Author
 |
Chao, Kuanglin - Kevin Chao |
|
TING, K - OHIO STATE UNIVERSITY |
|
Submitted to: Encyclopedia of Agricultural, Food, and Biological Engineering
Publication Type: Book / Chapter Publication Acceptance Date: 1/4/2002 Publication Date: N/A Citation: N/A Interpretive Summary: Model based control strategies are needed for predictive control of a system. The model used in the strategy is for correlating the input to and output from the system. This predictive power enables the control strategy to determine the change of input in anticipation of future output. The advantage is most obvious when time delay between the input and output of a asystem is a concern. Artificial neural networks are a good modeling tool for establishing the relationships between system input and output based on existing data. The structure of a neural network normally consists of an input layer, one or more hidden layer(s), and an output layer. Every layer is formed by a number of nodes. The number of nodes in the input layer or output layer is dictated by the number of input factors and output parameters specific to the system under consideration. The number of nodes in the hidden layers is determined based on certain criteria for effectiveness and accuracy of the network. The nodes in adjacent layers are fully connected. The function of a node is to accept information, process information, and generate new information for the nodes in the next layer. There is a "weight" associated with each piece of information coming to a node. Each node compiles all the incoming information, transforms the compiled information, and sends it forward. During the development of an artificial neural network to match input and output of a system, all the weights are adjusted following a systematic procedure in an effort to minimize the discrepancy between the predicted output and the observed output of the system. This article gives an overview of neural networks, their utility in model based control, and an application example. Technical Abstract: Quality is increasingly important in food and agricultural production as well as other manufacturing processes. In continuous processes, the goal is to keep the process composition steady and close to the optimum conditions. Uniform quality is also a required aspect of the process. Furthermore, there are frequently legal obligations that have to be fulfilled by product tcomposition, and in many cases the most economical product is the one closest to the legal limit. There are requirements for environmental protection, for production, and for plant safety. All of these require that the composition of various products be kept stable. These principles have been considered for many years in the development of control theory, controllers, and actuators in parallel with the growth of manufacturing industries. Because the behavior of most bioproduction processes are usually characterized by the interactions of many components, and by non-linearity, the modeling approach based on the stable composition principle is insufficient to represent the behavior of the process. On the other hand, measurements are usually available for each process cycle, which can be used to monitor the process. Neural networks have been shown to be good predictors of relations between process data and can be effectively used for modeling plant processes. They are thus an appropriate candidate for inclusion in process control designs. In section 2, process control designs are discussed with regard to why neural networks are appropriate to be included in process control. In section 3, the theory of feed-forward neural networks is presented along with a summary of model development procedures. Section 4 contains a design example of neuro-model based process control in a continuous food frying process. |
