Markov Parameters Based Detection And Processing Of Model Plant Mismatch

The scale of the process industries is increasing dramatically in recent years, which brings more uncertainties to the system and affects the performance significantly. Control performance assess-ment and diagnosis is capable of evaluating the performance potential, tracing and dealing with the sources, improving the performance and safety. Model plant mismatch, which is the difference between the control model in the design stage and the model of the current process, is a typical kind of uncertainty. It varies with the changes of the industrial processes and is inevitable in the field. Certain but limited amount of mismatch can be taken care of by the feedback scheme, which is also named the robustness of the control loop. However, it is conservative. Once the mismatch goes beyond the ability of the control loop, it may degrade the control performance and endanger the system stability. So it is crucial to study model plant mismatch in control systems.A framework based on the Markov parameters is proposed to detect and process the model plant mismatch. It consists of quantification, detection, isolation, consequence analysis and post disposal of model plant mismatch. In each part, a specific method is developed. The main contri-butions are as follows.(1) A measure based on Markov parameters is designed for the quantification of model plant mismatch. Based on the requirements of control performance assessment and diagnosis, four cri-teria for designing model plant mismatch measure are proposed. Then according to these criteria, several measures from the literature are analyzed. The designed measure has several advantages. First, it is capable of describing both the parametric mismatch and structural mismatch. Secondly, it can be estimated by subspace methods in closed loop. Thirdly, the relationship between this measure and parameters in transfer functions which have physical significance can be established. In addition, the consequences of the mismatch can be estimated based on it. After that, a compara-tive analysis between the Markov based mismatch measure and the traditional mismatch measures is given. It shows that the designed measure has better advantage in the terms of quantifiable abil-ity, closed-loop identifiability, physical interpretability and performance extendability. The work is meaningful to measure selection and lays foundation for studies on mismatch detection and processing.(2) A method based on improved subspace approach is developed for model plant mismatch detection. It utilizes the intermediate results of improved subspace approach to estimate the mis-match characteristics from excited historical data. This is robust to external disturbance due to the scheme borrowed from closed-loop identification. Both the SISO case and MIMO case can be dealt with this methodology. For MIMO case, the mismatched channel can also be isolated because of the decomposability of Markov parameters matrix. This would reduce the work for re-identification and aid the system maintenance.(3) An approach based on Markov parameters is proposed to isolate the mismatched param-eter in transfer functions. The parameters in transfer functions are usually attached with physical significance and are familiar by most engineers. Three signatures are established based on Markov parameters. Then for those SISO processes or sub-channels with the structure of first order plus time delay, the signatures can be used to isolate the mismatched parameters in transfer functions according to the relationship between the Markov parameters and transfer functions. Finally, these mismatches may be traced to possible physical problems in the field.(4) For consequence analysis, a methodology is proposed to quantify the relationship between the mismatch and the variation of control loop behavior. When a certain amount of mismatch happens, the change in control loop behavior including control performance and loop robustness can be calculated based on the loop structure and Laplace Transform. It is found that different parametric mismatch combinations have different patterns of impacts on control loop behavior, which means assessing the significance of mismatch by its magnitude is not practical. Also it is found that the processes with different dynamics have different patterns of mismatch-behavior relationships, which should be concerned during application of the proposed method. In addition, this approach can be extended to deal with setpoint tracking, stochastic disturbance, structural mismatch and controller design and tuning. And according to the relationship between Markov parameters and transfer function, it can be formulated in the Markov parameters based framework of mismatch monitoring.(5) For the post proposal of mismatch, the analysis on popular solutions is given followed by formulating the synthesized framework of model plant mismatch detection and processing. In the framework, the quantification of mismatch is helpful to compare the degree of different mismatch- es; the channel detection is capable of isolating the significantly mismatched channels, which is used to determine the area for model re-identification; the isolation of mismatched parameter in transfer function is useful in relating the mismatch with physical significance and aiding equipment maintenance; the consequence analysis of model plant mismatch is meaningful in determining the urgency of the accident; all these results can provide constructive suggestions for post proposal of mismatch.In the end, the work in the thesis and future work is summarized and discussed.