driving module for HCRs, gathering time-series vibration data until the module failed. In this study, to apply classification methods in the absence of post-failure data, the initial 50% of the collected data were designated as the normal section, and the data from the 10 h immediately preceding the failure were selected as the fault section. To generate additional data for the limited fault dataset, the Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) model was utilized and residual connections were added to the generator to maintain the basic structure while preventing the loss of key features of the data. Considering that the performance of image encoding techniques varies depending on the dataset type, this study applied and compared five image encoding methods and four CNN models to facilitate the selection of the most suitable algorithm. The time-series data were converted into image data using image encoding techniques including recurrence plot, Gramian angular field, Markov transition field, spectrogram, and scalogram. These images were then applied to CNN models, including VGGNet, GoogleNet, ResNet, and DenseNet, to calculate the accuracy of fault diagnosis and compare the performance of each model. The experimental results demonstrated significant improvements in diagnostic accuracy when employing the WGAN-GP model to generate fault data, and among the image encoding techniques and convolutional neural network models, spectrogram and DenseNet exhibited superior performance, respectively.
... Study in [15] proposed the diagnosis of bearing faults based on short-time Fourier transform and convolutional neural network. A deep learning mechanism reported by [16] in combination of cyclic spectral coherence and CNN to enhance the performance of rolling bearings. Te conversion of the vibration signal to grey scale image to establish convolutional neural network model for classifcation is reported in [17]. ...
Bearings are critical components in modern manufacturing, yet they are prone to failures in induction machines. Detecting these faults early can reduce repair costs. To achieve efficient and accurate fault detection, we explore vibration-based analysis. Traditional methods rely on manual feature extraction, which is time-consuming. In contrast, our work leverages deep learning, particularly convolutional neural networks, to automatically extract fault features from raw data. We investigate various image sizes (16 × 16, 32 × 32, 64 × 64, 128 × 128, 256 × 256) and their performance in bearing fault diagnosis. Our convolutional neural networks-based approach is compared to traditional methods such as support vector machine, nearest neighbors, and artificial neural networks. Results demonstrate the superior performance of our data-driven fault diagnosis using convolutional neural networks.
... These models use both temporal and spatial information in order to enhance the comprehensiveness of the input [27]. Widely used 2D conversion methods for bearing signals include the wavelet packet energy [28], short-time Fourier transform (STFT) [29], symmetrized dot pattern (SDP) [30], Cyclic Spectral Coherence (CSCoh) [31], and continuous wavelet transform (CWT) [32,33]. Ruan et al. [34] designed 2D CNN network parameters based on fault signal analysis such as fault characteristic frequency for bearing fault diagnosis. ...
This paper proposes a noise-robust and accurate bearing fault diagnosis model based on time-frequency multi-domain 1D convolutional neural networks (CNNs) with attention modules. The proposed model, referred to as the TF-MDA model, is designed for an accurate bearing fault classification model based on vibration sensor signals that can be implemented at industry sites under a high-noise environment. Previous 1D CNN-based bearing diagnosis models are mostly based on either time domain vibration signals or frequency domain spectral signals. In contrast, our model has parallel 1D CNN modules that simultaneously extract features from both the time and frequency domains. These multi-domain features are then fused to capture comprehensive information on bearing fault signals. Additionally, physics-informed preprocessings are incorporated into the frequency-spectral signals to further improve the classification accuracy. Furthermore, a channel and spatial attention module is added to effectively enhance the noise-robustness by focusing more on the fault characteristic features. Experiments were conducted using public bearing datasets, and the results indicated that the proposed model outperformed similar diagnosis models on a range of noise levels ranging from −6 to 6 dB signal-to-noise ratio (SNR).
... In contrast, deep learning techniques offer an alternative approach by enabling automated feature extraction, obviating the need for manual intervention, and have thus garnered significant attention among researchers in the domain of fault diagnosis in recent years. Currently, the prevailing landscape of deep learning methodologies encompasses primarily the convolutional neural network (CNN) [11][12][13], the auto-encoder (AE) [14][15][16], and the deep belief network (DBN) [17][18][19]. For instance, a study by Liu et al [13] ingeniously amalgamated the multiscale kernel approach with the CNN framework, yielding effective fault diagnosis outcomes for electric motors amid varying operational conditions. ...
In response to the challenge posed by traditional deep learning methods, which apply uniform nonlinear transformations to all vibration signals and thus struggle to address fault diagnosis under variable working conditions, a novel activation function called the convex global parametric rectifier linear unit (CGPReLU) is developed based on our prior research. Initially, an analysis of the numerical patterns governing the adaptive derivation process of GPReLU’s two slope parameters revealed the surprising observation that these convex parameter combinations invariably appear in pairs. This discovery serves as the primary motivation for the development of CGPReLU. Leveraging this convex regularity, we subsequently redesigned a lightweight convex sub-network for the adaptive derivation of the CGPReLU’s slope. Simultaneously, a deep residual shrinkage network with CGPReLU (DRSN-CGPReLU) is constructed for fault diagnosis. Furthermore, we introduce an innovative evaluation metric designed to measure the collective influence of diagnostic accuracy and computational complexity after the process of model lightweight. Finally, it is experimentally demonstrated that the developed method can maintain a better diagnostic performance while greatly improving the diagnostic efficiency under variable operating conditions compared to our previous work.
... It is worth mentioning that when there is only one group, GN becomes LN. When the number of groups equals the number of channels, GN becomes instance normalization (IN), where information between channels is not considered [27]. In general, GN leads to better training results and improved computational efficiency compared to LN and IN. ...
The integration of deep neural networks and cloud computing has become increasingly prevalent within the domain of medical image processing, facilitated by the recent strides in neural network theory and the advent of the internet of things (IoTs). This juncture has led to the emergence of numerous image segmentation networks and innovative solutions that facilitate medical practitioners in diagnosing lung cancer. Within the contours of this study, we present an end-to-end neural network model, christened as the “semi-residual Multi-resolution Convolutional Neural Network” (semi-residual MCNN), devised to engender precise lung nodule segmentation maps within the milieu of cloud computing. Central to the architecture are three pivotal features, each coalescing to effectuate a notable enhancement in predictive accuracy: the incorporation of semi-residual building blocks, the deployment of group normalization techniques, and the orchestration of multi-resolution output heads. This innovative model is systematically subjected to rigorous training and testing regimes, using the LIDC-IDRI dataset – a widely embraced and accessible repository – comprising a diverse ensemble of 1,018 distinct lung CT images tailored to the realm of lung nodule segmentation.
