Cross-domain remaining useful life prediction is a critical task for industrial applications, and existing domain adaptation (DA) techniques typically focus on mitigating domain shift by learning domain-invariant features. These methods help transfer knowledge from a labeled source domain to an unlabeled target domain, thus improving prediction accuracy and robustness. However, the alignment strategies used are often coarse-grained, which can lead to the disruption of inherent temporal dependencies in degradation data. In this article, we introduce the novel order-preserving kernel contrastive (OPKC) learning, which leverages the concept that samples with smaller label differences should exhibit higher kernel similarities in the reproducing Kernel Hilbert space (RKHS), irrespective of their domain origin. This kernel-enhanced, regression-aware contrastive learning technique enables fine-grained instance-level pairwise alignment between the source and target domains, ensuring that label difference information is preserved while maintaining the temporal order information inherent in the data. In addition, OPKC can be seamlessly integrated with adversarial DA methods to further enhance both prediction performance and training stability. Extensive experiments on two widely used industrial benchmark datasets demonstrate that the proposed framework significantly outperforms state-of-the-art transfer learning and contrastive learning methods, achieving relative improvement of 16.9% on the PHM 2012 dataset and 13.2% on the C-MAPSS dataset.

Data-driven methods for predicting remaining useful life (RUL) have received considerable attention in the field of degradation data analysis. The transfer learning (TL) method offers new possibilities for RUL tasks in various operational settings. However, in many engineering applications, challenges in TL arise mainly from the scarcity or high cost of labeled data in the target domain, coupled with incomplete degradation of RUL samples within the target domain. This article proposes an innovative model named deep cross-domain transfer learning for interpretable prediction The model effectively harnesses the advantages of domain adaptation (DA) techniques in mitigating domain distribution disparities and also uses the exceptional visualization capabilities inherent in the variational autoencoder (VAE) model. This method integrates the VAE framework with regression networks and utilizes DA techniques to align feature spaces, achieving cross-domain RUL prediction with unlabeled target domain data and cross-domain visualization of the entire degradation process. The reproducing kernel Hilbert space is considered in domain adaption to control the complexity of hypothesis space. The effectiveness of the proposed method is demonstrated by analyzing the real C-MAPSS dataset.

With the advancements in sensor technology, it is now possible to measure and record a multitude of features that reflect the health condition of complex systems. These measurements are stored in a sizable data matrix, enabling the detection of anomalies. Nevertheless, the presence of this large data matrix poses a significant computational burden. The dimension-reduction methods, such as non-negative matrix factorization (NMF), can efficiently reduce computational burden. However, their pure data-driven nature can lead to overfitting and biases in anomaly detection results. To address this shortcoming, we propose a physics-enhanced NMF (PNMF) method by incorporating physical knowledge into NMF with the help of graph technique. The graph technique organizes measurements and features into two graph objects, respectively, and the physical knowledge guides the formation of edges between nodes in the graph. This allows the PNMF to capture not only the data-driven patterns but also the physical structure inherent in the system. The closed-form update algorithm is developed for the PNMF model, which can guarantee the convergence of parameters estimation. The superior performance of the PNMF model in detecting anomalies is demonstrated by comparing prevailing methods in both public datasets and real-world applications.

Supervisory control and data acquisition (SCADA) data from a complex mechanical system, such as a high-speed train power bogie, nonpower bogie, and wind turbine, are widely used for anomaly detection and fault isolation. The SCADA data include measurements of process variables and exogenous covariates for key components in the system. The process variables refer to the performance characteristics of the key component while the exogenous covariates are working loads or working conditions of the complex mechanical system. Dominated by such physical mechanisms as dynamic motion laws of the system, there are complex relationships between the process variables and covariates, that complicate anomaly detection and fault isolation. To solve this problem, we propose a framework that integrates physical knowledge and statistical learning. We first build a spline model to capture the relationship between process variables and exogenous covariates. To make the model interpretable, we use physical knowledge to impose constraints on the model parameters. We then conduct anomaly detection at a system level based on the physical-statistical regression model. Once an anomaly is detected, we propose a Lasso-based method to isolate the faulty components. Our fault isolation method does not require historical failure data or knowing the true number of faulty components. Real-world case studies on power bogies from high-speed trains illustrate the advantages of our framework: the best benchmark achieves at least 2.50% lower F1-score in anomaly detection and 6.01% lower F1-score in fault isolation compared to our method.

Since the degradation process is dependent on the environmental stresses, degrading products operating under dynamic environments can have time-varying degradation rates. Existing studies generally exploit a random walk to model the time-varying degradation rate, considering the randomness of the environmental effects. The random walk is not stationary, while the real environments, although dynamic, are often stationary. The degradation process under a stationary environment would have a stationary degradation rate. Therefore, instead of the random walk, we propose to model the stochastic degradation rate by an autoregressive model. The autoregressive rate can accommodate the randomness and stationarity of the environmental effects. Conditional on the autoregressive degradation rate, a Wiener process is used to model the degradation process. We develop an Expectation-Maximization algorithm to perform maximum likelihood estimation of model parameters. Moreover, to facilitate remaining useful life prediction, we derive the explicit probability density function for the remaining useful life (RUL). We validate the proposed model by a simulation study and justify the applicability and performance of the proposed model by two real degradation datasets.

The evaluation of long-lifetime and high-reliability products has attracted much attention. Stochastic degradation modeling is one of the most popular methods. The classical stochastic processes are frequently employed to discuss degradation trajectories. Most current work assumes that the underlying probability model of a degradation process is known or fixed in the estimation and model selection procedures. However, the ground-truth degradation model is usually unavailable in engineering applications. This article proposes a feasible parameter estimation and model selection procedure by measuring the distribution divergence among the nonparametric estimated model and some candidate models. In the proposed methods, it is not necessary to assume the availability of a ground-true model, which is replaced by a nonparametric estimated model. The proposed methodologies are suitable for restricted independent and nonidentically distributed samples. We discuss the large sample property of the suggested estimators. We report the Monte Carlo simulation study and practical data analysis to demonstrate our methods.

