Multi-loop process controller
Fig. 8.26. Diagrammatic representation of a supervisory setpoint control (SSC) system for fermenters. This example illustrates a system controlling temperature by means of heating only, dissolved oxygen tension by stirrer speed and pH by the addition of acid and alkali. All control functions are performed by the intelligent process controller and the computer only communicates with this m order to log data and send new setpoints when instructed to do so by the user (Whiteside and Morgan, 1989).
process. The first level of control, which is already routinely used in the chemical industries, involves sequencing operations, such as manipulating valves or starting or stopping pumps, instrument recalibration, on-line maintenance and fail-safe shut-down procedures. In most of these operations the time base is at least in the order of minutes, so that high-speed manipulations are not vital. Two applications in fermentation processes are sterilization cycles and medium batching.
The next level of computer control involves process control of temperature, pH, foam control, etc. where the sensors are directly interfaced to a computer (Direct Digital Control (DDC); Fig. 8.25). When this is done separate controller units are not needed. The computer program determines the set point values and the control algorithms, such as PID, are part of the computer software package. Better control is possible as the control algorithms are mathematically stored functions rather than electrical functions. This procedure allows for greater flexibility and more precise representation of a process control policy. The system is not very expensive as separate electronic controllers are no longer needed, but computer failure can cause major problems unless there is some manual back-up facility.
The alternative approach is to use a computer in a purely supervisory role. All control functions are performed by an electronic controller using a system illustrated in Fig. 8.26 where the linked computer only logs data from sensors and sends signals to alter set points when instructed by a computer program or manually. This system is known as Supervisory Set-Point Control (SSC) or Digital Set-Point Control (DSC). When SSC is used, the modes of control are limited to proportional, integral and derivative because the direct control of the fermenter is by an electronic controller. However, in the event of computer failure the process controller can be operated independently.
Whiteside and Morgan (1989) have discussed some of the relative merits of DDC and SSC systems and given case histories of the installation and operation of both systems.
The most advanced level of control is concerned with process optimization. This will involve understanding a process, being able to monitor what is happening and being able to control it to achieve and maintain optimum conditions. Firstly, there is a need for suitable on-line sensors to monitor the process continuously. A number are now available for dissolved oxygen, dissolved carbon dioxide, pH, temperature, biomass (the bug meter, NADH fluorescence, near infra-red spec troscopy) and some metabolites (mass spectroscopy a H near infra-red spectroscopy). All these sensors hav been discussed earlier in this chapter. Secondly it^ important to develop a mathematical model that ad<T quately describes the dynamic behaviour of a process" Shimizu (1993) has stressed the vital role which these models play in optimization and reviewed the use "0f this approach in batch, fed-batch and continuous processes for biomass and metabolites. This approach with appropriate on-line sensors and suitable model programs has been used to optimize bakers' yeast production (Ramirez et al., 1981; Shi et al, 1989), an industrial antibiotic process and lactic acid production (Shi et al., 1989).
Although much progress has been made in the ability to control a process, few sensors are yet available to monitor on-line for many metabolites or other parameters in a fermentation broth thus delaying or making a fast response difficult for on-line control action. Also, it is possible that not all the important parameters in a process have been identified and the mathematical model derived to describe a process may be inadequate. Because of these limitations, an artificial neural network may be used to achieve better control (Karim and Rivera, 1992). These are highly interconnected networks of non-linear processing units arranged in layers with adjustable connecting strengths (weights).
Input Hidden Output layer layer layer
Input Hidden Output layer layer layer
Fio. 8.27. Two-layer neural network (not all the possible interconnections are shown).
In simpler neural networks there is one input layer, 0ne hidden layer and one output layer (Fig. 8.27). Unlike recognized knowledge-based systems, neural networks do not need information in the form of a series of rules, but learn from process examples from which they derive their own rules. This makes it possible to deal with non-linear systems and approximate or limited data.
When training a neural network the aim is to adjust the strengths of the interconnections (neurons) so that a set of inputs produces a desired set of outputs. The inputs may be process variables such as temperature, pH, flow rates, pressure and other direct or indirect measurements which give information about the state of the process. The process outputs obtained (biomass, product, etc.) produce the teacher signal(s) which trains the network. The difference between the desired output and the value predicted by the network is the prediction error. Adjustments are made to minimize the total prediction error by modifying the interconnection strengths until no further decrease in error is achieved. Commercial computer packages are now available to help to determine which of the input variables to use for training and to determine the optimum number of interconnections and hidden layers (Glassey et al., 1994). Readers requiring more detail of the theory of neural networks should consult Karim and Rivera (1992).
This method of control is still at an early stage of development, but it has already been used in a case study on ethanol production by Zymomonas mobilis (Karim and Rivera, 1992), in real-time variable estimation and control of a glucoamylase fermentation (Linko and Zhu, 1992) and recombinant Escherichia coli fermentations (Glassey et al., 1994).
In industrial systems where a significant amount of on-line and off-line process data may be available, but there are tight time restraints imposed on process optimization, the potential for developing a relatively accurate neural network model within short time scales becomes very attractive (Glassey et al., 1994).
Was this article helpful?
Discover How To Become Your Own Brew Master, With Brew Your Own Beer. It takes more than a recipe to make a great beer. Just using the right ingredients doesn't mean your beer will taste like it was meant to. Most of the time it’s the way a beer is made and served that makes it either an exceptional beer or one that gets dumped into the nearest flower pot.