Fault diagnosis of gearboxes has attracted increasing interest in recent decades due to their ubiquity and importance in the industry. Modern research trends focus on developing a diagnosis system that works automatically with the application of artificial intelligence. These previous studies have used the Deep Learning (DL) network without adequately addressing noise of the input data, requiring more data to achieve effective training. Thus, this work proposes a novel Transfer Learning method using the time–frequency representation of gear vibration signals, which enables more accurate classification in complex working conditions and reduces necessary input data to train. Using fine-tuning techniques proposed in this paper requires only a limited data set while ensuring acceptable classification results. An experiment test rig within different gear faults and load conditions was set up to evaluate the algorithm’s effectiveness.
With the continuous development of various industries, the diagnosis of industrial equipment faults has been receiving increasing attention in recent years. Considering the complex and variable working conditions, and the limited amount of fault data, transfer learning has become an effective solution for fault diagnosis. Data augmentation techniques, particularly generative adversarial networks, have achieved tremendous development within the field of transfer learning fault diagnosis. However, traditional data augmentation methods experience difficulty in extracting features conducive to fault diagnosis from fault data under complex operating conditions, particularly in the case of raw vibration data from bearings. Therefore, this study proposes a new multiple mixed augmentation-based transfer learning (MMATL) method for machinery fault diagnosis. First, an augmentation chain that dynamically adjusts data augmentation strategies in accordance with the model’s performance is constructed based on AutoAugment. Then, a multiple mixed augmentation strategy that integrates fault data into the augmented data from the augmentation chain to obtain enhanced data suitable for training is proposed. This strategy consists of multiple augmentations, augmentation mixing, and data mixing. Finally, experiments confirm the effectiveness of MMATL on the bearing datasets from the gearbox of the Chinese CRH380A high-speed train, the test rig at the University of Paderborn in Germany and the self-made bearing failure test platform. Results indicate that the method can adaptively extract features from fault data that are conducive to fault diagnosis under complex operating conditions.
The rolling elements of the induction motor are highly susceptible to faults. The detection and diagnosis of rolling element faults are accurate and reliable only when the extracted features are accurate. The paper proposes an approach for bearing and rotor fault diagnosis using deep optimal feature extraction and selection based on vibration signal analysis. The deep feature extraction is done using an ensemble deep models features extraction approach in which features are extracted from seven pretrained models are fused serially using serial-based feature fusion technique. This leads to a solution for a higher efficacy model, but at the cost of high processing time as the feature data set gets large. A unique approach termed Ensemble Feature Selection has been developed to address this issue and limit the harmful impact of unwanted features in data-driven diagnostics. The processing time is further reduced using the shallow classifier at the fully connected layer. The proposed model is tested using the data acquired in the laboratory and validated using the available online benchmark data sets.
The extraction and detection of weak faults of rolling bearings on rotating machinery are essential and crucial. However, traditional methods of Sparse Matrix Optimization have shortcomings of amplitude underestimation. Therefore, we propose a sparse optimization method based on sparse matrix and singular value vector. Due to the intrinsic features of the faulty bearing signals, we extract the sparsity and low-rank property from the matrix of time-frequency diagram. To improve the performance of the detection, we utilize the minimax concave penalty to both the elements and the singular value vector which respectively enhance the reconstruction accuracy of fault signal. Then, the convexity of the model is proved and the selection of the parameters is also discussed. To solve this convex optimization problem, an iterative algorithm based on Alternating Direction Method of Multipliers and Forward and Backward Splitting is adopted to attain the global optimal solution. Synthetic signals and experimental signals are utilized to verify the effectiveness and superiority of proposed method.
Deep learning has been used to enhance the efficiency of rolling bearing fault diagnosis. However, the complexity of working conditions in rolling bearings, coupled with varying data distribution, often results in the failure of most deep learning models. To solve this problem, a feature extractor with an attention mechanism is constructed, which allows the model to selectively study and preserve critical features relevant to fault information during the training process. Moreover, the Wasserstein distance is employed in adversarial neural networks to calculate the distribution discrepancy between data from different domains, which can avoid disappearing gradients or vanishing problems of the model. The experimental validation are conducted using the rolling bearing datasets from CWRU and Jiangnan University (JNU). Compared with other existing models, the proposed model has better performance under variable conditions, and this demonstrates the superiority and accuracy of this model.
Data-driven fault diagnosis methods have significantly contributed to the rapid development of prognostics and health management of gearboxes. However, these methods require sufficient labeled data for model training. In real industrial scenarios, labeled data is usually scarce. Thus, a residual attention temporal recurrent network is proposed for fault diagnosis of gearboxes under limited labeled data. More specifically, a fault diagnosis framework with self-supervised sample generation learning is explored in this paper. First, a small amount of labeled data is utilized to self-supervise the generation of new samples with similar semantics in adversarial generation modeling. Second, a temporal recurrent network with residual attention is constructed to extract fault features of vibration signals using two-stage temporal and recursive operations. In addition, the construction of the residual attention mechanism can facilitate the network to have a deeper semantic understanding of the vibration signal. Finally, the model can guarantee a more robust classifier in the proposed fault diagnosis framework using a small number of labeled samples. The extensive validation of the proposed method is performed on the gearbox dataset. The experimental results show that the proposed method can achieve competitive fault diagnosis performance under limited labeled data.
Fault diagnosis of rotating machines is essential for the safe and efficient operation of maritime vessels. It prevents potential failures in rotating machines in maritime systems. Simultaneously developed faults severely damage the machines, leading to unprecedented incidents and higher maintenance costs. Data availability for developing multi-fault diagnosis solutions is limited, creating maintainability problems. The unavailability of data also causes further obstacles in the accurate diagnosis of machines. The efficiency of maritime machinery can be improved with intelligent diagnosis by advanced machine learning algorithms. This paper proposes a methodology to diagnose multi-faults with balanced and imbalanced acoustic, vibration and current signals datasets. After acquiring the data from an experimental setup under various speeds, a time-frequency-based Stepping Frame Synchrosqueezed Fourier Transform (SFSFT) is deployed in each segmented signal prior to feature extraction. The extracted features are used for training the Radial Basis Function Neural Network (RBFNN). Different balanced and imbalanced datasets are created to check the multi-fault classification ability of the proposed methodology. The results show that SFSFT-RBFNN is a generalised methodology, which performed tremendously well with a maximum classification accuracy of 100 % for balanced and imbalanced datasets in all three modality datasets.
