Raw Vibration Signals Research Articles

A power information guided-variational mode decomposition (PIVMD) and its application to fault diagnosis of rolling bearing

As a powerful tool of mode component extraction, variational mode decomposition (VMD) has found its wide application in fault diagnosis. However, VMD still suffers from an unsolved problem. To be specific, how to adaptively determine the decomposition number, the initial center frequencies and the variable penalty factors in VMD according to the characteristics of vibration signal is still an unsolved problem. To address this problem, the authors propose a novel mode component extraction method, namely, the power information guided-variational mode decomposition (PIVMD). The essential idea of the proposed method is using the enveloped auto-regressive (AR) power spectrum of raw vibration signal to determine the VMD parameters (decomposition number, initial center frequencies and variable penalty factors). After determining these VMD parameters, an iterative search is conducted to obtain the optimal solution for constructed variational model. A noise reduction envelope spectrum is further used to enhance the weak fault features contained in the extracted mode component. A set of simulation signals representing the faulty characteristics of rolling bearing with inner/outer race faults is analyzed by the proposed PIVMD method to test its mode extraction ability. Then, an open dataset from Case Western Reserve University and a measured dataset from a steel mill in Ma Steel Co. as well as an experimental dataset collected from the authors' laboratory are analyzed to verify the fault diagnosis capability of the proposed PIVMD method. Meanwhile, the analyzed results are compared with those obtained from other mode decomposition methods to highlight the superiority of the proposed PIVMD method in terms of extracting mode components with rich fault information.

Intelligent Motor Bearing Fault Diagnosis Using Channel Attention-Based CNN

Many components of electric vehicles contain rolling bearings, and the operating condition of rolling bearings often affects the operating performance of electric vehicles. Monitoring the operating status of the bearings is one of the key technologies to ensure the safe operation of the bearings. We propose a channel attention-based convolutional neural network (CA-CNN) model for rolling bearing fault diagnosis. The model can directly use the raw vibration signal of the bearing as input to achieve bearing fault diagnosis under different operating loads and different noise environments. The experimental results show that, compared with other intelligent diagnosis methods, the proposed model CA-CNN achieves a high diagnostic accuracy under different load cases and still has advantages in different noisy environments. It is also beneficial to promote the intelligent fault diagnosis and maintenance of electric vehicles.

Impulse feature extraction via combining a novel voting index and a variational model penalized by center frequency constraint

A key step in rotating machinery fault diagnosis is extracting the subcomponent that best characterizes the fault features. To this end, a fault feature extraction methodology based on an adaptive spectrum segmentation method, a new voting index, and a new variational model is constructed in this paper. More specifically, the spectrum is divided into a series of sub-bands through a fast iteration filter by changing the width of the window. Then, by considering the amplitudes and locations of the fault feature spectral lines in squared envelope spectrum, a novel index based on voting mechanism is constructed to evaluate the periodic impulses in each sub-band. Finally, a new variational model penalized by a center frequency penalty item is proposed to extract the fault characteristics from the raw vibration signal by maximining the proposed index iteratively. The proposed methodology is applied to extract the fault characteristics from rolling bearing and planetary gearbox signals. The comparative results indicate the proposed method is superior to fast kurtogram, fast entrogram, and the proposed fault feature extraction methods based on other classical indices.

Bearing Fault Diagnosis Based on Randomized Fisher Discriminant Analysis.

In this paper, a novel randomized Fisher discriminant analysis (RFDA) based bearing fault diagnosis method is proposed. First, several representative time-domain features are extracted from the raw vibration signals. Second, linear Fisher discriminant analysis (FDA) is extended to nonlinear FDA named RFDA by introducing the random feature map to deal with the non-linearity issue. Specifically, the extracted time-domain features data are mapped onto a high-dimensional space using the random feature map function rather than kernel functions. Third, the time-domain features are fed into the built RFDA model to extract the discriminant features for diagnosis. Moreover, a Bayesian inference is employed to identify the class of the collected vibration signals to diagnose the bearing status. The proposed method uses random Fourier features to approximate the kernel matrix in the kernel Fisher discriminant analysis. Through employing randomized Fisher discriminant analysis, the nonlinearity issue is dealt with, and the computational burden is remarkably reduced compared to the kernel Fisher discriminant analysis (KFDA). To illustrate the superior performance of the proposed RFDA-based bearing fault diagnosis method, comparative experiments are conducted on two widely used datasets, the Case Western Reserve University (CWRU) bearing dataset and the Paderborn University (PU) bearing dataset. For the CWRU dataset, the computation time of RFDA is much shorter than KFDA, while the accuracy rate reaches the same level of KFDA. For the PU dataset, the accuracy rate of RFDA is slightly higher than KFDA, and the computation time is only 44.14% of KFDA.

