Search

Intelligent bearing diagnostic methods are developing rapidly, but they are difficult to implement due to the lack of real industrial data. A feasible way to deal with this problem is to train a network through laboratory data to mine the causality of bearing faults. This means that the constructed network can handle domain deviations caused by the change of machines, working conditions, noise, and so on which is, however, not a simple task. In response to this problem, a new domain generalization framework--Whitening-Net--was proposed in this article. This framework first defined the homologous compound domain signal as the data basis. Subsequently, the causal loss was proposed to impose regularization constraints on the network, which enhances the network's ability to mine causality. To avoid domain-specific information from interfering with causal mining, a whitening structure was proposed to whiten the domain, prompting the network to pay more attention to the causality of the signal rather than the domain noise. The results of diagnosis and interpretation proved the ability of Whitening-Net in mining causal mechanisms, which shows that the proposed network can generalize to different machines, even if the tested working conditions and bearing types are completely different from the training domains.

In industrial applications, different abnormal conditions of a machine or a machinery fleet typically occur sequentially instead of becoming initially known. Therefore, one of the most challenging tasks for the machinery fault diagnosis is to incrementally classify unknown abnormal patterns. For this reason, an incremental novelty identification (INI) method is proposed, on the strength of deep learning-based representation learning, novelty detection and incremental learning. In the INI framework, a convolutional neural network autoencoder is first developed for high-level feature extraction from multiple monitoring signals. An online sequential-extreme learning machine (OS-ELM) is then proposed to learn healthy features of the normal condition initially known. The newly collected data are employed for updating the OS-ELM with the quantification of the model modification for novelty detection. An incremental learning procedure is implemented by adding a new OS-ELM model representing the novel abnormal pattern. The repository of OS-ELMs, each of which represents a specific condition, is employed for incrementally diagnosing known abnormal patterns with the identification of unknown ones. The performance of the proposed INI is validated under experimental scenarios considering both a benchmark bearing case study and a delta 3D printer, and the results are compared with other state-of-the-art approaches.

Representation learning (RL) has gained increasingly considerable attention in intelligent fault diagnosis due to its great capability of automatically learning useful features. Most existing studies focus on developing various variants about RL through modifying loss function of original versions, whereas it is a challenging task. Using the promising sparse filtering as the basic module, this paper presents a simple and effective RL method called multi-scale and multi-pooling sparse filtering (MSMPSF). Instead of modifying loss function of sparse filtering, we simply introduce two fusion mechanisms into sparse filtering, i.e., multi-scale fusion and multi-pooling fusion. In detail, the former aims to learn different local features from the collected signals under multiple scales. The latter tries to fuse various local features using multiple poolings. With these two mechanisms, MSMPSF is capable of capturing complementary fault information hidden in raw signals of several scales and obtaining richly informative feature representations, hence it can perform better. The proposed method is evaluated through experiments on three datasets about gear and bearing. Extensive comparison results confirm that both of two mechanisms facilitate a significant improvement on diagnosis performance. Furthermore, our method receives very reliable and competitive results in terms of diagnosis accuracy and stability in comparison with existing related works.

In most practical applications of fault diagnosis methods, two problems will inevitably arise. First, limited by the monitored object itself and its environment, accelerators are difficult to install. Second, industrial applications lack data with fault labels, which limits the use of data-driven-based methods. To solve these problems, a current signal-based adaptive semisupervised framework (C-ASSF) is proposed. In C-ASSF, the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) is adopted to extract recognizable features from only normal current signals. Subsequently, since WGAN-GP pays too much attention to body signals and ignores the changes caused by faults, the line spectrum feature extraction (LSFE) technique is utilized to remove the main frequency component of the current signal specifically. Finally, an index indicating the degree of deviation from the normal distribution is introduced to identify external bearing faults in drivetrains. Two groups of different experimental data sets are applied to verify the performance of C-ASSF. The results show that C-ASSF is superior to existing methods, such as self-organizing map (SOM) and stack autoencoder (SAE), and can not only identify faults in drivetrains but also identify different fault classes.

