Stereo matching plays a crucial role in 3D reconstruction and autonomous driving; however, achieving a balance between accuracy and inference efficiency remains a significant challenge. In this paper, we present EL-IGEV++, a lightweight and accurate stereo matching network. We introduce two key improvements to the original IGEV++ architecture. First, we integrate an Efficient Multi-scale Attention (EMA) module to capture long-range dependencies and enhance feature representation without significant computational cost. Second, we replace the computationally expensive traditional 3D convolutions with 3D Depthwise Separable Convolutions (3D-DSC) to reduce parameter count and accelerate inference speed. Extensive experiments on the Scene Flow and KITTI datasets demonstrate that our method achieves state-of-the-art performance with lower latency. Furthermore, zero-shot generalization tests on the Middlebury and ETH3D datasets confirm that EL-IGEV++ exhibits strong robustness in real-world scenarios, particularly in handling fine-grained details and ill-posed regions.

This article studies the optimal cooperative tracking control problem of nonlinear multi-agent systems (MASs) in the presence of unknown dynamics and denial-of-service (DoS) attacks. It is more challenging to achieve the desired optimized performance under DoS attacks, especially when neural networks are utilized to identify the unknown dynamics online. Considering two types of DoS attacks (connectivity-maintained attacks and connectivity-broken attacks), the article presents a novel defense control strategy consisting of a resilient distributed observer and a neuro-adaptive optimal cooperative tracking controller. In specific, by introducing a label information algorithm, the resilient distributed observer for each follower agent is presented to reconstruct the leader's states under two types of DoS attacks. Subsequently, a neuro-adaptive optimal cooperative tracking controller is proposed under an identifier-critic-actor architecture in the presence of DoS attacks. It is proven that the output of all follower agents can synchronize with that of the leader under DoS attacks, while the local performance indexes reach the Nash equilibrium simultaneously. Finally, the simulation results show the effectiveness of the presented optimal cooperative tracking control methodology.

Rapid and accurate obstacle detection is crucial for reducing collision risk in urban rail transit. Given the long braking distance and limited onboard computing resources of urban rail vehicles, this paper proposes a real-time 3D object detection model based on a monocular-stereo cascaded architecture, named ARE-stereo. To mitigate the loss of critical stereo cues due to inaccurate depth estimates in the monocular stage, two novel optimization strategies are proposed: pseudo disparity guided enhancement (PDE) and adaptive region of interest augmentation (ARA). The PDE strategy warps right-view region-of-interest (ROI) features into the left image coordinate frame guided by pseudo disparity, and fuses them with left-view ROI features to increase the confidence of proposals suppressed in the monocular stage. During the stereo stage, the ARA strategy dynamically expands the ROIs, effectively preventing the loss of object-relevant stereo features without introducing excessive redundancy, achieving a balance between inference speed and accuracy. Experiments on the KITTI dataset demonstrate that ARE-stereo achieves competitive performance in both inference speed and accuracy. Furthermore, validation in real-world urban rail transit scenarios confirms the practical effectiveness of our model, providing a viable solution for high-precision obstacle detection in resource-constrained urban rail transit environments. Our code is available at https://github.com/Xliu1009/ARE-stereo.

A Fuzzy Reinforcement Learning-Adaptive Robust Predictive Control (FRL-ARPC) framework is proposed in this study for dynamic trajectory tracking, in which reliance on predefined control strategies is alleviated through the integration of model-free fuzzy reinforcement learning for autonomous trajectory generation with adaptive parameter optimization. A dual-flexibility architecture characterized by adaptive reward functions and online parameter tuning is employed in this framework, thereby system adaptability in dynamic industrial environments is significantly enhanced. Furthermore, a Lyapunov-based stability analysis is established to characterize the robustness and uniform ultimate boundedness of the resulting closed-loop system under bounded disturbances and model uncertainties. Validated by an industrial-scale case study on variable load control in coal-fired power plants within a steel manufacturing park, the proposed FRL-ARPC framework demonstrates superior tracking accuracy and convergence rate compared to state-of-the-art predictive controllers.

