Meta-Learning With Distributional Similarity Preference for Few-Shot Fault Diagnosis Under Varying Working Conditions.

Few-shot fault diagnosis is a challenging problem for complex engineering systems due to the shortage of enough annotated failure samples. This problem is increased by varying working conditions that are commonly encountered in real-world systems. Meta-learning is a promising strategy to solve this point, open issues remain unresolved in practical applications, such as domain adaptation, domain generalization, etc. This article attempts to improve domain adaptation and generalization by focusing on the distribution-shift robustness of meta-learning from the task generation perspective. In fact, few-shot fault diagnosis under varying working conditions allows to address the distribution shift problem in a natural way. An unsupervised across-tasks meta-learning strategy with distributional similarity preference is proposed, where the core is the distribution-distance-weighting mechanism. Differently from the naive random meta-train task generation strategy used in existing meta-learning methods, the source instances that present a more similar distribution with respect to the target instances gain larger weightings in the task generation. This strategy leads to a meta-task training set that is enough diverse, and at the same time can be easily learned due to the distribution similarity features of the source tasks. The proposed method introduces the concept of maximum mean discrepancy that is applied to derive the distribution distance of the measurements. Moreover, a model-agnostic meta-learning is applied to realize few-shot fault diagnosis under varying working conditions. The proposed solutions are verified and compared by considering two public datasets used for bearing fault diagnosis. The results show that the proposed strategy outperforms different related few-shot fault diagnosis methods under varying working conditions. Moreover, it is thus proved that, meta-learning with distribution similarity feature represents an effective approach for domain adaptation and generalization.

Model Fusion and Multiscale Feature Learning for Fault Diagnosis of Industrial Processes.

The data generated by modern industrial processes often exhibit high-dimensional, nonlinear, timing, and multiscale characteristics. Presently, most of the fault diagnosis methods based on deep learning only consider the part of the characteristics of industrial data, which will cause the loss of part of the feature information during training, thereby affecting the final diagnosis effect. In order to solve the above problems, this article proposes an end-to-end multiscale feature learning method based on model fusion, which can simultaneously extract multiscale spatial features and temporal features of data, effectively reducing the loss of feature information. First, this article combines the convolutional neural network (CNN) with residual learning and designs a multiscale residual network (MRCNN) to extract high-dimensional nonlinear spatial features of different scales in the data. Then, the extracted features are input into the long and short-term memory (LSTM) network to further extract the temporal features of the data. After the fully connected layer, it is input into the classifier for final fault classification. The residual learning in MRCNN can effectively avoid the problem of model degradation and improve the training efficiency of the model. Through the fusion of MRCNN and LSTM, we can significantly improve the feature extraction ability of the model, thereby greatly improving the diagnosis effect. In the final case experiment, the method improved the comprehensive diagnostic accuracy of the Tennessee-Eastman (TE) process and industrial coking furnace datasets to 94.43% and 97.80%, respectively, which was significantly better than the existing deep learning model and proves the effectiveness and superiority of this method.

An AVMD-DBN-ELM Model for Bearing Fault Diagnosis.

Rotating machinery often works under complex and variable working conditions; the vibration signals that are widely used for the health monitoring of rotating machinery show extremely complicated dynamic frequency characteristics. It is unlikely that a few certain frequency components are used as the representative fault signatures for all working conditions. Aiming at a general solution, this paper proposes an intelligent bearing fault diagnosis method that integrates adaptive variational mode decomposition (AVMD), mode sorting based deep belief network (DBN) and extreme learning machine (ELM). It can adaptively decompose non-stationery vibration signals into temporary frequency components and sort out a set of effective frequency components for online fault diagnosis. For online implementation, a similarity matching method is proposed, which can match the online-obtained frequency-domain fault signatures with the historical fault signatures, and the parameters of AVMD-DBN-ELM model are set to be the same as the most similar case. The proposed method can decompose vibration signals into different modes adaptively and retain effective modes, and it can learn from the idea of an attention mechanism and fuse the results according to the weight of MIV. It also can improve the timeliness of the fault diagnosis. For comprehensive verification of the proposed method, the bearing dataset from the University of Ottawa is used, and some recent methods are repeated for comparative analysis. The results can prove that our proposed method has higher reliability, higher accuracy and higher efficiency.

