Detection and diagnosis of control loop nonlinearities, valve stiction and data compression.

摘要

The field of controller performance monitoring has received much attention in the engineering research literature. However, the diagnosis of poor control performance remains an open area. Performance diagnosis requires identification of the cause(s) of poor control performance. Poor controller tuning, oscillatory external disturbances, process nonlinearities and valve nonlinearities are the primary causes of poor control performance.; Based on higher order statistical (HOS) theory, two new indices---the Non-Gaussianity Index (NGI) and the Non-Linearity Index (NLI)---have been developed to detect and quantify signal non-Gaussianity and nonlinearity. These indices, together with specific patterns in the mapping of process output (pv) and controller output (op), can be used to diagnose the causes of poor control loop performance.; Stiction is the most common problem in spring-diaphragm type valves. A generalized definition of valve stiction based on the investigation of real plant data is proposed in this thesis. A simple two parameter data-driven model of valve stiction is developed. The model is simple, yet powerful enough to properly simulate the complex valve stiction phenomena. Both open and closed loop results have been presented and validated to show the capability of the model.; Conventional invasive methods such as the valve travel test can detect stiction easily. However, they are expensive, time consuming, and tedious to use for examining thousands of valves in a typical process industry. A noninvasive method that can simultaneously detect and quantify control valve suction is presented. The method requires only routine operating process data. Over a dozen industrial case studies have demonstrated the wide applicability and practicality of this method as a useful diagnostic aid in troubleshooting poor control performance.; In chemical industrial practice, data are often compressed, for archival purposes, using various techniques. Compression degrades data quality and induces nonlinearity in the data. The issues of data quality degradation and nonlinearity induction due to compression are investigated in this thesis. An automatic method for detection and quantification of the compression present in the archived data has been presented. Compelling and quantitative analyses have been presented to end the practice of process data compression.