Modeling a new process can be challenging and time-consuming due to either a lack of understanding of the true behavior of the process, costly experimentation, or scarce data. Oftentimes, there are past processes related to but not exactly the same as the new process, in which abundant knowledge has already existed. Although the process is new, mathematical models from similar yet non-identical past processes can provide insights into the new process and, hence, careful integration of the existing models will potentially improve the new predictive model as well as reducing modeling effort. Over the last few years, little has been studied on the combination of past models and expert judgments in the discipline of chemical engineering. Therefore, this study will propose model mig...[ Read more ]

Modeling a new process can be challenging and time-consuming due to either a lack of understanding of the true behavior of the process, costly experimentation, or scarce data. Oftentimes, there are past processes related to but not exactly the same as the new process, in which abundant knowledge has already existed. Although the process is new, mathematical models from similar yet non-identical past processes can provide insights into the new process and, hence, careful integration of the existing models will potentially improve the new predictive model as well as reducing modeling effort. Over the last few years, little has been studied on the combination of past models and expert judgments in the discipline of chemical engineering. Therefore, this study will propose model migration methodologies to address the issue.

The purpose of this study is to use similarity, which is referred to as the phenomenon that structurally similar, yet not-identical processes perform the similar function or yield the similar output. Consequently, model migration aims to integrate data of the new process with existing model of the related past processes. This study systematically investigates model migration methodologies based on process similarity which widely exists in engineering problems and many others. The overall theme of the proposed methods is the statistical adjustments of the known past models, and the incorporation of prior knowledge into new process data under Bayesian framework. A series of migration implementations are studied where the objective is to develop a migration model for the new process that is better than using the data of the new process only. The merits of model migration are demonstrated on various test functions as well as on real-world chemical processes. It is shown that the proposed methods are promisingly effective and efficient modeling tools for complex chemical systems.