Extended Relevance Vector Machine-Based Remaining Useful Life Prediction for DC-Link Capacitor in High-Speed Train.

Remaining useful life (RUL) prediction is a reliable tool for the health management of components. The main concern of RUL prediction is how to accurately predict the RUL under uncertainties. In order to enhance the prediction accuracy under uncertain conditions, the relevance vector machine (RVM) is extended into the probability manifold to compensate for the weakness caused by evidence approximation of the RVM. First, tendency features are selected based on the batch samples. Then, a dynamic multistep regression model is built for well describing the influence of uncertainties. Furthermore, the degradation tendency is estimated to monitor degradation status continuously. As poorly estimated hyperparameters of RVM may result in low prediction accuracy, the established RVM model is extended to the probabilistic manifold for estimating the degradation tendency exactly. The RUL is then prognosticated by the first hitting time (FHT) method based on the estimated degradation tendency. The proposed schemes are illustrated by a case study, which investigated the capacitors' performance degradation in traction systems of high-speed trains.

Conditional Joint Distribution-Based Test Selection for Fault Detection and Isolation.

Data-driven fault detection and isolation (FDI) depends on complete, comprehensive, and accurate fault information. Optimal test selection can substantially improve information achievement for FDI and reduce the detecting cost and the maintenance cost of the engineering systems. Considerable efforts have been worked to model the test selection problem (TSP), but few of them considered the impact of the measurement uncertainty and the fault occurrence. In this article, a conditional joint distribution (CJD)-based test selection method is proposed to construct an accurate TSP model. In addition, we propose a deep copula function which can describe the dependency among the tests. Afterward, an improved discrete binary particle swarm optimization (IBPSO) algorithm is proposed to deal with TSP. Then, application to an electrical circuit is used to illustrate the efficiency of the proposed method over two available methods: 1) joint distribution-based IBPSO and 2) Bernoulli distribution-based IBPSO.

Deep Auto-Encoder and Deep Forest-Assisted Failure Prognosis for Dynamic Predictive Maintenance Scheduling.

Prognostics and health management (PHM) with failure prognosis and maintenance decision-making as the core is an advanced technology to improve the safety, reliability, and operational economy of engineering systems. However, studies of failure prognosis and maintenance decision-making have been conducted separately over the past years. Key challenges remain open when the joint problem is considered. The aim of this paper is to develop an integrated strategy for dynamic predictive maintenance scheduling (DPMS) based on a deep auto-encoder and deep forest-assisted failure prognosis method. The proposed DPMS method involves a complete process from performing failure prognosis to making maintenance decisions. The first step is to extract representative features reflecting system degradation from raw sensor data by using a deep auto-encoder. Then, the features are fed into the deep forest to compute the failure probabilities in moving time horizons. Finally, an optimal maintenance-related decision is made through quickly evaluating the costs of different decisions with the failure probabilities. Verification was accomplished using NASA's open datasets of aircraft engines, and the experimental results show that the proposed DPMS method outperforms several state-of-the-art methods, which can benefit precise maintenance decisions and reduce maintenance costs.

A data-driven prognostics method for explicit health index assessment and improved remaining useful life prediction of bearings.

Although bearings offer a broad extent of applications and rank among the most-used elements in rotating machinery they also are the most vulnerable to failure. Consequently, "prognostics and health management (PHM)" of bearings has gained awareness in both academia and industry. As it aims to predict future failure events, "remaining useful life (RUL)" prediction is an important process to ensure a reliable and safe operation of bearings in the course of their degradation. However, accurate RUL prediction can hardly be carried out without an explicit health index that fully reflects the bearing's dynamic performance degradation process. Thus, obtaining an explicit health index is a major concern. This paper advocates a novel method to solve this issue. The "proposed method" is based on the ensemble of "deep autoencoder (DAE)" and "locally linear embedding (LLE)". To begin with, secondary features are extracted from the original unprocessed data obtained from sensors. These secondary features are used as inputs to the DAE where they become compressed to a more compact, lower-dimension form. Accordingly, the dimensionally reduced features are evaluated based on a trend factor with which higher-trend features are selected to enhance the accuracy and computational efficiency of the subsequent RUL prediction. The selected features are used as inputs for the LLE algorithm to determine a truly representative explicit health index which fully reflects the bearing's dynamic performance degradation. Having obtained the health index by the "proposed method", the RUL is finally predicted by employing the "long short-term memory (LSTM)" neural network. The obtained results from the experiment, authenticates the "effectiveness and superiority" of the "proposed method".

