To balance environmental sustainability and production efficiency in modern computer-integrated manufacturing, the energy-efficient scheduling of automated systems with batch processing machines (BPMs) is essential. This study investigates a multi-resource constrained energy-efficient hybrid flow shop scheduling problem (EHFSP) featuring middle-stage BPMs, stage precedence flexibility, and heterogeneous worker assignments. To concurrently minimize the makespan, total energy consumption, and total carbon emissions, an integrated collaborative scheduling approach driven by an improved teaching-learning-based optimization (EHTLBO) algorithm is proposed. The EHTLBO incorporates a score electoral system-based hyper-heuristic mechanism to effectively handle the complex coupling between automated machine operations and multi-skilled workforce allocation. Specifically, low-level heuristics are dynamically recruited as independent voters to execute consensus-based resource dispatching. Subsequently, an adaptive multiple-class construction strategy and an inter-class collaborative teaching mechanism are designed to facilitate directional knowledge transfer and efficient rule assimilation. To strictly mitigate premature convergence, a dynamic perturbation mechanism featuring diverse neighborhood structures is implemented to revitalize stagnant solutions. In addition, a CPLEX-based verification is conducted on a small-scale linearized MILP instance to support the feasibility and consistency of the proposed mathematical formulation. Extensive computational experiments demonstrate that EHTLBO consistently provides high-quality non-dominated solutions with superior convergence and distribution over state-of-the-art meta-heuristics. Furthermore, a real-life application in an automated casting enterprise validates its practical effectiveness, proving that the proposed approach significantly reduces energy consumption and carbon emissions while improving production throughput compared to conventional industrial rules.

In the complex landscape of multi-dimensional coupling within intelligent manufacturing systems, optimizing decisions for the flexible job shop scheduling problem (FJSP) is influenced not only by the assignment of machine resources but also by the interplay with the transportation system and the availability of auxiliary resources. In this paper, a multi-objective FJSP model considering limited transportation and auxiliary resources is proposed in the context of hydraulic cylinder manufacturing, aiming to minimize the maximum completion time and semifinished product inventory cost. To address the problem, a reinforcement learning enhanced imperialist competitive algorithm (RLICA) is proposed. The algorithm has the following features: multiple hybrid initialization strategies are introduced to enhance population diversity and quality. Additionally, a dynamic parameter adaptation strategy based on reinforcement learning is designed for the assimilation stage to optimize the exploration-exploitation balance during the colonization assimilation process. Finally, a two-layer neighborhood search operator is implemented to improve the algorithm's local search capability. The experimental section evaluates the performance of the algorithm based on 33 benchmark instances and real enterprise data. The results demonstrate that RLICA significantly outperforms five other state-of-the-art algorithms regarding convergence, distribution, and robustness. Case study further demonstrates that the algorithm can effectively coordinate transportation, processing, and auxiliary resource systems, offering a robust scheduling optimization solution for complex manufacturing scenarios.

In manufacturing systems, unexpected machine failures disrupt production plans and degrade scheduling performance. Unlike existing works that rely on either overly conservative proactive plans or unstable purely reactive rules, this paper investigates the distributed heterogeneous flexible job shop resilient scheduling problem, combining a proactive strategy with preventive maintenance (PM) and a predictive-reactive mechanism. Specifically, this paper establishes model with the following phased optimization objectives: during the proactive scheduling phase, the aims are to minimize the makespan and maximize the number of PM within safety thresholds to probabilistically minimize expected downtime and failure risks; during the recovery phase, the goals are to minimize makespan deviation and minimize the number of machine reassignments to preserve system stability. Corresponding solution algorithms are specifically designed for two phases. For the proactive scheduling phase, an improved multi-objective artificial bee colony algorithm is developed, featuring a newly integrated simulated annealing neighborhood search targeting global bottlenecks. For the recovery phase, a deep reinforcement learning (DRL) framework is constructed with a novel three-dimensional state-feature representation to adaptively select rescheduling strategies. Experimental results show the proposed framework significantly improves system resilience. Proactive PM integration reduces unexpected breakdowns, limiting severe makespan delays to 7.0%, while the trained DRL agent achieves <5 ms strategy inference during recovery. Scientifically, this study advances the multi-objective optimization of stochastic manufacturing environments. Societally, it provides a viable pathway toward realizing resilient shop-floor scheduling, ensuring stability and continuous production.