The effective fault feature extraction is the core of rolling bearing fault diagnosis. However, rolling bearings usually operate in normal state and fault duration is very short, which will cause imbalance in fault diagnosis data, thus leading to difficulty in fault feature extraction and low diagnosis accuracy. Meanwhile, mutual interference between multiple fault responses will also lead to poor diagnosis performance. To solve these issues, a novel compound fault diagnosis method with imbalanced data based on frequency-domain Gramian angular field (FGAF) and convolutional neural networks optimized by instance normalization and efficient channel attention (IECNN) is proposed. Firstly, FGAF is adopted to map frequency-domain features of fault signals to the polar coordinate to obtain 2D FGAF feature spectrum. Secondly, an instance normalization module is established to reduce internal covariant shift caused by data distribution discrepancy and improve generalization ability. An efficient channel attention module is constructed to further excavate fault features and improve anti-interference ability. Finally, experiments are conducted under imbalanced dataset and imbalance intensified dataset, and the average accuracy of 99.91% and 99.92% were obtained, respectively, which shows the proposed method has better resistance to data imbalance.
Deep learning-based rolling bearing fault diagnosis method has made significant achievements, but its diagnostic performance is still limited by few samples. Aiming at this problem, a novel Intelligent Fault Diagnosis (IFD) method for rolling bearings, named Multi-scale Residual Anti-noise Network (MRANet) via interpretable Dynamic Recalibration Mechanism (DRM), is proposed. Firstly, the raw vibration signal is generated into a time-frequency diagram with more characteristic domains by Short-Time Fourier Transform (STFT). Then, the shallow mechanism and deep discriminable features are extracted using multi-branch dilated convolution and improved residual blocks. Simultaneously, the DRM assists the feature extractor to adaptively adjust the feature weights from the spatial position and the channel information ratio, respectively, to enhance the local impulse excitation. Furthermore, the corrective effect of DRM on the feature extractor is visualized, which improves the interpretability of the network. Comparative experiments are conducted with other popular IFD methods on public and Lanzhou University of Technology (LUT) bearing dataset, and the results show that MRANet can exhibit superior diagnostic performance with few samples under variable load and multi-speed conditions.
Intelligent compound fault diagnosis of rotating machinery plays a crucial role for the security, high-efficiency and reliability of modern manufacture machines, but identifying and decoupling the compound fault is still a great challenge. The traditional compound fault diagnosis methods focus on either bearing or gear fault diagnosis, where the compound fault is always regarded as an independent fault pattern in the process of fault diagnosis, and the relationship between the single fault and compound fault is not considered completely. To solve such problem, a novel method called Deep Decoupling Convolutional Neural Network (DDCNN) is proposed for intelligent compound fault diagnosis. Firstly, one-dimensional deep convolutional neural network (1D DCNN) is employed as the feature learning model, which can effectively learn the discriminative features from raw vibration signals. Secondly, multi-stack capsules are designed as the decoupling classifier (DC) to accurately identify and decouple the compound fault. Finally, the routing by agreement algorithm and the margin loss cost function are utilized to train and optimize the proposed model. The proposed method is validated by gearbox fault tests, and the experimental results demonstrate that the proposed method can effectively identify and decouple the compound fault.
A sparse deep learning method is proposed to overcome over-fitting risk of deep networks with large number of nodes and layers. Deep stacking network (DSN) is a classic and effective deep learning method, and its sparse form is presented to generate the sparse deep learning method. In DSN, output labels are encoded as a series consisted of 1 and 0. This coding strategy makes output labels to be sparse. However, sparsity of output labels is not considered in DSN model. To improve its effectiveness, sparse deep stacking network (SDSN) is developed in this paper. The SDSN extends tradition DSN in sparsity characterization using a sparse regularization term. By this term, predicted output label is constrained to be similar with ideal output label which is binary and consisted of continuous 1 and 0 with a sidestep shape. The sparse regularization term is used as a soft threshold strategy to set irrelevant element to be zero, by which effectiveness of SDSN is enhanced. Case studies about fault diagnosis of motor are used to validate performance of SDSN. Comparison between SDSN and commonly used deep networks is further conducted. The results show advance of SDSN for fault classification.
This paper presents a novel method for diagnosing incipient bearing defects under variable operating speeds using convolutional neural networks (CNNs) trained via the stochastic diagonal Levenberg-Marquardt (S-DLM) algorithm. The CNNs utilize the spectral energy maps (SEMs) of the acoustic emission (AE) signals as inputs and automatically learn the optimal features, which yield the best discriminative models for diagnosing incipient bearing defects under variable operating speeds. The SEMs are two-dimensional maps that show the distribution of energy across different bands of the AE spectrum. It is hypothesized that the variation of a bearing’s speed would not alter the overall shape of the AE spectrum rather, it may only scale and translate it. Thus, at different speeds, the same defect would yield SEMs that are scaled and shifted versions of each other. This hypothesis is confirmed by the experimental results, where CNNs trained using the S-DLM algorithm yield significantly better diagnostic performance under variable operating speeds compared to existing methods. In this work, the performance of different training algorithms is also evaluated to select the best training algorithm for the CNNs. The proposed method is used to diagnose both single and compound defects at six different operating speeds.
One of the significant tasks in data-driven fault diagnosis methods is to configure a good feature set involving statistical parameters. However, statistical parameters are often incapable of representing the dynamic behavior of planetary gearboxes under variable operating conditions. Although the use of deep learning algorithms to find a good set of features for fault diagnosis has somewhat improved diagnostic performance, the lack of domain knowledge incorporated into deep learning algorithms has limited further improvement. Accordingly, this paper developed a variant of deep residual networks, so-called “deep residual networks with dynamically weighted wavelet coefficients (DRN+DWWC)” to improve diagnostic performance, which takes a series of sets of wavelet packet coefficients on various frequency bands as an input. Further, the fact that no general consensus has been reached as to which frequency band contains the most intrinsic information about a planetary gearbox's health status calls for “dynamic weighting layers” in the DRN+DWWC and the role of the layers is to dynamically adjust a weight applied to each set of wavelet packet coefficients to find a discriminative set of features that will be further used for planetary gearbox fault diagnosis.