In the field of reliability engineering, the gamma process plays an important role in modeling degradation processes. However, extracting lifetime information from product degradation observations has long been suffering from both ineffective modeling techniques and inefficient statistical inference methods. To overcome these challenges, we propose a reparameterized gamma process with random effects in this article. Compared with the classical gamma process, the proposed model has a more intuitive physical interpretation. In addition, statistical inference for the model can be readily done through the variational Bayesian algorithm. Combining with the Gauss–Hermite quadrature and the Laplace approximation, the algorithm yields closed-form variational posteriors for the proposed model. Its superiority over two other inference methods (expectation maximization and Monte Carlo Markov Chain) in terms of computational efficiency and estimation accuracy is demonstrated by simulation.

Event causality identification is an important task in natural language processing. However, this task is highly challenging due to the high dependency of event context, text semantic ambiguity and insignificant causality features between text events. These issues lead to the low precision of causal relationship identification between events. We propose an Event Causality Identification Based on Feature Fusion (ECIFF) to improve the causality identification precision between events by integrating the context, semantics, and syntax of natural language. Firstly, we utilize BERT to capture the contextual features of events in natural language, enhancing the contextual embedding of events in different contexts. Secondly, based on an adversarial generative graph representation method, ECIFF learns a massive amount of causal relationships in the CauseNet, which can enhance the semantic representation of causes and effects of events. Next, we exploit the shortest dependency path to shorten the length of sentences and inductively learn all possible syntactic dependency relationships. Finally, the contextual, semantic and syntactic features are fused to synthetically determine the causal relationships among events. The experimental results indicate that our proposed approach significantly outperforms the state-of-the-art method LSIN: on the CTBank dataset, the precision, recall and F1-score of our approach are improved by 1.6%, 3.2% and 2.4%; on the ESL dataset, the precision, recall and F1-score of our approach are improved by 4.0%, 4.7% and 4.3%.

The great achievement of IoT and the wide use of edge devices have brought explosive growth in data. The quality and scale of data determine the performances of machine learning models. Federated learning has attracted widespread attention for its ability to use isolated data and protect data privacy. Models can represent excellent generalization capabilities through federated training. However, the large number of devices and complex models involved in federated training exacerbate the communication costs and degrade the performance of the global model. Although existing approaches can reduce communication costs, they ignore the degradation of global model accuracy in a heterogeneous environment. To alleviate the huge communication costs in federated learning, this paper focuses on reducing upstream and downstream communication frequency while ensuring global model accuracy. We propose a Two-way Delayed Updates method with Model Similarity in Communication-Efficient Federated Learning (FedTDMS). FedTDMS employs personalized local computation to improve global model accuracy on heterogeneous data. Combining 10-cal update relevance check and global model compensation, FedTDMS reduces the communication frequency in Federated Learning. We conduct experiments on the MNIST-FL and CFAR-10-FL datasets. Results show that FedTDMS can greatly optimize communication efficiency while maintaining good global model accuracy.

In many engineering systems, aside from the main component fulfilling the essential functions, a number of auxiliary components are configured to protect the main component and improve the reliability of the system. In actual operation, the failure or state change of the auxiliary components may affect the reliability both the main component and the remaining operational auxiliary components. However, the structure and dependence between the auxiliary components has been ignored in the existing studies. To fill this gap, we consider a system with a main component and a protective auxiliary subsystem. The latter is a load-sharing$\bm{k}$-out-of-$\bm{n}$system, that is, there is dependence between the auxiliary components. For such a system, an opportunistic inspection and preventive maintenance strategy is proposed. Then, we derive the system reliability using the Laplace transforms and the matrix method. The long-run average cost of the system is then derived, based on which the optimal maintenance problem is formulated and solved by an enumeration method. A numerical example, together with sensitivity studies of some model parameters, shows how the evolution of the parameters influences the optimal maintenance strategy. Finally, the model is extended by introducing periodic inspection and preventive maintenance strategy for main component, and the two strategies are compared.

In degradation tests, it is common to see that the initial degradation levels of test units are heterogeneous. Moreover, the degradation rate of a path may also be correlated with the initial value of the degradation measure. Motivated by this observation, in this paper, we introduce a time shift to the traditional stochastic process models, which presumes that the product has experienced some degradation at the beginning of the test. Such a modeling technique can also capture the correlation between the initial degradation and the degradation rate, when the degradation rate of each path does vary from unit to unit. We apply this technique to the three popular stochastic process models, i.e., the Wiener process, the gamma process, and the inverse Gaussian process, and develop the corresponding parameter inference procedures. Monte Carlo simulations are implemented to validate the proposed models and the estimation procedures. Applications to the degradation analysis of block error rates data and GaAs laser data reveal good performance of the proposed models.

Modern products are usually designed with high reliability and have complex structures, and degradation analysis of the complex systems with two or multiple performance characteristics is still a challenge. In this paper, we propose a new bivariate degradation model based on the Wiener process. There are three main merits of the proposed model: it can describe the common factor affecting the degradation of the two performance characteristics and unit-to-unit variation simultaneously, the reliability functions of the system and the remaining useful life of the system have analytic forms, and the model parameters and the missing values can be estimated by the Bayesian method and data augmentation. The simulation study and data analysis show that the Bayesian method and the proposed model have satisfactory performance.