Intelligent Automation for Fermentation, Part 2
By Bonnie Haferkamp, Senior Solutions Engineer, Gensym Corp.
Back to Intelligent Automation for Fermentation, Part 1
Process Monitoring, Validation, Diagnostics and Advising
Many process monitoring applications contain elements of fault detection, sensor validation and alarming. Fermentations pose unique challenges in this area due to the non-linear nature of the process. This problem can be addressed through G2 Fermentation Expert by using non-linear classification schemes and fuzzified inferences for fault detection and sensor validation. Systems can be configured and trained to detect process, system and equipment faults using historical or on-line process data, and to extract cause-and-effect relationships. Sensor validation can be implemented with regard to the current state of the fermentation for increased confidence of on-line measurements.
An important aspect of the system is its ability to minimize nuisance and sympathy alarms through the use of rule-based reasoning and fuzzy logic. By prioritizing rules, incorporating fault detection information and using fuzzified characterizations of process values and trends, the system is able to reason about the most likely cause of a fault, and advises operators of the most appropriate corrective action to take to resolve the source of alarms. This is particularly useful for ensuring that consistent action is taken in response to the same fault, and for providing expert advice to less experienced operators. Default alarm configurations can be defined, or alarms can be configured individually for a particular fermentation.
Monitoring and alarm components are implemented as configurable logic blocks that can be "mixed and matched" according to the requirements of each fermentation. Alarms can be activated or deactivated during specific process or growth phases, enabling phase-based alarming.
Soft Sensors and Physiological State Identification
Typically, the parameters of interest in a fermentation process cannot be directly measured. Such parameters include biomass, substrate and product concentrations. Measurements such as these can be referred to as quality measurements of the system, and are generally performed as off-line assays. The soft sensor capabilities of G2 Fermentation Expert estimate and predict quality measurements on-line using models developed from historical quality and process data. During model training, subtle correlations are extracted from quality and process measurement data and incorporated into the soft sensor model. When used with on-line process measurements (the same process parameters used to train the model), the soft sensor produces on-line quality measurement estimates. These estimates can be used for determining more optimal control actions and transfer times during the course of a fermentation.
Process variable trends of a fermentation exhibit distinct patterns, which are often repeated in similar fermentations. Fermentation cultures pass through a sequence of growth phases during the fermentation process. Growth phases are typically marked by distinct patterns in key process variables, such as oxygen uptake rate (OUR). Shifts in substrate utilization can also produce distinct patterns in process variable trends. G2 Fermentation Expert provides capabilities for identifying growth phase and substrate shifts through neural networks, fuzzy logic and rule-based reasoning. Automatically detected and predicted growth phase and substrate shifts allow for improved control decisions. Control strategies can be activated or deactivated according to growth phase or substrate activity.
Intelligent Control Strategies
Sensor validation confirms the accuracy of process measurements. Validated sensor readings can then be used to produce soft sensor and state identification estimates. Once the fermentation state is estimated from sensor readings of sufficient confidence, intelligent control strategies can be used to coerce the system to a preferred state. This sequence toward optimizing fermentation monitoring and control is depicted in Figure 3.
Figure 3: Optimizing Fermentation
G2 Fermentation Expert provides both basic control strategies and the means to easily develop more sophisticated, proprietary strategies specific to a particular fermentation. Control is supervisory; in a typical G2 fermentation application basic control loops are held in the PLC or DCS while setpoints are determined by G2. Control can be implemented in a variety of ways, including open-loop, closed-loop, automatic or strictly advisory.
Control strategies are created in the same manner as other system components: clone, connect, and configure. Control strategies can be activated or deactivated according to events that occur in the system, including growth phase changes, substrate shifts and alarm conditions. Because G2 is a graphical environment, the current processing state of a controller can be visually observed. Control strategies can also be modified, deleted or added at any time during the execution of a fermentation.
Scheduling and Transfer Optimization
In a fermentation operation, scheduling decisions must be made for transferring the contents of one vessel into another: seed tank into production tank and production tank into harvest equipment, including partial and complete tank transfers. Such decisions are often based on goals that may conflict with conditions within a facility. For example, production tank transfers, if based solely on predicted yields, may conflict with downstream processing capacities. Thus, there may be trade-offs between the desired goals of the process (increased yields, higher throughput, reduced variability) and the process capacity constraints (equipment utilization, volumes and downtime).