Traditional feature extraction and selection is a labor-intensive process requiring expert knowledge of the relevant features pertinent to the system. This knowledge is sometimes a luxury and could introduce added uncertainty and bias to the results. To address this problem a deep learning enabled featureless methodology is proposed to automatically learn the features of the data. Time-frequency representations of the raw data are used to generate image representations of the raw signal, which are then fed into a deep convolutional neural network (CNN) architecture for classification and fault diagnosis. This methodology was applied to two public data sets of rolling element bearing vibration signals. Three time-frequency analysis methods (short-time Fourier transform, wavelet transform, and Hilbert-Huang transform) were explored for their representation effectiveness. The proposed CNN architecture achieves better results with less learnable parameters than similar architectures used for fault detection, including cases with experimental noise.
In modern industries, machine health monitoring systems (MHMS) have been applied wildly with the goal of realizing predictive maintenance including failures tracking, downtime reduction and assets preservation. In the era of big machinery data, data-driven MHMS have achieved remarkable results in the detection of faults after the occurrence of certain failures (diagnosis) and prediction of the future working conditions and the remaining useful life (prognosis). The numerical representation for raw sensory data is the key stone for various successful MHMS. Conventional methods are labor-extensive as they usually depend on handcrafted features, which require expert knowledge. Inspired by the success of deep learning methods that redefine representation learning from raw data, we propose local feature-based gated recurrent unit networks (LFGRU). It is a hybrid approach that combines handcrafted feature design with automatic feature learning for machine health monitoring. Firstly, features from windows of input time series are extracted. Then, an enhanced bi-directional GRU network is designed and applied on the generated sequence of local features to learn the representation. A supervised learning layer is finally trained to predict machine condition. Experiments on three machine health monitoring tasks: tool wear prediction, gearbox fault diagnosis and incipient bearing fault detection verify the effectiveness and generalization of the proposed LFGRU.
Intelligent fault diagnosis techniques have replaced time-consuming and unreliable human analysis, increasing the efficiency of fault diagnosis. Deep learning models can improve the accuracy of intelligent fault diagnosis with the help of their multilayer nonlinear mapping ability. This paper proposes a novel method named Deep Convolutional Neural Networks with Wide First-layer Kernels (WDCNN). The proposed method uses raw vibration signals as input (data augmentation is used to generate more inputs), and uses the wide kernels in the first convolutional layer for extracting features and suppressing high frequency noise. Small convolutional kernels in the preceding layers are used for multilayer nonlinear mapping. AdaBN is implemented to improve the domain adaptation ability of the model. The proposed model addresses the problem that currently, the accuracy of CNN applied to fault diagnosis is not very high. WDCNN can not only achieve 100% classification accuracy on normal signals, but also outperform the state-of-the-art DNN model which is based on frequency features under different working load and noisy environment conditions.
Early detection of the motor faults is essential and Artificial Neural Networks (ANNs) are widely used for this purpose. The typical systems usually encapsulate two distinct blocks: feature extraction and classification. Such fixed and hand-crafted features may be a sub-optimal choice and require a significant computational cost that will prevent their usage for realtime applications. In this paper, we propose a fast and accurate motor condition monitoring and early fault detection system using 1D Convolutional Neural Networks (CNNs) that has an inherent adaptive design to fuse the feature extraction and classification phases of the motor fault detection into a single learning body. The proposed approach is directly applicable to the raw data (signal) and thus eliminates the need for a separate feature extraction algorithm resulting in more efficient systems in terms of both speed and hardware. Experimental results obtained using real motor data demonstrate the effectiveness of the proposed method for real-time motor condition monitoring.
Vibration-based rolling element bearing diagnostics is a very well-developed field, yet researchers continue to develop new diagnostic algorithms quite frequently. Over the last decade, data from the Case Western Reserve University (CWRU) Bearing Data Center has become a standard refer-ence used to test these algorithms, yet without any recognised benchmark it is difficult to properly assess the performance of any proposed diagnostic methods. There is, then, a clear need to exam-ine the data thoroughly and to categorise it appropriately, and this paper intends to fulfil that ob-jective. To do so, three established diagnostic techniques are applied to the entire CWRU data set, and the diagnostic outcomes are provided and discussed in detail. Recommendations are given as to how the data might best be used, and also on how any future benchmark data should be generat-ed. Though intended primarily as a benchmark to aid in testing new diagnostic algorithms, it is also hoped that much of the discussion will have broader applicability to other bearing diagnostics cases.
Multilayer neural networks trained with the back-propagation
algorithm constitute the best example of a successful gradient based
learning technique. Given an appropriate network architecture,
gradient-based learning algorithms can be used to synthesize a complex
decision surface that can classify high-dimensional patterns, such as
handwritten characters, with minimal preprocessing. This paper reviews
various methods applied to handwritten character recognition and
compares them on a standard handwritten digit recognition task.
Convolutional neural networks, which are specifically designed to deal
with the variability of 2D shapes, are shown to outperform all other
techniques. Real-life document recognition systems are composed of
multiple modules including field extraction, segmentation recognition,
and language modeling. A new learning paradigm, called graph transformer
networks (GTN), allows such multimodule systems to be trained globally
using gradient-based methods so as to minimize an overall performance
measure. Two systems for online handwriting recognition are described.
Experiments demonstrate the advantage of global training, and the
flexibility of graph transformer networks. A graph transformer network
for reading a bank cheque is also described. It uses convolutional
neural network character recognizers combined with global training
techniques to provide record accuracy on business and personal cheques.
It is deployed commercially and reads several million cheques per day
In the era of the so called 4th industrial revolution, the Factory of the Future and the Industrial Internet of Things, the industrial mechanical systems become continuously more intelligent and more complex. Therefore, there is a clear need for research and development on data driven methodologies and condition monitoring techniques which are able to achieve fast, reliable and high-quality diagnosis in an automatic manner. In this paper, a novel fault diagnosis approach integrating Convolutional Neural Networks (CNN) and Extreme Learning Machine (ELM) is proposed, consisting of three main stages. Firstly, the Continuous Wavelet Transform (CWT) is employed in order to obtain pre-processed representations of raw vibration signals. Secondly, a CNN with a square pooling architecture is constructed to extract high-level features. The model does not require extra training and fine-tuning, which can effectively reduce computational cost. Finally, ELM as a strong classifier is further utilized to improve the classification performance on the diagnosis framework. Two datasets, including a gearbox dataset and a motor bearing dataset, have been collected and used to verify the effectiveness of the proposed method. A comprehensive comparison and analysis with widely used algorithms is also performed. The results demonstrate that the proposed method can detect different fault types and outperforms other methods in terms of classification accuracy.
