System Control and Diagnosis in Data-Rich Environment

由于计算机和网络技术的快速发展，如今的过程和控制系统通过来自多个来源和位置的众多传感器的丰富测量数据得到增强。在大规模制造和化工过程中，分布式控制系统（DCS）的应用创造了这样一个数据丰富的环境。实时测量以及来自不同位置和来源的历史数据可用于处理。同时，由于缺乏从海量数据中挖掘相关有用信息的有效工具，行业正面临着所谓的“数据丰富，信息贫乏”的现象。****在一些情况下，数据对决策至关重要。首先，系统和过程大多是在良好控制的约束条件下运行的，这些约束条件在初始设计阶段就被很好地理解了。对于许多系统和过程，第一性原理模型和过程知识是可用的，可用于在正常运行条件下的控制和性能分析/评价。然而，当意外情况出现时，例如与故障相关或其他时变操作，系统运行偏离其最初设计的条件，并且这些模型的保真度受到影响。在这种情况下，数据对于系统性能评估、故障诊断、控制律更新甚至基于状态监测的维护都变得非常重要。如何从多个数据源中提取有用的最新信息，并将其与物理模型的知识“融合”，已经引起了工业界和研究界的关注。此外，在许多实际系统中，传统的时间序列测量以及替代形式的测量，例如分布函数和光谱数据，都是可用的。它们都携带着系统的重要信息。如何从所有可用的不同形式的数据中提取有用的信息用于控制和系统故障诊断是一个新的挑战。当系统和流程非常复杂，并且物理模型甚至不有效时，数据是必不可少的另一种情况。在这种情况下，数据驱动的建模和重要的过程分析几乎成为唯一的手段。****在本研究计划中，我们将重点关注传统系统控制和故障诊断方法在新的数据丰富范式中面临的一些新挑战。长期目标旨在开发有效的设计和分析工具，用于控制和诊断能够处理不同形式和/或来自多个来源（及时和空间）的数据的系统/过程。这些工具对于开发新一代信息系统和新工业标准（如工业物联网（IIoT））的控制/仪表设备至关重要。* * * * *

Owing to the fast development of computer and network technology, nowadays processes and control systems are enhanced by rich measurement data from numerous sensors of multiple sources and locations. In large-scale manufacturing and chemical processes, application of distributed control systems (DCS) has created such a data-rich environment. Real-time measurements as well as historical data from different locations and sources are available for processing. Meanwhile, due to the lack of effective tools digging in the massive amount of data for relevant and useful information, industries are facing the so-called “data rich but information poor” syndrome.****There are several situations where data becomes essential and critical in decision-making. First of all, systems and processes are mostly operated under well-controlled constraints that are well understood in the initial design stage. For many systems and processes, first-principle models and process knowledge are available and can be used for control and performance analysis/evaluation under the normal operation conditions. However, when unexpected conditions emerge, e.g. fault related or other time-varying operations, system operations deviate from its initially designed conditions, and fidelity of those models suffers. In this case, data becomes important for system performance evaluation, fault diagnosis, updating of the control law and even condition monitoring based maintenance. How to extract useful up-to-date information from multiple sources of data and ‘fuse' it with knowledge from the physical model has raised attention from industries and research communities. Furthermore, in many practical systems, traditional time-series measurement, as well as measurement of alternate forms, e.g. distribution functions, and spectroscopic data, are available. They all carry vital information about the system. How to extract useful information from all available data of different form for control and system fault diagnosis is a new challenge. Another situation where data is essential arises when the system and process is highly complex and the physical model is not even valid. In this case, data-driven modeling and important process analytics become almost the only means.****In this research program, we will focus on some of the new challenges facing conventional system control and fault diagnosis methodologies in the new data-rich paradigm. The long-term objective is aimed at developing effective design and analysis tools for control and diagnosis of systems/processes that are capable of processing data of different forms and/or from multiple sources (timely and spatially). These tools are essential for the development of the new generation information system and control/instrumentation devices for the new industry standard, such as Industrial Internet of Things (IIoT). *** **