Under the Industrial Internet environment, the high interconnectivity of data has significantly driven the advancement of process control. However, due to factors such as adjustments in production targets and the variability of raw materials, industrial processes often operate in a state of frequent fluctuations. This leads to a mismatch between the prediction model and the industrial process, making it difficult for control methods based on fixed prediction models to achieve satisfactory results. To address this puzzle, this paper innovatively proposes a nonlinear predictive control method based on bridge-AE with knowledge embedding transfer and optimal input design. Specifically, a nonlinear state space model is constructed to characterize the different types of mismatches between the prediction model and the industrial process. Then, an optimal input design method based on control accuracy constraints is proposed to improve the quality of operating condition transition samples, laying the foundation for the rapid update of the model after changes in operating conditions. Finally, to address the model mismatch caused by changes in operating conditions, a model update strategy based on knowledge embedding transfer is proposed, enabling precise control throughout the entire process. To further validate the superiority of the proposed method, experiments were conducted through numerical simulations and industrial control system. The experimental results demonstrate that, compared with some state-of-the-art methods, the proposed method achieves better control performance across operating conditions.

With the rapid development of Industrial 4.0 and Industrial Internet of Things, the data collection with multi-source has significantly improved. How to effectively fuse these data for various engineering applications is still an open and challenge issue. To this end, we propose the canonical correlation guided deep neural network (CCDNN), a novel deep learning architecture, to learn a correlated representation for multi-source data fusion. Unlike the linear canonical correlation analysis (CCA), kernel CCA and deep CCA, in the proposed method, the optimization formulation is not restricted to maximize correlation, instead we make canonical correlation as a constraint, which preserves the correlated representation learning ability and focuses more on the engineering tasks endowed by optimization formulation, such as reconstruction, classification and prediction. Furthermore, to reduce the redundancy induced by correlation, a redundancy filter is designed. We illustrate its data fusion ability via correlated representation learning and superior performance on various engineering tasks. In experiments on MNIST dataset, the results show that CCDNN has better reconstruction performance in terms of mean squared error and mean absolute error than deep CCA and deep canonically correlated autoencoders (DCCAE). Also, we present the application of the proposed network to industrial fault diagnosis and remaining useful life cases for the classification and prediction tasks accordingly. The proposed method demonstrates approving performance in both tasks when compared to existing methods. Extension of CCDNN to much more deeper with the aid of residual connection is also presented in Appendix.

The performance of chillers significantly impacts the operating efficiency of heating, ventilation, and air conditioning systems. Accurately capturing and monitoring chiller performance changes is essential for ensuring efficient operation and implementing effective maintenance strategies. However, existing performance monitoring methods are sensitive to fluctuations in operating conditions, have difficulty perceiving subtle performance degradation trends, and insufficiently utilize knowledge of equipment degradation mechanisms. To this end, this paper proposes a degradation knowledge-guided health indicator for chiller performance monitoring based on variational information bottleneck (VIB). First, the steady-state operation data are screened and divided into conditional and state variables. Then, the VIB model predicts the state variables, and the residual is calculated to obtain the health status features, which are independent of operating conditions. Subsequently, a degradation trend is constructed using equipment degradation knowledge, which guides the design of a feature extraction structure comprising a shared encoder and an independent decoder to identify common degradation patterns across maintenance cycles. Finally, the extracted degradation features are subjected to dimensionality reduction to derive health indicators, which are combined with cumulative sum analysis to achieve performance monitoring. Validated using nearly five years of operating data from chillers in an actual building, the proposed method achieves monotonicity of approximately 0.80, correlation exceeding 0.92, and robustness exceeding 0.98 over three maintenance cycles, all of which exceed state-of-the-art benchmarks. Furthermore, the method can detect the onset of degradation and issue warnings over 80 hours earlier than traditional engineering practice, demonstrating its superior reliability and practical value for real-world chiller performance monitoring.

In modern corporate logistics transportation, effectively allocating railway platform freight space and optimizing loading route planning are vital for improving transportation performance and minimizing logistics costs. These aspects must be jointly optimized considering fluctuating time-sensitive preferences and the dynamic availability of resources. This dual objective poses a complex multi-objective optimization challenge, as achieving efficiency while reducing costs involves a non-convex, non-differentiable optimization space under real-time constraints. Existing algorithms often fall short of addressing these challenges with sufficient efficiency. In this paper, we develop a comprehensive mathematical framework for the Freight Space Allocation and Railcar Pickup-Delivery Route Planning problem (FSA-PDP). The model aims to enhance transportation efficiency while simultaneously reducing costs. To solve this, we employ a novel multi-objective reinforcement learning (MORL) algorithm. This innovative algorithm is designed to learn a general policy adaptable across diverse railway platform settings and varying preference spaces. By doing so, it effectively balances transportation performance and cost efficiency. Our methodology is validated through extensive experiments utilizing a real-world dataset derived from the large-scale Salt Lake chemical industry. The findings underscore the superior performance and adaptability of MORL, demonstrating its capacity to outperform methods in railway freight logistics.