A deep capsule neural network with data augmentation generative adversarial networks for single and simultaneous fault diagnosis of wind turbine gearbox

The fault diagnosis (FD) of wind turbine gearbox (WTG) is of special importance for keeping the wind turbine drivetrain working normally and safely. However, owing to the limited training data and the mutual interference of various mechanical parts, it is of great difficulty to realize the simultaneous-fault monitoring task of WTG using existing intelligent FD methods or manual inspection-based approaches. To tackle the issue, a deep capsule neural network with data augmentation generative adversarial networks, named ST-DAGANs-CapNet, is developed for the single and simultaneous FD of WTG by integrating capsule neural network (CapsNet) with Stockwell transform (ST) and data augmentation generative adversarial networks (DAGANs). The proposed ST-DAGANs-CapNet method mainly consists of three steps. First of all, ST is adopted to extract two-dimension (2-d) image features of time–frequency domain from raw time-domain vibration signals of WTG. Then, DAGANs are employed for generating more fake image samples to address the problem of lacking training data. At last, the built CapsNet model is utilized to diagnose the single and compound faults of WTG by the primary 2-d feature images and the made fake 2-d feature images in training set. Two experimental studies are implemented to prove the effectiveness of the proposed method, and the result is compared with some existing intelligent FD of WTG. It indicates that DAGANs are effective in helping to tackle the issue of limited and unbalanced training samples in real FD of WTG, and the diagnosis result of the proposed approach in test sample set is better than that of several commonly used FD methods in literatures.

M2FN: An end-to-end multi-task and multi-sensor fusion network for intelligent fault diagnosis

Intelligent fault diagnosis based on multi-sensor fusion has gained considerable attention in various modern industrial applications. However, it is still challenging to extract discriminative features from multi-sensor data to provide an accurate and reliable diagnosis. For this purpose, this paper proposes a new multi-task multi-sensor fusion network (M2FN) to improve fault diagnosis performance. The proposed method first uses convolutional neural networks to extract and fuse features from raw vibration and current signals. After that, to improve the discriminative ability of the learned features, a multi-task learning module (MTL) is designed which contains a classification task and a deep metric learning task. Our proposed M2FN model is evaluated on a bearing dataset and a gearbox dataset. Experimental results show that our proposed M2FN method significantly outperforms the compared single-sensor-based and single-task-based methods in terms of diagnosis accuracy, and the learned features present better inter-class discriminability and intra-class concentration through the feature visualization analysis.

Bolt-Loosening Detection Using 1D and 2D Input Data Based on Two-Stream Convolutional Neural Networks.

At present, the detection accuracy of bolt-loosening diagnoses is still not high. In order to improve the detection accuracy, this paper proposes a fault diagnosis model based on the TSCNN model, which can simultaneously extract fault features from vibration signals and time-frequency images and can precisely detect the bolt-loosening states. In this paper, the LeNet-5 network is improved by adjusting the size and number of the convolution kernels, introducing the dropout operation, and building a two-dimensional convolutional neural network (2DCNN) model. Combining the advantages of a one-dimensional convolutional neural network (1DCNN) with wide first-layer kernels to suppress high-frequency noise, a two-stream convolutional neural network (TSCNN) is proposed based on 1D and 2D input data. The proposed model uses raw vibration signals and time-frequency images as input and automatically extracts sensitive features and representative information. Finally, the effectiveness and superiority of the proposed approach are verified by practical experiments that are carried out on a machine tool guideway. The experimental results show that the proposed approach can effectively achieve end-to-end bolt-loosening fault diagnoses, with an average recognition accuracy of 99.58%. In addition, the method can easily achieve over 93% accuracy when the SNR is over 0 dB without any denoising preprocessing. The results show that the proposed approach not only achieves high classification accuracy but also has good noise immunity.

