As large-scale industrial processes evolve toward greater complexity, the increasing interdependence of networked and dynamic process data has a critical impact on product quality, creating significant challenges for quality-related fault diagnosis. Causal graphs (CGs) are effective in modeling structural relationships among nodes in large-scale industrial processes. However, traditional causal discovery methods are limited in their ability to represent hierarchical and dynamic causal structures with spatiotemporal features. To overcome these limitations, a dynamic causal entropy (DCE)-spatiotemporal convolutional network is designed in this article. First, the proposed DCE method enables the construction of hierarchical dynamic CGs that accurately represent dynamic interactions among process variables, effectively mitigating confounding factors and enhancing interpretability. Second, a 3-D squeeze-and-excitation (SE) convolutional neural network is designed to adaptively recalibrate channel-wise information and deeply analyze the spatiotemporal characteristics embedded in the hierarchical dynamic CGs. Furthermore, a local-global quality-related fault detection approach is introduced, along with a novel causal anomaly vector that facilitates precise recognition of fault root causes across multiple hierarchical levels. Finally, the effectiveness and practical advantages of the proposed method are thoroughly demonstrated using both numerical simulations and real-world data from a hot strip mill process (HSMP), achieving a fault detection accuracy of 95.78%.

In the aerospace field, equipment failures can lead to substantial economic losses and pose significant safety risks, making effective fault diagnosis crucial. Traditional fault diagnosis methods typically require large, precisely labeled datasets, which are challenging to obtain in aerospace applications due to the rarity and unpredictability of faults. To overcome these limitations, this article proposes a novel conditioned multiscale generative adversarial networks (GANs) approach designed to enhance fault diagnosis performance under small-sample conditions. Initially, raw vibration signals undergo preprocessing using the short-time Fourier transform, which expands frequency-domain features while preserving essential time-frequency characteristics. Subsequently, conditioned multiscale GANs are trained on these limited datasets, employing multiscale convolutional kernels to extract and fuse rich features, thus generating high-quality synthetic samples. Finally, these synthetic samples are combined with the original dataset to train a convolutional neural network offline, which can subsequently perform real-time online fault diagnosis. Extensive validation on two aerospace-related datasets demonstrates that the proposed method significantly enhances fault diagnosis accuracy and efficiency, even when the available training data is severely limited.

The networked structure of sensors emerges in large-scale industrial processes. Causal graphs can reveal the underlying mechanisms. However, due to the constraints of material and information flows, industrial process data exhibit complex spatio–temporal characteristics. Traditional causal discovery results include redundant information and the spatio–temporal features are not sufficiently mined, affecting the accuracy of fault diagnosis. To address the above problems, an integrated distributed fault diagnosis framework is proposed. First, a new method combining mechanism knowledge and correlation is proposed to construct a spatio–temporal causal graph, which highlight spatio–temporal causal information. Second, an embedded time convolutional network-based autoencoder is designed to extract spatio–temporal features simultaneously. Then, the local-global fault detection scheme is performed. On this basis, a new anomaly status information matrix is designed by decoder and spatial features to achieve root cause recognition. Finally, the effectiveness of the proposed method is validated using actual data from the hot strip mill process, achieving a fault detection accuracy of 98.3$\%$.

Cascading failures represent a significant issue in production lines, as they can lead to process defects and safety incidents. An accurate risk assessment of cascading failures is crucial for ensuring both safety and operational efficiency. However, existing methods for assessing cascading failures typically focus only on exposed failures, neglecting hidden failures. Hidden failures are functional faults not apparent under normal operating conditions; they often remain undetected until triggered by another failure event. Considering solely exposed failures thus provides an incomplete picture, insufficient for accurately assessing cascading failure risks. To address this limitation, this article proposes a novel virtual node-based framework designed to assess cascading failure risks explicitly accounting for hidden failures. A Bayesian network approach, enhanced by leveraging connectivity information, is employed to effectively model the structure of the production line. Within this Bayesian network, a virtual node is integrated, thus representing the background impact of hidden failures. Specifically, the interactions between this virtual node and other network nodes explicitly capture the dynamics and mechanisms underlying hidden failures. Building upon this framework, we propose the virtual node-assisted inverse PageRank algorithm. The algorithm is rigorously defined, with mathematically guaranteed properties including positivity, convergence, and an analytical solution. The methodology is validated using a real-world case study involving an aerospace impeller production line. Experimental results demonstrate that the proposed algorithm successfully identifies hidden failures, delivering superior performance compared to traditional risk assessment approaches.

Guaranteed system performance under various circumstances continues to be a challenge in technique and practice. Based on this, this article investigates the optimal fault-tolerant control strategy for a large-scale interconnected system with the intermittent actuator faults. Since the subsystem state is enforced to a restricted range, an asymmetric integral barrier Lyapunov function is incorporated into the principle of Bellman optimality to avoid the violation of state constraints. Also, it can conquer a conservative limitation that the bounds of the transformed error-constraints are known. Subsequently, the critic-actor–identifier framework is constructed in the backstepping step to evaluate the objective function, control behavior and unknown dynamic, respectively, wherein the decentralized controller derived from the learning process and the fault-tolerant controller are separated by introducing an intermediate controller. Meanwhile, it is illustrated that the trajectory tracking errors will approach to a small region nearby the origin, and the system states may not beyond the given asymmetric constraint bounds, even in the presence of faults. Finally, results are presented to exhibit the effectiveness and the advantage of the optimal approach through appropriate comparative simulations.

