Enhancement of the Overall Equipment Effectiveness of Factories – Development of Resilient Agent-Based Automation Systems for Machine and Plant Manufacturing Industry

全球市场竞争激烈，生产系统的有效运行（通常使用整体设备效率（OEE）进行衡量）对盈利能力至关重要。Resi4MPM将开发一种分散的基于代理的方法，以提高生产系统的弹性，从而提高工厂的OEE。这将通过将智能现场级设备与我们在NCKU的台湾研究合作伙伴的基于云的数据分析方法相结合来实现，其重点是智能预测维护系统。Resi4MPM将研究三种相互交织的方法来积极改善OEE。首先，它将提出行动空间方法，该方法限制机器或组件的自主行为，以确保确定性决策。该方法使得能够自动更换现场级设备，例如，基于设备的健康状态和评估的软设备质量，通过软传感器检测传感器。考虑无量纲，并进行累加，精确表示上级构件的安全动作空间。我们将特别处理关键的研究问题，即为实地一级和有效的决策适当地提供信息，这是恢复的先决条件。其次，在机器层面，Resi4MPM旨在支持（半）自动重启，以最大限度地减少工厂停机时间。为此，将研究创建状态机的方法，包括从可用工程数据转换的条件，并研究利用这些状态机的回溯方法。如果机器由于故障而停止，不允许恢复，这种方法允许机器识别如何达到安全状态，从安全状态可以恢复正常操作。第三，将开发一种新的异常分析方法，该方法将云中部署的好处（可用计算能力）与现场部署的好处（来自驱动器的高频数据、预处理、预过滤）结合起来。将研究算法，以直接在现场识别异常并过滤错误警报，以提高计算效率，并提供快速反应以启动恢复的能力（第一种方法）。此外，在云环境中确定洪水警报的根本原因，更新并发送到现场级别作为资源有效的模型，以增强和发展工厂及其智能。由于所有这些方法都依赖于适当的知识表示，Resi4MPM将研究从工程文档中提取信息和机器学习方法，从操作数据中学习，以生成和更新知识库。通过这种方式，提高弹性的方法可以实现工业应用，减少人工劳动。因此，将与积极加入Resi4MPM的机器和工厂制造业的相关行业合作伙伴一起完善要解决的用例。

Global markets are extremely competitive, making an effective operation of production systems, oftentimes measured using the Overall Equipment Effectiveness (OEE), crucial for profitability. Resi4MPM will develop a decentralized agent-based method to improve the resilience of production systems, thus increasing a factory’s OEE. This will be achieved by combining intelligent field-level devices with cloud-based data analysis methods from our Taiwanese research partners at NCKU, whose focus is the Intelligent Predictive Maintenance system.Resi4MPM will research three intertwined approaches to actively improve OEE. First, it will advance the action space method, which restricts the autonomous behavior of a machine or a component to ensure deterministic decision-making. This method enables the automatic replacement of field-level devices, e.g., sensors via soft sensors, based on the device's health status and the assessed soft device's quality. Uncertainties will be considered and accumulated to represent the safe action space of superordinate components precisely. We will especially address the key research issues of appropriately representing information for the field-level and efficient decision-making as a prerequisite for recovery. Secondly, on the machine level, Resi4MPM aims to support (semi-)automatic restart to minimize factory downtime. For this, methods to create state machines including conditions of transitions from available engineering data will be investigated and a retracing method will be researched that leverages these state machines. In case a machine stops due to a fault that does not allow recovery, this method allows the machine to identify how it can reach a safe state, from which normal operation can be resumed. Thirdly, a novel anomaly analysis method will be developed, which combines the benefits of deployment in the cloud (available computational power) with those of deployment at the field-level (high-frequency data from drives, pre-processing, pre-filtering). Algorithms will be investigated to identify anomalies and filter false alarms directly in the field for high calculation efficiency, also providing the ability to react quickly to initiate recovery (1st approach). Also, root-causes of alarm floods are determined in the cloud environment, updated, and sent to the field-level as resource-efficient models to enhance and evolve the plant and its intelligence.Since all these approaches are dependent on appropriate knowledge representations, Resi4MPM will research information extraction from engineering documents and machine learning methods for learning from operation data to generate and update the knowledge bases. This way, the methods for increasing resilience are enabled for industrial applicability reducing manual efforts.Consequently, the use cases to be addressed will be refined with the associated industry partners from machine and plant manufacturing, who will actively join Resi4MPM.