To address the critical issues of inefficient coordination and suboptimal global performance in Heterogeneous Robot Order Fulfillment Systems (HROFS), this study investigates the Heterogeneous Robot Pick-up and Delivery (HRPD) problem by mapping it as an Open Shop Scheduling problem with Sequence-dependent Setup, Transportation Time, and Transportation Waiting Time (OSS-SSTTWT). A multi-objective scheduling model is introduced to enhance collaboration between Autonomous Mobile Robots (AMRs) and picking robots by simultaneously minimizing Total Completion Time, transportation time, and transportation waiting time. To achieve Heterogeneous Robot Adaptive Collaborative Task Planning (HR-ACTP), an Adaptive Multi-Objective Deep Q-Network improved with an Upper Confidence Bound strategy (UCB-AMDQN) is proposed. Experimental results demonstrate that the UCB-AMDQN significantly outperforms industrial baselines such as FIFO, as well as DDQN and combined scheduling rules, by discovering synergistic collaboration patterns that transcend human-designed heuristics and existing cognitive frameworks. Furthermore, the research analyzes the impact of velocity differences and partitioning strategies on coordination efficiency. By explicitly incorporating oftenneglected transportation metrics into a novel OSS-SSTTWT framework, this work provides a comprehensive and effective pathway for optimizing complex heterogeneous robot collaborative task planning in smart logistics. (c) 2026 by the authors; licensee Growing Science, Canada

In the context of mass customization, manufacturing services require frequent adjustment and configuration of resources. Enterprises lack a unified model to describe the reliability of manufacturing service encapsulation under complex resource combinations. Meanwhile, during the encapsulation process, the latent relationships between heterogeneous resources are challenging to extract, making it difficult to guarantee the reliability of manufacturing services, ultimately resulting in production efficiency and product quality failing to meet customer demands. To address the issues above, this paper proposes A novel adaptive optimization method for reliable encapsulation of manufacturing service based on a graph convolution network with multi-dimensional feature fusion. First, a novel adaptive optimization method for manufacturing service reliability encapsulation (AO-MSRE) is constructed to characterize encapsulation reliability under diverse resource combinations. Subsequently, based on the characteristics of this model, the graph convolutional network (GCN) algorithm was improved, and the improved GCN algorithm was combined with the edge graph neural network (EGNN). A novel multi-dimensional feature fusion graph convolutional network (MFFGCN) framework-specifically, the reliability characterization network (RCN)-was designed. This framework integrates node, edge, and edge-graph feature information to comprehensively analyze the impacts of heterogeneous resources, resource relationships, and implicit interference between relationships on encapsulation reliability. Experimental results demonstrate that RCN achieves outstanding performance in optimizing manufacturing service reliability encapsulation, with a mean absolute percentage error (MAPE) as low as 1.95% while maintaining robustness under noise interference. This work provides a theoretical foundation and practical tools for reliable manufacturing service encapsulation in dynamic production environments.

In the context of the growing demand for service-oriented manufacturing and collaborative production, social manufacturing (SM) emerges as an effective paradigm for addressing the dynamic collaboration challenges among small and medium-sized enterprises (SMEs) and individual manufacturers. Swarm intelligence-based self-organizing collaboration mechanisms are critical for integrating distributed enterprises into manufacturing communities (MCs). However, due to the partial observability of community states and the non-stationary nature of the SM environment, traditional centralized or rule-based methods are inadequate for such complex collaboration problems, particularly when enterprise data privacy must be preserved. Therefore, this paper proposes a federated deep reinforcement learning (DRL) approach to guide the collaboration and production strategies of manufacturing nodes within MCs, aiming to achieve swarm collaborative manufacturing while protecting data privacy under enterprise independence. Specifically, a multi-participant SM system is modeled as a multi-agent system (MAS), and four fundamental modules are established to simulate a realistic SM environment. Subsequently, the optimization objectives for this collaborative problem are defined, and then the problem is formulated as a decentralized partially observable Markov decision process (Dec-POMDP). A federated learning-based proximal policy optimization algorithm (FedPPO) is developed to solve the Dec-POMDP, with detailed designs of the local training process and the federated aggregation mechanism. Simulation results demonstrate that the proposed DRL-based method significantly improves production efficiency and reduces collaboration costs in MCs compared with benchmark paradigms.

