|
The talk will outline the principles of a
model/observer-based methodology for designing and implementing digital
control units (DCU). Control algorithms and devices are developed around a
discrete-time (DT) state equation model of the plant to be controlled: the
embedded model (EM). The EM consists of a pair of interweaved dynamics: the
controllable dynamics (commands-to-measures) and the observable disturbance
dynamics. The latter must be driven by unpredictable signals (the driving
noise).
The essential DCU is made by three real-time algorithms:
1) the measurement law estimating the last occurred driving noise, 2) the
trajectory generator translating higher-level commands like operator requests
into reference trajectories for each controllable state, 3) the command law
combining trajectory errors and disturbance rejection into commands.
The main design principles are:
1) one-to-one
relation from EM to control algorithms: degrees-of-freedom in the MIMO case
are suppressed by decomposition,
2) feedback
parsimony: plant measurements only serve to estimate the last occurred
driving noise,
3) feedback
stability: robust stability and performance are guaranteed by forcing the
measurement law to discriminate unmodelled dynamics from driving noise,
4) performance
prediction: performance are predicted against a fine plant simulator and then
in-field refined, if necessary.
The methodology has been tested on different
applications. As an introductory example model and control of a Mars landing
vehicle will be treated.
|