The application of autonomous systems plays a crucial role in ensuring the safe and stable operation of industrial processes. Control methods grounded in artificial intelligence (AI) present new prospects for this application. However, the implementation of AI-based control methods typically depends on complete observation data. When unobservable states exist, it becomes challenging to achieve autonomous control. On the other hand, the process generally operates under dynamic working conditions, and the model needs to be updated in time to avoid model mismatch, which further increases the difficulty of autonomous control. To address these challenges, an autonomous control method based on a joint spatial-temporal model (AC-JSTM) is proposed to achieve autonomous control of unobservable point states of industrial processes under dynamic working conditions. Specifically, to solve the problem of model mismatch and poor control effect caused by the dynamic conditions of industrial processes, a condition identifier based on orthogonal test design (CI-OTD) is first proposed. The data set is constructed through typical condition parameter design, and the conditions are distinguished based on the correspondence between data distribution and condition labels. Then, considering the spatial distribution characteristics between unobservable and observable points and the dynamic correlation of observable points in the time dimension, a JSTM is established. The features of spatial and dynamic dimensions are integrated by a joint training method to achieve the prediction of unobservable points based on observable points. Finally, combining CI-OTD and JSTM, a dynamic working condition control (DWCC) framework is established to achieve autonomous control of unobservable points under dynamic working conditions. To verify the superiority and effectiveness of the proposed method, control experiments are designed for both catalytic rods and tubular reactors. The results show that the proposed method can achieve accurate control of unobservable points under dynamic conditions, and the control accuracy is improved by 18.69% compared with the control framework based on the spatial-temporal model. Note to Practitioners-Motivated by the fact that the control of DPSs is usually based on complete observations and a single predictive model, it is difficult to apply under observation-limited and multi-working conditions. This paper proposes a predictive control method for large-scale and multi-working condition industrial processes with unobserved states based on a JSTM, achieving accurate control under different conditions.

The complex industrial environment and poor data quality inherent in electrolytic copper foil production often lead to inaccurate energy consumption predictors, and consequently unreliable energy optimization. Given the difficulty in further improving the accuracy of energy consumption predictors, this article proposes a high-reliability stochastic prediction-based ensemble optimization (SPEO) method for energy saving of copper foil electrodeposition. The approach begins by introducing stochastic prediction to establish a standard stochastic prediction optimization framework, wherein filtering operations are applied to mitigate the adverse effects of prediction errors on the optimization process. Subsequently, reliability analysis methods are employed to derive reliability metrics based on optimization results generated by energy consumption predictors of varying accuracy. Finally, these metrics are comprehensively incorporated to integrate multiple optimizers, along with their respective optimal result feedback, thereby enhancing the generalizability and stability of the method. Experimental evaluations using real-world data from an industrial electrodeposition process demonstrate the superiority of the proposed SPEO method.

Deploying EtherCAT in container-based virtualized programmable logic controllers (vPLCs) faces critical challenges regarding driver portability, network determinism, and synchronization scalability. This article proposes a container-native architecture named vECAT by leveraging the data plane development kit (DPDK) for high-performance user-space packet processing, which eliminates kernel dependencies and ensures cross-platform portability. We develop a latency model that characterizes the end-to-end packet processing pipeline and guides systematic optimizations at the task, OS, and hardware levels. We also introduce an enhanced synchronization algorithm to address error accumulation in large-scale networks. Validated on x86, ARM, and RISC-V platforms running PREEMPT_RT Linux, the proposed solution achieves maximum jitters of 11, 12, 24 & micro;s at a 125 & micro;s cycle time on three typical types of vPLCs CPUs under various kinds of stress loads. Furthermore, the optimized synchronization mechanism eliminates the linear accumulation of synchronization error, achieving bounded precision validated on networks with up to 8 subdevices.

