Pub Date : 2026-05-01Epub Date: 2026-02-02DOI: 10.1016/j.conengprac.2026.106816
Yunpeng Guo , Qifu Chen , Jianqi An , Zhuang Li , Jinhua She
As the smelting intensity (SI) affects both the chemical reactions and physical changes inside blast furnaces (BF), the relationship between the gas utilization rate (GUR) and blast supply parameters varies according to different levels of SI. This paper introduces a GUR prediction model that considers SI classification information. First, the impact of SI on the state parameters of a BF is evaluated from the perspective of the molten iron smelting mechanism. Then, a fuzzy C-means clustering method (FCM) is presented to classify SI based on state parameters. Subsequently, an SI-aware GUR prediction model is constructed using principal component analysis (PCA) and an extreme learning machine (ELM) to predict GUR development trends. Finally, the model is used to predict real-world GUR data under different SI levels. Analysis of real-world production data shows that the proposed method accurately predicts GUR and outperforms methods that do not account for SI classification.
{"title":"Prediction model of the gas utilization rate in a blast furnace considering smelting intensity classification","authors":"Yunpeng Guo , Qifu Chen , Jianqi An , Zhuang Li , Jinhua She","doi":"10.1016/j.conengprac.2026.106816","DOIUrl":"10.1016/j.conengprac.2026.106816","url":null,"abstract":"<div><div>As the smelting intensity (SI) affects both the chemical reactions and physical changes inside blast furnaces (BF), the relationship between the gas utilization rate (GUR) and blast supply parameters varies according to different levels of SI. This paper introduces a GUR prediction model that considers SI classification information. First, the impact of SI on the state parameters of a BF is evaluated from the perspective of the molten iron smelting mechanism. Then, a fuzzy C-means clustering method (FCM) is presented to classify SI based on state parameters. Subsequently, an SI-aware GUR prediction model is constructed using principal component analysis (PCA) and an extreme learning machine (ELM) to predict GUR development trends. Finally, the model is used to predict real-world GUR data under different SI levels. Analysis of real-world production data shows that the proposed method accurately predicts GUR and outperforms methods that do not account for SI classification.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106816"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191318","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-02-10DOI: 10.1016/j.conengprac.2026.106837
Junjie Kang, Jinjun Shan
Unmanned aerial vehicles with cable-suspended payloads suffer from underactuation, strong coupling, and disturbances. These factors often cause payload swing and degrade tracking performance. This paper proposes a robust control method based on the fully actuated system (FAS) framework. The system dynamics are reformulated into a reduced-order FAS form with virtual constraints. The remaining states are arranged into a cascade structure for ensuring stability. On this basis, FAS controllers are designed for both outer-loop dynamics (UAV position and payload swing) and inner-loop attitude dynamics. Disturbance observers are introduced to handle external disturbances. Stability is proved through cascade analysis. Simulations and experiments confirm that the proposed controllers achieve robust waypoint tracking with improved swing suppression.