Coke dry quenching (CDQ) in steel enterprises faces critical control challenges arising from dynamic complexity, inherent nonlinearity, operating uncertainties, and sparse feedback signals. To address these problems, a temporal-fuzzy enhanced reinforcement learning (TFERL) framework is proposed in this study, in which exploration and fuzzy-objective shaping are balanced to improve main steam temperature stability. Specifically, temporal dependencies between observed states and latent dynamics are captured by the long short-term memory, and time aware latent states are embedded into both the policy and value networks to enhance dynamic representation. On this basis, an adaptive reward mechanism based on a tradeoff between imitation-guided intrinsic rewards and interval type-2 fuzzy objective rewards is designed to mitigate uncertainty and encourage exploration under sparse rewards during coke loading and discharge interruptions. Reinforcement learning regularized by expert priors closes the control loop by learning a policy from the generated fuzzy rewards, achieving stable regulation of the CDQ main steam temperature. A simulation study based on real operational data from a CDQ plant of a steel industrial in China is conducted to evaluate TFERL for process regulation. Superior regulation performance over state-of-the-art methods is verified under multiple disturbances and conditions, reflected by smaller steady-state error and overshoot, and higher convergence speed.

Accurate prediction of operation rhythm and blast furnace gas (BFG) consumption for hot blast stoves group are critical for energy conservation and consumption reduction in steel industry. Considering the spatio-temporal non-stationarity, the multi-modal, and the strong nonlinearity characteristics caused by the alternating uncertain operation of multiple stoves, this paper proposes a process-constrained spatio-temporal graph neural network (PC-STGNN) prediction model. By integrating process mechanisms, the model extracts the hot blast stove operation rhythm, the remaining state time, and the flow rate change treating as process-constrained features, i.e., the industrial expert knowledge embedded in the data input layer. In terms of architectural design, a serial spatio-temporal fusion feature extraction mechanism is constructed, which utilizes the graph attention network (GAT) to dynamically capture the spatial nonlinear coupling weights that evolve with inter-stove conditions. To address the sharp fluctuations in energy consumption patterns, a mixture of experts (MoE)-based hierarchical prediction strategy is introduced to achieve adaptive regression of consumption patterns under different modes such as on-blast, combustion, and stove-switching. Real-world data from the SCADA system of a large-scale steel enterprise in China are employed in the experiments. It demonstrates that the PC-STGNN model significantly outperforms traditional time series models. The proposed method effectively improves the prediction performance when the sudden change occurs.

针对立式储罐内壁出现内壁损伤，导致罐体发生局部泄露破坏，探索了基于光纤光栅的储罐内壁损伤识别方法，实现不停产情况下罐体内壁损伤精确定位。提出了一种加权质心的FBG空间阵列损伤识别方法，将FBG光栅传感器应用到立式储罐上形成空间阵列，将罐壁分成不同的检测区域，在温差变化的情况下得到各测点波长变化特征。利用孤立森林算法得到波长变化异常基准，判定波长异常测点，定位损伤局部区域，通过局部损伤加权定位算法，实现在役罐体内壁损伤的定位。结果表明：温差变化下的FBG传感器在损伤位置检测到明显波长变化，确定了异常基准为0.01408 nm，检测到5处异常测点，质心定位得到每个检测区域的损伤弧面与空间位置，与实测损伤位置误差为4.8 %，验证了此识别定位方法的可靠性。实际应用中，在1000 m3大型立式储罐上布置FBG传感器空间阵列，检测温差变化下的罐壁波长变化数据，得到0.176 nm的异常基准，确定3处异常测点，经由加权质心定位算法进行局部损伤定位，得到损伤具体弧面与空间位置。研究方法实现了储罐内壁多耦合作用场景下的损伤检测与监测， 可用于储罐内壁早期损伤的快速在线定位识别。

The modeling of coal-fired units, regarded as a key technology for improving efficiency, reducing emissions, ensuring safety, and achieving intelligent control, is confronted with dual challenges: the high computational complexity of physics-based models and the limited generalization capability of data-driven methods. Although Physics-Informed Neural Networks (PINNs) have promoted progress in hybrid modeling, difficulties remain in the allocation of physical-data loss weights and in noise resistance. To address these issues, an innovative Dual-Network PINN architecture is proposed, in which a physics network is designed for mechanistic equation construction and a data network is developed for feature extraction. A dynamically weighted loss function is constructed, along with gradient backpropagation from global outputs is applied to achieve joint optimization. Simulation results demonstrate that the proposed method outperforms conventional single-network models in accuracy and robustness, as well as effectively resolve the loss-weight allocation problem, especially considering complex environment of steel industry.