Playing a vital role in the safe operation of equipment, intelligent fault detection has been extensively applied in mechanical systems. However, the complicated mapping of signal symptoms and multicomponent failure modes under sharp speed variation makes it hard. The classification error is large due to insufficient samples in the fault condition. Aiming at these problems, a novel multitask redundant lifting adversarial network (MRLAN) is proposed. First, inspired by redundant second-generation wavelet transform, a reconstruction network is established to generate high-quality time–frequency representation of vibration signal, thus eliminating speed fluctuations and expanding training samples. The reconstruction network consists of transposed convolution layer, resume update layer, resume predict layer, and merge layer. Then, multitask category discriminator, which forms dual discriminator together with distance discriminator, is constructed to separate the features of multiple components, thereby realizing multiple tasks (multicomponent fault location and single-component health assessment). Finally, the model is alternately optimized through adversarial process between the generation network and the double discriminators. The proposed method is validated by two cases and other methods are also established to show its superiority. The results show satisfactory performance of MRLAN in multicomponent fault diagnosis under sharp speed fluctuation and little training data.

In recent years, the identification of out-of-distribution faults has become a hot topic in the field of intelligent diagnosis. Existing researches usually adopt domain adaptation methods to complete the generalization of diagnostic knowledge with the aid of target domain data, but the acquisition of fault samples in real industries is extremely time-consuming and costly. Moreover, most researches focus on samples with fixed fault levels, ignoring the fact that system degradation is a continuous process. In response to the above intractable problems, this article proposed a causal disentanglement network (CDN) to realize cross-machine knowledge generalization and continuous degradation mode diagnosis. In CDN, multitask instance normalization and batch normalization structure was proposed to learn task-specific knowledge and enhance the informativeness of the extracted features. On this basis, a causal disentanglement loss was proposed, which minimized the mutual information of features between subtask structures and captured the causal invariant fault information for better generalization. The experimental results proved the superiority and generalization ability of CDN, and the visualization results proved the performance of CDN in causality mining.

The synchrosqueezing transform (SST) and its ameliorated methods can enhance the energy concentration of the vibration signals in time-frequency domain, but it does not show enough effect for the weak feature extraction. Besides, many large-scale rotating machines such as aerospace engines and gas engines often work under complex vibrations, and this increases the difficulty for health monitoring based on vibration signals. This article presents a time-frequency analysis (TFA) named bilateral second-order SST (BFSST2). It can sharpen the instantaneous frequencies (IFs) like second-order SST (FSST2), and more importantly, it can extract the weak features such as amplitude-modulation frequency-modulation IFs and small-amplitude time-varying IFs which are easily ignored on time-frequency plane. To achieve the aforementioned target, the IFs are first obtained by employing FSST2, and then a bilateral function representing a novel negative-positive TFA is structured by combining the hyperbolic tangent function, exponential function, and sign function together. After that the large-amplitude IFs are transformed into small-negative amplitude ones by the product with the bilateral function, and this can not only enhance the original weak feature but also avoid its confusion with the transformed small-negative amplitude ones. The proposed BFSST2 is validated by a numerical simulation. At last, two case studies are given to illustrate its effectiveness in aerospace engine vibration monitoring.

Anomaly detection is crucial to the safety of complex electromechanical equipment. With the rapid accumulation of industrial data, intelligent methods without human intervention have become the mainstream of anomaly detection. Among them, variational autoencoder (VAE) performs well in anomaly detection with missing fault samples due to the self-supervised learning paradigm. However, the data from electromechanical equipment is usually non-Gaussian, making it difficult for the standard VAE based on Gaussian distribution to recognize the abnormal states. To solve the above problems, we proposed multi-mode non-Gaussian VAE (MNVAE) to detect anomalies from unknown distribution vibration signals without fault samples or prior knowledge. Firstly, the encoder maps the input to a Gaussian mixture distribution in latent space and samples a latent variable from it, after which the Householder Flow is applied to the latent variable to capture more abundant features. Finally, to describe the non-Gaussianity of the signal, Weibull distribution serves as the likelihood function of the reconstructed signal output from the decoder and as the basis for anomaly discrimination. In comparison to 6 related methods, our method yields the best results across various datasets. Through further experiments, the robustness of our method is proved and the proposed improvements are effective.

Fault identification for bearings of special electromechanical equipment is significant to avoid catastrophic accidents. However, spectral aliasing and nonstationarity resulted from variable speed condition make this task difficult. In this paper, a modified self-organizing maps (SOM) based on nonlinear neural weight updating way is proposed to solve the problem of bearing fault severity identification in variable speed condition. Firstly, a multi-domain features extraction method based on angular re-sampling technique is introduced. Then considering the nonlinear relationship between fault severity and fault features, the traditional Euclidian distance of SOM is substituted with the geodesic distance when update the neural weight and select the best-matching cell, which can improve the nonlinear identification ability of proposed method. Finally, two cases are performed and the results show that the method can identify bearing fault with different severities effectively and have practical significance when considering both accuracy and time cost compared with other methods.

