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

Abstract

One of the limitations of today's knowledge-based ( K B) IntroductionDuring the last several years, the knowledge-based (KB) approach has found useful application in the control of fermentation processes. So far, KB systems have been successfully used in the control of various bioprocesses, such as wastewater treatment, production of yeast and bacterial biomass, biosynthesis of enzymes, antibiotics, and amino acids . These applications have shown that by exploiting KB techniques, many of the shortcomings of conventional control methodology can be, at least partially, overcome. KB systems are capable of informal "physiological" interpretation of on-line measurements, monitoring and handling of complex phenomena which remained "invisible" to conventional systems, and flexible alteration of the control policy according to the real situation in the plant. It is expected that the KB approach will have growing importance in the field of bioprocesses control (Cooney et al., 1988; Aynsley et al., 1 990).Despite these positive results, however, KB control of ferCorrespondence concerning this article should be addressed to K. B. Konstantinov who is currently at the Department of Chemical Engineering, University of Delaware, Newark, DE 19716. mentation processes is still a new, rather poorly developed, area of biotechnology, with many unsolved theoretical and practical problems. These problems are related to shortcomings in current (real-time) KB technology on the one hand, and to the extreme complexity of biological plants on the other. Recent studies have shown that KB systems used in bioprocess control must possess a number of specific characteristics that are nontrivial and are not supported by the KB shells available today . Among these, of primary importance are the temporal reasoning capabilities which allow KB systems to account for the time dimension, significantly broadening their scope (see Figure 1). The availability of a consistent set of temporal reasoning features is considered to be a requisite in highly dynamic, complex environments (Moore et al., 1990;Laffey, 1991).In the field of fermentation processes, a particular aspect of temporal reasoning that deserves special attention is the capability of the system to reason about the temporal behavior of the process variables over specific (usually recent) episodes of the process history. In many cases, the underlying phenomena that have to be identified by the KB system leave their AIChE JournalNovember 1992 Vol. 38, NO. 11 1703 Figure 1. Scope of KB control systems with and without temporal reasoning.KB systems provided with temporal reasoning capabilities can reason about past (and possibly future) events and phenomena, and their temporal relationships.own unique "stamp" on the process history, expressed in the form of characteristic profiles of the related variables. TO detect and diagnose adequately such events, KB systems should be able to reason about these profiles. This will result in a number of advantages: KB systems will be capable of more relia...

TRACE, a graphical tool for the acquisition and detection of signal patterns

“…Combining the different possible qualitative values of the first and second derivative gives rise to seven representative primitives, which Cheung terms triangular episodes, and with which it is possible to represent the imprecision in the evolution of a parameter in the sections between two consecutive points, but not at the points themselves, which are still defined in a totally precise manner. Konstantinov and Yoshida (1992) make a completely qualitative description of the evolution of a parameter by using the sign of the first and second derivatives. This is the first work in which a solution for the acquisition of expert knowledge is proposed: a library -which can easily be extended by the user -comprising 20 representative primitives that arise from the concatenation of more basic primitives, similar to those employed by Cheung.…”

Fermentation seed quality analysis with self-organising neural networks

“…The extensions of PCA and PLS to nonlinear PCA and PLS has enabled faults to be detected in batch chemical processes (Dong and McAvoy, 1996). Other methodologies for deriving the process condition by extracting the trends from process data have been developed and applied to both chemical and bioprocesses (Konstantinov and Yoshida, 1992;Bakshi and Stephanopoulos, 1994). Trend extraction using feedforward ANNs (FANNs) has also been considered (Rengaswamy and Venkatasubramanian, 1995).…”

An approach to the linguistic summarization of time series using a fuzzy quantifier driven aggregation

“…For an example of a language for the representation and extraction of temporal patterns, based on the changes of the first and second derivative, and the analysis of the increasing/decreasing, and convex/concave functions, see, e.g., Stockman, Kanal, Kyle, Yager and Zadrożny, 28 Cheung and Stephanopoulos 29 or Konstantinov and Yoshida. 30 Boyd 31 and Batyrshin and Wagenknecht 25 propose a system-generating linguistic descriptions of time series as IF-THEN rules, with fuzzy intervals as values in the case of the latter work. A rule-based description of times series via linguistic (fuzzy) summaries by using the so-called moving approximation transform is given by Batyrshin et al 26,27 Baldwin et al 32 propose a model time series by using linguistic shape descriptors represented by parametrized functions and use the FRIL system based on evidential reasoning for prediction.…”