The internal temperature field of a blast furnace, vital for steelmaking, impacts energy use, emissions, and smelting costs, ensuring operational stability. Direct measurement of inner wall temperatures is infeasible due to the furnace’s enclosed, dusty, high-temperature, and high-pressure environment. Current approaches thus rely on sparse outer wall thermocouple measurements to reconstruct the internal temperature field via mathematical models. However, the large furnace volume coupled with limited outer wall data lead to a highly ill-posed inverse heat conduction problem, hindering accuracy and efficiency for industrial needs. To address this, we propose Diffusion-PINN, a Diffusion-Physics-Informed Neural Networks, which transforms the temperature field inversion into a diffusion-based generation process for temperature distribution maps. For the inverse heat conduction problem, the diffusion model is designed to fit the joint distribution of inner and outer wall temperature fields, directly generating coupled temperature distributions to circumvent the multi-solution issue of inverse problems. To address sparse measurement points and enhance solution accuracy, physical heat transfer information is embedded as conditional constraints in the diffusion process, progressively narrowing the constraint range during training. A dynamic loss-pulling algorithm is further proposed to enhance both solution accuracy and training speed. For the challenge of large spatial scales, the computationally intensive inversion process is reduced to a highly parallelizable map generation task, enabling rapid reconstruction of large-scale temperature fields. Simulations demonstrate that this method effectively addresses ill-posed inverse heat conduction problems. When applied to blast furnace smelting, it achieves the accuracy and efficiency required for industrial applications, offering substantial practical value.

Coal is a vital energy resource and chemical raw material. The price of coal is highly susceptible to fluctuations due to various complex and uncertain factors like supply-demand relationships and the macroeconomic environment. To accurately obtain coal price information, this paper proposes a novel multi-feature-aware gated residual deep network (MFA-GRDN) model to address the dynamic and multifaceted nature of coal price forecasting. The model integrates a gated residual network (GRN) for feature interaction modeling, a variable selection network (VSN) for adaptive factor weighting, and a gated recurrent neural network (GRNN) for temporal dependency learning, enabling dynamic factor importance assessment and enhanced long-term trend capture. Extensive experiments on real-world coal price datasets from Qinhuangdao Port demonstrate that the proposed MFA-GRDN model achieves the lowest root mean square error (RMSE) and highest goodness-of-fit ( R^2 ) among all benchmark models. Specifically, compared to the best-performing baseline, MFA-GRDN reduces RMSE to 36 R^2 consistently across both datasets, underscoring its superior accuracy and robustness in volatile coal markets.

BACKGROUND:Laser-induced breakdown spectroscopy (LIBS) enables rapid, in situ quantitative compositional analysis, yet accurate quantification of minor-content elements remains difficult due to weak spectral signatures, noise, matrix effects, and line interference. Physics-informed handcrafted features are often incomplete, while purely data-driven models may learn redundant bands and spurious correlations that suppress trace signals. Two-stage pipelines further depend heavily on preprocessing and can propagate errors. Therefore, selective feature enhancement and representation learning are needed for robust minor-element quantification in LIBS. RESULTS:A novel feature-enhanced dual-input transformer for LIBS quantitative analysis of minor-content elements is proposed. First, a cross-attention mechanism is employed to align tabulated single-atom emission spectra with complex-object LIBS spectra, so that physical prior knowledge can rapidly guide feature band. Second, Self-attention is applied to real, interference-contaminated LIBS spectra to learn the relative contributions of spectrally distant wavelength regions, thereby improving discrimination between informative signals and noise or spectral interference. Moreover, a gated attention-based fusion strategy is devised to balance data-driven and physics-guided, to emphasize feature bands associated with the target elements while suppressing those strongly affected by line overlap or matrix effects. The performances of the feature-enhanced dual-input transformer model proposed for quantitative analysis of minor-content elements are verified in two LIBS datasets, and key indicators such as R2, RMSE, and MAE are significantly improved compared with other methods. SIGNIFICANCE:The proposed feature-enhanced dual-input transformer effectively integrates physics-guided with data-driven modeling to enhance the features of minor-content elements, thereby improving the accuracy of target element quantification. It offers a novel strategy to improve the performance of LIBS-based quantitative analysis of minor-content elements, thereby enabling more reliable measurements of complex materials across various application domains.

Precise control of the outlet density in the alumina evaporation process is of critical importance for ensuring stable system operation. However, due to the pronounced nonlinearity, strong coupling, and frequent operating condition fluctuations inherent in this process, outlet density control still faces significant challenges, including insufficient modeling accuracy, frequent operational disturbances, and difficulties in control parameter tuning. To address these issues, this paper proposes a mechanism-data-driven hybrid model-based adaptive model predictive control (MDH-AMPC) method for the alumina evaporation process. First, a hybrid model integrating heat and mass transfer mechanisms with data-driven approach is constructed to enhance the representation of the dynamic characteristics of the evaporation process, thereby providing a reliable prediction basis for model predictive control. Then, based on the hybrid model, a model predictive control framework is developed, in which a disturbance compensation mechanism is incorporated to improve the system's disturbance rejection capability. Furthermore, the JAYA algorithm is employed to adaptively adjust control parameters, enabling the controller to maintain stable operation and efficient response under practical operating conditions. Simulation results under high-concentration outlet section demonstrate that the proposed MDH-AMPC outperforms conventional MPC, PID, and manual operation in terms of control stability and response speed, thereby exhibiting its excellent comprehensive control performance and engineering application potential.

