| Summary: | Thesis (Ph. D.)--Massachusetts Institute of Technology, Dept. of Chemical Engineering, 2010. === Cataloged from PDF version of thesis. === Includes bibliographical references (p. 205-209). === Technology selection is a complex decision problem that is often faced in process engineering. This has been a particularly important problem recently in the energy field, in which many new technologies have been proposed. Typically only point estimates of the chosen metrics are used in the evaluation, with uncertainty often overlooked. However, uncertainty can have a significant effect on the conclusions and decisions to be made. This work investigates the issues surrounding the uncertainty in process engineering models. Model complexity, selection bias and information gain are examined. Existing model selection methods, including Information Criteria Methods and Hypothesis Testing are analyzed, with an emphasis on how they address issues surrounding uncertainty in models. Bayes' methods are investigated in detail because they offer a mathematically sound and very flexible alternative to traditional techniques. A framework is proposed for evaluating the information difference between competing process engineering models involving uncertainties. This framework can be applied when there are competing processes (e.g. a technology selection problem) or when there are competing models for the one process (e.g. several models of the one process with different levels of complexity). The framework uses the Deterministically Equivalent Modeling Method (DEMM) and Bayes' model selection methods and consequently can be applied to black box models. The methods chosen allow assumptions required in other methods to be relaxed, while keeping computation time minimal. In particular, assumptions about output distributions are relaxed, which is important in process engineering models because equilibrium and theoretical limits can cause output distributions to be highly irregular. A major challenge has been applying Bayes' model selection methods to cases where experimental output data does not exist, which occurs when assessing new technologies. Modifications to existing model selection have been developed to address these cases. Applying this framework will give the information difference between models, and identify which parameters are driving the overall. These results can be used in a sequential decision making process, facilitating decisions over the best use of resources. This may include helping to shape experimental programs or further refinement of the models. The framework has been applied to three case studies. The first involves competing hydrogen producing thermochemical cycles. It was found that the best use of resources was to further investigate the separations involved, rather than the reactions. The second involved two versions of a refinery process. The overall uncertainty was driven by uncertainty in the fitted parameters, and consequently if a difference is to be observed then the uncertainty in these fitted parameters need to be reduced. The third case study involved competing technologies for warm syngas cleanup. The excel-based tool has been constructed so that this framework can be applied by others in the future. This tool calls Matlab to complete the required calculations, but only requires the user to enter the required inputs in Excel, making it easy for the user. === by Ingrid Berkelmans. === Ph.D.
|
|---|