{"title":"Robust control of aerial cable-suspended payload transportation via fully actuated system approach","authors":"Junjie Kang, Jinjun Shan","doi":"10.1016/j.conengprac.2026.106837","DOIUrl":"10.1016/j.conengprac.2026.106837","url":null,"abstract":"<div><div>Unmanned aerial vehicles with cable-suspended payloads suffer from underactuation, strong coupling, and disturbances. These factors often cause payload swing and degrade tracking performance. This paper proposes a robust control method based on the fully actuated system (FAS) framework. The system dynamics are reformulated into a reduced-order FAS form with virtual constraints. The remaining states are arranged into a cascade structure for ensuring stability. On this basis, FAS controllers are designed for both outer-loop dynamics (UAV position and payload swing) and inner-loop attitude dynamics. Disturbance observers are introduced to handle external disturbances. Stability is proved through cascade analysis. Simulations and experiments confirm that the proposed controllers achieve robust waypoint tracking with improved swing suppression.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106837"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191316","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-02-05DOI: 10.1016/j.conengprac.2026.106794
Karim Nadim , Ahmed Ragab , Hakim Ghezzaz
Industrial plants are equipped with several local controllers with a high degree of interaction. Controllers in complex systems tend to operate in a competitive way to achieve their own objective, which can negatively impact other controllers and consequently the global KPI. In addition, the rapid changes in process dynamics, the variations, and fluctuations in the process conditions and production targets introduce major challenges in optimizing the whole process. As a result, operators struggle to adjust the controllers’ setpoints to optimize the process operation. Therefore, there is a clear need for an approach that captures the controllers’ interdependence and optimizes the setpoints in real-time to ensure energy-efficient operations. This paper proposes an intelligent decentralized supervisory control approach based on multi-agent deep reinforcement learning (MADRL) to recommend the optimal combinations of controllers’ setpoints that maintain desired process operation. Multiple agents are developed based on the deep deterministic policy gradient algorithm to collaborate and control different interconnected subsystems. The agents are trained via interacting with a process simulation, where each agent performs actions (setpoint changes) and observes certain rewards (global KPI to be maximized) and states (measured variables) from the simulation. The approach is validated on a case study based on a heat recovery network of a thermomechanical pulp mill comprising four different subsystems. The proposed decentralized approach was compared to two centralized approaches: a baseline control set by the process expert and a single DDPG agent. The multi-agent approach was able to reduce the steam flow consumption by 6.7 % compared to the experts’ baseline and 5.3% compared to the single agent with faster convergence. Two possible strategies were proposed to implement this approach in the industry, depending on the criticality of the process and the degree of fidelity of its process simulation.
{"title":"Multi agent deep reinforcement learning for supervising local controllers in energy-intensive industrial processes","authors":"Karim Nadim , Ahmed Ragab , Hakim Ghezzaz","doi":"10.1016/j.conengprac.2026.106794","DOIUrl":"10.1016/j.conengprac.2026.106794","url":null,"abstract":"<div><div>Industrial plants are equipped with several local controllers with a high degree of interaction. Controllers in complex systems tend to operate in a competitive way to achieve their own objective, which can negatively impact other controllers and consequently the global KPI. In addition, the rapid changes in process dynamics, the variations, and fluctuations in the process conditions and production targets introduce major challenges in optimizing the whole process. As a result, operators struggle to adjust the controllers’ setpoints to optimize the process operation. Therefore, there is a clear need for an approach that captures the controllers’ interdependence and optimizes the setpoints in real-time to ensure energy-efficient operations. This paper proposes an intelligent decentralized supervisory control approach based on multi-agent deep reinforcement learning (MADRL) to recommend the optimal combinations of controllers’ setpoints that maintain desired process operation. Multiple agents are developed based on the deep deterministic policy gradient algorithm to collaborate and control different interconnected subsystems. The agents are trained via interacting with a process simulation, where each agent performs actions (setpoint changes) and observes certain rewards (global KPI to be maximized) and states (measured variables) from the simulation. The approach is validated on a case study based on a heat recovery network of a thermomechanical pulp mill comprising four different subsystems. The proposed decentralized approach was compared to two centralized approaches: a baseline control set by the process expert and a single DDPG agent. The multi-agent approach was able to reduce the steam flow consumption by 6.7 % compared to the experts’ baseline and 5.3% compared to the single agent with faster convergence. Two possible strategies were proposed to implement this approach in the industry, depending on the criticality of the process and the degree of fidelity of its process simulation.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106794"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper quantitatively compares and evaluates fault-tolerant control strategies to ensure vehicle-level safety under complete steer-by-wire (SbW) fault conditions. Previous studies were often limited to specific failure types or lacked systematic strategy comparisons, making it difficult to clearly identify their applicability and limitations. In this study, a unified fault-tolerant control framework addressing two types of complete SbW failures-fixed steering angle (FSA) and loss of steering torque (LST)–was developed. Within this framework, various strategies including rear-wheel steering (RWS), torque vectoring (TV), and their combinations were implemented, and performance was analyzed using standard evaluation scenarios: slow-ramp, sine-sweep, step, and 1-period sine steer. Simulation results indicate that under FSA failure, RWS-based strategies are relatively effective, with all strategies achieving near-nominal vehicle performance at high speeds. In contrast, LST failure leads to significant performance degradation due to unintended front-wheel steering, making nominal-level cornering unattainable. RWS-only control exhibits severe limitations, while partial compensation is achieved when combined with TV, demonstrating the benefit of multi-actuator coordination under fault conditions. These findings were further validated through real-vehicle tests, confirming the practical applicability of the proposed SbW fault-tolerant controller. By systematically comparing multiple strategies across both FSA and LST failure types under complete SbW conditions, the study provides fundamental insights for designing fault-tolerant controllers that account for failure-specific characteristics, establishing a foundation for future real-vehicle implementation and application research.
