The manufacturing industry is facing the challenge of meeting the growing demand for personalized products, which requires enhanced agility, flexibility, reconfigurability, and sustainability on the shop floor. To tackle these requirements, one possible solution is to foster collective intelligence (CI) by sharing data and knowledge among human operators, machines, and workpieces, thereby improving resource utilization and enabling informed decision-making. However, the implementation of CI in manufacturing systems poses several challenges, including interoperability issues and the complexity of communication and coordination between heterogeneous manufacturing resources. These challenges can be addressed by integrating cyber-physical-social systems (CPSS) and distributed artificial intelligence. Therefore, this paper presents a framework for enabling the emergence of CI in industrial CPSS to achieve collaborative task allocation and defect detection. The framework infuses intelligence into physical manufacturing resources through CPSS configuration and establishes real-time collaborative communication coupled with decentralized decision-making using a multi-agent system. Additionally, an online deep Q-network (DQN) is employed to train a smart scheduler agent for self-organized assignment of manufacturing tasks to machines. The proposed method is implemented in a 3D printing factory testbed. Experimental findings demonstrate the practicality and effectiveness of the proposed method, which has been deployed in the 3D printing factory for a few months and has reduced production costs, defects, and nonconformities in 3D printed parts.

制造业面临着满足个性化产品日益增长的需求的挑战，这需要增强车间的敏捷性、灵活性、可重构性和可持续性。为了满足这些要求，一种可能的解决方案是通过在人类操作员、机器和工件之间共享数据和知识来培养集体智慧（CI），从而提高资源利用率并实现明智的决策。然而，在制造系统中实施 CI 带来了一些挑战，包括互操作性问题以及异构制造资源之间通信和协调的复杂性。这些挑战可以通过集成网络物理社会系统（CPSS）和分布式人工智能来解决。所以，本文提出了一个框架，使工业 CPSS 中 CI 的出现能够实现协作任务分配和缺陷检测。该框架通过 CPSS 配置将智能注入物理制造资源，并使用多代理系统建立实时协作通信和分散决策。此外，还采用在线深度 Q 网络 (DQN) 来训练智能调度代理，以自组织将制造任务分配给机器。所提出的方法在 3D 打印工厂测试台中实施。实验结果证明了该方法的实用性和有效性，该方法已在 3D 打印工厂部署了几个月，并降低了 3D 打印零件的生产成本、缺陷和不合格品。