Uncertainty in the production process is the primary factor affecting the execution of hot rolling scheduling plans. Although robust optimization methods are commonly employed to address this challenge, existing approaches struggle to quantify the schedulable production capacity (time margin) over longer time scales. Furthermore, they fail to adequately consider planning delays, the capacity constraints of upstream and downstream processes, and various other uncertainties. Therefore, a weak-robust modeling and knowledge-driven solution framework for scheduling margin calculation (WRoKS-SMC) is proposed. To tackle the difficulties of solving continuous uncertainty scenarios, the weak-robust optimization problem is reformulated into a discretized Markov decision process (MDP) model. Considering multidimensional slab features such as material specifications, time margin, and logistics among multiple processes, a multihierarchical graph convolutional network (MH-GCN) is proposed to encode these features into scalable graph feature vectors that serve as the state representation in the MDP. In order to integrate the hard constraints of production scheduling into the MDP model, a knowledge-driven imitation learning framework (KDIL) is proposed, where the feasible solution is approximated by minimizing the cross-entropy loss between the expert strategies and policy network decisions, ensuring the satisfaction of hard constraints. To verify the effectiveness of the proposed method, experiments were conducted using practical data from a domestic steel enterprise. The results show that the proposed method outperforms the state-of-the-art approaches in terms of scheduling costs and margin calculation under uncertainty scenarios.

Industrial energy systems typically operate under a hierarchical decision-making architecture, where the upper scheduling layer optimizes economic energy allocation and the lower control layer tracks the resulting references. However, neglecting closed-loop dynamics in existing model-based and data-driven optimization approaches often results in overly conservative or aggressive scheduling schemes. To address this issue, a data-driven closed-loop dynamic optimization framework that deeply integrates scheduling and control is proposed. First, a two-stage data-driven representation for nonlinear conversion units is established. Furthermore, at the scheduling layer, a distributed architecture decomposes the system-level problem into energy network and conversion unit subproblems. For each unit, a data-driven optimization with embedded closed-loop prediction is developed, which embeds nonparametric, data-driven predictive models representing tracking behavior into the scheduling problem. Meanwhile, at the control layer, data-driven predictive control is deployed for reference tracking. Case studies based on real industrial data from a steel enterprise in China demonstrate that the proposed method reduces total operating costs by 14.70% compared with conventional model-based distributed optimization and by 4.49% relative to a data-driven distributed optimization without closed-loop embedding. In addition, it maintains favorable robustness and tracking performance under load fluctuations, model mismatch, and measurement noise.

The stable control of the dense medium separation (DMS) process is vital for improving coal beneficiation efficiency and final product quality. However, achieving high-precision and robust dynamic control with traditional methods is challenging due to inherent difficulties in industrial settings, such as multivariable coupling, significant time delays, strong noise interference, and time-varying operating conditions. To address these challenges, an improved proportional-integral generalized predictive control (PI-IGPC) with active disturbance rejection control (ADRC) is proposed for DMS. This method converts multivariable systems into independent control channels through feedforward decoupling, thereby an adaptive step factor is designed to reconstruct control increment sequence that avoids matrix inversion to improve computational efficiency and numerical stability. Additionally, a linear extended state observer (LESO) is introduced to estimate the unmodeled dynamic errors of the system and external disturbances as total disturbances, enabling feedforward compensation. This significantly enhances the disturbance rejection capability and robustness of the system. Experimental results based on actual measurement data from a coal preparation plant show that the proposed method outperforms other approaches in response speed, overshoot, steady-state accuracy, and disturbance rejection capability. Moreover, ablation experiments and time-delay robustness tests validated the effectiveness and synergistic interaction of each module.

In this article, we studied consensus tracking for second-order nonlinear multi-agent systems with unmeasurable states under switching topologies. State observers are designed with a combination of neural networks to approximate the unknown function. Using the backstepping technique to develop an adaptive control strategy. The average dwell time is applied to the stability analysis of switching topology. It has been proven that the given control approach for tracking is reachable under a switching topology. Finally, the simulation results are attached to illustrate the effectiveness of the proposed control protocol.

Timely detection of sliding bearing oil-film faults is critical for preventing costly failures and ensuring equipment reliability. Yet, diagnosis accuracy is often hindered by data scarcity and severe class imbalance in industrial environments. To overcome these challenges, this study proposes a physics-informed generative framework that couples sliding bearing mechanics with data-driven modeling. The Reynolds equation is embedded into a physics-informed neural network (PINN) to generate physically consistent samples for data augmentation, while a self-attention-enhanced convolutional neural network (CNN) is used to extract features from 3-D oil-film pressure distributions. To the best of our knowledge, this is the first work to integrate physics-informed generative modeling with self-attention mechanisms for sliding bearing fault diagnosis. Comparative experiments against conventional finite difference and other imbalanced approaches demonstrate that the proposed method not only solves the oil-film pressure field with higher efficiency but also improves fault classification accuracy on imbalanced datasets by up to 37.5%, thereby offering a new paradigm for robust diagnosis under limited data conditions.

