M.N. Karim

Adaptive pole placement control algorithm for DO‐control in β‐lactamase production

The dissolved oxygen (DO) is an important variable in aerobic fermentations and affects the cell growth and product formation. Dissolved oxygen control is difficult in batch fermentations because of the time-varying conditions, time delays, and the probe dynamics. Modeling of the various patterns of biological activity in fermentations and their impact on the DO process dynamics is essential to both achieve a satisfactory control and to track the aforementioned patterns. An adaptive pole placement algorithm with time-delay compensation was used for controlling the DO, coupled with system identification using recursively estimated autoregressive models with exogeneous inputs (ARX). The flow rate of O2 in a constant flow rate gas inlet mixture is used as the manipulated variable. Supervision and coordination techniques are applied to improve the control performance. The control performance is affected by the accuracy of the model prediction and the selected time delay. The effect of DO level on the productivity of beta-lactamase using Bacillus subtilis under oxygen-limited conditions is investigated. Beta-lactamase stability is improved under prolonged growth conditions with low DO levels.

Control relevant identification of a pH waste water neutralisation process using adaptive radial basis function networks

An adaptive radial basis function network is developed for non-linear and time-varying processes based on control-relevant-identification of a pH waste water neutralisation processes. Most adaptive control techniques rely on fixed model structures and variable parameter estimates but these often lack the ability to control time-varying systems. A model structure and parameter updating algorithm is applied for the adaptive training of a radial basis function network (RBFN), to create an adaptive RBFN. Both the network structure (centres of radial basis function) and the related parameters (weights of centres) are updated on-line using an exponential window. Candidate centres are generated and eliminated on-line according to the operation of the system. The selection of centres is based on the contribution these candidate centres make to the output at each selection step, so that the network can track changes in both the system structure and parameters. Suitability of the identification schemes for model predictive control is demonstrated by means of a multiple step ahead prediction of a simulated, noise corrupted, pH neutralisation process.

Application of Neural Networks in Bioprocess State Estimation

The application of artificial neural networks to the estimation of bioprocess variables will be discussed. In fermentation processes, direct on-line measurements of primary process variables usually are unavailable. The state of the cultivation, therefore, has to be inferred from measurements of secondary variables and any previous knowledge of process dynamics. This research investigates the learning, recall and generalization characteristics of neural networks trained to model the nonlinear behavior of a fermentation process. Two different neural network methodologies are discussed, namely, feed-forward and recurrent neural networks, which differ in their treatment of time dependence. The neural networks are trained by backpropagation using a conjugate gradient technique, which provides a dramatic improvement in the convergence speed. The objective is to use environmental and physiological information available from on-line sensors to estimate concentrations of species in a bioreactor. Results of the neural network estimators are presented, based on experimental data available from the ethanol production by Zymomonas mobilis fermentation. The feed-forward and recurrent neural network methodologies are demonstrated to perform suitably as unmeasurable state estimators. Both networks offer comparable abilities of recall, but recurrent networks perform better than feed-forward networks in generalization.

Use of Recurrent Neural Networks for Bioprocess Identification in On-line Optimization by Micro-Genetic Algorithms

The use of recurrent neural networks in bioprocess identification and optimization is investigated. A recurrent neural network is trained on a set of fermentation data, and there-after used as a nonlinear process model to estimate nonmeasurable process states at different conditions. With the bioprocess state variable information available, an optimization technique can be used to generate optimum controls settings to improve the process performance. This paper explores the use of Micro-Genetic Algorithms as a technique for bioreactor optimization. Simulation results will be discussed based in the fermentative ethanol production by the anaerobic bacteria Zymomonas mobilis.

Towards intelligent process supervision: Industrial penicillin fermentation case study

Abstract Due to the their biological nature and inherent variability bioprocesses place significant demands on even the most advanced supervision approaches. Sophisticated supervision software, such as G2 from Gensym (a real-time knowledge based system) is already finding wide industrial bioprocess application. However, in order to overcome the severe challenges posed a straightforward ‘coding’ of operators and engineers knowledge is not sufficient. What is required is a methodology which attempts to maximise the information available for supervision purposes. This paper describes the development of an application in which feature extraction and data based methodologies are integrated with sophisticated physiological models to considerably enhance a rule-based supervisory system. Together within the G2 real-time knowledge based system framework they offer the potential for bioprocess operational improvement.

