The addition of an expert system can advise a process controller when to make adjustments. These devices use pattern recognition algorithms, calculate control loop performance on-line, and readjust operator intervention. Expert systems can help monitor a process and notify operators of an out-of-specification situation. Self-tuning controllers can also use model-based references where the difference in response between the process and the model is used to tune the controller and also update the model. A classic expert system for process control was designed for advisory control of thermal processing in a major soup company (3). The expertise of a retiring staff member for troubleshooting, start-up, and shutdown of hydrostatic and rotary sterilizers was captured in an expert system program, saving the company up to US$3 million per year. An expert system can respond to user questions, query the user for additional information, and provide a report of the interaction along with a process and instrumentation diagram. Fault analyzers respond to alarms and other process inputs to diagnose abnormal conditions. They can find a previously determined pattern in a set of alarm messages. Rather than receiving a number of alarms, the operator will receive a simplified message stating the recommendation. Corrective action can then be taken.
A widely used rule-based expert system from Gensym Corp (Cambridge, Mass.), among other applications, has been useful in controlling wastewater treatment. The arrangement of equipment and controls is a complex, time-consuming assignment and many factors need to be considered. With control-system configuration, operating conditions and parameters, objectives, and constraints such as equipment limitations can be entered into a process model. The configuration best satisfying all requirements is then presented to the user.
Neural networks can predict patterns in process data, which in turn can be used to control the process as desired. With neural network pattern recognition software, responses are observed and necessary control adjustments are made. Applications have been reported across a wide variety of industries. Pattern recognition methods have been used for the identification of flours (4). BrainMaker (Nevada City, Calif.) has reported a wide variety of applications for its neural network product including investment advising, heart attack diagnosis, beverage quality testing, chemical structure recognition, and protein sequence pattern recognition. The application of neural networks in the dairy industry was recently discussed in an Australian seminar (5). Reports included neural networks in ultrahigh temperature (UHT) operations, natural language computer control of crucial steps in cheese making, and fermentation processes. The effects of transmembrane pressure and cross-flow velocity on cross-flow filtration were modeled using a neural network approach. Modeling was obtained after only five experimental trials for either raw cane sugar remelt or natural gum solution (6). The use of neural networks for modeling instrumental-sensory relationships has been investigated. The advantages and disadvantages of using neural networks were compared with multivariate linear methods of principal components re gression (PCR) and partial least-squares (PLS) regression. It was concluded that neural networks cannot replace PCR and PLS for linear relationships but do offer potential for modeling nonlinear relationships (7).
Questions regarding expansion of neural networks when necessary have prompted research on network configuration. Learning rates for neural networks, which were expanded by coupling modules, rather than simply directly expanding the existing network were improved. Coupled neural networks were found to improve food starch identification (8).
The relationships between process parameters and flow rates through ultrafiltration membranes have been examined using fuzzy mathematics (9). Main parameters in the permeate flow were identified. Fuzzy logic has been used in many familiar applications including, home appliances, video cameras, automobiles, robots, and aircraft. One company has developed a hybrid product combining neural networks, fuzzy logic, genetic algorithms, and chaos theory (Neuristics, Baltimore, Md.). The company maintains this design overcomes limitations of neural networks, explaining the logic behind its forecasts.
Regarding business applications, neural networks have been successfully used for bank loan applications. A program called Countrywide Loan Underwriting Expert System, or CLUES (Brightware, Inc, Novato, CA), evaluates a loan file's strengths and weaknesses in human terms. Loans can be approved in less than 2 min compared with 50 min for human underwriters. Benefits include personnel cost savings and better quality and consistency in underwriting. Neural networks can be used to set up intelligent agents for information services. These agents are able to automatically locate and retrieve competitive intelligence information. Consumer electronics industries have developed software programs with the ability to operate autonomously, performing operations ranging from electronic shopping to getting data over the Internet. The growth of the Internet may create new artificial intelligence applications.
The many facets of artificial intelligence are able to reduce labor costs, improve productivity and quality, and remove human bias. Expert systems can act as training tools, both during the process of development, as well as during implementation. Advanced computer-aided systems can improve food safety, reduce processing times, and achieve higher quality and yields at minimal costs if applied correctly to the food processing system (10). Human talents and abilities can be better utilized and even assisted as these programs are made available for process control and analysis.
The large number of computing techniques within the very broad terms of artificial intelligence will continue to drive many of the advanced applications in science, industry, and consumer products. Products will continue to improve in data assimilation and communication, replacing an increasing number of tasks previously requiring human intervention.
Artificial intelligence will continue in importance as it becomes embedded in mainstream systems. The different components of artificial intelligence are each finding their own particular embedded applications.
Was this article helpful?
Your heart pumps blood throughout your body using a network of tubing called arteries and capillaries which return the blood back to your heart via your veins. Blood pressure is the force of the blood pushing against the walls of your arteries as your heart beats.Learn more...