Design of Reset control by using PWQ Lyapunov function

Jaepil Ban1 Sang Woo Kim2

12Department of Electrical Engineering

Pohang University of Science and Technology (POSTECH)

[email protected]

The 15th IEEE International Conference on Control and Automation

July 19, 2019
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

𝐴𝐴𝑐𝑐
+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
𝑥𝑥𝑐𝑐+
𝐵𝐵𝑐𝑐

2/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
Improves transient response!
Step Response
2

No Reset

+
1.5 Reset at e=0
1
1 𝐴𝐴𝑐𝑐 𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
y

0.5

0
𝑥𝑥0
0 5 10 15 20 25 30 35 40 45 50 𝐵𝐵𝑐𝑐
1

0.5
Control input

-0.5

-1
0 5 10 15 20 25 30 35 40 45 50
Time (s)

2/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
Improves transient response!
Step Response
2

No Reset

+
1.5 Reset at e=0
1
1 𝐴𝐴𝑐𝑐 𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
y

0.5

0
𝑥𝑥0
0 5 10 15 20 25 30 35 40 45 50 𝐵𝐵𝑐𝑐
1

0.5
Control input

-0.5

-1
Less L2 gain (Guanglei Zhao et. al., 2016)
0 5 10 15 20 25
Time (s)
30 35 40 45 50
Phase lead (Yuqian Guo et. al., 2009)

2/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
𝑥𝑥0
𝐵𝐵𝑐𝑐

3/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
▲ Baseline Linear Controller
𝑥𝑥0
 Parameters of baseline linear controller 𝐵𝐵𝑐𝑐

3/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
▲ Baseline Linear Controller
𝑥𝑥0
 Parameters of baseline linear controller 𝐵𝐵𝑐𝑐

▲ Reset law
 Where to reset?
 When to reset?

3/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
▲ Baseline Linear Controller
𝑥𝑥0
 Parameters of baseline linear controller 𝐵𝐵𝑐𝑐

▲ Reset law
 Where to reset? 𝐴𝐴𝑟𝑟 is given (e.g., 0)

 When to reset? When 𝑢𝑢𝑢𝑢 ≥ 0

3/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
▲ Baseline Linear Controller
𝑥𝑥0
 Parameters of baseline linear controller 𝐵𝐵𝑐𝑐

▲ Reset law
 Where to reset? 𝐴𝐴𝑟𝑟 is given (e.g., 0)
1) In practical application, a simple strategy is usually demanded
 When to reset? When 𝑢𝑢𝑢𝑢 ≥ 0

3/18
▣ Reset Controller
𝑟𝑟 𝑒𝑒 𝑢𝑢 𝑦𝑦
ℛ 𝒫𝒫
𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑒𝑒, 𝑢𝑢, 𝑒𝑒 ∉ 𝑅𝑅 +
𝑥𝑥𝑐𝑐+ = 𝐴𝐴𝑟𝑟 𝑥𝑥𝑐𝑐 𝑢𝑢, 𝑒𝑒 ∈ 𝑅𝑅 −
𝑢𝑢𝑐𝑐 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐

▣ What to design? 𝐴𝐴𝑐𝑐

+ 1
𝐶𝐶𝑐𝑐
𝑠𝑠 ⨅
+
▲ Baseline Linear Controller
𝑥𝑥0
 Parameters of baseline linear controller 𝐵𝐵𝑐𝑐

▲ Reset law
 Where to reset? 𝐴𝐴𝑟𝑟 is given (e.g., 0)
1) In practical application, a simple strategy is usually demanded
 When to reset? When 𝑢𝑢𝑢𝑢 ≥ 0 2) Theoretically difficult

3/18
▣ Usual design approach for controller
▲ Design in frequency domain
 Based on describing function
 Bode plot (Guo, Yuqian, et al., IEEE Trans. Control Syst. Technol., 2009)
 Nyquist plot (Van Loon, S. J. L. M., et al., Automatica, 2017)

▲ Design in time domain with linear matrix inequalities (LMIs)

 Based on the stability analysis with Lyapunov stability theorem
 Obtain controller parameters numerically in polynomial time
 Associate with useful performance index such as 𝐻𝐻2 and 𝐻𝐻∞ norm

Goal: Derive LMI-based design condition of reset controller

4/18
▣ Usual design approach for controller
▲ Design in frequency domain
 Based on describing function
 Bode plot (Guo, Yuqian, et al., IEEE Trans. Control Syst. Technol., 2009)
 Nyquist plot (Van Loon, S. J. L. M., et al., Automatica, 2017)

▲ Design in time domain with linear matrix inequalities (LMIs)

 Based on the stability analysis with Lyapunov stability theorem
 Obtain controller parameters numerically in polynomial time
 Associate with useful performance index such as 𝐻𝐻2 and 𝐻𝐻∞ norm

Goal: Derive LMI-based design condition of reset controller

4/18
▣ Usual design approach for controller
▲ Design in frequency domain
 Based on describing function
 Bode plot (Guo, Yuqian, et al., IEEE Trans. Control Syst. Technol., 2009)
 Nyquist plot (Van Loon, S. J. L. M., et al., Automatica, 2017)