As one of effective ways to improve the reliability and extend the lifetime of power converters, thermal stress control (TSC) methods have drawn increasing attention. However, the existing TSC methods mainly consider the normal condition, but do not take into account the fault conditions, which will increase the risk of secondary faults. For this reason, a thermal distribution estimation-based thermal stress tolerant control method, which is based on two-tier thermal weighting (TTW) finite-control-set model predictive control, is proposed for traction inverters. Using the proposed method without additional sensors or hardware redundancy, better reliability of traction inverters can be achieved under both normal and open-circuit (OC) fault conditions. Firstly, the electrical model and thermal model of the system under fault conditions are built and analyzed. Secondly, the energy loss-based thermal distribution estimator (TDE) is proposed to evaluate the thermal stress distribution within traction inverters. Then, a TTW-based reconfigurable reward function and a TDE-based dynamic thermal weighting model are proposed to dynamically reconfigure the control strategies and their parameters, which is used to improve the thermal distribution of traction inverters under various OC fault conditions. Finally, the effectiveness and merits of the proposed method are verified and discussed.

The volatilization kiln, a major carbon-emitting unit in zinc smelting, recovers metals from leaching residue under high-temperature redox conditions while generating carbon dioxide. Achieving efficient metal recovery with reduced carbon emissions via precise reaction atmosphere control is essential for optimal operation of the volatilization kiln. This article proposes a multiobjective adaptive dynamic optimization method to address this challenge. First, to overcome modeling difficulties arising from complex physicochemical reactions involved in the volatilization kiln, a two-stage coordinated accurate identification method is proposed, enabling high-precision process dynamics characterization through critical features selection and information criterion guided recovery mechanism. Then, to handle the infinite-dimensional spatiotemporal distribution of the reaction atmosphere, an optimal control method for key points based on the spatiotemporal response characteristics is developed. By selecting the spatial anchor points with the most significant dynamic gain of the system as key points, the multiobjective optimization control of the reaction atmosphere is realized. Finally, a memory-based dynamic optimization method is introduced to ensure optimal performance under varying operating conditions (OCs). Directed knowledge transfer from historical nondominated solution sets enables efficient optimization of control objectives under dynamic OCs, thereby ensuring stable and efficient operation over the entire operation range. To verify the effectiveness of the proposed method, comprehensive experiments are designed, and the experimental results indicated that the proposed method reduces carbon emissions by 20.45% while maintaining the zinc recovery rate at a high level, significantly improving the low-carbon optimal operation level of the volatilization kiln.

In ternary cathode materials (TCMs) manufacturing, the roller kiln temperature field directly affects product quality. Existing control methods still face difficulties caused by complex spatiotemporal dynamics, time delays, environmental disturbances, and limited online resources for monitoring and updating. This paper proposes a self-triggered H∞ control method based on adaptive dynamic programming for the time-delay sintering temperature field. First, considering perturbations and historical temperatures in the roller kiln, a time-delay H∞ performance index is constructed to formulate the robust control problem of the temperature field. Then, a time-delay triggered state is formulated, and a self-triggered mechanism is designed using the temperature at the current triggering instant and historical temperatures. Based on this, a self-triggered adaptive dynamic programming (ST-ADP) algorithm is proposed to obtain the triggered control law by solving the Hamilton-Jacobi-Isaacs equation and to predict the next control update time, thereby reducing unnecessary updates. Theoretical analysis proves closed-loop stability, excludes Zeno behavior, and establishes an upper bound for the performance index. Finally, the proposed ST-ADP method is deployed for the TCM sintering process. Industrial validation shows that ST-ADP regulates the temperature field around the setpoint under external disturbances while reducing the control update frequency.

