Experimental implementation of a dual neural network based optimal controller for a heat diffusion system is presented. Using the technique of proper orthogonal decomposition (POD), a set of problem-oriented basis functions are designed taking the experimental data as snap shot solutions. Using these basis functions in Galerkin projection, a reduced-order analogous lumped parameter model of the distributed parameter system is developed. This model is then used in an analogous lumped parameter problem. A dual neural network structure called adaptive critics is used to obtain optimal neurocontrollers for this system. In this structure, one set of neural networks captures the relationship between the states and the control, whereas the other set captures the relationship between the states and the costates. The lumped parameter control is then mapped back to the spatial dimension, using the same basis functions, which results in a feedback control. The controllers are implemented at discrete actuator locations. Modeling aspects of the heat diffusion system from experimental data are discussed. Experimental results to reach desired final temperature profiles are presented.

Meeting Name

2003 American Control Conference, 2003


Mechanical and Aerospace Engineering

Keywords and Phrases

Galerkin Method; Galerkin Projection; Adaptive Critics; Basis Functions; Distributed Parameter Systems; Feedback; Feedback Control; Finite Difference Methods; Finite Difference Model; Heat Diffusion System; Lumped Parameter Control; Neural Network; Neurocontrollers; Optimal Control; Orthogonal Decomposition; Reduced Order Analogous Lumped Parameter Model; Reduced Order Systems

International Standard Serial Number (ISSN)


Document Type

Article - Conference proceedings

Document Version

Final Version

File Type





© 2003 Institute of Electrical and Electronics Engineers (IEEE), All rights reserved.

Publication Date

01 Jan 2003