Comparative analysis of qualitative models when the model changes

Abstract

An important application of the recently developed techniques in qualitative mathematical modeling is to qualitatively predict how changes in the operation of chemical units affect their behavior. Weld (1987, 1988a,b) has developed a series of comparative techniques that analyze the effect of perturbations to the parameters of a given qualitative unit model. In this article we demonstrate a system, based on the qualitative process theory of Forbus ( 19841, which extends comparative analysis in two ways. First, it predicts the effects of changes in the qualitative equations and in parameter values. Secondly, it compares physical descriptions, rather than comparing models directly, and builds and compares the associated models automatically. IntroductionReasoning with qualitative models of process units is an important component of intelligent systems which reason with "deep" or "first principles" knowledge of chemical plants (Dalle Molle et al., 1988;Oyeleye and Kramer, 1988;Rich and Venkatasubramanian, 1987). Comparative analysis is a branch of qualitative reasoning that predicts how changes to the inputs of a unit model affect the behavior of model parameters. It can be viewed as the qualitative analog of sensitivity or perturbation analysis of a set of equations. Comparative analysis is required for many intelligent systems. For example, in "generate and test" troubleshooting systems it is necessary to determine how hypothesized faults would change the behavior of process units (Grantham and Ungar, 1990), and in design it is necessary to compare the relative effect on process behavior of alternative design scenarios (Grantham, 1990).Comparative analysis systems demonstrated to date are limited to analyzing the effect of changes to parameter values; the actual qualitative constraints that define the structure of the model must remain constant. Thus, although these systems can predict that an increase in the input temperature to a liquid preheater results in a higher output temperature they cannot predict that some of the liquid may vaporize. Neither can they predict that the addition of catalyst to the input would result in an exothermic reaction which would increase the exit temperature of the liquid and change its composition.To make such deductions, comparative analysis systems must be extended in two ways. First, they must recognize that changes in a physical description may result in a change in the structure of the qualitative model, i.e., they must take on the task of modifying a model to reflect changes in process conditions rather than requiring it to be specified apriori. Secondly, they must be able to compare the modified and original models to determine how the structural changes affect behavior. It is, of course, still necessary to be able to analyze parameter changes within a single model.A comparative analysis system is presented here which combines automatic model modification with model-model comparison. This enables the system to automatically determine the effects of changes in process condit...

Real‐time qualitative analysis of the temporal shapes of (bio) process variables

“…As has been reported elsewhere (Grantham and Ungar, 1991;Konstantinov and Yoshida, 1991b), such constructions have high expressive power and can be used to create rules with high efficiency. Such logic can be very useful for the early detection of slow phenomena, either of physical or physiological origin (Moore et al, 1990).…”

A Method of Trend Analysis using Latent Dirichlet Allocation

“…A trend is defined as a method of describing changes over a long period of time and can predict future using past patterns [1]. Qualitative trend analysis uses the opinions of experts to extract trends in each period and then analyzes the trends by comparing the similarities of the trends [2]. Interval-Halving algorithm is used to extract trends and timeweighted average is calculated to compare the similarity of two trends [3].…”

Model‐based approach to automated hazard identification of chemical plants

“…Thirdly, in the current version of QHI, hazards such as over-pressure and over-temperature are examined only when the model does not change by comparing the value of variables within the same quantity space. To identify these types of hazards when the model changes, a more advanced comparative analysis method, such as Grantham and Ungar (1991), is required since the value of the variable found by simulating the original design model will no longer be directly related to the value of the variable found by simulating the fault model.…”