Context-Sensitive Modeling and Analysis of Cyber-Physical Manufacturing Systems for Anomaly Detection and Diagnosis
Context-Sensitive Modeling and Analysis of Cyber-Physical Manufacturing Systems for Anomaly Detection and Diagnosis
复制标题
DOI：
10.1109/tase.2019.2918562
复制
发表时间：
2020-01
影响因子：
5.6
通讯作者：
Miguel Saez;F. Maturana;K. Barton;D. Tilbury
中科院分区：
文献类型：
--
作者：
Miguel Saez;F. Maturana;K. Barton;D. Tilbury
信息物理制造系统（CPMS）可以通过将控制、网络通信和计算与物理制造过程集成来定义。在这项工作中，我们提出了一个混合模型的CPMS结合传感器数据，上下文信息和专家知识。我们使用了全局操作状态的识别和多模型框架，以提高异常检测和诊断。异常检测是基于上下文敏感的自适应阈值限制。根本原因诊断是基于分类模型和专家知识。所提出的方法是使用物联网（IoT）从计算机数控机床中提取数据来实现的。结果表明，使用上下文敏感的建模策略允许结合联合收割机基于物理和数据驱动的模型进行残差分析，以检测零件，机器或过程中的异常。通过在分类模型中添加上下文信息来识别磨损或破损的工具和错误的材料，从而改进了根本原因的识别。从业人员注意事项-制造设备的异常检测和诊断是一个复杂的问题。一些挑战是复杂的机器动力学和非平稳的操作条件。本文描述了一个框架，用于建模制造设备使用传感器数据，上下文信息和系统知识的组合。所提出的建模框架是用来提高异常检测诊断使用上下文敏感的策略。这项工作旨在通过识别机器、零件或工艺中的问题来支持更有效的维护措施。建模和异常检测策略被用于识别计算机数控机床中的异常，并可以扩展到工厂车间的其他设备。
Cyber-physical manufacturing systems (CPMS) can be defined by the integration of control, network communication, and computing with a physical manufacturing process. In this work, we present a hybrid model of CPMS combining sensor data, context information, and expert knowledge. We used the identification of global operational states and a multimodel framework to improve anomaly detection and diagnosis. The anomaly detection is based on context-sensitive adaptive threshold limits. Root cause diagnosis is based on classification models and expert knowledge. The proposed approach was implemented using the Internet of Things (IoT) to extract data from a computer numerical control machine. Results showed that using a context-sensitive modeling strategy allowed to combine physics-based and data-driven models for residual analysis to detect an anomaly in the part, machine, or process. The identification of root cause was improved by adding context information in classification models to identify worn or broken tools and wrong material. Note to Practitioners—Anomaly detection and diagnosis of manufacturing equipment is a complex problem. Some of the challenges are complex machine dynamics and nonstationary operating conditions. This paper describes a framework for modeling manufacturing equipment using a combination of sensor data, context information, and system knowledge. The proposed modeling framework is used to improve anomaly detection for diagnostics using a context-sensitive strategy. This work aims to support more effective maintenance actions by identifying problems in the machine, part, or process. The modeling and anomaly detection strategy was used to identify anomalies in computer numerical control machines and can be extended to other equipment on the plant floor.