{"title":"Comparative study of fault-tolerant control strategies for complete steer-by-wire failures","authors":"Yunchul Ha , Seunguk Jeon , Aldo Sorniotti , Seunghoon Woo","doi":"10.1016/j.conengprac.2026.106812","DOIUrl":"10.1016/j.conengprac.2026.106812","url":null,"abstract":"<div><div>This paper quantitatively compares and evaluates fault-tolerant control strategies to ensure vehicle-level safety under complete steer-by-wire (SbW) fault conditions. Previous studies were often limited to specific failure types or lacked systematic strategy comparisons, making it difficult to clearly identify their applicability and limitations. In this study, a unified fault-tolerant control framework addressing two types of complete SbW failures-fixed steering angle (FSA) and loss of steering torque (LST)–was developed. Within this framework, various strategies including rear-wheel steering (RWS), torque vectoring (TV), and their combinations were implemented, and performance was analyzed using standard evaluation scenarios: slow-ramp, sine-sweep, step, and 1-period sine steer. Simulation results indicate that under FSA failure, RWS-based strategies are relatively effective, with all strategies achieving near-nominal vehicle performance at high speeds. In contrast, LST failure leads to significant performance degradation due to unintended front-wheel steering, making nominal-level cornering unattainable. RWS-only control exhibits severe limitations, while partial compensation is achieved when combined with TV, demonstrating the benefit of multi-actuator coordination under fault conditions. These findings were further validated through real-vehicle tests, confirming the practical applicability of the proposed SbW fault-tolerant controller. By systematically comparing multiple strategies across both FSA and LST failure types under complete SbW conditions, the study provides fundamental insights for designing fault-tolerant controllers that account for failure-specific characteristics, establishing a foundation for future real-vehicle implementation and application research.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106812"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081621","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-01-30DOI: 10.1016/j.conengprac.2026.106808
Penglong Lian , Penghui Shang , Jianxiao Zou , Shicai Fan
Open-set fault diagnosis in rotating machinery is critically hindered by substantial inter-class similarity between unknown and known fault classes, leading to unreliable recognition. Although significant advances have been made using various adaptation and classification techniques, current open-set methods still struggle to resolve fine-grained distinctions and class ambiguities in open-set environments, often resulting in misclassifications and higher maintenance costs. To address these challenges, we propose an adaptive dual-stage framework that integrates a novel tri-branch network and dynamic contrastive learning (Ds-TBN). Specifically, the tri-branch network integrates a base feature branch, a similarity-sensitive branch, and a global feature enhancement branch to collaboratively extract complementary and discriminative representations. Dynamic contrastive learning is then applied to enforce intra-class compactness and explicitly enhance inter-class separability, significantly improving feature discriminability. Building on these enhanced representations, the dual-stage recognition framework first utilizes an adaptive Weibull distribution to detect boundary outliers for accurate identification of unknown fault classes. Subsequently, the second stage further refines classification probabilities using a meta-recognition module, adaptively resolving ambiguities between highly similar known and unknown faults. Extensive experiments across diverse similarity-based open-set diagnostic tasks on the CWRU, Gearbox, and our self-developed Drivetrain Prognostics Simulator (DPS) test bench show that the proposed method Ds-TBN achieves average H-scores of 96.65%, 90.43%, and 93.58%, respectively. These results significantly surpass existing approaches and highlight the framework’s robustness and practical applicability for real-world industrial fault diagnosis.