Plate-type fuel elements are critical components in advanced nuclear reactors, prized for their high power density and efficient heat transfer. However, understanding their damage mechanisms and safety under accident conditions remains a significant challenge for next-generation nuclear systems. This review systematically examines the damage behavior and assessment methodologies for plate-type fuel elements under Reactivity Insertion Accidents (RIA), Loss of Flow Accidents (LOFA), and Loss of Coolant Accidents (LOCA). It highlights the synergistic effects of multiphysics factors—such as irradiation swelling, oxidative corrosion, thermal-mechanical stress, and coolant boiling—on failure modes like blistering, cracking, and cladding rupture. Recent advances in numerical modeling (e.g., finite element analysis, CFD, multiphysics coupling tools) and experimental techniques are critically evaluated. Key limitations persist, including insufficient cross-scale modeling, lack of validation under extreme multi-fault conditions, and scarce experimental data on irradiated fuel behavior. Future work should focus on developing integrated multiphysics frameworks, constructing extreme-condition experimental platforms, and incorporating AI-assisted safety assessment tools to advance inherently safe and intelligent nuclear energy systems.

Blast furnace gas is a significant category of byproduct energy source produced during the ironmaking process, and its rational utilization is crucial to improving energy efficiency in steel plants. However, frequent operator interventions continually create new operating scenarios, which makes it difficult for traditional methods to maintain effective scheduling. Existing scheduling methods based on optimization and generative adversarial networks (GANs) rely excessively on historical scenarios and fail to capture the explicit causal relationships among the key factors, thus restricting their applicability for scheduling under unknown conditions. To tackle such an issue, an optimal scheduling method for BFG system based on an improved causal generative model, which is capable of generating diverse and physically consistent scenarios, is proposed in this study. Each scenario is characterized by three interpretable factors, i.e. gas tank level, generation-consumption flow difference, and consumption of adjustable units. A causal conditional Wasserstein GAN (Causal-CWGAN) is then constructed by embedding a process-informed adjacency matrix and a differentiable acyclicity constraint into the WGAN-GP framework. In addition, a correction model combined with mechanism-based rationality rules is adopted to further optimize the consumption and filters unreasonable scenarios. Subsequently, the tank-level prediction is performed to update the generated scenario set and derive practical adjustment suggestions. Experimental results on real data from a steel enterprise show that, compared with the WGAN and the MGAN methods, the proposed one yields smaller Wasserstein distances, generates more rational scenarios, and provides adjustment strategies that can stably keep the gas tank level within the safety operating range.

Accurate remaining useful life (RUL) prediction is a prerequisite for reliable prognostics and health management. However, under variable operating conditions, frequent condition switching leads to nonstationary degradation, which poses a major challenge to RUL prediction. To address this issue, a novel RUL prediction method is proposed to learn trend-based degradation characteristics. Specifically, in the spatial domain, condition-modulated graph propagation is performed on a structure-prior graph, where condition information is embedded into message passing to capture condition-dependent spatial dependencies. In the temporal domain, a condition-aware low-rank state-space update mechanism is proposed to model the recursive evolution of degradation features and alleviate condition-induced nonstationarity. In addition, a contraction-regularized manifold optimization method is introduced to preserve the geometric consistency of low-rank basis matrices during constrained parameter learning. Experiments on the FD002 and FD004 subsets of C-MAPSS demonstrate that the proposed method achieves lower RMSE and competitive Score compared with baselines, while reducing computational and storage overhead. Ablation studies further confirm the contribution of each component.

3D occupancy prediction plays a pivotal role in the realm of autonomous driving, as it provides a comprehensive understanding of the driving environment. Most existing methods construct dense scene representations for occupancy prediction, overlooking the inherent sparsity of real-world driving scenes. Recently, 3D superquadric representation has emerged as a promising sparse alternative to dense scene representations due to the strong geometric expressiveness of superquadrics. However, existing superquadric frameworks still suffer from insufficient temporal modeling, a challenging trade-off between query sparsity and geometric expressiveness, and inefficient superquadric-to-voxel splatting. To address these issues, we propose SuperOcc, a novel framework for superquadric-based 3D occupancy prediction. SuperOcc incorporates three key designs: (1) a cohesive temporal modeling mechanism to simultaneously exploit view-centric and object-centric temporal cues; (2) a multi-superquadric decoding strategy to enhance geometric expressiveness without sacrificing query sparsity; and (3) an efficient superquadric-to-voxel splatting scheme to improve computational efficiency. Extensive experiments on the SurroundOcc and Occ3D benchmarks demonstrate that SuperOcc achieves state-of-the-art performance while maintaining superior efficiency. The code is available at https://github.com/Yzichen/SuperOcc.