Liquid level of internal silicone oil is about the safety operation of high voltage porcelain bushing type (PBT) terminals, and its regular inspection can effectively prevent the security risks caused by oil leaks. In this paper, a deep learning detection method based on local wavelet features and unsupervised feature fusion is proposed to quantitatively detect the internal liquid level of high voltage PBT terminals. Firstly, ultrasonic guided wave signals are divided into many segments by a sliding window and all segments are transformed into wavelet domain to catch the local time-frequency information. After preliminary selection according to the monotonic trends, the features are fed into a deep learning method named auto-encoder networks for unsupervised feature fusion. Finally, a two-layer neural network is employed to regress the liquid level based on fused features. The experiments were conducted in different liquid level, and detection data are divided into model training set and testing set. The mean detection error of proposed method is only 0.034 m when the accuracy of training set is 0.1 m, and 0.0543 m when the accuracy of training set is 0.2 m. In liquid level detection experiments, proposed method also shows good robustness in limited training samples condition and low label accuracy condition, and better detection performance than PCA and STFT method. Experimental results demonstrate that proposed method can directly diagnosis the internal liquid level in PBT terminals and provide an effective maintenance policy.
Bearings are one of the most crucial components in many industrial machines. Effective bearing fault diagnosis is essential for normal and safe machine operation. Existing fault diagnosis methods are mostly limited to manual feature characterization and traditional machine learning schemes such as support vector machines (SVM) and k-nearest neighbors (k-NN) algorithms. Unfortunately, the interpretation and engineering of such features require substantial human expertise. This paper proposes an adaptive deep convolutional neural network (ADCNN) that utilizes a two-dimensional visualization of the raw acoustic emission (AE) signal to provide bearing health state information, which serves to automate and better generalize the feature extraction and classification process. This 2D visualization tool applies a discrete wavelet packet transform (WPT) and quantifies each sub-band of the signal by defining a new evaluation metric—the degree of defectiveness ratio (DDR)— to precisely represent each fault condition, henceforth called a DDRgram. The motivation for using this DDRgram-based preprocessing scheme is that valuable information regarding rotating components is distributed across discrete frequencies, and thus bearing health conditions can be revealed by those frequency spectra. Furthermore, the proposed ADCNN is trained using an adaptive learning method to achieve improved diagnostic performance. The efficiency of the proposed fault diagnosis methodology (DDRgram + ADCNN) is verified using AE data, collected from a benchmark bearings testbed. The experimental studies demonstrate that the proposed approach outperforms existing state-of-the-art algorithms for the multi-fault classification of bearings with good accuracy.
Bearing is commonly used in rotating machinery, and it is significant to monitor bearing running states to ensure machine safety. Performance degradation assessment is an important work in bearing condition-based maintenance (CBM) and predictive maintenance. In this paper, a data-driven bearing performance degradation assessment method based on long short-term memory (LSTM) recurrent neural network (RNN) is proposed to comprehensively utilize the fault propagation information. Firstly, universal degradation simulation model based on vibration response mechanism is constructed for feature verification. A new proposed indicator "waveform entropy (WFE)" is developed and validated by this simulation model. Then waveform entropy and other conventional indicators are input into the LSTM RNN to identify the bearing running state, while particle swarm optimization method is applied to optimize the network structure parameters simultaneously. Experimental results demonstrate that LSTM RNN can effectively identify the bearing degradation states and accurately predict the remaining useful life.
Fault diagnosis is vital for modern industry, and an increasing number of intelligent methods have been proposed for the fault diagnosis. However, most of the studies focus on distinguishing different fault patterns while ignoring fault deterioration. In this paper, a new hierarchical convolutional neural network (HCNN) is proposed as the two-level hierarchical diagnosis network, and it has two characteristics: 1) the fault pattern and fault severity are modeled as one hierarchical structure and 2) the fault pattern and fault severity can be estimated at the same time. Based on these, a new structure of HCNN is designed, which has two classifiers. Then, a two-stage training method is developed for HCNN to train these two classifiers at once training. The proposed HCNN is conducted on three case studies and has achieved state-of-the-art results. The results show that HCNN outperforms traditional two-layer hierarchical fault diagnosis network, and other machine learning and deep learning methods.
Fault diagnosis of rotating machinery is crucial to improve safety, enhance reliability and reduce maintenance cost. The manual feature extraction and selection of traditional fault diagnosis methods depend on signal processing skills and expert experience, which is labor-intensive and time-consuming. As a typical intelligent fault diagnosis method, the convolutional neural network automatically learns features from original data, but it is extremely difficult to design and train a deep network architecture. This paper proposes a fault diagnosis scheme combined of hierarchical symbolic analysis (HSA) and convolutional neural network (CNN), which achieves laborsaving and timesaving preliminary feature extraction and accomplishes automatically feature learning with simplified network architecture. Firstly, hierarchical symbolic analysis is employed to extract features from original signals. The extracted features are able to identify different health conditions under various operating conditions. Then, convolutional neural network instead of human labor is used to learn the complex non-linear relationship between features and health conditions automatically. The architecture of CNN diagnosis model is simple and convenient to implement. Finally, a centrifugal pump dataset and a motor bearing dataset are adopted to validate the effectiveness of the proposed method. The diagnosis results show that the proposed method exhibits superior performance compared with shallow methods and deep learning methods.
One of the main aims of second order cyclostationary (CS2) analysis is the estimation of the full spectral correlation, allowing the identification of different CS2 components in a signal and their characterisation in terms of both spectral frequency f and cyclic frequency α. Unfortunately, traditional estimators of the full spectral correlation (e.g. averaged cyclic periodogram) are highly computationally expensive and hence their application has been quite limited. On the other hand, fast envelope-based CS2 indicators (e.g. cyclic modulation spectrum, CMS) are bound by a cyclic-spectral form of the uncertainty principle, which limits the extent of the cyclic frequency axis αmax at approximately the value chosen for the spectral frequency axis resolution Δf. A recent work has however introduced a ground-breaking approach resulting in a fast algorithm for the calculation of the spectral correlation. This approach is based on the calculation of a series of CMS-like quantities, each scanning a different cyclic-frequency band, given a certain spectral frequency resolution. The superposition of all these quantities allows covering a larger α-band breaking the constraint between maximum cyclic frequency αmax and spectral frequency axis resolution Δf, at a limited computational cost. In this paper a new algorithm for the calculation of the same fast spectral correlation is introduced, resulting in a further computational efficiency gain, and a simplification of the computational procedure.