A New Dynamic Balance Framework Based on Blind Source Separation under Multiple Fault Conditions

Accordingly, the mass unbalance of the rotors is usually the major cause of excessive vibration. The information extracted at the fundamental frequency is often employed to fix the unbalance issue. However, other rotor faults like rotor bending and bearing-failure effect also generate additional components to the characteristics. Thus, it is necessary to isolate the corresponding features and obtain the intrinsic causes of the multiple failures. In this paper, a productive hybrid method is successfully developed to deal with the root mass unbalance problem with additional force interference by integrating the superiority of different methods, including Ensemble Empirical Mode Decomposition (EEMD) and Nonnegative Matrix Factorization (NMF), where EEMD is used to obtain sensitive IMFs and NMF is employed to acquire the inherent source signal, respectively. Meanwhile, a root dynamic balancing and implemental framework is also developed to accomplish the task of vibration reduction. For verification, a serial of simulations and experimental investigations have been analysed to demonstrate the preferable potentialities of the proposed method. In particular, a standard Bently Nevada rotor rig with a specifically designed device was employed to simulate appended faults by adjusting the additional forces during the experimental steps. The analysis results show that the proposed method can isolate and extract the unbalance faults from the raw vibration signals and achieve accurate correction balancers, where a nearly identical correction angle has been achieved, which indicates that the optimal installation position has been successfully figured out.

Domain adaptation network base on contrastive learning for bearings fault diagnosis under variable working conditions

Unsupervised domain adaptation (UDA)-based methods have made great progress in bearing fault diagnosis under variable working conditions. However, most existing UDA-based methods focus only on minimizing the discrepancy of two working conditions. The similarity of fault features extracted from the bearing vibration signal is ignored. The samples near the distribution boundaries learned by the network might be misclassified. As a result, even if the marginal distributions is aligned well, the diagnosis result may not be satisfactorily. Therefore, this paper proposes a domain adaptation network base on contrastive learning (DACL) to achieve the aim of bearing fault diagnosis cross different working conditions and reduce the probability of samples being classified near or on the boundary of each class to improve diagnosis accuracy. The method is made up of a feature mining module and an adversarial domain adaptation module. In the feature mining module, a one-dimensional Convolutional Neural Network (1-D CNN) is utilized to extract features from raw vibration signals. The adversarial domain adaptation module followed is designed to learn domain-shared discriminant features for aligning marginal distribution. Meanwhile, the contrastive estimation term is designed to quantize the similarity of data distribution and increase the distance between samples of different health conditions, declining the probability of samples near the boundary and improving diagnosis performance. At last, an adaptive factor is introduced to measure the relative importance of transferring and discriminating abilities of the method. The effectiveness of the proposed method is confirmed by examining various fault diagnosis scenarios with domain discrepancies across the source and target domains, using experimental data from two bearing systems.

Application potential of vibration sensors for online process monitoring of hydrocyclones

ABSTRACT An attempt has been made to develop a non-intrusive and robust technique for online process monitoring of hydrocyclones using commercially available vibration sensors. Vibration signals were captured first at the spigots of 50.8 mm and 76.2 mm diameter hydrocyclones while using water only at different operating conditions. For denoising the captured raw vibration signals, a wavelet-based method coupled with median filtering-based technique was used. The Root Mean Square (RMS) values of vibration accelerations induced in the hydrocyclone were found to be proportional to the total volumetric feed rate. Based on this observation, it has been demonstrated how suitable empirical correlations can be developed which can ultimately be used as an online process monitoring tool for hydrocyclones.