▲ Design in time domain with linear matrix inequalities (LMIs)

 Based on the stability analysis with Lyapunov stability theorem
 Obtain controller parameters numerically in polynomial time
 Associate with useful performance index such as 𝐻𝐻2 and 𝐻𝐻∞ norm

Goal: Derive LMI-based design condition of reset controller

4/18
▣ Usual design approach for controller
▲ Design in frequency domain
 Based on describing function
 Bode plot (Guo, Yuqian, et al., IEEE Trans. Control Syst. Technol., 2009)
 Nyquist plot (Van Loon, S. J. L. M., et al., Automatica, 2017)

▲ Design in time domain with linear matrix inequalities (LMIs)

 Based on the stability analysis with Lyapunov stability theorem
 Obtain controller parameters numerically in polynomial time
 Associate with useful performance index such as 𝐻𝐻2 and 𝐻𝐻∞ norm

Goal: Derive LMI-based design condition of reset controller

4/18
▣ Motivation
▣ Background
▲Hybrid dynamical system
▲PWQ Lyapunov function
▲Linear matrix inequalty
▣ LMI-based design of Reset Control
▣ Numerical Example
▣ Conclusion

5/18
▣ Hybrid dynamical system framework
𝑥𝑥̇ ∈ 𝐹𝐹 𝑥𝑥 , 𝑥𝑥 ∈ ℱ
𝑥𝑥 + ∈ 𝐺𝐺 𝑥𝑥 , 𝑥𝑥 ∈ 𝒥𝒥

6/18
▣ Hybrid dynamical system framework
𝑥𝑥̇ ∈ 𝐹𝐹 𝑥𝑥 , 𝑥𝑥 ∈ ℱ
𝑥𝑥 + ∈ 𝐺𝐺 𝑥𝑥 , 𝑥𝑥 ∈ 𝒥𝒥

Bouncing ball example

𝑥𝑥̇ 1 = 𝑥𝑥2 ℱ ≔ 𝑥𝑥|𝑥𝑥1 > 0
𝑥𝑥̇ 2 = −𝑔𝑔
𝑥𝑥1+ = 0 𝒥𝒥 ≔ 𝑥𝑥| 𝑥𝑥1 = 0
𝑥𝑥2+ = −𝑐𝑐𝑥𝑥2

𝑥𝑥1 : Height of the ball

𝑥𝑥2 : Velocity of the ball

6/18
▣ Piecewise quadratic Lyapunov function
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑷𝑷(𝒙𝒙)𝑥𝑥
𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑗𝑗 𝑥𝑥, ∀𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗
𝜃𝜃𝑁𝑁 𝜃𝜃𝑁𝑁−1 𝜃𝜃𝑖𝑖
1
𝑃𝑃𝑁𝑁 𝜃𝜃𝑖𝑖−1
0.8
𝑃𝑃𝑖𝑖
0.6
𝜃𝜃2
0.4
𝑃𝑃0 𝑃𝑃2
𝜃𝜃1
0.2
𝑃𝑃1
𝑥𝑥2 0 𝜃𝜃0
𝑃𝑃0
-0.2

-0.4

-0.6

𝑃𝑃𝑖𝑖
-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

𝑥𝑥1
7/18
▣ A linear matrix inequality (LMI) in the variable 𝑥𝑥 ∈ ℝ𝑛𝑛 has the
form
𝐹𝐹 𝑥𝑥 ≔ 𝐹𝐹0 + 𝑥𝑥1 𝐹𝐹1 + 𝑥𝑥2 𝐹𝐹2 + ⋯ + 𝑥𝑥𝑛𝑛 𝐹𝐹𝑛𝑛 ≽ 0(≼ 0),

where 𝐹𝐹0 ∈ ℝ𝑚𝑚×𝑚𝑚 , … , 𝐹𝐹𝑛𝑛 ∈ ℝ𝑚𝑚×𝑚𝑚 are symmetric matrices.

8/18
▣ A linear matrix inequality (LMI) in the variable 𝑥𝑥 ∈ ℝ𝑛𝑛 has the
form
𝐹𝐹 𝑥𝑥 ≔ 𝐹𝐹0 + 𝑥𝑥1 𝐹𝐹1 + 𝑥𝑥2 𝐹𝐹2 + ⋯ + 𝑥𝑥𝑛𝑛 𝐹𝐹𝑛𝑛 ≽ 0(≼ 0),

where 𝐹𝐹0 ∈ ℝ𝑚𝑚×𝑚𝑚 , … , 𝐹𝐹𝑛𝑛 ∈ ℝ𝑚𝑚×𝑚𝑚 are symmetric matrices.

Example of LMIs (Lyapunov stability)

𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 < 0, 𝑃𝑃 > 0
where, 𝑃𝑃 is matrix variables. (𝐴𝐴 is stable if and only if there exist such 𝑃𝑃.)

