Automatic Modeling using Chemical Plant Data

点击翻译按钮获取中文摘要

Abstract - Ungar & Seider - 9632992 Process design engineers spend much of their time developing and refining process models. Such models are typically built using graphical editors to assemble model fragments (for individual units or processes from code and equation libraries) into simulators. In addition, extensive historical plant data is also available. Process models often fail to accurately predict this plant data, due to inaccurate parameters in the model, incorrect model equations and structure, or faults, disturbances, or inaccurate measurements. The PIOs plan to address two related questions: (a) how to use available data effectively to improve process models, and (b) how to use data and models to detect and diagnose abnormalities in plant operation. The goal is to build intelligent simulators which will aid in understanding and improving models based on measured data. This project is to develop methods for automating the process of preparing simulations, using chemical process simulators like ASPEN PLUS and SPEEDUP, and interpreting their results. Its long-range goal is to develop techniques for the automatic generation, modification, and interpretation of dynamic simulations and to apply them to data reconciliation, fault diagnosis, plant and controller design, and operator training. In this work, the PIOs plan to concentrate on an initial stage of the project: the development of a designerOs aid that can prepare a simulation model, compare the results of a simulation with plant data and guide the process of improving the model. This involves the development of methodologies for (1) the detection of changes and outliers, (2) the representation of information describing the model fragments, (3) the initialization of dynamic models, (4) the selection of model fragments, and (5) the generation of explanations. Historical data, from operating plants, as well as reference models, will be obtained from the CIM/21 system.