{"title":"Dual-stage recognition framework for open-set fault diagnosis in rotating machinery considering varying inter-class similarity","authors":"Penglong Lian , Penghui Shang , Jianxiao Zou , Shicai Fan","doi":"10.1016/j.conengprac.2026.106808","DOIUrl":"10.1016/j.conengprac.2026.106808","url":null,"abstract":"<div><div>Open-set fault diagnosis in rotating machinery is critically hindered by substantial inter-class similarity between unknown and known fault classes, leading to unreliable recognition. Although significant advances have been made using various adaptation and classification techniques, current open-set methods still struggle to resolve fine-grained distinctions and class ambiguities in open-set environments, often resulting in misclassifications and higher maintenance costs. To address these challenges, we propose an adaptive dual-stage framework that integrates a novel tri-branch network and dynamic contrastive learning (Ds-TBN). Specifically, the tri-branch network integrates a base feature branch, a similarity-sensitive branch, and a global feature enhancement branch to collaboratively extract complementary and discriminative representations. Dynamic contrastive learning is then applied to enforce intra-class compactness and explicitly enhance inter-class separability, significantly improving feature discriminability. Building on these enhanced representations, the dual-stage recognition framework first utilizes an adaptive Weibull distribution to detect boundary outliers for accurate identification of unknown fault classes. Subsequently, the second stage further refines classification probabilities using a meta-recognition module, adaptively resolving ambiguities between highly similar known and unknown faults. Extensive experiments across diverse similarity-based open-set diagnostic tasks on the CWRU, Gearbox, and our self-developed Drivetrain Prognostics Simulator (DPS) test bench show that the proposed method Ds-TBN achieves average H-scores of 96.65%, 90.43%, and 93.58%, respectively. These results significantly surpass existing approaches and highlight the framework’s robustness and practical applicability for real-world industrial fault diagnosis.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106808"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081726","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-02-13DOI: 10.1016/j.conengprac.2026.106833
Haoan Wang , Ying Meng , Jinji Sun , Lu Zhang , Haifeng Zhang , Shiqiang Zheng
To address the dual challenges of low-frequency spatial magnetic field fluctuations and mid-frequency noise in compact magnetically shielded rooms (MSRs) for optically pumped magnetometer-based magnetoencephalography (MEG), a dual-coil cooperative control method is proposed. The method integrates external compensation (EC) coils and internal compensation (IC) coils to establish a joint compensation mechanism in frequency and space within the target region. Each coil employs a modular error-based linear active disturbance rejection controller (e-LADRC) to simplify engineering applications. Specifically, the EC subsystem reduces low-frequency magnetic fluctuations in the target region by 95.0% through anti-phase spatial compensation of non-uniform disturbance fields. Concurrently, the IC subsystem actively suppresses mid-frequency noise via feedback control, decreasing the average power spectral density (PSD) in the 1–40 Hz by 4.22 dB. This work provides an engineering-oriented solution for compact, high-precision magnetic measurement systems, with potential applications in fields requiring low-noise magnetic sensing.
