Sem.

2
Introduction to robust control 2019/2020

Formulation of the H∞ problem

 Consider the plant model with feedback as shown below

 Here, Δ(s) represents plant parameter uncertainty, P(s) represents the plant
nominal transfer function and K(s) represents feedback control
 One of the objectives is to make the system insensitive to external disturbances.
This is equivalent to make z as independent of w as possible
 If the plant perturbation Δ(s) is ignored, then the model simplifies to

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 13

 Sem. 2
Introduction to robust control 2019/2020

 With the feedback law u=K(s)y, u and y can be eliminated resulting to

 To minimize the error z due to the external inputs w, the function Fl(P, K) must be
minimized

Mixed performance and robustness objective

 The following set of characteristics are possible:
‒ Achievement of good disturbance rejection from external signals in the low-
frequency region. ➠ This can be achieved by making the sensitivity S = (I + PK)-1
small as ω → 0
‒ Make the closed loop transfer function small at high frequencies limit excitation by
noise. ➠This can be achieved by making T = I - S =I - (I + PK)-1 small as ω → ∞.
‒ Guard against instability from parameter variations. This is achieved by
minimizing K(I + PK)-1
 Then the H ∞ problem is formulated as the minimization of the function

 Where W1 and W3 are frequency-dependent matrices

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 14

 Sem. 2
Introduction to robust control 2019/2020

Glover-Doyle Algorithm in H∞ formulation

 This algorithm solves a family of stabilizing controllers such that

 The aim is to find the lowest value of for which the above equation has a solution
 One possibility is to start with the LQG solution and then to reduce it using a binary
search
 The plant equations in state-space form is

 In a packed matrix form the above equations are represented as

 The following assumptions must be satisfied to ensure a solution

‒ The pair (A, B2) must be stabilizable and the pair (C2, A) detectable
‒ With the dimensions of dim x = n, dim w = m1, dim u = m2, dim z =p1 and dim y = p2,
then the Rank D12= m2 and Rank D21=p2 to ensure that they controllers are proper
and the transfer function from w to y is non-zero at high frequencies (i.e. all-pass)

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 15

 Sem. 2
Introduction to robust control 2019/2020

‒ Rank for all frequencies

‒ Rank for all frequencies

‒ D11=0 and D22=0 simplifies the equations and implies that the transfer functions
from u to y and from w to z rolls off at high frequency.
 So, the problem simplifies to

Solution to H∞ problem
 This requires the solving of two Ricatti equation, one for the controller and the
other for the observer
 The control law is given by

 The state estimator equation

 Where,

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 16

 Sem. 2
Introduction to robust control 2019/2020

 The controller gain is Kc as for the LQG case, and the estimator gain is Z∞ke instead
of Ke as for the LQG case, with

 And

 The terms X∞ and Y∞ are solutions to the controller and estimator Ricatti equations

 Note: These calculations are not carried out by hand-MATLAB robust control
toolbox is used instead.

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 17

 Sem. 2
Introduction to robust control 2019/2020

Properties of H∞ controller
 The stabilizing feedback law u(s)=Ky2 minimizes the norm of the closed-loop transfer
function

‒ Optimal H2 control: min||Tyu||2

‒ Optimal H∞ control: min||Tyu||∞
‒ Standard H∞ control:
 All H∞ cost function Tyu is all-pass i.e., for all values of ɷ
 The H∞ optimal controller for an n-state augmented plant have at most n-1 states
 The H∞ sub-optimal controller for an n-state augmented plant have exactly n states
 In the weighted mixed sensitivity problem formulation, the H∞ controller always
cancels the stable poles of the plant with its transmission zeroes
 In the weighted mixed sensitivity problem formulation, the unstable poles of the
plant inside the specified bandwidth will be shifted to its mirror image once a H∞ or
H2 feedback loop is closed

Weighted mixed sensitivity problem

 This technique allows very precise frequency-domain loop shaping via suitable
weighting strategies
 If you augment the plant with frequency dependent weights W1 to W, then the
MATLAB script hinf or newer hinfsyn or mixsyn will find a controller that ”shapes”
the signals to the inverse of these weights, if it exists.

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 18

 Sem. 2
Introduction to robust control 2019/2020

 The MATLAB function augw.m forms the augmented plan

Example
 Consider the case of the double integrator

 This plant violates the rules for a solution (poles on imaginary axis)
 The equation set, in state space form, with the addition of a ”disturbance” term
representing uncertainty d, is
Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 19
 Sem. 2
Introduction to robust control 2019/2020

Example (cont’)  Corresponding equations

 The equation set, in state space form,
with the addition of a ”disturbance”
term representing uncertainty d, is

 The regulated output (note the

inclusion of the control signal to bound
it) given by  In standard format, the block diagrams
of the double integrator is shown below

 The measurement equation

 The ”noise” term n may include

measurement errors or unmodelled
high
 Note: The rank condition of D21 must be
met

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 20

 Sem. 2
Introduction to robust control 2019/2020

Example (cont’)  The solution to this problem by

 In H∞ format, the block diagrams of the computer (pre-shifting the poles at the
double integrator is shown below origin and post-shifting the controller
poles back) gives

 With the closed loop poles at

 Collecting the equations in packed

matrix form

Dr.-Ing. Jackson G. Njiri EMT 3104 L-5: Introduction to robust control 21