Mechanical fault datasets are always highly imbalanced with abundant common mechanical fault samples but a paucity of samples from rare fault conditions. To overcome this weakness, the simulation of ...rare fault signals is proposed in this paper. Specifically, frequency spectra are employed as model signals, then Wasserstein generative adversarial network (WGAN) is implemented to generate simulated signals based on a labeled dataset. Finally, the real and artificial signals are combined to train stacked autoencoders (SAE) to detect mechanical health conditions. To validate the effectiveness of the proposed WGAN-SAE method, two specially designed experiments are carried out and some traditional methods are adopted for comparison. The diagnosis results show that the proposed method can deal with imbalanced fault classification problem much more effectively. The improved performance is mainly due to the artificial fault signals generated from the WGAN to balance the dataset, where the signals that are lacking in training dataset are effectively augmented. Furthermore, the learned features in each layer of the generator network are also analyzed via visualization, which may help us understand the working process of the WGAN.
As mechanical fault diagnosis enters the era of big data, the traditional fault diagnosis methods under variable working condition are difficult to be applied because of the massive computation cost ...and excessive reliance on human labor. For the application of intelligent fault diagnosis under variable working conditions, the crucial difficulty is that the variable speed or load can cause smearing and skewing of classable feature. It is the key to break the predicaments by extracting the features which are irrelevant to the working conditions and contain fault information. This paper propose a new intelligent fault diagnosis framework under variable working conditions called Data-driven Fault feature Separation Method (DFSM) which can eliminate the working condition features from all the information and employ the rest fault information for diagnosis. In our DFSM, classification loss ensures the basic classified ability, first. Second, uncorrelation loss increases the discrepancy between fault features and working condition features. Then, reference loss guides the working condition encoder only extracting the working condition features. Finally, autoencoder loss ensures that all the information has been extracted. It should be noticed that our DFSM is trained only using the dataset under a certain working condition and can diagnose faults with high accuracy under variable even time-varying working condition, which means that it is easy to be applied. The experimental results of rolling bearing dataset show that the proposed DFSM can not only break the limitation of existing methods, but also achieve a superior performance comparing with related method. Besides, through the visual understanding, the proposed DFSM is certified that it is able to eliminate the working condition information and extract precise classable feature for fault diagnosis.
Quickly detecting and accurately diagnosing early bearing faults is the key to ensuring the stable operation of high-precision equipment. In actual industrial applications, it is common to face the ...issues of big data and poor fault identification accuracy. To accurately and automatically realize the diagnostics of rolling bearings, a convolutional neural network algorithm and fault feature enhancement method is proposed. A two-dimensional space feature extraction method based on the Cyclostationary theory and wavelet transform shows good results in noise suppression. Firstly, the cyclic demodulation of wavelet transform coefficients is performed on bearing vibration signals to convert one-dimensional vibration data into a two-dimensional spectrogram for enhancing the weak fault feature. Secondly, the image segmentation theory is introduced, which can obtain more data and improve the calculation accuracy and efficiency on the basis of data dimension reduction. Finally, the augmented 2D spectrograms are inputted into a convolutional neural network. Through the analysis of the actual planetary gearbox bearing data, and compared with other mainstream intelligence algorithms, the effectiveness and superiority of this method are verified.
Wind power forecasting plays a significant role in regulating the peak and frequency of the power system, which can improve the wind power receiving capacity. Despite plenty of forecasting methods ...have been proposed to fortify the accuracy of forecasting, the existing forecasting models do not consider the reconstruction of missing data and can not extract the spatiotemporal features from the wind power data. To address these issues, this study proposes an improved long short-term memory (LSTM) network based forecasting method to reconstruct the missing data and capture the spatiotemporal features from wind power data. In order to fortify the wind power forecasting accuracy of the model, the multiple imputation technique (MIT) is first developed to fill up the missing samples with reconstructed data samples by analyzing the correlation among variables in the raw wind power data. Secondly, to exploit the spatial and temporal features and reduce the low computation complexity, a new parallel convolutional network involving dilated convolution and causal convolution is established for features extraction. Finally, to further improve the wind power forecasting performance, the LSTM is applied to extract the long-term trends and reveal the internal relations of the derived spatiotemporal features. The experimental results on a benchmark dataset and an experimental dataset both demonstrate that the proposed LSTM based wind power forecasting can obtain the better forecasting performance.
Intelligent fault diagnosis is a promising tool for processing mechanical big data. It can quickly and efficiently process the collected signals and provide accurate diagnosis results. However, ...rotating machinery often works under various speed conditions, which makes it difficult to extract fault features. Inspired by speech recognition, the nuisance attribute projection method in speech recognition is introduced into fault diagnosis to solve the problem of feature extraction in variable speed signals. Based on the idea of unsupervised feature learning, the loss function of nuisance attribute projection is added to the loss function of convolutional neural network (CNN) to learn fault features from original data. Health status is classified according to the learned characteristics and projection matrix P. A special designed bearing dataset is employed to verify the effectiveness of the proposed method. The results show that the proposed method has a higher accuracy and a simpler framework, which is superior to the existing methods in bearing fault diagnosis.