Sequence-dependent setup times often cause capacity losses in elevator component manufacturing. This study targets the flexible flow shop scheduling problem (FFSP) with stage-skipping to mitigate these issues. To overcome the bottlenecks of logistic blockages and equipment idleness inherent in traditional process batch transfer, a mathematical model aimed at minimizing Makespan is established, and an Improved Genetic Algorithm integrated with a lot streaming strategy is proposed. By employing heuristic initialization and greedy decoding mechanisms, the algorithm achieves intelligent clustering and parallel processing of sub-lots. Simulation experiments based on real-world data indicate that at the optimal splitting granularity, the proposed scheme effectively balances setup penalties and parallel gains. It significantly shortens production cycles and reduces work-in-process inventory, validating the effectiveness of the strategy for lean scheduling in discrete manufacturing.

Multi-objective optimization plays an important role in modern design and complex engineering applications. However, achieving an effective balance between the convergence and diversity of Pareto-optimal solutions remains challenging. This paper proposes a Sobol-driven Multi-objective Whale Migration Algorithm (SMOWMA), which extends the Whale Migration Algorithm within a non-dominated sorting and elite-selection framework. A maximin scrambled Sobol initialization scheme is first employed to improve the distribution of the initial population. An archive-guided adaptive Student-t flight mechanism is then incorporated into the leader-whale position update to dynamically balance global exploration and local exploitation. In addition, archive crowding information and archive-entry success feedback are jointly used to adjust the search behavior according to both environmental diversity and recent search performance. SMOWMA is evaluated on five widely used multi-objective benchmark suites, namely ZDT, DTLZ, WFG, UF, and CF, using four performance indicators: generational distance (GD), inverted generational distance (IGD), spacing (SP), and hypervolume (HV). The results, together with Friedman tests and Holm-adjusted Wilcoxon tests, demonstrate that SMOWMA achieves competitive overall performance in terms of convergence, diversity, and objective-space coverage, although its relative advantage remains problem-dependent. The practical applicability of SMOWMA is further examined using multi-objective welded-beam design formulations, a bi-objective four-bar truss design problem, and a five-objective car side-impact design problem. The engineering results show that SMOWMA can obtain competitive and stable approximation sets for constrained design problems with different numbers of objectives, supporting its effectiveness and applicability in multi-objective engineering optimization.

PurposeAs goods-to-person systems become increasingly common in warehouses, multiload automated guided vehicles (AGVs) face growing challenges in avoiding path conflicts and deadlocks during collaborative operations. This study aims to propose a conflict-free path planning strategy to enhance overall efficiency and system stability.Design/methodology/approachThe warehouse is modeled as a uniform cellular grid, with AGVs represented as dynamic cells possessing state attributes. Task sequencing is optimized via dynamic programming based on the Traveling Salesman Problem. An improved A* algorithm - integrating turning penalties and dynamic weights - generates initial AGV paths. These serve as guides for a cellular automaton-based simulation that resolves Head-on, Intersection and Parking Conflicts through a Von Neumann neighborhood rule set and a hierarchical decision-making mechanism.FindingsSimulation results demonstrate that the proposed method achieves optimal single-AGV path, improves coordination among multiple AGVs and enhances system robustness. It effectively prevents collisions and deadlocks, significantly boosting warehouse operational efficiency.Originality/valueThis study introduces the CA*-MLACFP framework: a conflict-free, multistage path planning method for multiload AGVs that integrates improved A* search with a cellular automaton and a layered conflict resolution strategy. The approach offers high practical applicability and scalability.