A Data-Driven Aero-Engine Degradation Prognostic Strategy.

Degradation prognostics of aero-engine are a well-recognized challenging issue. Data-driven prognostic techniques have been receiving attention because they rely on neither expert knowledge nor mathematic model of the system. But they are highly dependent on the quantity and quality of degradation data. To solve the problems caused by unlabeled, unbalanced condition monitoring (CM) data and uncertainties of the prognostics process, a novel data-driven aero-engine degradation prognostic strategy is proposed in this article. First, two indicators are defined to remove redundant degradation features. Then, the number of discrete states of health is determined by a fuzzy c -means algorithm, and the health state labels can be automatically assigned for health state estimation, where the uncertain initial condition and the uncertainty of health state's transition are fully considered. Finally, a multivariate health estimation model and a multivariate multistep-ahead long-term degradation prediction model are proposed for remaining useful life estimation for aero-engines. Verification results using the aero-engine data from NASA can show that the proposed data-driven degradation prognostic strategy is effective and feasible.

Adaptive relevant vector machine based RUL prediction under uncertain conditions.

Engineering systems often suffer with many uncertainties during their performance degradation processes, such as the inherent uncertainties associated with the degradation progression over time and the inevitable uncertainties caused by change of loading, operation and usage conditions. In order to improve the accuracy of remaining useful life (RUL) prediction, this study takes these common uncertainties into consideration via an improved relevance vector machine (RVM) approach, which can describe accurately the degradation process from fault to failure. Firstly, based on historical data, a multi-step RVM regression model is established offline, in which the uncertainties are represented by the variances of Gaussian distributions of parameters and then are quantified as time-varying variables. Then, an adaptive RVM model is trained and the time-varying variables are updated by the expectation-maximization (EM) algorithm. For on-line prediction, given the real-time data, the RUL is forecasted by the first hitting time (FHT) method in probability perspective. The proposed method is demonstrated by two case studies on a high-speed train's traction system. The results can show the effectiveness of the proposed method.

An improved incipient fault detection method based on Kullback-Leibler divergence.

This paper presents an improved incipient fault detection method based on Kullback-Leibler (KL) divergence under multivariate statistical analysis frame. Different from the traditional multivariate fault detection methods, this methodology can detect slight anomalous behaviors by comparing the online probability density function (PDF) online with the reference PDF obtained from large scale off-line data set. In the principal and residual subspaces obtained via PCA, a symmetric evaluation function is defined for both single variate and multivariate cases. The uniform form of probability distribution and fault detection thresholds associated with all eigenvalues are given. In addition, the robust performance is analyzed with respect to a wide range of Signal to Noise Ratio (SNR). Case studies are conducted with three types of incipient faults on a numerical example; combining with two nonlinear projections, the proposed scheme is successfully used for incipient fault detection in non-Gaussian electrical drive system. The results can demonstrate the superiority of the proposed method than several other methods.

Multiple incipient sensor faults diagnosis with application to high-speed railway traction devices.

This paper deals with the problem of incipient fault diagnosis for a class of Lipschitz nonlinear systems with sensor biases and explores further results of total measurable fault information residual (ToMFIR). Firstly, state and output transformations are introduced to transform the original system into two subsystems. The first subsystem is subject to system disturbances and free from sensor faults, while the second subsystem contains sensor faults but without any system disturbances. Sensor faults in the second subsystem are then formed as actuator faults by using a pseudo-actuator based approach. Since the effects of system disturbances on the residual are completely decoupled, multiple incipient sensor faults can be detected by constructing ToMFIR, and the fault detectability condition is then derived for discriminating the detectable incipient sensor faults. Further, a sliding-mode observers (SMOs) based fault isolation scheme is designed to guarantee accurate isolation of multiple sensor faults. Finally, simulation results conducted on a CRH2 high-speed railway traction device are given to demonstrate the effectiveness of the proposed approach.