B. Romein

Classification, error detection, and reconciliation of process information in complex biochemical systems.

Bioprocess identification starts with collection of process information. Usually there is a variety of information available, consisting of actual measurement data, historical data, empirical kinetic and yield correlations, and general knowledge available from literature. A central problem is to find out how the various pieces of information should be integrated. In addition, one should know how to deal with missing, inconsistent, or too inaccurate data. Recently, a general systematic method for dealing with these problems, based on conservation constraints, was published, and application shown to simple black box systems. In this article, the scope is generalized by including metabolic network data and dispersed process information of diverse type and nature, such as multiple sources of the value of one particular quantity, use of kinetic expressions, analytical problems, cometabolism or mixed substrate utilization, and chemical reactions. The alkalophilic bacterium Acinetobacter calcoaceticus is used as a model organism, growing on acetate and converting xylose into xylonolactone. It is shown that all relevant pieces of information can be straightforwardly and systematically treated, by considering them as constaints. In general, it is illustrated how the search for directed process improvements starts with an optimal selection of information sources, followed by an accurate analysis of possible metabolic bottlenecks. In this particular case it is shown that the yield of A. calcoaceticus on acetate at varying xylose/acetate feed ratios can be accurately predicted using dispersed process information.

MACROBAL: A Program for Robust Data Reconciliation and Gross Error Detection

Abstract A set of linear constraint equations provides the mathematical basis for relating conversion rates in (bio-)chemical processes. In the computer program MACROBAL, running on any type IBM-PC, such equations can be defined by summing up all relevant compounds and the conserved quantities in an ‘elemental’ composition matrix. MACROBAL uses the concept of the redundancy matrix for analyzing the system of equations. If possible, measured rates will be balanced and non-measured rates will be calculated. A Chi-square test is performed for detecting errors in the system description and/or measured rates. The serial elimination method can be used to trace the error location.

Error Diagnosis and Data Reconciliation Using Linear Conservation Relations

In biotechnological conversion processes, black box process descriptions can easily be derived and are most useful to process measured data. Such descriptions relate the net compound flows to and from the process by means of the conservation laws for chemical elements, charge, enthalpy, ... When several compound flows (conversion rates) are measured, other conversion rates can generally be calculated this way. If a calculated conversion rate is also measured, one has obtained more than one estimate of that rate. By comparing these estimates, measurement errors or errors in the black box description can be detected and sometimes even be located. When the estimates are combined, a more accurate estimate for the conversion rate is obtained. This chapter provides an introduction to these techniques for fault detection and data reconciliation.

Steady-state analysis of hybridomas in chemostat culture

Summary The metabolic activities of a hybridoma cell line was studied in chemostat culture. Using data sets of all measured culture parameters macroscopic balances were constructed for the elements C, H, N, and O. From the balances a non-measured culture parameter, the oxygen consumption rate, was estimated together with its accuracy. In addition the accuracy and credibility of the measurements were improved by this method. A statistical test was performed to check the consistency of the chemostat data. On basis of the estimated oxygen consumption rates estimates were made for the energy requirements of hybridoma biomass synthesis Y′ ATP. The Y′ATP was 5.0–6.8 g biomass * mol ATP−1 which is low in comparison to most microorganisms.

Monitoring and Modelling Hybridoma Cultures

Possibilities for process design, optimisation and control are determined by reproducibility of cell cultivation and quality of the process model. Since the latter can only be judged with respect to the modelling aims, those must be specified beforehand. If reproducible culturing results cannot be obtained, modelling prospects are worsened. In the present work, standardisation of pre-culturing procedures proved to deliver suitable results. High-frequency sampling provided the necessary information for the formulation of an unstructured model. This could be used to predict process behaviour under various conditions.

Control of Cell Concentration in Fedbatch Hybridoma Cultures

In a fedbatch process cell life can be extended as compared to batch cultures.The control of cell concentrations in fedbatch hybridoma cultures may form one step in the optimisation with respect to product formation, if the relation between growth and product formation is known.

Substrate level control in fed-batch hybridoma cultures

Summary By controlling the substrate concentrations in fed-batch cultures yields of hybridoma cells and product can be improved. Possibilities for process control are determined by reproducibility of cell cultivation, quality of the process model, and the amount and quality of available measurement data. These aspects are studied by investigation of literature data and a limited number of experiments.

A Program for Error Diagnosis using a Sequential Vector Comparison Test

Abstract Verification of data measured in biotechnological processes is essential in research and for industrial monitoring and control. The newly developed “vector comparison test” (van der Heijden, 1991) is more powerful in detecting and locating errors than the serial elimination method, especially when it is applied sequentially. Both methods use linear constraint equations based on the conservation principles. A program (running on IBM-PC’s) is under development as an implementation of the vector comparison test. Process specific information must still be changed in the source code. The program was used successfully for the detection and location of errors in industrial fermentation data. The error diagnosis is represented numerically and graphically.

Error Detection and Diagnosis

Abstract The accuracy and reliability of measurement information can be improved by means of data reconciliation methods. Normally a number of constraints is used, arising from conservation principles of e.g. chemical elements. These methods often suffer from low error sensitivity. Furthermore, the sensitivity of error detection is improved by mediating the results of series data reconciliation with time, and accurate statistical elaborations. As a result, measurement bias of about 2% to 5% can be detected. If inconsistencies are detected, the cause should be revealed. They originate from three types of error: 1) measurement errors, 2) errors in the constraints, and 3) errors in the translation from primary measurement data to observed conversions. For this, a very flexible error diagnosis method has been developed: the vector comparison test. This new method directly detects errors in the primary measurements, while the well-known “serial elimination method”, is restricted to errors in the observed rates. Specific errors in the constraints can also be dealt with. An example in which industrial data are analyzed is provided. Errors were detected for two similar process-runs. A faulty nitrogen content in the specified biomass composition was found to be the cause. Several future possibilities for advanced error diagnosis are indicated.

An Improved Method for Measurement Reconcilation and Gross Error Diagnosis for (Bio)Chemical Processes

Abstract For gross error diagnosis and balancing of fermentation data in terms of conversion rates, a set of linear equations constraining the system is represented by the elementary composition matrix. An improved method is presented to 1) detect and calculate the balanceable and calculable rates, given the set of measured rates and 2) detect, locate and estimate possible gross errors.