In the increasingly fierce market competition, the demand for mass personalized production in the printed circuit board (PCB) manufacturing industry is rapidly increasing. To address these needs, enterprises integrate and pack age workshop manufacturing resources into manufacturing services, and carry out a series of operations, such as matching, optimization, combination, and scheduling, to complete related tasks. However, with the exponential growth of personalized customer orders, manufacturing services must undergo frequent restructuring to generate new solutions to complete the tasks. A lack of effective collaboration between services can lead to poor task-service alignment, ultimately compromising PCB quality and failing to meet customer expectations. Therefore, this article proposes a reliable collaborative optimization method for manufacturing services based on a cascade effect integrated graph convolutional network (CGCN), which considers the dynamic changes of PCB workshop resources and solves the problem of manufacturing service collaboration. Firstly, a manufacturing service col laboration network (MSC-Net) was constructed to characterize the physical properties of workshop production. Then, the graph data on MSC-Net are used to quantify the dynamic changes in service collaboration. Finally, the global embedding features of MSC-Net are processed through CGCN, and regression prediction is performed to optimize the service collaboration. The experimental results show that, taking the production process of PCB as an example, the performance of CGCN is superior to other typical algorithms.

As manufacturing transitions towards the Industry 5.0 paradigm, simultaneously enhancing production efficiency while ensuring environmental sustainability and system resilience presents a critical managerial decision-making challenge in discrete manufacturing. This paper addresses the trade-off between production makespan and resource-environmental impact in hydraulic cylinder manufacturing systems, particularly where uncertain processing times significantly undermine the resilience of production operations. To align with the Industry 5.0 objectives of sustainable and robust manufacturing, we propose a novel green fuzzy scheduling framework grounded in a real-world enterprise context. First, processing time uncertainty is modeled using interval type-2 fuzzy sets to enhance the robustness of production planning. A multi-objective optimization model is then developed to simultaneously minimize the makespan and the integrated resource environmental impact factor, thereby harmonizing economic viability with ecological sustainability. To solve this NP-hard problem, a Nonlinear Grey Wolf Optimizer incorporating Simulated Annealing is constructed. This approach integrates global search capabilities with local refinement strategies to effectively navigate the complex decision space. Empirical validation is conducted through a case study at a hydraulic cylinder enterprise. The results demonstrate that the proposed method outperforms existing practical scheduling solutions, reducing the makespan by 4.74% and the integrated resource environmental impact factor by 10.60%. These findings provide actionable managerial insights for fostering resilient, green, and efficient manufacturing systems in the Industry 5.0 era.

To improve the efficiency of mobile robots carrying shelves for material handling in manufacturing enterprises, this study considers multiple material requirements of orders in a coordinated manner and investigates the transfer of multi-load shelves between storage areas and picking workstations. Three types of resources, shelves, automated guided vehicles (AGVs), and storage locations, are jointly considered, and the task allocation and path planning problem for AGVs is studied within an integrated framework. With the objective of minimizing the overall order picking completion time, a mixed-integer programming model is established to simultaneously represent decisions including resource allocation, path selection, and task sequencing, enabling the optimization of multiple interdependent decisions within a unified scheduling framework. To solve the proposed model, an improved genetic algorithm is developed. The algorithm adopts a chromosome encoding scheme based on the "pick-sort-deliver" shelf-handling task unit and constructs multi-granularity crossover operators at the order level, AGV level, and task level to effectively address the multiresource and multi-level decision structure of the problem. In addition, five neighborhood search operators are designed to form a hierarchical neighborhood search mechanism, further enhancing local solution quality. Experimental results show that the hybrid genetic algorithm significantly outperforms the comparative methods in maximum order completion time, with average improvements of 19.06% and 16.38% over the greedy algorithm and greedy-local search algorithm, respectively. The algorithm also exhibits satisfactory stability and robustness across various problem scales. (c) 2026 by the authors; licensee Growing Science, Canada

This paper studies the shop scheduling problem of elevator buffer components in a spinning-area production line. A bi-objective scheduling model is established to minimize makespan and improve equipment utilization under process precedence and setup-time constraints. Based on actual production data, a genetic algorithm with constraint-preserving encoding is applied to solve the model. Simulation results show that the average equipment utilization reaches 97.03%, and the overall completion time is reduced compared with the original scheduling strategy.