{"title":"Dual-coil active disturbance rejection control for compact MEG measurement systems","authors":"Haoan Wang , Ying Meng , Jinji Sun , Lu Zhang , Haifeng Zhang , Shiqiang Zheng","doi":"10.1016/j.conengprac.2026.106833","DOIUrl":"10.1016/j.conengprac.2026.106833","url":null,"abstract":"<div><div>To address the dual challenges of low-frequency spatial magnetic field fluctuations and mid-frequency noise in compact magnetically shielded rooms (MSRs) for optically pumped magnetometer-based magnetoencephalography (MEG), a dual-coil cooperative control method is proposed. The method integrates external compensation (EC) coils and internal compensation (IC) coils to establish a joint compensation mechanism in frequency and space within the target region. Each coil employs a modular error-based linear active disturbance rejection controller (e-LADRC) to simplify engineering applications. Specifically, the EC subsystem reduces low-frequency magnetic fluctuations in the target region by 95.0% through anti-phase spatial compensation of non-uniform disturbance fields. Concurrently, the IC subsystem actively suppresses mid-frequency noise via feedback control, decreasing the average power spectral density (PSD) in the 1–40 Hz by 4.22 dB. This work provides an engineering-oriented solution for compact, high-precision magnetic measurement systems, with potential applications in fields requiring low-noise magnetic sensing.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106833"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191393","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-01-31DOI: 10.1016/j.conengprac.2026.106805
Niccolò La Rosa , Samuele Moscato , Luigi Fortuna , Maide Bucolo , Massimo Camarda
Maintaining sub-micrometer stability of synchrotron X-ray beams is essential for the accuracy and repeatability of cutting-edge scientific and medical experiments. Traditional beamline stabilization systems, based on mechanical actuation of optical elements, are inherently limited in speed due to physical constraints like friction and inertia. This study introduces an innovative control strategy based on electrical actuation, directly influencing the bending magnet responsible for steering the beam into the beamline. This approach unlocks the potential for significantly higher control frequencies, comparable to those used for the electron beam stabilization. A laboratory-scale replica was developed to validate the feasibility and robustness of this method. A Proportional-Integral (PI) controller has been implemented to stabilize the electron beam and compensate for disturbances. Experimental results demonstrate that this strategy enables precise, high-frequency beam stabilization, even in the presence of typical disturbances such as position drift occurring during X-Ray Absorption Spectroscopy (XAS) experiments. This work lays the groundwork for next-generation control systems in synchrotron facilities, aiming to enhance performance and open the door to more advanced experimental capabilities.
{"title":"Electrically actuated control system for the stabilization of synchrotron X-ray beams","authors":"Niccolò La Rosa , Samuele Moscato , Luigi Fortuna , Maide Bucolo , Massimo Camarda","doi":"10.1016/j.conengprac.2026.106805","DOIUrl":"10.1016/j.conengprac.2026.106805","url":null,"abstract":"<div><div>Maintaining sub-micrometer stability of synchrotron X-ray beams is essential for the accuracy and repeatability of cutting-edge scientific and medical experiments. Traditional beamline stabilization systems, based on mechanical actuation of optical elements, are inherently limited in speed due to physical constraints like friction and inertia. This study introduces an innovative control strategy based on electrical actuation, directly influencing the bending magnet responsible for steering the beam into the beamline. This approach unlocks the potential for significantly higher control frequencies, comparable to those used for the electron beam stabilization. A laboratory-scale replica was developed to validate the feasibility and robustness of this method. A Proportional-Integral (PI) controller has been implemented to stabilize the electron beam and compensate for disturbances. Experimental results demonstrate that this strategy enables precise, high-frequency beam stabilization, even in the presence of typical disturbances such as position drift occurring during X-Ray Absorption Spectroscopy (XAS) experiments. This work lays the groundwork for next-generation control systems in synchrotron facilities, aiming to enhance performance and open the door to more advanced experimental capabilities.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106805"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191311","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-02-10DOI: 10.1016/j.conengprac.2026.106823
Jiliang Luo , Zexuan Lin , Sijia Yi , Zhaoyu Ye , Fei-Yue Wang
In order to bridge the gap between the virtual and real worlds, this paper presents a novel model known as the parallel Petri net, specifically designed for the digital twin modeling and real-time coordination of cyber physical systems, where each operation is assumed to require a constant processing time. Unlike traditional Petri nets, the parallel Petri net incorporates new elements, such as actor and promoter functions in association with traditional places. In this framework, actors may function as either discrete-event or continuous-state controllers. The transition firing rules are defined to create an execution algorithm to drive the cyber physical system being modeled and co-run the corresponding parallel Petri net in parallel as a cyber model of the physical system, thereby applying its semantic control specifications, including manufacturing process sequences and recipes, to real-world actions. Due to its extensive modeling capabilities, the parallel Petri net offers a wide range of transition firing options in most states of practical operations. This enables simultaneous addressing of scheduling and control challenges to enhance the efficiency of physical systems. To this end, a reinforcement learning approach is developed based on the simulations of parallel Petri nets. It is demonstrated that the state-action value function can reliably predict the minimum time required to reach a goal state following a transition. Additionally, a deep Q-learning algorithm is presented, where the parallel Petri net serves as the operational environment, to train a neural network model for real-time scheduling. As a result, the parallel Petri net is capable of making intelligent decisions for cyber-physical systems through the neural network. Finally, the implementation framework for parallel Petri nets has been detailed, and experiments conducted in a manufacturing plant have verified the validity of the proposed techniques.
