Albuquerque, New Mexico
June 24, 2001
June 24, 2001
June 27, 2001
6.65.1 - 6.65.11
A Neural Network Lab Experiment
Robert Lynn Mueller The Pennsylvania State University New Kensington Campus
Neural networks are becoming widely used in complex control problems. Many academic exercises approach neural network applications using only software simulations; however, simulations alone do not give students a full appreciation of the power and complexity of neural network-based controls. This paper describes a laboratory experiment that uses a temperature and airflow process simulator to demonstrate neural network control applications. The simulator is fundamentally a temperature controller in which large-scale changes in forced airflow produce significant changes in heat load.
The initial labs use PID control techniques to solve the temperature control problem and to demonstrate the problem that PID controllers have with large disturbances. The following labs address the same problem using a neural network control strategy. An actual neural network controller is built and used to perform the same temperature control as the classical PID system. Capabilities and drawbacks of neural network control are demonstrated.
The classical PID feedback control system is shown in Figure 1. The setpoint (SP) is the system’s input and the process variable (PV) is the output. The PID controller uses the error signal (SP minus the PV) to calculate a value for the manipulated variable (MV). In turn, the value of the MV then determines the value of the process variable. Unexpected changes to the process output can be caused by system disturbances, and it is the PID controller’s job to adjust the MV to account for these changes.
Based on the characteristics of the process, the PID controller is tuned to achieve the desired system response. Unfortunately, when the disturbance input varies over a large range, the PID controller’s ability to achieve the desired system response is greatly reduced. For these situations, a more advanced control scheme is necessary. One possible method to overcome large disturbance variations is a neural network (NN)-based control strategy.
Muelller, R. (2001, June), A Neural Network Lab Experiment Paper presented at 2001 Annual Conference, Albuquerque, New Mexico. https://peer.asee.org/9598
ASEE holds the copyright on this document. It may be read by the public free of charge. Authors may archive their work on personal websites or in institutional repositories with the following citation: © 2001 American Society for Engineering Education. Other scholars may excerpt or quote from these materials with the same citation. When excerpting or quoting from Conference Proceedings, authors should, in addition to noting the ASEE copyright, list all the original authors and their institutions and name the host city of the conference. - Last updated April 1, 2015