Data driven approaches to modeling and analysis of bioprocesses: some industrial examples

Data-generated models find numerous applications in areas where the speed of collection and logging data surpasses the ability to analyze it. This work addresses some of the challenges and difficulties encountered in the practical application of these methods in an industrial setting, and more specifically in the bioprocess industry. Neural networks and principal component models are the two topics that are covered in detail in this paper. A review of these modeling technologies as applied to bioprocessing is provided, and three original case studies using industrial fermentation data are presented that utilize these models in the context of prediction and monitoring of bioprocess performance.

Mass balance and thermodynamic description of solid state fermentation of lignocellulosics byPleurotus ostreatus for animal feed production

This study was focused primarily on the degradation of lignin in water hyacinth and barley straw for animal-feed production. The experiment was performed in a 1.5-L Applikon fermenter for 30 days, varying the air flow rate from 0.022 VVM/0.047 VMM to 0.048 VVM/0.102 VMM. A novel approach was introduced for prediction of a kinetic model by using instantaneous respiratory quotient (RQ) measurements and steady state elemental balances. Growth kinetics were determined for the fungus in a 30-day fermentation with a mixture of barley straw and water hyacinth as the substrate. The instantaneous heat-interaction profile was predicted from steady state balances. Fermentation data were checked for consistency using the entropy balance inequality, and thermodynamic efficiency was calculated to show that degradation of lignocellulosics byPleurotus ostreatus followed more than one metabolic pathway during the course of the fermentation. Growth ofP. ostreatus on lignocellulosics, such as water hyacinth and barley straw, was di-auxic or possibly tri-auxic during the 30 days of fermentation.

Use of an Extended Kalman Filter and development of an automated system for xylose fermentation by a recombinantEscherichia coli

SummaryAn automated system was developed for on-line monitoring and control of xylose fermentation by a recombinantEscherichia coli. A 7-L fermenter was interfaced with a personal computer. Control circuits were constructed and a software was developed to estimate the states of the fermentation using an Extended Kalman Filter. The automated system combined with the Extended Kalman Filter provided a satisfactory way to obtain on-line information regarding estimation of fermentation parameters.

Use of Micro-Genetic Algorithms in Bioprocess Optimization

Abstract A hybrid neural network-genetic algorithm approach for bioprocess optimization is proposed and evaluated. A recurrent neural network is supervised-trained on a set of available fermentation data, and used to predict the species concentrations which are difficult to measure on-line. With the state variable information available, a micro-genetic algorithm optimization technique is used to generate the optimum control settings to improve the process performance. The methodology is applied to optimize the fermentative ethanol production by Zymomonas mobilis in batch mode. The objective is to find the best environmental conditions, namely temperature, which can maximize the product yield. A simulation study shows that the proposed technique is capable of modelling and optimizing the bioprocess.

Neural Networks in Estimation and Control of Antibody Production Using Hybridoma Cells in Fed-Batch Cultures

Abstract The production of antibody in animal cell culture is usually accomplished by fed-batch operation using both glutamine and glucose feeds. The antibody production is related to the number of viable cells which in turn is related to the optimum levels of glutamine, glucose and the lactate which is produced in the cultivation process. However, these variables are not available on-line, hence in order to implement feeding policies for the maintenance of these variables at the desired values, on-line estimates are needed. Different neural networks are used to train three sets of fed-batch data for the estimation of concentrations of lactate, glucose, glutamine, ammonia and antibody. Also estimated are the number of total cells and the viable cells. These estimates are generated from the on-line measurements of variables such as base addition, turbidity Oaser measurement), dissolved oxygen, oxygen partial pressure, carbon dioxide concentration in the exit gas, flow rates of air and oxygen into the fermentor, feed rates of glucose and glutamine. Different neural network structures are required for different physiological states.