{"title":"Parallel Petri nets with reinforcement learning for intelligent decision-making of digital twins in cyber physical systems","authors":"Jiliang Luo , Zexuan Lin , Sijia Yi , Zhaoyu Ye , Fei-Yue Wang","doi":"10.1016/j.conengprac.2026.106823","DOIUrl":"10.1016/j.conengprac.2026.106823","url":null,"abstract":"<div><div>In order to bridge the gap between the virtual and real worlds, this paper presents a novel model known as the parallel Petri net, specifically designed for the digital twin modeling and real-time coordination of cyber physical systems, where each operation is assumed to require a constant processing time. Unlike traditional Petri nets, the parallel Petri net incorporates new elements, such as actor and promoter functions in association with traditional places. In this framework, actors may function as either discrete-event or continuous-state controllers. The transition firing rules are defined to create an execution algorithm to drive the cyber physical system being modeled and co-run the corresponding parallel Petri net in parallel as a cyber model of the physical system, thereby applying its semantic control specifications, including manufacturing process sequences and recipes, to real-world actions. Due to its extensive modeling capabilities, the parallel Petri net offers a wide range of transition firing options in most states of practical operations. This enables simultaneous addressing of scheduling and control challenges to enhance the efficiency of physical systems. To this end, a reinforcement learning approach is developed based on the simulations of parallel Petri nets. It is demonstrated that the state-action value function can reliably predict the minimum time required to reach a goal state following a transition. Additionally, a deep Q-learning algorithm is presented, where the parallel Petri net serves as the operational environment, to train a neural network model for real-time scheduling. As a result, the parallel Petri net is capable of making intelligent decisions for cyber-physical systems through the neural network. Finally, the implementation framework for parallel Petri nets has been detailed, and experiments conducted in a manufacturing plant have verified the validity of the proposed techniques.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106823"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191319","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-05-01Epub Date: 2026-02-09DOI: 10.1016/j.conengprac.2026.106815
Bikang Hua , Zhaofeng Du , Tianhao Liu , Runqi Chai , Jiping Xu , Senchun Chai
This paper presents and validates an efficient dual-layer trajectory planning framework for the dynamic environment. The upper-layer utilizes offline global planning to generate an optimal global trajectory in a static obstacle environment, while the lower-layer performs online local planning, enabling real-time obstacle avoidance by predicting the future states of the dynamic obstacle and employing appropriate avoidance strategies. The main novelty lies in the following aspects: firstly, a fault-tolerant dynamic obstacle avoidance strategy, along with an LSTM-based trajectory prediction network, enables flexible velocity planning or local trajectory replanning based on the situation; secondly, a convex optimization-based parallel stitching strategy for local trajectory replanning, where candidate trajectories are generated through parallel computation and the optimal stitching solution is greedily selected. During the parallel problem-solving process, the original problem is transformed into a convex optimization problem via linearization and convexification to enhance solution efficiency. Iterative numerical solving is applied, with Line Search steps introduced between iterations to reduce deviation from original constraints and further minimize approximation errors. Simulation and experimental results validate the effectiveness and practicality of the proposed framework.