Gated Adaptive Hierarchical Attention Unit Neural Networks for the Life Prediction of Servo Motors

The remaining useful life (RUL) prediction of servo motors plays a greatly indispensable role in preventing abnormal and inefficient operation of servo drive system and even major safety accident. In this article, a novel variant of recurrent neural networks, called gated adaptive hierarchical attention unit (GAHAU), is constructed to efficiently predict the RULs of servo motors, which can not only pay attention to the outputs of reset and update gates but also adaptively divide the hierarchy of the sequence information in attention gate. The updating formulas of GAHAU are derived. With the raw vibration signals of servo motors, a quadratic function-based deep convolutional autoencoder is used for automatically generating the health indicator vector. Based on the constructed HI vector, the RULs of servo motors are accurately predicted by GAHAU. The experiments on servo motors demonstrate that the presented GAHAU has remarkable prediction performance especially for the long-term prediction. Compared to the existing typical RUL prediction methods, the proposed methodology has stronger predictive ability according to the comparative results.

Probability-driven identification mechanism for degradation of magnetic drive pumps

Pump-state prediction and health management have entered the intelligent era. Data analysis, feature extraction, and automatic classification are the critical stages of the state self-recovery regulation of machines. To explore the identification mechanism of degraded states in magnetic drive pumps, the wavelet packet transform is utilised to filter the raw vibration signals. A classification model is subsequently established based on K-means clustering analysis. The highly sensitive characteristic parameters are accessed via a corresponding pre-processing procedure. Herein, clustering points are acquired, and the detected states are classified. Moreover, the probability of operating states over time is ascertained using hidden Markov models. Thus, the healthy machine states are validated via comparison of the calculated results, indicating that the trigger mechanism can recognise the degraded machine states successfully. The proposed probability-driven identification mechanism makes the automatic identification and intelligent decision-making of self-recovery systems possible, and may be used to provide technical details for application to other rotating machinery systems.

Intelligent Diagnosis Method for Mechanical Faults of High-Voltage Shunt Reactors Based on Vibration Measurements

Aiming at the difficulty of accurately identifying latent mechanical faults inside high-voltage shunt reactors (HVSRs), this paper proposes a new method for HVSR state feature extraction and intelligent diagnosis. The method integrates a modified complementary ensemble empirical mode decomposition (CEEMD)–permutation entropy–CEEMD (MCPCEEMD) method, mutual information theory (MI), multiscale fuzzy entropy (MFE), and an improved grasshopper optimization algorithm to optimize the probabilistic neural network (IGOA-PNN) model. First, we introduce MCPCEEMD for suppressing modal aliasing to decompose the HVSR raw vibration signals. Then, the correlation degree between the obtained intrinsic mode function (IMF) components and the HVSR original vibration signals is judged through MI, and the IMF with the highest correlation is selected for feature extraction. Furthermore, this study uses MFE to quantify the selected IMF. Finally, we employ piecewise inertial weights to improve GOA to select the best smoothing factor for PNN, and use the optimized IGOA-PNN model to identify feature subsets. The experimental results show that the proposed method can successfully diagnose different types and degrees of HVSR mechanical faults, and the identification accuracy rate reaches more than 98%. The high recognition accuracy of the proposed method is helpful for the state detection and field application of HVSRs.

Ensemble Capsule Network with an Attention Mechanism for the Fault Diagnosis of Bearings from Imbalanced Data Samples.

In order to solve the problem of imbalanced and noisy data samples for the fault diagnosis of rolling bearings, a novel ensemble capsule network (Capsnet) with a convolutional block attention module (CBAM) that is based on a weighted majority voting method is proposed in this study. Firstly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method was used to decompose the raw vibration signal into different IMF signals, which are noise reduction signals. Secondly, the IMF signals were input into the Capsnet with CBAM in order to diagnose the fault category preliminarily. Finally, the weighted majority voting method was utilized so as to fuse all of the preliminary diagnosis results in order to obtain the final diagnostic decision. In order to verify the effectiveness of the proposed ensemble of Capsnet with CBAM, this method was applied to the fault diagnosis of rolling bearings with imbalanced and different SNR data samples. The diagnostic results show that the proposed diagnostic method can achieve higher levels of accuracy than other methods, such as single CNN, single Capsnet, ensemble CNN and an ensemble capsule network without CBAM and that it has stronger immunity to noise than an ensemble capsule network without CBAM.