Accurate time-series prediction of molten iron composition is essential for proactive quality assessment and operational control in blast furnace ironmaking. However, existing data-driven approaches often overlook target autocorrelation and distribution shift challenges. We propose a Transformer-based time-series prediction framework that integrates target autoregression and parameter adaptation in a collaborative manner. Specifically, an encoder-decoder architecture captures both process-target interactions and autoregressive target dependencies through multihead attention mechanisms. Building on this architecture, we further introduce a multitask and dual-path learning framework. During training, the model jointly optimizes a primary autoregressive task and an auxiliary masked-reconstruction task, thereby enabling the encoder to learn robust representations. At inference, the reconstruction error is used to adapt the encoder parameters online, maintaining feature-space consistency under distribution shift. Experiments on real industrial data demonstrate that the proposed method substantially outperforms existing approaches in both overall prediction accuracy and multistep hit rates.

The burden surface topography of a blast furnace is the main basis for judging the furnace conditions and plays an important role in adjusting the charging system and ensuring the stable progress of the ironmaking process. Visible light imaging technology has the potential to capture real-time high-resolution images of the burden surface, providing a wealth of burden surface topography information. However, capturing high-quality burden surface videos in a sealed environment with extremely uneven light distribution remains an urgent problem to be solved. To this end, this paper develops a novel type of industrial endo-scope with an adaptive analog gain to obtain information-rich burden surface images under complex lighting conditions. Firstly, a signal conversion model of the burden surface imaging process is constructed to analyze the impact of lighting on imaging. Based on the analysis, an imaging optical system with a large relative aperture and a long optical path imaging structure is developed to address the problem of weak illumination in the burden surface area. On this basis, an automatic exposure control system with an adaptive analog gain is designed to suppress the interference of dynamic strong light on imaging. Finally, experimental and application results demonstrate that the developed industrial endoscope can significantly enhance the effect of burden surface imaging and increase the amount of burden surface topography information obtained.

The state transition algorithm (STA), as an intelligent optimization method grounded in constructivist learning, has been demonstrated to be highly effective in solving complex optimization problems. However, the standard STA suffers from slow convergence, particularly in the later stages when dealing with flat landscapes. Additionally, users are required to set the maximum number of iterations based on intuition. To address these issues, an enhanced STA with guaranteed optimality is introduced. This improvement involves three key components. First, novel translation transformations (TTs), inspired by predictive modeling, are developed to generate a broader set of candidate solutions by leveraging historical data. Second, adaptive parameter control strategies are incorporated to accelerate convergence. Finally, a dedicated termination condition is designed to ensure that the algorithm converges at the optimal solution, analogous to the zero gradient condition in mathematical programming. The comprehensive experimental results validate the effectiveness and superiority of the proposed method. The source codes for ESTA and EXSTA will be publicly available at https://github.com/tiezhongyu2005/ESTA

The rapid proliferation of artificial intelligence in the industrial sector is catalyzing a smart manufacturing paradigm driven by industrial AI agents. In contrast to IT counterparts, future industrial AI agents operate under unique constraints that demand both adaptive intelligence and strictly deterministic communication with ultra-low latency and ultra-high reliability. To address these stringent requirements, the convergence of reinforcement learning (RL) and Time-Sensitive Networking (TSN) has emerged as a critical enabler. This paper presents the fundamentals of RL, TSN, and industrial AI agent communication, and establishes a mapping between AI agent cognitive behaviors, communication requirements, and deterministic networking mechanisms. We propose a four-dimensional framework encompassing intelligent scheduling, dynamic resource management, network convergence and mobility, and application-aware networking to systematically analyze existing RL-for-TSN research and elucidate challenges and requirements for future industrial AI agent communication. Finally, we outline a research roadmap spanning constrained RL algorithms and agent-network co-design to pave the way for deterministic industrial edge intelligence.

Residual oxygen detection in sealed pharmaceutical glass vials is critical for assessing container closure integrity and ensuring the quality and safety of packaged drug products. Under online inspection-line conditions, the limited absorption path and open measurement configuration make 2f/1f harmonic amplitudes susceptible to local perturbations, which restricts the accuracy of peak-based concentration inversion. To improve the detection accuracy of residual oxygen measurement, this article proposes a multiorder harmonic geometric-feature fusion (MHGF) method. Harmonic components from 2f/1f to nf/1f, each carrying concentration-related information, are first demodulated from the absorption spectra. Stable geometric features in harmonic signals of different orders are then explored, from which a compact 16-dimensional set of geometric descriptors with strong concentration discrimination capability is identified through quantitative evaluation. Experimental results obtained on a real pharmaceutical-vial inspection line show that, compared with the conventional 2f/1f peak method, the MHGF method reduces the root-mean-square error from 1.67 to 1.13, corresponding to a 32.31% reduction in detection error. In particular, under the weak-absorption condition at 0% oxygen, the error is reduced by 42.14%, indicating the effectiveness of MHGF for high-precision and online residual oxygen detection in pharmaceutical vials under industrial inspection-line conditions.