This paper proposes a pattern recognition method for the fault diagnosis of machinery based on an enhanced trace ratio linear discriminant analysis (ETR-LDA) algorithm. It is an extension of TR-LDA that pursues to maximize the between-class distance and minimize the within-class distance. Since TR-LDA focuses on the total between-class distance and ignores the attention to local features, it may lead to deficiencies in the classification of specific patterns/classes, which is defined as short board in this paper. To cope with this limitation, the proposed ETR-LDA makes a reconsideration of the relationship between the total between-class distance and the global separability. More specifically, a new objective function is derived in ETR-LDA by taking the smallest between-class distance into account. The optimal solution of this objective function is proved to improve the smallest between-class distance with no change to the convergence and the global optimality of the algorithm, and the separability of different classes is increased eventually. Both synthetic data and hard disk drive wearing experimental data are employed to verify the efficiency of the proposed method. The results show that ETR-LDA is able to distinguish those fault categories and outperform TR-LDA and other dimensionality reduction methods.

Operating deflection shape analysis allows investigating the dynamic behaviour of a structure during operation. It normally requires simultaneous, multi-point measurements to capture the response from an unknown excitation source (unknown-input and multiple-output), which can complicate its usage for structures without ease of access. A novel vibration pattern testing method is proposed based on a roving continuous random excitation employing a small robotic Hexbug device and a single-point measurement. The Hexbug introduces a random excitation in consecutive locations while roaming over the structure. The resulting multi-modal, time and location dependent response of the system is captured in a single location, and then analysed with a newly developed method based on empirical wavelet transform, multiscale morphological filtering and optimization to extract the excited vibration patterns. The efficiency of the proposed method is experimentally demonstrated on a free-free and a cantilevered beam with comparison to mode shapes extracted by hammer test. The validation highlights its ability to extract several vibration patterns from a long slender structure with good accuracy and robustness, with the general ability to expand the usability of an operating deflecting shape analysis.

Intelligent fault diagnosis with small training samples plays an important role in the safety of mechanical equipment. However, affected by sharp speed variation, fault feature is extremely weak, which raises difficulty for fault diagnosis. The mutual coupling of multi-component fault features further increases the difficulty. Considering the ability of redundant second generation wavelet transform in non-stationary feature extraction, a multi-branch redundant adversarial net (RedundancyNet) is proposed to address the above issues. The Net consists of discriminator, the generator based on redundant reconstruction, and the classifier based on redundant decomposition. Firstly, through adversarial training process, the generator fuses multi-scale features to generate the signal with varying speeds, thereby expanding training data. Secondly, through layer-by-layer multi-resolution feature enhancement, the classifier boosts weak fault features of vibration signals at variable speeds. Finally, a multi-branch framework is proposed to realize multi-component fault location and damage identification. The proposed method is validated on two cases. The average classification accuracy in the two cases reach 97.14% and 98.33% respectively. However, other end-to-end intelligent fault diagnosis methods for varying speeds or small samples can only reach the highest classification accuracy of 95.14% in Case 1 and 93.59% in Case2, which is much less than RedundancyNet. The analysis results highlight the effectiveness of the net under drastically variable speeds and small faulty training samples. Besides, the proposed classifier is easy to understand, which reveals the process of feature learning and the extracted feature under varying speeds.

It has been challenging to correctly separate sources from their mixtures with large noise and unknown source number in underdetermined cases. To address this problem, a novel underdetermined blind source separation (UBSS) method is proposed using synchrosqueezing transform (SST) and improved density peaks clustering (DPC). First, SST is employed to reassign time-frequency (TF) coefficients of short time Fourier transform (STFT) of mixed signals from their original position to the center of gravity along frequency direction, contributing to sparser TF representation than STFT. Second, a single source point (SSP) identification method is proposed by directly searching the identical normalized TF vectors, which considers linear representation relations among TF vectors and therefore can identify SSPs more accurately in noisy cases. Third, DPC is improved to automatically identify source number, with which the mixing matrix can be estimated. After that, sources are finally recovered using l1-norm decomposition method. The effectiveness of the proposed method is validated with some numerical studies and experimental studies. The results show that the proposed method can estimate source number correctly, and recover sources accurately even in noisy cases.