A transformer model with enhanced feature learning and its application in rotating machinery diagnosis

Deep learning has become the prevailing trend of intelligent fault diagnosis for rotating machines. Compared to early-stage methods, deep learning methods use automatic feature extraction instead of manual feature design. However, conventional intelligent diagnosis models are trapped by a dilemma that simple models are unable to tackle difficult cases, while complicated models are likely to over-parameterize. In this paper, a transformer-based model, Periodic Representations for Transformers (PRT) is proposed. PRT uses a dense-overlapping split strategy to enhance the feature learning inside sequence patches. Combined with the inherent capability of capturing long range dependencies of transformer, and the further information extraction of class-attention, PRT has excellent feature extraction abilities and could capture characteristic features directly from raw vibration signals. Moreover, PRT adopts a two-stage positional encoding method to encode position information both among and inside patches, which could adapt to different input lengths. A novel inference method to use larger inference sample sizes is further proposed to improve the performance of PRT. The effectiveness of PRT is verified on two datasets, where it achieves comparable and even better accuracies than the benchmark and state-of-the-art methods. PRT has the least FLOPs among the best performing models and could be further improved by the inference strategy, reaching an accuracy near 100%.

Multiscale One-Class Classification Network for Machine Health Monitoring

Currently, most machine health monitoring methods require labeled abnormal or faulty samples to build a supervised classification model. However, it is difficult and expensive to collect enough faulty samples for accurate and reliable fault detection and diagnosis. To address this issue, we formulate machine health monitoring as a one-class classification problem and propose a new unsupervised health monitoring model named multiscale one-class classification network (MOCN) only relying on normal data without any label information. It consists of two sequential learning modules: a multiscale feature representation learning module to extract effective features from raw one-dimensional vibration signals and a one-classifier training module to build a hypersphere model to contain all normal samples with a minimum radius. Unlike traditional two-step one-class classification methods, both modules in our proposed MOCN are jointly optimized, and therefore it can simultaneously learn compact representations and train a deep one-class classifier. Our proposed MOCN method is evaluated through experiments on three datasets, and experimental results have proved the superiority of our proposed MOCN model compared with several existing methods.

Maximal overlap discrete wavelet packet transforms-based bipolar neutrosophic cross entropy measure for identification of rotor defects

This study propounds a novel methodology for automatic identification of rotor defects severity, when the machine is operated at constant speed, through maximal overlap discrete wavelet packet transforms (MODWPT) and proposed cross entropy measures of bipolar neutrosophic sets, single valued neutrosophic sets and fuzzy sets respectively. After the 3rd level of decomposition of raw vibration signals into eight frequency bands (called as WPT3 bands) through MODWPT, the energy reading in each WPT3 band is monitored. Thereafter, the lower (here called as truth membership degrees) and upper energy bounds from the training and testing samples of various rotor defect conditions at each WPT3 band are extracted and then normalized to lie in [0,1]. Finally, the energy intervals are constructed and thereafter converted into the desired forms of fuzzy sets, bipolar neutrosophic sets and single valued neutrosophic sets consecutively. The minimum cross entropy measure values between the fuzzy sets (Method 1), single valued neutrosophic sets (Method 2) and bipolar neutrosophic sets (Method 3) of training and testing samples of rotor defect conditions are computed and utilized to identify defect conditions such as angular/parallel misalignment of 10 Mils, unbalance, rub and defect free respectively. The novelty of the proposed rotor defect identification methodology lies in the fact that all the proposed variants of bipolar neutrosophic cross entropy measure have been found skilled enough to identify the desired rotor defects under study. On another hand, the existing cosine similarity measure [1] does not possess the necessary capability of identifying all rotor defects, especially, the rub defect condition. Furthermore, like the existing cosine similarity measure, our proposed variants of bipolar neutrosophic cross entropy measures have been found equally compatible for fairly furnishing the consistent and feasible results under intuitive analysis.