Wind industry experiences a tremendous growth during the last few decades. As of the end of 2016, the worldwide total installed electricity generation capacity from wind power amounted to 486,790 MW, presenting an increase of 12.5% compared to the previous year. Nowadays wind turbine manufacturers tend to adopt new business models proposing total health monitoring services and solutions, using regular inspections or even embedding sensors and health monitoring systems within each unit. Regularly planned or permanent monitoring ensures a continuous power generation and reduces maintenance costs, prompting specific actions when necessary. The core of wind turbine drivetrain is usually a complicated planetary gearbox. One of the main gearbox components which are commonly responsible for the machinery breakdowns are rolling element bearings. The failure signs of an early bearing damage are usually weak compared to other sources of excitation (e.g., gears). Focusing toward the accurate and early bearing fault detection, a plethora of signal processing methods have been proposed including spectral analysis, synchronous averaging and enveloping. Envelope analysis is based on the extraction of the envelope of the signal, after filtering around a frequency band excited by impacts due to the bearing faults. Kurtogram has been proposed and widely used as an automatic methodology for the selection of the filtering band, being on the other hand sensible in outliers. Recently, an emerging interest has been focused on modeling rotating machinery signals as cyclostationary, which is a particular class of nonstationary stochastic processes. Cyclic spectral correlation and cyclic spectral coherence (CSC) have been presented as powerful tools for condition monitoring of rolling element bearings, exploiting their cyclostationary behavior. In this work, a new diagnostic tool is introduced based on the integration of the cyclic spectral coherence (CSC) along a frequency band that contains the diagnostic information. A special procedure is proposed in order to automatically select the filtering band, maximizing the corresponding fault indicators. The effectiveness of the methodology is validated using the National Renewable Energy Laboratory (NREL) wind turbine gearbox vibration condition monitoring benchmarking dataset which includes various faults with different levels of diagnostic complexity.
Fault diagnosis of rotating machinery plays a significant role for the reliability and safety of modern industrial systems. As an emerging field in industrial applications and an effective solution for fault recognition, artificial intelligence (AI) techniques have been receiving increasing attention from academia and industry. However, great challenges are met by the AI methods under the different real operating conditions. This paper attempts to present a comprehensive review of AI algorithms in rotating machinery fault diagnosis, from both the views of theory background and industrial applications. A brief introduction of different AI algorithms is presented first, including the following methods: k-nearest neighbour, naive Bayes, support vector machine, artificial neural network and deep learning. Then, a broad literature survey of these AI algorithms in industrial applications is given. Finally, the advantages, limitations, practical implications of different AI algorithms, as well as some new research trends, are discussed.
Vibration signals of gearboxes working under time-varying conditions are non-stationary, which causes difficulties to the fault diagnosis. Based on the techniques of signal sparse decomposition and order tracking, a novel method is proposed to extract fault features from non-stationary vibration signals of gearboxes. The method contains two key procedures, the quasi-steady component separation in angle domain and the impact resonance component extraction in time domain. The sparse dictionary including quasi-steady sub-dictionary and impact sub-dictionary is specifically designed according to the time-frequency characteristics of steady-type fault and impact-type fault. The former sub-dictionary consists of cosine functions and is based on the order spectrum information of angle domain signal. The latter sub-dictionary consists of the unit impulse response of multiple-degree-of-freedom vibration system whose modal parameters are self-adaptively recognized by the method of correlation filtering. An improved matching pursuit algorithm on segmental signal is designed to solve sparse coefficients and reconstruct steady-type fault components and impact-type fault components. The simulation analyses show that the proposed method is capable to process the signal with 30% speed fluctuation and −1.5 dB signal-to-noise ratio (SNR), in which the SNR of impact-type fault components is as low as −14.6 dB. The effectiveness is further verified by experimental tests on a fixed-shaft gearbox and a planetary gearbox.
Deep learning has attracted attentions in intelligent fault diagnosis of machinery because it allows a deep network to accomplish the tasks of feature learning and fault classification automatically. Among deep learning models, convolutional neural networks (CNNs) are able to learn features from mechanical vibration signals and thus several studies have applied CNNs in intelligent fault diagnosis of machinery. However, these studies suffer from the following weaknesses. (1) The imbalanced distribution of machinery health conditions is not considered. (2) What CNNs have learned is not clear. Therefore, in this paper, a framework called deep normalized convolutional neural network (DNCNN) is proposed for imbalanced fault classification of machinery to overcome the first weakness. Meanwhile, neuron activation maximization (NAM) algorithm is developed to handle the second weakness. To verify the proposed methods, three bearing datasets containing single faults and compound faults are constructed with different imbalanced degrees. The classification accuracies of the three datasets demonstrate that DNCNN is able to deal with the imbalanced classification problem more effectively than the commonly used CNNs. By analyzing the kernels of the convolutional layers of DNCNN via NAM algorithm, we find that these kernels act as filters and they become complex when the layers go deeper. This result may help us understand what DNCNN has learned in intelligent fault diagnosis of machinery.
This paper proposes a new signal feature extraction and fault diagnosis method for fault diagnosis of low-speed machinery. Statistic filter (SF) and wavelet package transform (WPT) are combined with moving-peak-hold method (M-PH) to extract features of a fault signal, and special bearing diagnostic symptom parameters (SSPs) in a frequency domain that are sensitive to bearing fault diagnosis are defined to recognize fault types. The SF is first used to adaptively cancel noises, and then fault detection is performed by exploiting the optimum symptom parameters in a time domain to identify a normal or fault state. For precise diagnosis, the SSPs are calculated after the signals are processed by M-PH and WPT. A decision tree is used to structure intelligent diagnosis rules in each step until the states are fully and automatically detected. The efficacy of this method was confirmed by applying it to an experimental low-speed rotation machine.
Fault diagnosis is vital in manufacturing system, since early detections on the emerging problem can save invaluable time and cost. With the development of smart manufacturing, the data-driven fault diagnosis becomes a hot topic. However, the traditional data-driven fault diagnosis methods rely on the features extracted by experts. The feature extraction process is an exhausted work and greatly impacts the final result. Deep learning (DL) provides an effective way to extract the features of raw data automatically. Convolutional neural network (CNN) is an effective DL method. In this research, a new CNN based on LeNet-5 is proposed for fault diagnosis. Through a conversion method converting signals to 2-D images, the proposed method can extract the features of converted 2-D images and eliminate the effect of handcrafted features. The proposed method which is tested on three famous datasets, including motor bearing dataset, self-priming centrifugal pump dataset, axial piston hydraulic pump dataset, has achieved prediction accuracy of 99.79%, 99.481% and 100% respectively. The results have been compared with other DL and traditional methods, including adaptive deep CNN, sparse filter, deep belief network, and support vector machine. The comparisons show that the proposed CNN based data-driven fault diagnosis method has achieved significant improvements.