Identifying impact fault features from fault vibration signal is significantly meaningful for the fault diagnosis and condition monitoring of rotating machinery. Given defects and the working ...conditions, impact features are covered by background noise. A new method named empirical wavelet transform (EWT) has been receiving attention from the researchers and engineers. However, detecting boundaries by using the local maxima method from Fourier spectra and capturing the impact features through Meyer wavelet are the two crucial drawbacks of EWT. The former might be invalidated by the influence of non-stationary and noise frequency, and the latter is inappropriate for impact signal features. Therefore, reinforced empirical Morlet wavelet transform (REMWT) is proposed to overcome these shortcomings and efficiently diagnose fault features. In this method, the frequency spectrum boundaries are adaptively detected from the inner product of spectral kurtosis and Gaussian function via scale space representation, which can enhance the frequency character of impact features in vibration signals. Then, the constructed empirical Morlet wavelet serves as the adaptive filter bank for decomposing the signal into several empirical modes on the basis of spectrum boundaries. The meaningful component is selected via the maximum Pearson correlation coefficient method, and the envelope spectrum is used to accurately extract the fault features. The proposed method is then used to diagnose the fault features from the collected vibration signals. The results show its effectiveness and outstanding performance.
Transfer learning based models have been employed for intelligent fault diagnosis under different working conditions. However, an actual and important problem is neglected in existing intelligent ...fault diagnosis methods, which is that target domain mechanical fault datasets are always highly imbalanced with abundant normal condition mechanical samples but a paucity of samples from rare fault conditions. To solve this actual problem, this paper proposed a novel adaptive cross-domain feature extraction (ACFE) method which can automatically extract similar features between different feature spaces. ACFE wants to obtain more information for determining the category of each training target sample, so it avoids the distribution adaptation and is suitable for the imbalanced problem. Specifically, high dimensional distance through kernel method is employed for ensuring the strong identify ability first. Then, for automatically extracting cross-domain feature, we calculate the posterior probability of category of each target domain training sample based on high dimensional distance, and employ entropy loss to capture the cross-domain information. Besides, we propose the guide loss to avoid the features of a category overall falling into false category caused by imbalanced dataset. Based on ACFE, the intelligent fault diagnosis method for dealing with the imbalanced target dataset is described. To verify the effectiveness, we carry out two specially designed experiments, and the results shows that, comparing with related method, the proposed method achieve a superior performance.
...machinery fault diagnosis systems, which focus on the detection of health conditions after the occurrence of certain faults, have attracted considerable attention 3,4. ...the vibration signal is ...separated into M samples to compose the training set {xi}i=1M , where xi∈ℜN×1 is a training sample which contains N data points. ...the energy of frequency is dispersed, learning to unclear learned features, which results in lower testing accuracy. 3.3. ...these row vectors exhibit time-frequency properties and show similarities to one-dimensional Gabor functions, which serve as good band-pass bases for mechanical signals.
Rolling element bearings are widely used in rotating machines and their faults can lead to heavy investment and productivity losses so that the fault diagnosis of bearing is very important for ...guaranteeing a high performance transmission. However, despite the marvelous success of data-driven models, the most existing intelligent methods have a great limitation: the training data and testing data are under the same working condition. Therefore, to promote the successful applications of intelligent fault diagnosis, domain adaptation is employed in this paper. Specifically, we proposed a framework based on multilayer multiple kernel variant of Maximum Mean Discrepancy. For guaranteeing the stable results and improving the accuracy, the kernel method is introduced for replacing the high dimensional map of Maximum Mean Discrepancy. This makes features from different domains are close to each other in the Reproducing Kernel Hilbert Space. Besides, the features from two different domains of each feature layer both participate in the domain adaptation. Two bearing datasets are adopted to verify the effectiveness of the proposed method. The experimental results show that the proposed method can not only break the limitation of existing methods, but also achieve a superior performance comparing with related method.
Normal operation of bearing is the key to ensure the reliability and security of rotary machinery, so that bearing fault diagnosis is quite significant. However, the large amount of data collected by ...modern data acquisition system and time-varying working conditions make it hard to diagnose the fault using traditional methods To break the predicaments, we propose a new intelligent fault diagnosis framework inspired by the infinitesimal method. The proposed model including three parts can ignore the effect of different rotational speeds. Firstly, the sample is segmented and every segment dimension is extended by input network to ensure the adequate information memory space. Secondly, the classification information is stored and transferred in the long short-term memory (LSTM) network and output to the third part. In this process, the working condition information is ignored because of the gate units function. Finally, the likelihood is given by output network to classify the health conditions. Besides, we propose a loss function combining all the output of every time step and employ dropout to train the model, which increase the training efficiency and diagnosis ability. The bearing datasets under time-varying speeds and loads are used to verify the proposed method. The application result shows that our method has higher accuracy with simpler structure, and is superior to the traditional method in bearing fault diagnosis. Moreover, we give a physical interpretation of the proposed model.