To accurately estimate source signals from their post-nonlinear mixtures, a post-nonlinear blind source separation (PNLBSS) method with kurtosis constraints is proposed based on augmented Lagrangian particle swarm optimization (PSO). First, an improved contrast function is presented by combining mutual information of the separated signals and kurtosis ranges of source signals. Second, an augmented Lagrangian multiplier method is used to convert PNLBSS into an unconstrained pseudo-objective optimization problem. Then, improved PSO is applied to update the parameters in complex nonlinear spaces. Finally, numerical case studies and experimental case studies are provided to evaluate the performance of the proposed method. By adding the kurtosis ranges constraints, the estimation accuracy of source signals could be improved, which would benefit vibration and acoustic monitoring and control.

Tacholess order tracking (TLOT) is a promising tool for mechanical fault diagnosis in speed variation conditions without a tachometer. However, TLOT is still a challenging task due to the inaccurately acquired instantaneous phase of shaft and noise corrupted resample signal in angle domain. Therefore, an improved TLOT technique is proposed in this paper to deal with those limitations for mechanical fault feature extraction. In this method, the modified numerical integration (MNI) algorithm is proposed to estimate the instantaneous phase by integrating instantaneous frequency (IF), which is estimated by a time-frequency representation (TFR) approach. During the MNI, Romberg integration is investigated and performed to improve the estimated accuracy of instantaneous phase. Subsequently, a flexible synchronous average (FSA) algorithm based on Chirp-Z transform in angle domain is further employed to reduce noise interference for the resampled signal. The effectiveness and efficiency of the improved TLOT technology is verified by both simulations and experiments. Analysis results show that the proposed method has an outstanding performance for the mechanical fault diagnosis in the condition of speed fluctuation.

This paper presents an intelligent combustion fault diagnosis approach for V-type marine diesel engines with uneven firing intervals based on the instantaneous crankshaft speed (ICS). First, experimental ICSs of a 16-cylinder marine diesel engine under normal and different degrees of fault conditions are investigated to obtain the most sensitive fault features. A new polar representation method named cyclic-polar-map (CPM) is proposed to effectively demonstrate the fault features. The quantitative radii of the CPM are used to indicate the location and degree of fault. Then, a support vector machine (SVM)-based intelligent fault diagnosis system is developed for automatically detecting the combustion faults with the features extracted from the proposed CPM and the frequency analysis of ICS. In addition, for the SVM training set is not always available in practice, a dynamic model of the crankshaft system is developed to simulate different fault patterns. Finally, the efficiency of the proposed intelligent diagnosis approach is validated by experimental data. The results show that the proposed SVM-based intelligent diagnosis system using both CPM features and frequency domain features of the ICS can obtain good accuracies under different fault degrees at various speed and load conditions.

Remaining useful life (RUL) is a critical metric in prognostics and health management (PHM) because it reflects the future health status and fault progression of products. Most RUL estimation methods are based on degradation data. In practice, due to changing degradation mechanisms during products’ whole life cycle, the degradation data may consist of two or more distinct phases, and the time points of these mechanisms switching are usually nondeterministic. This property makes RUL estimation a difficult task. To solve this problem, this paper proposes a switchable state-space degradation model to characterize degradation paths with nondeterministic switching manner dynamically. To update the model parameters by newly available data, a novel statistical procedure based on Rao-Blackwellized filter/smoother and an expectation maximization algorithm is derived. To improve the robustness and efficiency of the RUL prediction, a semianalytic prediction model is developed, which can avoid significant fluctuation in RUL estimation. The developed methodologies can automatically track different degradation phases and adaptively update parameters related to prior distributions. Two real products degradation cases are used to verify our methodologies.

Novelty detection is a challenging task for the machinery fault diagnosis. A novel fault diagnostic method is developed for dealing with not only diagnosing the known type of defect, but also detecting novelties, i.e. the occurrence of new types of defects which have never been recorded. To this end, a sparse autoencoder-based multi-head Deep Neural Network (DNN) is presented to jointly learn a shared encoding representation for both unsupervised reconstruction and supervised classification of the monitoring data. The detection of novelties is based on the reconstruction error. Moreover, the computational burden is reduced by directly training the multi-head DNN with rectified linear unit activation function, instead of performing the pre-training and fine-tuning phases required for classical DNNs. The addressed method is applied to a benchmark bearing case study and to experimental data acquired from a delta 3D printer. The results show that it is able to accurately diagnose known types of defects, as well as to detect unknown defects, outperforming other state-of-the-art methods.