By utilizing the predictive soft sensor capabilities of G2 Fermentation Expert, estimates of yield rates of a current fermentation can be made. These yield rates can then be used to schedule tanks with lower predicted yield rates in front of tanks with higher predicted rates. Preferred transfer times can also be estimated for reducing variability in production vessels and downstream equipment. When equipment constraints become a priority over transferring tanks according to yield or variability estimates, control strategies can be initiated to modify the state of a fermentation, thus changing the preferred transfer time, to accommodate scheduling constraints. Fermentation Expert's object-oriented modeling capabilities, predefined palettes of fermentation and downstream equipment, soft sensors and rule-based reasoning combine to provide scheduling and transfer logic which can help maximize production capacity and yield.
Batch and Recipe Management
G2 Fermentation Expert provides recipe management functionality that can be used stand-alone or interfaced to an existing system. When interfaced to an existing recipe management system, information flow between the existing system and G2 Fermentation Expert can be configured to be completely transparent, allowing the user to interface exclusively with the existing system. Recipe management is implemented in G2 Fermentation Expert using the same look and feel as other capabilities, by cloning from palettes, connecting and configuring logic blocks. By combining the soft sensor and event detection (such as a growth phase or substrate shift) capabilities of G2 Fermentation Expert with the configurable, intelligent control and alarm strategies, metabolically-based monitoring and control can be achieved. This can lead to significant gains in batch consistency in production operations. In research and pilot operations, metabolically-based monitoring and control combined with the inherent flexibility of G2 Fermentation Expert, including on-line modifications to recipes, can help reduce the timeline for process development and strain evaluation.
G2 Fermentation Expert's batch management functionality provides configurable capabilities for recording, summarizing and retrieving information about a particular batch. A wide range of information can be recorded through this facility, such as:
- Setpoint changes, automatic and manual
- Operator actions
- Off-line data entry, such as assay results
- Faults detected
- Events detected, including growth and substrate shifts
- Derived values, such as totalized substrate feeds or process variable derivatives
Historical Trend Analysis
Trend charts are an important element of a fermentation monitoring system. In G2 Fermentation Expert, trend charts provide engineers and operators a means to rapidly view on-line comparisons between current fermentations, previous fermentations, and aggregates of previous fermentations. For example, on-line OUR readings for a specified fermentor can be displayed against the OUR readings of other currently active fermentors, the OUR readings of the previous batch in that same fermentor, or against the average OUR for the last month. In addition to the standard trend charts provided by G2 software, G2 Fermentation Expert is pre-configured for the following trend charts:
- Batch trend charts displaying several process variables and setpoints for one process unit--single batch trend
- Batch trend charts displaying one process variable or setpoint for many process units--multi-batch trend
- PID loop trend charts displaying process variable, setpoint and percent output for a specific process loop
Future System Developments
G2 Fermentation Expert has been designed as modular, object-oriented system to allow it to be easily extended into other application areas and enhanced capabilities.
Currently, G2 Fermentation Expert is focused toward the processing requirements of pharmaceutical fermentations. Future developments of the system may be towards the fermentation capabilities of other industries, in particular the food and beverage industry. These developments would include additional equipment, sensor and logic blocks that are specific to the food and beverage industry.
Genetic algorithms for optimizing fermentation processes are a likely enhancement to G2 Fermentation Expert. At the Kyushu Institute of Technology in Japan, a 14 percent increase in productivity was achieved in a research-scale ethanol fermentation by using a genetic algorithm to determine the optimal temperature profile. Genetic algorithms are already used in other G2-based applications for driving processes toward optimal operating conditions.
Because of the substantial work done by Gensym for developing robust scheduling methodologies, scheduling optimization is perhaps a key enhancement area. G2 Fermentation Expert's current strategies for scheduling tank transfers are based on predicted tank performance and basic equipment constraints. Such considerations can improve product output, but do not produce a true optimized schedule. In fermentation facilities with capacity issues or simultaneous multi-product campaigns, a more rigorous, optimized transfer schedule could provide substantially increased throughput while maximizing yields.
For more information contact Bonnie Haferkamp, Senior Solutions Engineer, Gensym Corp., 125 Cambridge Park Dr., Cambridge, MA 02140. Tel: 617-588-9241.
Reference, Part 2
1. Shimizu, K. and Ye, K. Development of Intelligent Control Systems for Bioreactors. IFAC Computer Applications in Biotechnology, Garmisch-Partenkirchen, Germany, 1995.
Back to text