In recent years, intelligent fault diagnosis algorithms using machine learning technique have achieved much success. However, due to the fact that in real world industrial applications, the working load is changing all the time and noise from the working environment is inevitable, degradation of the performance of intelligent fault diagnosis methods is very serious. In this paper, a new model based on deep learning is proposed to address the problem. Our contributions of include: First, we proposed an end-to-end method that takes raw temporal signals as inputs and thus doesn’t need any time consuming denoising preprocessing. The model can achieve pretty high accuracy under noisy environment. Second, the model does not rely on any domain adaptation algorithm or require information of the target domain. It can achieve high accuracy when working load is changed. To understand the proposed model, we will visualize the learned features, and try to analyze the reasons behind the high performance of the model.
The operation conditions of the rotating machinery are always complex and variable, which makes it difficult to automatically and effectively capture the useful fault features from the measured vibration signals, and it is a great challenge for rotating machinery fault diagnosis. In this paper, a novel deep autoencoder feature learning method is developed to diagnose rotating machinery fault. Firstly, the maximum correntropy is adopted to design the new deep autoencoder loss function for the enhancement of feature learning from the measured vibration signals. Secondly, artificial fish swarm algorithm is used to optimize the key parameters of the deep autoencoder to adapt to the signal features. The proposed method is applied to the fault diagnosis of gearbox and electrical locomotive roller bearing. The results confirm that the proposed method is more effective and robust than other methods.
To assess health conditions of rotating machinery efficiently, multiple accelerometers are mounted on different locations to acquire a variety of possible faults signals. The statistical features are extracted from these signals to identify the running status of a machine. However, the acquired vibration signals are different due to sensor's arrangement and environmental interference, which may lead to different diagnostic results. In order to improve the fault diagnosis reliability, a new multisensor data fusion technique is proposed. First, time-domain and frequency-domain features are extracted from the different sensor signals, and then these features are input into multiple two-layer sparse autoencoder (SAE) neural networks for feature fusion. Finally, fused feature vectors can be regarded as the machine health indicators, and be used to train deep belief network (DBN) for further classification. To verify the effectiveness of the proposed SAE-DBN scheme, the bearing fault experiments were conducted on a bearing test platform, and the vibration data sets under different running speeds were collected for algorithm validation. For comparison, different feature fusion methods were also applied to multisensor fusion in the experiments. Experimental results demonstrated that the proposed approach can effectively identify the machine running conditions and significantly outperform other fusion methods.
In most current intelligent diagnosis methods, fault classifiers of electric machine are built based on complex handcrafted features extractor from raw signals, which depend on prior knowledge and is difficult to implement intelligentization authentically. What’s more, the increasingly complicated industrial structures and data make handcrafted features extractors less suited. Convolutional neural network (CNN) provides an efficient method to act on raw signals directly by weight sharing and local connections without feature extractors. However, effective as CNN works on image recognition, it does not work well in industrial applications due to the differences between image and industrial signals. Inspired by the idea of CNN, we develop a novel diagnosis framework based on the characteristics of industrial vibration signals, which is called dislocated time series convolutional neural network (DTS-CNN). The DTS-CNN architecture is composed of dislocate layer, convolutional layer, sub-sampling layer and fully connected layer. By adding a dislocate layer, this model can extract the relationship between signals with different intervals in periodic mechanical signals, thereby overcome the weaknesses of traditional CNNs and is more applicable for modern electric machines, especially under non-stationary conditions. Experiments under constant and nonstationary conditions are performed on a machine fault simulator to validate the proposed framework. The results and comparison with respect to the state of the art in the field is illustrated in detail, which highlights the superiority of the proposed method in industrial applications.
Traditional artificial methods and intelligence-based methods of classifying and diagnosing various mechanical faults with high accuracy by extracting effective features from vibration data, such as support vector machines and back propagation neural networks, have been widely investigated. However, the problems of extracting features automatically without significantly increasing the demand for machinery expertise and maximizing accuracy without overcomplicating machine structure have to date remained unsolved. Therefore, a novel hierarchical learning rate adaptive deep convolution neural network based on an improved algorithm was proposed in this study, and its use to diagnose bearing faults and determine their severity was investigated. To test the effectiveness of the proposed method, an experiment was conducted with bearing-fault data samples obtained from a test rig. The method achieved a satisfactory performance in terms of both fault-pattern recognition and fault-size evaluation. In addition, comparison revealed that the improved algorithm is well suited to the fault-diagnosis model, and that the proposed method is superior to other existing methods.
Influenced by factors such as speed fluctuation, rolling element sliding and periodical variation of load distribution and impact force on the measuring direction of sensor, the impulse response signals caused by defective rolling bearing are non-stationary, and the amplitudes of the impulse may even drop to zero when the fault is out of load zone. The non-stationary characteristic and impulse missing phenomenon reduce the effectiveness of the commonly used demodulation method on rolling element bearing fault diagnosis. Based on sparse representation theories, a new approach for fault diagnosis of rolling element bearing is proposed. The over-complete dictionary is constructed by the unit impulse response function of damped second-order system, whose natural frequencies and relative damping ratios are directly identified from the fault signal by correlation filtering method. It leads to a high similarity between atoms and defect induced impulse, and also a sharply reduction of the redundancy of the dictionary. To improve the matching accuracy and calculation speed of sparse coefficient solving, the fault signal is divided into segments and the matching pursuit algorithm is carried out by segments. After splicing together all the reconstructed signals, the fault feature is extracted successfully. The simulation and experimental results show that the proposed method is effective for the fault diagnosis of rolling element bearing in large rolling element sliding and low signal to noise ratio circumstances.
It is a big challenge to identify the most effective features for enhancement of fault classification accuracy in rotating machines due to nonstationary and nonlinear vibration characteristics of the machines under varying operating conditions. To find discriminative features, a novel dimension reduction algorithm, referred to as the nearest and farthest distance preserving projection (NFDPP), is proposed for machine fault feature extraction and classification. With the NFDPP, both the nearest and farthest samples of the data manifold can be analyzed simultaneously to identify features leading to fault classification. Additionally, we denoise the features directly in the feature space to save computation time and storage space, and prove its equivalence to denoising the signals in the time domain. Through analysis of measured vibration data for bearings with different defects, it is demonstrated that the proposed NFDPP approach can effectively classify different bearing faults and identify the severity of the bearing ball defect, and the direct denoising of features yield a significant improvement in fault classification. The effectiveness of the proposed method is further validated in identifying compound faults in locomotive bearings in an industrial setting.