The active fault diagnosis (AFD) problem is investigated for linear parameter-varying (LPV) systems subject to unknown disturbances, including both bounded and unbounded types. In this paper, a two-step-based set-membership observer approach is presented by integrating an unknown input observer with reachability analysis. This approach decouples partial unknown disturbances from the state error dynamics and further mitigates their impact on the error reachable sets. Then, the observer gain is derived based on the F-W -radius criterion of zonotopes in order to minimize the size of the error reachable sets. Furthermore, an optimization problem is formulated to obtain the optimal auxiliary signal, and the event-triggered AFD scheme is designed to limit the frequency of the signal injection into the system. The main objective of this work is to achieve set-membership estimation for the system state via the designed observer, while separating the estimated state sets of different system modes with the aid of the auxiliary signal. Finally, the effectiveness of the proposed AFD method is illustrated through a numerical example and an industrial gas pipeline model. Note to Practitioners-In practical industrial systems, there are many unknown disturbances whose boundaries are unknown or difficult to determine in advance, caused by large-scale fluctuations of industrial loads, which are referred to as unbounded disturbances in this paper. However, most existing deterministic AFD methods assume that the system disturbances and noise are unknown but bounded. To address this issue, this paper proposes a novel AFD method based on a two-step set-membership observer. In step 1, an unknown input observer is designed to achieve the decoupling of unbounded disturbances from the state error dynamics. In step 2, the error reachable sets are derived through reachability analysis, thereby obtaining the state set-membership estimation results. Moreover, the event-triggered AFD scheme is developed to limit the frequency of the auxiliary signal injection into the system, thereby mitigating its impact on system operation. Experimental results from both a numerical example and an industrial gas pipeline model demonstrate the feasibility and effectiveness of the proposed method.

Loads in the industrial gas pipeline network (IGPN) are typically adjusted in real time according to production requirements, which potentially affects the security of the whole energy system. This paper proposes the concept and construction method of gas dynamic security region with variable load (V-GDSR), so as to provide a secure range for gas adjustment. Firstly, a set of equations for assessing the gas supply security is established, which incorporates the IGPN mechanism model considering the dynamic characteristics of gas transmission. Then, an initial V-GDSR is constructed as an axis-aligned hyper-rectangle, whose boundaries are determined by the vertices of the gas dynamic security region under the same initial conditions. By scaling the initial hyper-rectangle and judging the feasibility of the equation system, the boundary of V-GDSR is continuously approached, thereby providing support for process control and system optimization. Simulation results based on real-world data in a steel plant in China demonstrates the effectiveness of the proposed method.

The steam heating network (SHN) serves as a critical component of the integrated energy system (IES), where precise transient characterization is essential for enhancing IES flexibility and stability. This paper proposes a physics-guided Bayesian framework for sequential inference of network states and spatio-temporal hidden thermodynamic properties, which are explicitly identified as latent property fields embedded within the PDE-based transition model. In contrast to conventional methods which typically treat thermodynamic properties as lumped parameters or non-interpretable model weights, the proposed method models probabilistic Gaussian process (GP) priors as surrogates for these latent properties, including the derivative-related terms required by the governing dynamics. By integrating thermal-hydraulic governing equations directly as hard constraints within the physics-guided transition model, this framework designs GP-ICM priors as surrogates for latent spatio-temporal properties to capture spatial correlations and heterogeneity across pipelines. Since thermodynamic properties are inseparable from state dynamics, the joint posterior becomes strongly coupled and generally non-Gaussian. The proposed joint sequential inference method combines a moment-matched Gaussian approximation for the state posterior with Monte Carlo marginalization for the thermodynamic posterior is presented, providing an efficient solution for estimating both state and property fields. To verify the performance of the proposed method, a single pipeline system and a real-world industrial superheated steam network are employed. The experimental results demonstrate that the proposed method improves accuracy in both state estimation and thermodynamic-property inference under hard physical constraints, achieving thermodynamically consistent and interpretable results.