Nonlinear and non-Gaussian characteristics are common in industrial processes. Artificial neural correlation analysis (ANCA) is a good nonlinear process monitoring algorithm, which combines classical correlation analysis with artificial neural networks. However, its performance is not very satisfactory for industrial processes with non-Gaussian characteristics. To solve non-Gaussian problems, almost all the existing process monitoring algorithms only consider the effect of kurtosis. Nevertheless, both kurtosis and skewness affect the data distribution. To improve the limitations of existing algorithms, this study proposes a new process monitoring algorithm named Jarque–Bera-based ANCA. This new algorithm makes many improvements to ANCA scheme, and the designed loss function combines the influence of both kurtosis and skewness on the data distribution, which not only maintains the advantages of the ANCA algorithm in solving nonlinear problems, but also provides superior monitoring performance in non-Gaussian processes. Furthermore, the superior performance of the proposed new algorithm is verified through simulations using non-Gaussian and nonlinear numerical examples, the Tennessee Eastman process, and catalytic cracking units.

The wind farm power loss due to wake interaction becomes more pronounced with a more compact farm layout. The steady-state wake model, for example, the Gaussian wake model, can help optimize the wake deflection and reduce the power loss. However, accurately estimating the farm-level ambient wind, like wind speed, direction, and turbulence intensity, is challenging, especially under time-varying inflow conditions. An artificial neural network-based (ANN) wind farm control strategy is proposed for real-time wake steering control. This algorithm only requires the measurement data of each turbine without estimating the farm’s ambient wind. Firstly, numerous steady-state simulation scenarios are designed by setting different ambient winds and yaw deflection. Then, these simulation measurement data and optimal yaw offsets are used to train the neural network as the farm control strategy. This method utilizes the nonlinear mapping capabilities and knowledge compression characteristics of neural networks. Finally, the effectiveness of the control strategy is validated in a dynamic simulation setting, which includes fluctuating wind direction in the inflow.

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.

The absence of fault data in certain categories presents a significant challenge in data-driven fault diagnosis, as obtaining a complete fault dataset is often unfeasible. Zero-shot learning has emerged as a viable solution to this problem. Nonetheless, it often encounters problem of unreliable diagnosis results due to domain shift. In this article, a feature generating network with attribute-consistency is developed for zero-shot fault diagnosis, which introduces the attribute consistency constraint and feature transformation with attribute information. The implementation process comprises two parts, unseen fault class generation and discriminative feature transformation. The attribute consistency constraint adopted in data generation can make the generated data represent their attribute well. For feature transformation, a concatenation operation is used to transforming the generated samples into more discriminative representations. The effectiveness of the proposed method is verified using a public dataset for fault diagnosis purpose. Results indicate that the proposed method outperforms the state-of-art zero-shot diagnosis method.

The interest on the use of renewable energy resources is increasing, especially wind and hydro powers. To this aim, fuzzy control techniques represent viable strategies that can be employed for this purpose, thanks to the features of these nonlinear dynamic processes working over a wide range of operating conditions, driven by stochastic inputs, excitations and disturbances. Based on past investigations carried out by the authors about the control of wind turbines and hydroelectric power plants, this paper provides some guidelines on the design and application of the control strategies suitable to these energy conversion systems. The working conditions of these energy conversion systems are also taken into account to highlight the reliability and robustness characteristics of the developed control strategies.

This study proposed a model based fault detection and isolation (FDI) method using multi-layer perceptron (MLP) neural network. Detection and isolation of realistic faults of an industrial gas turbine engine in steady-state conditions is mainly centered. A bank of MLP models which are obtained by nonlinear dynamic system identification is used to generate the residuals, and also simple thresholding is used for the intend of fault detection while another MLP neural network is employed to isolate the faults. The proposed FDI method was tested on a single-shaft industrial gas turbine prototype and it have been evaluated using non-linear simulations based on the real gas turbine data. A brief comparative study with other related works in the literature on this gas turbine benchmark is also provided to show the benefits of proposed FDI method.

In this study, a model-based procedure exploiting analytical redundancy for the detection and isolation of faults of a gas turbine system is presented. The diagnosis scheme is based on the generation of so-called "residuals" that are errors between estimated and measured variables of the process. The work is completed under both noise-free and noisy conditions. Residual analysis and statistical tests are used for fault detection and isolation, respectively. The final section shows how the actual size of each fault can be estimated using a multilayer perceptron neural network used as a nonlinear function approximator. The proposed fault detection and isolation tool has been tested on a single-shaft industrial gas turbine model.