MidSys |
Identification
Software for Multi-input/Multi-output
Systems
on PC and Workstations
PURPOSE
Most of the digital control
systems (DCS) are able to implement multi-variable
algorithms for tracking and regulation, in order to take
into account the eventual interactions between several
loops. To adjust such algorithms, it is necessary to have a
global model, which underlines interactions between the
various control variables and the various process
measurements.
The implementation of complex mechanical systems, with
several degrees of freedom, like positioning of platforms,
vehicles, flexible structures, robots, also needs a
multi-input/multi-output model, showing the interactions in
order to design the control laws.
It is of a great importance to identify directly a
multi-input/multi-output model in order to analyse the
coupling effect between the various inputs and outputs. This
is possible with MidSys, but is impossible when using a
concatenation of multi-input/single output models (as do
other softwares which claim to be multi-variable).
MidSys brings an other advantage : the software allows to
directly estimate the structure of the model to be
identified from the experimental data. As a consequence, it
is no more required to have an ?a priori? knowledge of the
structure of the model.
This software was developped in cooperation with the
Laboratoire d'Automatique de Grenoble (C.N.R.S./I.N.P.G.)
FIELDS OF
APPLICATION
MidSys software can be used for
any application where multi-input/multi-output control is
involved : distillation columns (chemical, petro-chemical),
engines (cars, aeronautics), mechanical systems with several
degrees of freedom, temperature and dryness plant
management, thermal systems, paper industry, cement
industry, food processes, teaching, research, ...
GENERAL DESCRIPTION
This software is made of
several modules, allowing the following functionalities :
* Data
management
This module allows to select from a data file the different
inputs and outputs of the process. It also provides the
following functions :
- removing continuous components of signals (which is
necessary for identifying dynamical models),
- scaling the data signals to allow a regular convergence
speed for identification algorithms,
- Filtering high frequencies or drifts from these signals,
- Under-sampling of data signals (useful when too high
acquisition frequency has to be used, or when a digital
filter is used for anti-aliasing purpose),
* Model structure estimation
This module allows the estimation (from data) of the
observability indices of multi-output systems whose sum
gives the minimal state dimension of the system. These
indices enables to define precisely the structure of the
multi-input/multi-output model to be identified (because the
number of states is not enough).
This structural estimation is made through the use of a
least squares type criterion weighted by the dimension of
the observability indices and using measured inputs and
outputs or instrumental variables.
Recently developped structural estimation algorithms are
included (the D-L. Alg.- IEEE Transactions in Automatic
Control, January 1994).
* Parametric identification
This module provides several methods to identify parameters
of a multi-input/multi-output model, because there is no a
unique method which gives the best results for all types of
measurement noise which may be encountered in practice.
Furthermore, several kinds of parametrisation of the
multi-variable models are proposed, related to the control
algorithm to be used later with the model.
* Validation of identified model
The aim of this module is to perform statistical tests for
the validation of the model, which are more significative
than visual comparisons of the plant and model outputs
(because of the presence of noise). These tests are based on
whiteness tests or uncorrelation tests, depending upon the
identification method used.
* Analysis of the model
This module provides methods for the analysis of the model
in time domain and in the frequency domain, as well as for
the study of the coupling between the various transfers.
Furthermore, this module computes the poles of the
multi-input/multi-output model, and converts the model into
a state space form (observability or controllability
canonical form).
* Simulation
This module allows to simulate the plant outputs using
different signals applied to each input.
DISTRIBUTION KIT
- floppy disks containing
program, help file and example data files,
- user's manual (with theory, operating instructions and
examples).
2 VERSIONS
- Independant complete software under WindowsTM
(compatible with PIMTM, MatlabTM and Program CCTM)
- Toolbox for MatlabTM software (version 4.2 or higher is
needed) containing functions for Model structure estimation,
Parametric identification, and Validation of identified
model.
Available on PC and Workstations
MINIMAL HARDWARE
CONFIGURATION
PC and compatibles
Memory : minimum 1 Mo (2 Mo recommended)
1 hard disk + 1 floppy disk driver
Windows version 3.1 or more
EGA or VGA graphic screen
Mouse (highly recommended)
(Opt.) printer
(Opt.) mathematic coprocessor |
|
|