{"title":"Convex optimization-based parallel trajectory stitching in dynamic environments: A dual-layer trajectory planning framework","authors":"Bikang Hua , Zhaofeng Du , Tianhao Liu , Runqi Chai , Jiping Xu , Senchun Chai","doi":"10.1016/j.conengprac.2026.106815","DOIUrl":"10.1016/j.conengprac.2026.106815","url":null,"abstract":"<div><div>This paper presents and validates an efficient dual-layer trajectory planning framework for the dynamic environment. The upper-layer utilizes offline global planning to generate an optimal global trajectory in a static obstacle environment, while the lower-layer performs online local planning, enabling real-time obstacle avoidance by predicting the future states of the dynamic obstacle and employing appropriate avoidance strategies. The main novelty lies in the following aspects: firstly, a fault-tolerant dynamic obstacle avoidance strategy, along with an LSTM-based trajectory prediction network, enables flexible velocity planning or local trajectory replanning based on the situation; secondly, a convex optimization-based parallel stitching strategy for local trajectory replanning, where candidate trajectories are generated through parallel computation and the optimal stitching solution is greedily selected. During the parallel problem-solving process, the original problem is transformed into a convex optimization problem via linearization and convexification to enhance solution efficiency. Iterative numerical solving is applied, with Line Search steps introduced between iterations to reduce deviation from original constraints and further minimize approximation errors. Simulation and experimental results validate the effectiveness and practicality of the proposed framework.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106815"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Wireless sensor networks (WSNs) are an important component of cyber-physical systems (CPS). They consist of low-resource, low-power nodes that perform local measurement and communication. While the CPS framework relies on complex computation and optimization over a network, this approach is not feasible in remote areas lacking stable network infrastructure. Under such situations, computation and optimization must be executed locally where the WSN nodes are installed. Using the IoT devices of the WSN nodes for performing computation and optimization is a possible approach, but the limited computational resources of these devices make it impossible to implement advanced control laws, such as Model Predictive Control (MPC). One solution for this is to decompose a single MPC problem into small subproblems, each of which is implemented on a low resource IoT device. In this paper, we focus on a parallel MPC method and derive an analytic form of unconstrained solutions of subproblems. With this derivation, we propose improved implementation algorithms of the parallel MPC for multiple low-resource IoT devices. Compared to existing time-splitting parallel MPC schemes, our parallel MPC scheme requires a smaller amount of data exchanged among subproblems, making it suitable for implementations using wireless communication capabilities of low-resource IoT devices. The proposed method is experimentally validated using multiple micro:bits with wireless communications.
{"title":"Experimental validation of parallel model predictive control on multiple low-resource IoT devices","authors":"Shunta Yamamoto, Naoyuki Hara, Keiji Konishi, Yoshiki Sugitani","doi":"10.1016/j.conengprac.2026.106818","DOIUrl":"10.1016/j.conengprac.2026.106818","url":null,"abstract":"<div><div>Wireless sensor networks (WSNs) are an important component of cyber-physical systems (CPS). They consist of low-resource, low-power nodes that perform local measurement and communication. While the CPS framework relies on complex computation and optimization over a network, this approach is not feasible in remote areas lacking stable network infrastructure. Under such situations, computation and optimization must be executed locally where the WSN nodes are installed. Using the IoT devices of the WSN nodes for performing computation and optimization is a possible approach, but the limited computational resources of these devices make it impossible to implement advanced control laws, such as Model Predictive Control (MPC). One solution for this is to decompose a single MPC problem into small subproblems, each of which is implemented on a low resource IoT device. In this paper, we focus on a parallel MPC method and derive an analytic form of unconstrained solutions of subproblems. With this derivation, we propose improved implementation algorithms of the parallel MPC for multiple low-resource IoT devices. Compared to existing time-splitting parallel MPC schemes, our parallel MPC scheme requires a smaller amount of data exchanged among subproblems, making it suitable for implementations using wireless communication capabilities of low-resource IoT devices. The proposed method is experimentally validated using multiple micro:bits with wireless communications.</div></div>","PeriodicalId":50615,"journal":{"name":"Control Engineering Practice","volume":"170 ","pages":"Article 106818"},"PeriodicalIF":4.6,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191309","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号