|
Adaptive Control
Astrom & Wittenmark begin their book:"Adaptive control" (Addison-Wesley, 1995) in the following way:
In everyday language, "to adapt" means to change a behavior to conform to new
circumstances. Intuitively, an adaptive controller is thus a controller that can modify its behavior in response to changes in the dynamics of the process and the character of the
disturbances.
In practice this implies that an adaptive controller is a controller with adjustable parameters, which is tuned on-line
according to some mechanism in order to cope with time-variations in process dynamics and changes in the environment.
This can be approached in different ways, but at IAU we focus primarily on the so-called indirect self-tuning regulators, illustrated below:
|
|
Block diagram of an adaptive control system. The "extra" loop constituted by the blocks denoted "identification" and
"design" is what separates the adaptive controller from a conventional one. The identification block contains some kind of recursive estimation algorithm
which aims at determining the best model of the process at the current instant. The design block then applies this model to produce a controller according to some
design strategy. This could for example be a pole placement, minimum variance, or predictive control design. |
|
What is the character of our work?
At IAU we believe we have a quite "pragmatic" approach to adaptive control. We do not spend much time on stability
proofs, but our research is primarily governed by implementation issues. We are continuously working on developing a set
of software tools to facilitate practical implementation of adaptive controllers. Furthermore we are always interested in
trying out adaptive methods on "real world" systems. This is also reflected in the DTU course "50360 Adaptive Control", taught by the department.
The IRCST tool
The IRCST tool was originally implmented in an adaptive control context, but it has since then developed way beyond that.
Today the tool contains an adaptive controller "template" and a "library" of building blocks for use in adaptive control. The
library contains a number of utilities for control design, filtering, recursive estimation, and utilization of physical insight. Some of the features are listed below:
Close connection to Matlab. Matlab is an integrated part of the tool: filters, matrices, vectors, parameters, etc. are
initialized in MATLAB and then send to the Real-time program. The results are then returned to MATLAB either during or after a run.
Design. The tool mainly supports pole placement type control. In order to solve the Diophantine equation, different
possibilies are available: 1) The equation can be solved symbolically beforehand, and the solution be programmed as a
function; 2) a Mathematica tool can solve the equation and write the necessary C-code; 3) a generic diophantine equation solver can be used for solving the equation on-line.
Identification. It is possible to work with discrete models described using either time-shift or delta operator or to use
continuous model descriptions. When working with continuous models, the so-called discrete state variable filter
technique is used for generating the regressors. A number of different recursive estimation algorithms are provided and
others are easily incorporated. Physical insight about the process is easily specified, and will be utilized in order to reduce the number of parameters to be estimated.
Some of our recent projects have been on adaptive control of biogas reactors and flexible robots.
 31360 Fuzzy, Neural, and Adaptive Control
Contact Ole Ravn (homepage) <or@oersted.dtu.dk >
|
