Intelligent Automation for Fermentation, Part 1

By Bonnie Haferkamp, Senior Solutions Engineer, Gensym Corp.

Editor's Note: This two-part article is excerpted from Bonnie Haferkamp's talk, which will be delivered at the InterPhex Show later this week.

Introduction
Fermentations are difficult processes to monitor and control, particularly with conventional control systems. These systems generally aren't designed to perform the sophisticated pattern detection, analysis and state prediction that is necessary to effectively monitor and control a fermentation process. It is often impossible to directly measure the state of a fermentation process, particularly on-line. This problem is magnified by the inconsistencies imposed on the system by media, culture, raw material and process variability.

To overcome deficiencies in typical fermentation control systems, many operations rely on manual monitoring and control. This causes systems to be prone to operator error, inconsistent decision-making, and missed actions as a result of human scan rates. The lack of appropriate monitoring and control systems for fermentation can result in lower yields, lower production rates and increased downstream process variability. In research and development environments, typical control systems often don't provide the necessary flexibility to rapidly implement new monitoring and control schemes to support strain evaluation and process development, slowing the development process. Systems are needed that can help:

• Increase product yield
• Decrease product degradation
  
• Decrease downstream losses
  
• Decrease production costs
  
• Maximize production capacity
  
• Minimize batch-to-batch variation
  
• Improve process development and strain evaluation turnaround

G2 Fermentation Expert is a monitoring and supervisory control system being developed to meet the needs of fermentation. It is a Gensym G2-based system comprised of a suite of configurable and extensible components specifically designed for fermentation processes. Custom developed G2-based systems have been proven successful in supporting production, pilot and research scale fermentations. Eli Lilly improved antibiotic yields and reduced process variability by implementing a fermentation monitoring system utilizing the rule-based reasoning and statistical process control (SPC) functionality of G2¹. Novo Nordisk also used this functionality of G2 for sensor validation and improved control to drive fermentation cultures toward optimal performance². Osaka University applied the neural network functionality of G2 for producing on-line soft sensor substrate and metabolite estimates for antibody production in an animal cell culture³. Other systems based on neural networks and advanced technologies have been used to predict production tank performance from seed tank data⁴, and there are many references to predictive biomass estimates based on neural network models. As reported in a recent publication⁵ on the state of fermentation monitoring and control, neural networks, fuzzy logic and rule-based reasoning clearly provide benefits in fermentation quality and state estimates, fault detection and control loops.

System Overview
G2 Fermentation Expert is an object-oriented, graphical development system layered on G2 and Gensym layered products. It provides high-level functionality for rapidly configuring and implementing fermentation process monitoring and control solutions. Its hierarchical, modular structure is designed for maximizing flexibility and future extensibility, even to other process areas beyond fermentation. Figure 1 depicts the high-level system architecture.

Figure 1: G2 Fermentation Expert architecture

Some of the specific capabilities of the system include:

• Real-time growth phase identification and characterization
• Multivariate and metabolically-based alarming and advising
• Fault detection
  
• Soft sensor estimates
  
• Sensor validation
  
• Intelligent control
  
• Historical trend analysis
  
• Transfer optimization
  
• Rapid, flexible recipe development

FermentationExpert's functionality is built upon an underlying substrate of intelligent core technologies, including neural networks, fuzzy logic, SPC and rule-based reasoning. This enables the use of the most appropriate technology to solve individual fermentation problems while maintaining an integrated environment for deploying solutions. For example, within the same application, neural networks may be used to detect faults, and rule-based technology may be used to implement logic for appropriately responding to the fault.

G2 Fermentation Expert includes palettes of predefined equipment, sensor, alarm and logic blocks for configuring an application for a specific process (Figure 2).

Figure 2: Palette Examples

Applications are developed graphically by cloning objects from palettes, connecting objects together to form logical process sequences, and configuring objects for the specific requirements of each fermentation.

The system interfaces to most standard external data sources and repositories on-line. This feature enables real-time, bi-directional communication with programmable logic controllers (PLCs), distributed control systems (DCSs) and data historians for process variable monitoring and supervisory control. Recipe and batch management data in G2 Fermentation Expert can be transferred between external databases or recipe management systems. Intelligent instrumentation can be interfaced to the system for incorporating on-line analyses.

System Functionality
G2 Fermentation Expert utilizes both on- and off-line process measurements and historical process data for learning the dynamics of a specific fermentation to provide inferential, predictive monitoring and intelligent, supervisory control. Typically, fermentation facilities store large amounts of historical process data. This historical data and data generated on-line can be used by G2 Fermentation Expert to develop neural network models of the system and rules for sensor processing and operations advise. G2 Fermentation Expert also allows the knowledge of experts-process operators, process technicians, engineers and scientists-to be embedded within the application in the form of rules.

Go to Intelligent Automation for Fermentation, Part 2

References, Part 1

1. Alford, J. S., et al. Development of Real-Time Expert System Applications for the On-Line Analysis of Fermentation Data. American Chemical Society Conference Proceeding Series, Harnessing Biotechnology for the 21st Century, 1992.
   Back to text
2. Aldridge, Susan. Chemical Engineering Meeting Spotlights New Technologies for Bioprocess Control. Genetic Engineering News, Vol. 14, No. 18, 1994.
   Back to text
3. Oh, G.S., et al. Neural Networks in Estimation and Control of Antibody Production Using Hybridoma Cells in Fed-batch Cultures. IFAC Computer Applications in Biotechnology, Garmisch-Partenkirchen, Germany, 1995.
   Back to text
4. Ignova, M., et al. Seed Data Analysis for Production Fermentor Performance Estimation. IFAC Computer Applications in Biotechnology, Garmisch-Partenkirchen, Germany, 1995.
   Back to text
5. Montague, G. Monitoring and Control of Fermentors. Institute of Chemical Engineers, Warwickshire, UK, 1997.
   Back to text