Abstract Addressing the coupling of production scheduling and mold maintenance in precision forging, this paper formulates a bi-objective single-stage parallel machine scheduling model to minimize makespan and total operating cost. The joint model integrates maintenance thresholds, opportunistic maintenance, and group maintenance. To ensure fair scheme evaluations, a discrete event-driven dynamic evaluator incorporating the Common Random Number (CRN) mechanism is designed. To solve the problem, a hybrid memetic algorithm (MDNS-MA-NSGA-II) featuring mixed initialization and multi-dimensional variable neighborhood search is proposed. Experimental results demonstrate that MDNS-MA-NSGA-II outperforms standard NSGA-II in Pareto set quality, convergence, and distribution. Compared with traditional corrective maintenance and sequential decision-making strategies, it achieves a superior makespan-cost balance, providing valuable decision support for efficient precision forging operations.

This study proposes an end-to-end deep reinforcement learning framework to solve workshop scheduling problems. First, an improved heterogeneous graph attention network is introduced to explicitly model topological relationships between heterogeneous processes and machines. Then, a type-aware projection matrix is employed to handle diverse edge semantics and fuse global-local features, aligning local operational constraints with global optimization objectives. Finally, the agent is trained using the Proximal Policy Optimization algorithm. Comparative evaluations against priority scheduling rules and other deep reinforcement learning algorithms on common benchmark datasets demonstrate superior performance in minimizing completion time and achieving zero-shot scalability.

Distributed energy-efficient hybrid flow shop scheduling problem (DEHFSP) with batch processing machines (BPMs) is rarely considered, let alone DEHFSP with BPMs and uncertainty. In this study, a fuzzy DEHFSP with BPMs at a middle stage and no precedence between some stages is presented, and a dynamic artificial bee colony (DABC) is proposed to simultaneously optimize the total agreement index, fuzzy makespan, and fuzzy total energy consumption. To produce high quality solutions, Metropolis criterion is used, dynamic employed bee phase based on neighborhood structure dynamic selection is implemented, and group-based onlooker bee phase with bidirectional communication is given. Migration operator is also adopted to replace scout bee phase. Extensive experiments are conducted, and the optimal combination of key parameters for DABC is decided by the Taguchi method. Comparative results and statistical analysis show that new strategies of DABC are effective, and DABC is highly competitive in solving the considered fuzzy DEHFSP.

With the advancement of economic globalization, the distributed heterogeneous factory environment has become the mainstream in manufacturing enterprises. Scheduling flexible job shops in such a production environment holds practical value. However, due to the high complexity of certain jobs, the transfer of jobs between different factories are often required in practical production to balance machine load rates. Accordingly, this study addresses the distributed heterogeneous assembly flexible job shop scheduling problem with transfers, aiming to minimize both the makespan and total energy consumption. First, a multi-objective optimization model is formulated to define the problem, wherein knowledge of factory assignment and processing sequence for operations is summarized. Subsequently, given the complexity of this problem, a Q-learning-based improved multi-objective genetic algorithm (QL-IMOGA) is proposed as an effective approach. Within the proposed algorithm, a hybrid population initialization method is designed, considering factory load balancing and the earliest product completion time, to generate a high-quality initial population. Furthermore, two types of crossover operators, four types of mutation operators, and six objective-oriented neighborhood search operators are devised to enhance the algorithm's exploration and exploitation capabilities. Q-learning is employed for adaptive adjustment of key parameters to improve both convergence speed and solution quality. The effectiveness of the proposed population initialization method and neighborhood search operators is validated through 15 test cases. The results demonstrate that the proposed algorithm significantly outperformed four advanced meta-heuristic algorithms. Furthermore, it is observed that the solution employing the job transfer strategy led to an average reduction of 7.5% in makespan, a 3.9% decrease in total energy consumption, and an 8.4 % improvement in factory load rates compared to the solution using the job no-transfer strategy.

