Fault Diagnosis for Chemical Process Plants

该研究项目的目标是开发能够检测和解释化工厂运行中的故障的计算机算法，供工厂人员在工厂颠覆期间做出运营决策时使用。将开发专家系统，允许将工艺工程师和操作员积累的经验、判断和启发式知识纳入自动推理系统。专家系统将通过过程控制计算机访问工厂测量数据，从而实时应用这些知识。该系统将以模块化形式设计，以便可以分解复杂的工厂，单独开发单个单元的计算机逻辑(并在随后的迭代中重新组合)。本项目将解决的具体问题包括：(1)考虑噪声和时变数据等现实因素的干扰定位和识别的新算法的理论发展；(2)基于不确定推理技术的诊断理论和方法的发展，以取代目前基于布尔逻辑的理论；(3)研究用于集成启发式和基于模型的推理的专家系统体系结构；(4)基础研究，以制定最佳过程仪表指南，以提高新诊断系统的有效性；(5)定性故障模拟的理论研究，以对故障对工厂运行的影响提供准确的定性预测；(6)在更复杂和现实的模拟中实施理论思想，测试方法，并就如何改进和扩展当前的诊断系统提供进一步的见解。

The goal of this research project is to develop computer algorithms that can detect and interpret faults in operating chemical plants, to be used by plant personnel making operating decisions during plant upsets. Expert systems will be developed which allow the accumulated experience, judgment, and heuristics of process engineers and operators to be incorporated into an automated reasoning system. The expert system would apply this knowledge in real time by accessing plant measurements through the process control computer. This system will be designed in modular form so that complex plants can be decomposed and computer logic for individual units be developed separately (and recombined in subsequent iterations). The specific problems that will be addressed in this project include: (1) theoretical development of new algorithms for disturbance location and discrimination which account for realistic factors such as noisy and time-varying data; (2) the development of a theory and methodology of diagnosis based on uncertain reasoning techniques, to replace the current theory based on Boolean logic; (3) research on expert system architectures for integration of heuristic and model-based inferences; (4) basic research leading to development of guidelines for optimal process instrumentation to increase the effectiveness of the new diagnostic systems; (5) a theoretical study of qualitative fault simulation to provide accurate qualitative predictions of the effects of malfunctions on plant operations; and (6) the implementation of theoretical ideas in more complex and realistic simulations, to test methodology, and to give further insight on how to improve and extend the current diagnostic system.