Training Deep Neural Networks is complicated by the fact that the
distribution of each layer's inputs changes during training, as the parameters
of the previous layers change. This slows down the training by requiring lower
learning rates and careful parameter initialization, and makes it notoriously
hard to train models with saturating nonlinearities. We refer to this
phenomenon as internal covariate shift, and address the problem by normalizing
layer inputs. Our method draws its strength from making normalization a part of
the model architecture and performing the normalization for each training
mini-batch}. Batch Normalization allows us to use much higher learning rates
and be less careful about initialization. It also acts as a regularizer, in
some cases eliminating the need for Dropout. Applied to a state-of-the-art
image classification model, Batch Normalization achieves the same accuracy with
14 times fewer training steps, and beats the original model by a significant
margin. Using an ensemble of batch-normalized networks, we improve upon the
best published result on ImageNet classification: reaching 4.9% top-5
validation error (and 4.8% test error), exceeding the accuracy of human raters.
This paper presents an approach to implement vibration, pressure, and current signals for fault diagnosis of the valves in reciprocating compressors. Due to the complexity of structure and motion of such compressor, the acquired vibration signal normally involves transient impacts and noise. This causes the useful information to be corrupted and difficulty in accurately diagnosing the faults with traditional methods. To reveal the fault patterns contained in this signal, the Teager–Kaiser energy operation (TKEO) is proposed to estimate the amplitude envelopes. In case of pressure and current, the random noise is removed by using a denoising method based on wavelet transform. Subsequently, statistical measures are extracted from all signals to represent the characteristics of the valve conditions. In order to classify the faults of compressor valves, a new type of learning architecture for deep generative model called deep belief networks (DBNs) is applied. DBN employs a hierarchical structure with multiple stacked restricted Boltzmann machines (RBMs) and works through a greedy layer-by-layer learning algorithm. In pattern recognition research areas, DBN has proved to be very effective and provided with high performance for binary values. However, for implementing DBN to fault diagnosis where most of signals are real-valued, RBM with Bernoulli hidden units and Gaussian visible units is considered in this study. The proposed approach is validated with the signals from a two-stage reciprocating air compressor under different valve conditions. To confirm the superiority of DBN in fault classification, its performance is compared with that of relevant vector machine and back propagation neuron networks. The achieved accuracy indicates that the proposed approach is highly reliable and applicable in fault diagnosis of industrial reciprocating machinery.
A hybrid two stage one-against-all Support Vector Machine (SVM) approach is proposed for the automated diagnosis of defective rolling element bearings. The basic concept and major advantage of the method, is that its training can be performed using simulation data, which result from a well established model, describing the dynamic response of defective rolling element bearings. Then, vibration measurements, resulting from the machine under condition monitoring, can be imported and processed directly by the already trained SVM, eliminating thus the need of training the SVM with experimental data of the specific defective bearing. A key aspect of the method is the data preprocessing approach, which among others, includes order analysis, in order to overcome problems related to sudden changes of the shaft rotating speed. Moreover, frequency domain features both from the raw signal as well as from the demodulated signal are used as inputs to the SVM classifiers for a two-stage recognition and classification procedure. At the first stage, a SVM classifier separates the normal condition signals from the faulty signals. At the second stage, a SVM classifier recognizes and categorizes the type of the fault. The effectiveness of the method tested in one literature established experimental test case and in three different industrial test cases, including a total number of 34 measurements. Each test case includes successive measurements from bearings under different types of defects, different loads and different rotation speeds. In all cases, the method presents 100% classification success.
This paper addresses the spectral analysis of cyclostationary (CS) signals from a generic point of view, with the aim to provide the practical conditions of success in a wide range of applications, such as in mechanical vibrations and acoustics. Specifically, it points out the similarities, differences and potential pitfalls associated with cyclic spectral analysis as opposed to classical spectral analysis. It is shown that non-parametric cyclic spectral estimators can all be derived from a general quadratic form, which yields as particular cases “cyclic” versions of the smoothed, averaged, and multitaper periodograms. The performance of these estimators is investigated in detail on the basis of their frequency resolution, cyclic leakage, systematic and stochastic estimation errors. The results are then extended to more advanced spectral quantities such as the cyclic coherence function and the Wigner–Ville spectrum of CS signals. In particular an optimal estimator of the Wigner–Ville spectrum is found, with remarkable properties. Several examples of cyclic spectral analyses, with an emphasis on mechanical systems, are finally presented in order to illustrate the value of such a general treatment for practical applications.
A K-means clustering approach is proposed for the automated diagnosis of defective rolling element bearings. Since K-means clustering is an unsupervised learning procedure, the method can be directly implemented to measured vibration data. Thus, the need for training the method with data measured on the specific machine under defective bearing conditions is eliminated. This fact consists the major advantage of the method, especially in industrial environments. Critical to the success of the method is the feature set used, which consists of a set of appropriately selected frequency-domain parameters, extracted both from the raw signal, as well as from the signal envelope, as a result of the engineering expertise, gained from the understanding of the physical behavior of defective rolling element bearings. Other advantages of the method are its ease of programming, simplicity and robustness. In order to overcome the sensitivity of the method to the choice of the initial cluster centers, the initial centers are selected using features extracted from simulated signals, resulting from a well established model for the dynamic behavior of defective rolling element bearings. Then, the method is implemented as a two-stage procedure. At the first step, the method decides whether a bearing fault exists or not. At the second step, the type of the defect (e.g. inner or outer race) is identified. The effectiveness of the method is tested in one literature established laboratory test case and in three different industrial test cases. Each test case includes successive measurements from bearings under different types of defects. In all cases, the method presents a 100% classification success. Contrarily, a K-means clustering approach, which is based on typical statistical time domain based features, presents an unstable classification behavior.
Vibration based condition monitoring of planetary gearboxes operating under speed varying operating conditions based on cyclo-non-stationary
- A Mauricio
- J Qi
- W Smith
- R Randall
- K Gryllias
A. Mauricio, J. Qi, W. Smith, R. Randall, K. Gryllias, Vibration based condition monitoring of planetary gearboxes operating under speed varying
operating conditions based on cyclo-non-stationary, Analysis 61 (2019) 265-279.