In modern intelligent manufacturing workshops, researchers increasingly integrate the transportation of Automated Guided Vehicles (AGVs) with production scheduling to enhance overall efficiency. However, in real-world production scenarios, such integrated scheduling systems are highly susceptible to stochastic disturbances stemming from unexpected equipment failures, thereby significantly undermining operational efficiency. This study focuses on the dynamic lot-streaming hybrid flowshop scheduling problem with automated guided vehicles (DLSHFSP-AGV) under a disruption-prone environment. A multi-objective mixed-integer linear programming model that accounts for machine and AGV failures is developed. Based on this model, an event-driven partial rescheduling strategy is proposed, in which the disrupted operations and delivery tasks are classified into three categories: retained, continued, and reconstructed. On the framework of NSGA2-MDDQN (NSGA-II-Multi-objective double-depth Q learning algorithm) algorithm, which is the basis of existing research, the dynamic encoding mechanism and multi-stage decoding strategy are innovatively introduced to realize the collaborative optimisation of the machine allocation, AGV scheduling, and process sequencing of the remaining tasks after the perturbation. Experimental results demonstrate that, compared to combined scheduling rules, NSGA-II, and DDQN algorithms, the proposed method achieves improvements of 18.59%, 41.05%, and 4.26% in makespan, machine idle time, and AGV travel distance, respectively. These enhancements significantly improve the robustness and optimization performance of the scheduling scheme under dynamic perturbations, offering a reliable dynamic scheduling solution for intelligent manufacturing systems. (c) 2026 by the authors; licensee Growing Science, Canada

The printed circuit board (PCB) industry is currently facing the challenge of mass customization demands, which places an urgent need for efficient scheduling in PCB production. Due to the production process’s complexity and the environment’s variability, traditional scheduling algorithms often fail to achieve optimal performance in practical applications. This paper establishes a dynamic multi-objective flexible PCB shop scheduling model to address the challenges above. The model uses total tardiness, maximum completion time, and average machine utilization as optimization objectives. Moreover, a rule-embedded deep Q-network (R-DMDQN) algorithm is developed to address the complex dynamic characteristics of the PCB production process. The algorithm integrates characteristics of PCB production, extracting seven selected features to describe the system state. Simultaneously, it embeds six composite scheduling rules developed and guided by specialized knowledge to enhance the interpretability of learned strategies, and to augment the adaptability and flexibility of the algorithm. Through extensive experimental verification, the results show that the R-DMDQN model proposed in this study has significant superiority and stability in improving scheduling performance compared to the existing well-known scheduling rules and the NSGA-II algorithm. The research provides an innovative approach to the automation and optimization of scheduling in the PCB industry. It is expected to promote the application of related technologies in other complex production systems.

Food pickling is an essential technique for maintaining flavour and texture. Due to the concentrated harvest period of raw materials, the demand for pit resources increases during peak seasons. Meanwhile, significant volume shrinkage of raw materials during the pickling leads to pit underutilisation, creating a bottleneck in production capacity expansion. To improve production capacity, it is crucial to address the challenges posed by raw material shrinkage. This study investigates the hybrid flow shop lot streaming scheduling problem with resource constraints (HFSLSP-RC) and proposes a Q-learning-based fruit fly optimisation algorithm (QFOA) to minimise the makespan. First, four distinct initialisation strategies are employed to enhance the quality of the initial population. Next, the Q-learning is integrated with the smell search stage to improve the local search capability of the algorithm. Additionally, a population merging strategy is introduced in the visual search stage to enhance the global search performance. Finally, the effectiveness of QFOA is validated through comparative experiments, ablation studies, and an industrial case study. The case study, which involved processing eight types of food across 90 pits, demonstrates that QFOA reduced the makespan by 10.0% compared to FOA and by 7.3% compared to IGWO.