Bearing fault diagnosis using transfer learning and self-attention ensemble lightweight convolutional neural network

The rapid development of big data leads to many researchers focusing on improving bearing fault classification accuracy using deep learning models. However, implementing a deep learning model on a limited resources platform such as the smartphone or STM32 includes two difficulties: making the model as lightweight as possible and reducing the dependence on large training samples. To this end, a self-attention ensemble lightweight model combined with the transfer learning (SLTL) method is proposed to solve these intractable problems, which are “small, light, and fast.” Firstly, the raw vibration signal is constructed into time–frequency images by continuous wavelet transform (CWT). Secondly, we build a self-attention lightweight convolutional neural network (SLCNN) model by integrating a self-attention mechanism (SAM) into the optimized SqueezeNet model. Then, based on a well-trained SLCNN in ImageNet, rich parameter knowledge is transferred from the pre-trained model to the target model. Finally, the fever training samples are used to fine-tune the target model. Experimental results on two bearing datasets validate the effectiveness of the SLTL method, which achieves 99.5% of classification accuracy with fewer training samples than other conventional CNN models. More importantly, the model parameters of SLTL are 0.95 M, and the floating-point operations (FLOPs) are 0.11 M, indicating that SLTL possesses high accuracy while maintaining lightweight, which benefits the platform with limited resources.

Deep transfer learning-based vehicle classification by asphalt pavement vibration

Deep transfer learning (TL) has great potential for a wide range of applications in civil engineering. This work aims to propose a deep transfer learning-based method for vehicle classification by asphalt pavement vibration. This work first used the pavement vibration IoT monitoring system to collect raw vibration signals and performed the wavelet transform to obtain denoised vibration signals. The vibration signals were then represented in two different ways, including the time-domain graph and the time-frequency graph. Finally, two deep transfer learning-based methods, namely Method Ⅰ (Time-domain & TL) and Method Ⅱ (Time-frequency & TL), were applied for vehicle classification according to the two different representations of vibration signals. The results show that the CNN model had a satisfactory performance in both methods with the accuracy of Method Ⅰ exceeding 0.94 and Method Ⅱ exceeding 0.95. The CNN model in Method Ⅱ performed better in the accuracy metrics with considering label imbalance, but worse in the accuracy metrics without considering label imbalance than that in Method Ⅰ. The differences between these two methods have been investigated and discussed in detail in terms of input types, accuracy metrics, and application prospects. The CNN model with deep transfer learning could be an effective, accurate, and reliable technique for vehicle classification based on asphalt pavement vibration.

Adaptive MOMEDA model based variational mode decomposition for Pelton wheel fault detection

A critical step in fault diagnosis is determining the frequency of faults. Variational mode decomposition (VMD) is extensively employed for this purpose since it can describe the signal in the time-frequency domain. On the other hand, the VMD frequently fails to analyse non-stationary data containing low-frequency disturbances/noises. A multipoint optimal minimal entropy deconvolution adjusted (MOMEDA) is used with VMD in this research to improve defect detection performance in the presence of low-frequency disturbances. The filter length has a strong influence on the output of MOMEDA thus choosing the right one is a critical step in recovering a periodic pulse in the event of a weak defective signal. Improved grey wolf optimization (GWO) adaptively selects the appropriate filter length using the autocorrelation energy as its fitness function. The GWO is improved by introducing a gaussian mutation strategy which maintains the proper balance between the exploration and exploitation process. The proposed method has been applied to investigate the bucket defects of the Pelton wheel. The raw vibration signal is first decomposed into a series of modes using VMD. Second, the MOMEDA model is used to purify each mode by reducing low-frequency noise interference. The modes processed by MOMEDA are reconstructed again into a pure signal. Finally, the Hilbert envelop spectrum of the pure signal is obtained to determine the fault frequency. The same is verified from the theoretical fault feature frequency of the turbine bucket. The performance of the proposed method in extracting fault frequency accurately is also compared with other two models: (a) MOMEDA model with ensemble empirical mode decomposition (EEMD) and (b) MOMEDA model with complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). The comparison results proved the efficacy and superiority of the proposed method.