Scholarlink

<p>In the era of Industry 5.0, manufacturing systems are increasingly required to integrate human-centric principles with real-time adaptability and responsiveness. Mixed-model assembly lines designed to accommodate multiple product variants play a pivotal role in this context, yet must overcome the challenges posed by frequent product changes, diverse human capabilities, and fluctuating operational conditions. Ensuring balanced and efficient operations under these constraints demands a dynamic approach to plan production and system configuration, one that incorporates the variability inherent in human work while maintaining steady throughput. This study proposes a dynamic assembly line rebalancing model that integrates deep reinforcement learning (DRL) with a high-fidelity discrete event simulation (DES). By framing the rebalancing process as a sequential decision problem, the DRL agent continuously adjusts task allocations across manual and automated stations, accommodating stochastic factors such as learning-forgetting effects and shifting production demands. The DES model developed based on a learning factory representing a reconfigurable assembly system ensures realistic modeling of both material flow and human-system interaction, capturing unpaced assembly processes. Comparative experiments demonstrate that the DRL-based strategy outperforms conventional heuristics in productivity, cycle-time reduction, and workload balance without retraining in varied operational scenarios. These findings highlight the framework’s scalability and generalization capacity, as well as its potential to optimize resource utilization and reduce work-in-progress levels. By merging DRL with advanced simulation, this approach enables self-adaptive scheduling that aligns with Industry 5.0 principles. Moreover, it offers a viable pathway toward resilient, human-focused assembly systems capable of real-time, data-driven responsiveness to complex manufacturing dynamics.&#xa0;</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Smart adaptable assembly line rebalancing based on reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年5月27日

查看全文
  • Mohammadreza Nikkerdar,
  • Waguih ElMaraghy,
  • Hoda ElMaraghy

摘要

In the era of Industry 5.0, manufacturing systems are increasingly required to integrate human-centric principles with real-time adaptability and responsiveness. Mixed-model assembly lines designed to accommodate multiple product variants play a pivotal role in this context, yet must overcome the challenges posed by frequent product changes, diverse human capabilities, and fluctuating operational conditions. Ensuring balanced and efficient operations under these constraints demands a dynamic approach to plan production and system configuration, one that incorporates the variability inherent in human work while maintaining steady throughput. This study proposes a dynamic assembly line rebalancing model that integrates deep reinforcement learning (DRL) with a high-fidelity discrete event simulation (DES). By framing the rebalancing process as a sequential decision problem, the DRL agent continuously adjusts task allocations across manual and automated stations, accommodating stochastic factors such as learning-forgetting effects and shifting production demands. The DES model developed based on a learning factory representing a reconfigurable assembly system ensures realistic modeling of both material flow and human-system interaction, capturing unpaced assembly processes. Comparative experiments demonstrate that the DRL-based strategy outperforms conventional heuristics in productivity, cycle-time reduction, and workload balance without retraining in varied operational scenarios. These findings highlight the framework’s scalability and generalization capacity, as well as its potential to optimize resource utilization and reduce work-in-progress levels. By merging DRL with advanced simulation, this approach enables self-adaptive scheduling that aligns with Industry 5.0 principles. Moreover, it offers a viable pathway toward resilient, human-focused assembly systems capable of real-time, data-driven responsiveness to complex manufacturing dynamics.