Yuan Tangxiao, Adjallah Kondo Hloindo, Sava Alexandre, Wang Huifen, Liu Linyan
School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China.
LCOMS, University of Lorraine, 57078 Metz, France.
Sensors (Basel). 2024 Oct 12;24(20):6586. doi: 10.3390/s24206586.
The ability and rapid access to execution data and information in manufacturing workshops have been greatly improved with the wide spread of the Internet of Things and artificial intelligence technologies, enabling real-time unmanned integrated control of facilities and production. However, the widespread issue of data quality in the field raises concerns among users about the robustness of automatic decision-making models before their application. This paper addresses three main challenges relative to field data quality issues during automated real-time decision-making: parameter identification under measurement uncertainty, sensor accuracy selection, and sensor fault-tolerant control. To address these problems, this paper proposes a risk assessment framework in the case of continuous production workshops. The framework aims to determine a method for systematically assessing data quality issues in specific scenarios. It specifies the preparation requirements, as well as assumptions such as the preparation of datasets on typical working conditions, and the risk assessment model. Within the framework, the data quality issues in real-time decision-making are transformed into data deviation problems. By employing the Monte Carlo simulation method to measure the impact of these issues on the decision risk, a direct link between sensor quality and risks is established. This framework defines specific steps to address the three challenges. A case study in the steel industry confirms the effectiveness of the framework. This proposed method offers a new approach to assessing safety and reducing the risk of real-time unmanned automatic decision-making in industrial settings.
随着物联网和人工智能技术的广泛传播，制造车间获取执行数据和信息的能力及速度得到了极大提升，实现了设施和生产的实时无人集成控制。然而，该领域普遍存在的数据质量问题引发了用户对自动决策模型应用前稳健性的担忧。本文探讨了自动化实时决策过程中与现场数据质量问题相关的三个主要挑战：测量不确定性下的参数识别、传感器精度选择以及传感器容错控制。为解决这些问题，本文提出了一种适用于连续生产车间的风险评估框架。该框架旨在确定一种系统评估特定场景下数据质量问题的方法。它规定了准备要求，以及诸如典型工作条件下数据集的准备、风险评估模型等假设。在该框架内，实时决策中的数据质量问题被转化为数据偏差问题。通过采用蒙特卡罗模拟方法来衡量这些问题对决策风险的影响，建立了传感器质量与风险之间的直接联系。此框架定义了应对这三个挑战的具体步骤。钢铁行业的案例研究证实了该框架的有效性。所提出的方法为评估工业环境中实时无人自动决策的安全性和降低风险提供了一种新途径。
