Process models developed by using fundamental principles explain and describe the behavior of the process. This provides the opportunity to assess the importance of various fundamental phenomena such as steps in the metabolic pathway, effects of various modes of mass transfer, or limitations in energy exchange. The user can postulate the existence or lack of some phenomena, modify the model accordingly and compare the predictions of the model with data to determine if the assumptions made could be supported. Often, the process may be too complex or the information may be too limited to develop fundamental models. In addition, the fundamental models developed may be too large to be used in process monitoring, fault diagnosis, and control activities. These activities require fast execution of the models so that regulation of process operation can be made in a timely manner. The alternative model development paradigm is based on developing relations based on process data.
Statistics and system identification literature provide a large number of methods for developing models to represent steady-state and dynamic relar tions that describe process behavior. Powerful model development software is available to build models that relate process inputs to process and quality variables or relate process variables and product properties with ease and speed. There are two important tradeoffs. First, the model describes the process behavior as it is captured in data used for model development. Unlike first principles models, datarbased models cannot provide information on behavior that has not been observed (in the sense of capturing in data). Data-based models should not be used for extrapolation, and nonlinear data-based models should be used with caution for interpolation as well. The second tradeoff is the loss of the ability to link physical or biochemical phenomena directly with some part of the model. Hence, the capability to explain the mechanisms in play for the observed behavior is severely limited. One has to rely on sensitivity analysis between inputs and outputs, and expert knowledge to extract information that sheds light on fundamental phenomena that are important in a specific process. However, one should not underestimate the power of good data-based models. In diverse fields such as the stock market, aircraft and ship navigation, oceanography, agriculture, and manufacturing, datarbased models have played a significant role. Some data-based modeling methods are built with algorithms that are useful in mining historical databases to elucidate hidden relations in process and product variables. Datarbased modeling techniques such as principal components regression and subspace state-space models provide good insight about the largest directions of variation in data and most influential variables. This information can be valuable in enhancing process understanding and developing fundamental models.
Data-based models are frequently used in process monitoring and control, quality control or fault diagnosis activities. Many case studies included in the book illustrate the value and power of datarbased models in supervising process operations. Data-based models are discussed in Chapters 4 and 5, their uses in monitoring, control, and fault diagnosis are illustrated in Chapters 6, 7, and 8, respectively.
Was this article helpful?