Modeling, identification and control of a cold flow circulating fluidized bed

Modeling, Identification and Control of a Cold Flow Circulating Fluidized Bed Rupendranath Panday Circulating fluidized bed (CFB) is used extensively in petrochemical industries especially for fluid catalytic cracking, coal combustion or gasification and various other chemical processes. Modeling will help identify the sensitivity of the performance of a CFB to variation in different operating conditions and design parameters. Mathematical models also have a more practical purpose – the development of engineering and design tools which will help calculation, and design of real plants. From the practical engineering point of view, a compact and accurate description of the dynamic behavior of the system under consideration is needed and mathematical models derived based on this requirement can be used for calculations, and design and operation of real systems. New 2-region measurement model describing the bed height in the standpipe of a cold model present at National Energy Technology Laboratory (NETL), US Department of Energy, Morgantown, WV, is formulated using the total pressure drop across the standpipe, pressure drop across the dense region and the static pressure contribution due to solids in the lean region of the standpipe. On the other hand, modeling a system of interest directly from observed input-output data is referred to as System Identification. Identification of a NETL cold flow circulating fluidized bed (CFCFB) is carried out using a multiple model approach. Under this technique, the CFCFB is considered as a nonlinear device. From the system theoretical point of view, any nonlinear model can be decomposed into multiple sub models to cover certain operating range of a given system and these multi-models can then be combined together through some weighting functions to encompass its wide operating range. To achieve this wide operating range model validity, a white noise experiment was conducted in the CFCFB using glass beads bed material with the objective in mind that the model trained on random data sets would be sufficient enough to be utilized in other simple operating conditions. In this work, these data are used to identify CFCFB’s multiple sub models and to combine them into a single nonlinear model such that solids circulation rate can be estimated from the move air flow and riser aeration fed to the device, and the total pressure drop developed across the riser at extremely different experimental conditions. However, when a cold flow circulating fluidized bed is believed to operate linearly in a single operating regime, it is reasonable to approximate the given system by a linear model in order to predict solids flow rate. In reality, any measurable variables may be corrupted by noise and it is sensible to arrive at results that back up the initial assumption regarding the basic relationship between the variable of interest and the independent variable that is presumed during the model development phase, which would otherwise be difficult due to the use of noisy measurements. Similarly, it is also helpful to analyze the stochastic processes that corrupt the measurements, using data from one particular experiment such that future inference could be made on the errors acting on those measurements if the same machine is operated under similar condition and/or in any other industrial plants that exhibit dynamics similar to that of the laboratory-based equipment like the NETL CFCFB at that operating condition. The present work begins with a complete black box model of a state-space description arising from the system identification and converts it into a model without any fictitious variable such that the interaction among the variables under consideration can be analyzed. Furthermore, this concept separates a state into stochastic and deterministic components which gives the nature of noise acting on the measurement device and rationalizes if there exists a certain relationship between independent and dependent variable. In this thesis, the state is a solids circulation rate. Independent parameters that comprise of aerations flow rates including move air flow, riser aeration and loop seal fluidization air are used to obtain deterministic component of a measured solids circulation rate. On the other hand, easily measurable dependent variables like the pressure drops across various sections of the machine are used to predict its stochastic counterpart. A real time pressure drop model based on the Recursive Prediction Error Method (RPEM) is built to predict the split of move air flow between the standpipe and L-valve. The split estimate is of paramount importance while simulating the phenomenological model of the standpipe or in other applications, if required. Additional aeration fed across the various sections of standpipe act as the fluidization bias and their routes determination within the component may help to maintain their required level to assist in solids movement during operation while minimizing excessive flows. The path determination is also predicted using RPEM on a discrete time pressure drop model such that the user can operate them at the desired intensity according to their operating requirements. Generally, a PID controller is not “portable”, i.e., a controller designed for one plant is usually not applicable to another plant. To resolve this long-standing issue of portable controllable design, the controller scaling method can be used to control similar plants that are different only in gain and frequency scales, thus avoiding tedious control redesign. However, there are always differences in design specifications and constraints for different control problems. For the sake of simplicity and optimized controller design for each application such that an engineer get the most performance out of a given set of hardware and software, bandwidth parameter is selected as the measure of performance. To avoid the repetitive tuning of this scaling parameter based on a controller performance, adaptive algorithm is designed. Furthermore, to predict the characteristics behavior of a complex system lacking a reliable mathematical model, a recursive least squares estimation (RLS) algorithm is utilized to find autoregressive moving average (ARMA) models for single-input single-output or multi-input singleoutput case. The adaptive PID control algorithm is then tested on the benchmark NETL CFCFB plant by controlling solids circulation rate according to the reference solids flow rate obtained from the Knowlton’s correlation utilizing average voidage in a moving bed condition and the move air flow. The optimal control of solids circulation rate affecting the heat and mass transfer characteristics which in turn impacts the efficiency of various chemical processes is necessary in CFB units. An example might be the catalytic systems that recirculate catalyst in a reaction/recirculation cycle. In the case of such units in which the addition of catalyst is small and need not be steady, the main solids flow-control problem is to maintain balanced inventories of catalyst in and controlled flow from and to the reactor and regenerator. This flow of solids from an oxidizing atmosphere to a reducing one, or vice versa, usually necessitates stripping gases from the interstices of the solids as well as gases absorbed by the particles. Steam is usually used for this purpose. The point of removal of the solids from the fluidized bed is usually under a lower pressure than the point of feed introduction into the carrier gas. The pressure is higher at the bottom of the solids draw-off pipe due to the relative flow of gas counter to the solids flow. The gas may either be flowing downward more slowly than the solids or upward. The standpipe may be fluidized, or the solids may be in moving packed bed flow with no expansion. Gas is introduced at the bottom (best for group B) or at about 3-m intervals along the standpipe (best for group A). The increasing pressure causes gas inside and between the particles to be compressed. Unless aeration gas is added, the solids could defluidize and become a moving fixed bed with a lower pressure head than that of fluidized solids. Thus, this observation leads to the fact that the gas velocity in the standpipe might be the main parameter to control the solids circulation rate. Acknowledgements This thesis is the result of my research experiences during the period of 2004 – 2008 at the West Virginia University, the National Energy Technology Laboratory and my home at Nepal. Many people have contributed directly or indirectly to this thesis whom I would like to express my sincere gratitude. Although unfortunately I am able to mention just a few, firstly I would like to thank my promoter Dr. P. Famouri for providing a valuable research project during my stay at WVU during 2004 – 2007. I am grateful to Dr. R. Turton for his critical remarks regarding one of my conference papers. I want to express my deepest gratitude to my research advisor, Dr. B. D. Woerner, for his guidance, encouragement and support that he provided me at the time I wanted most. Dr. M. Choudhry has encouraged me all the time; the support he provided me at the last moment at WVU is highly appreciable. I am greatly indebted to him and I will always be. I would like to thank Dr. L. J. Shadle, Dr. J. C. Ludlow and the NETL research team for conducting two test series on my request required for my thesis preparation. Dr. Shadle’s continuous thirst for new ideas keeps amazing me. Dr. J. C. Ludlow taught me how to perform research and undoubtedly, showed me the path where an engineer should in reality be able to put his (mathematical) knowledge in practice. Their excellent guidance and critical remarks to the presentation of one of my works at NETL are unforgettable. Thanks to both of you for opening the door to an incredible educational journey which I have always been looking for! I am greatly indebted to Dr. E. J. Boy