8/18
▣ A linear matrix inequality (LMI) in the variable 𝑥𝑥 ∈ ℝ𝑛𝑛 has the
form
𝐹𝐹 𝑥𝑥 ≔ 𝐹𝐹0 + 𝑥𝑥1 𝐹𝐹1 + 𝑥𝑥2 𝐹𝐹2 + ⋯ + 𝑥𝑥𝑛𝑛 𝐹𝐹𝑛𝑛 ≽ 0(≼ 0),

where 𝐹𝐹0 ∈ ℝ𝑚𝑚×𝑚𝑚 , … , 𝐹𝐹𝑛𝑛 ∈ ℝ𝑚𝑚×𝑚𝑚 are symmetric matrices.

Example of LMIs (Lyapunov stability)

𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 < 0, 𝑃𝑃 > 0
where, 𝑃𝑃 is matrix variables. (𝐴𝐴 is stable if and only if there exist such 𝑃𝑃.)
Example of bilinear matrix inequality (Static output feedback control case)
𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 + 𝑷𝑷𝐵𝐵𝑲𝑲𝐶𝐶 + 𝐶𝐶 𝑇𝑇 𝑲𝑲𝑇𝑇 𝐵𝐵𝑇𝑇 𝑷𝑷 < 0, 𝑃𝑃 > 0 where 𝑃𝑃, 𝐾𝐾 are matrix variables

8/18
▣ A linear matrix inequality (LMI) in the variable 𝑥𝑥 ∈ ℝ𝑛𝑛 has the
form
𝐹𝐹 𝑥𝑥 ≔ 𝐹𝐹0 + 𝑥𝑥1 𝐹𝐹1 + 𝑥𝑥2 𝐹𝐹2 + ⋯ + 𝑥𝑥𝑛𝑛 𝐹𝐹𝑛𝑛 ≽ 0(≼ 0),

where 𝐹𝐹0 ∈ ℝ𝑚𝑚×𝑚𝑚 , … , 𝐹𝐹𝑛𝑛 ∈ ℝ𝑚𝑚×𝑚𝑚 are symmetric matrices.

Example of LMIs (Lyapunov stability)

𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 < 0, 𝑃𝑃 > 0
where, 𝑃𝑃 is matrix variables. (𝐴𝐴 is stable if and only if there exist such 𝑃𝑃.)
Example of bilinear matrix inequality (Static output feedback control case)
𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 + 𝑷𝑷𝐵𝐵𝑲𝑲𝐶𝐶 + 𝐶𝐶 𝑇𝑇 𝑲𝑲𝑇𝑇 𝐵𝐵𝑇𝑇 𝑷𝑷 < 0, 𝑃𝑃 > 0 where 𝑃𝑃, 𝐾𝐾 are matrix variables

LMIs can be solved by the semi-definite programming in polynomial time

Useful SDP solvers: cvx (http://cvxr.com/cvx/), MPT toolbox(https://www.mpt3.org/), MATLAB LMI toolbox

8/18
▣ Motivation
▣ Background
▣ LMI-based design of Reset Control
▲Difficulties in deriving LMI-based design condition
▲Our Method
▣ Numerical Example
▣ Conclusion

10/18
▣ Stability condition by using PQLF (Luca Zaccarian et. al., 2005, Nesic et. Al., 2011)
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑷𝑷(𝒙𝒙)𝑥𝑥 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴 ℱ𝑖𝑖 ≔ 𝑥𝑥 𝑥𝑥 𝑇𝑇 𝑀𝑀𝑖𝑖 𝑥𝑥 ≥ 0
𝑉𝑉̇ < 0, 𝑥𝑥 ∈ ℱ 𝑥𝑥 + = 𝐴𝐴𝑟𝑟 𝑥𝑥 𝒥𝒥 ≔ 𝑥𝑥|𝑥𝑥 𝑇𝑇 𝑀𝑀0 𝑥𝑥 ≤ 0
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 ≤ 0, 𝑥𝑥 ∈ 𝒥𝒥 𝐴𝐴 =
𝐴𝐴𝑝𝑝
𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝
𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝑨𝑨𝒄𝒄 𝑀𝑀𝑖𝑖 = Θ𝑖𝑖−1 Θ𝑇𝑇𝑖𝑖 + Θ𝑖𝑖 Θ𝑇𝑇𝑖𝑖−1 , 𝑖𝑖 = 1, … , 𝑁𝑁
𝐼𝐼 0 𝑀𝑀0 = Θ𝑁𝑁 Θ𝑇𝑇0 + Θ0 Θ𝑇𝑇𝑁𝑁
𝐴𝐴𝑟𝑟 = cos 𝜃𝜃𝑖𝑖
0 0 Θ𝑖𝑖 =
𝑇𝑇 sin 𝜃𝜃𝑖𝑖
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇

𝜃𝜃𝑁𝑁 𝜃𝜃𝑁𝑁−1
=0 degrees
1

0.8 𝑃𝑃𝑁𝑁 𝑃𝑃𝑖𝑖

0.6
𝜃𝜃2
0.4

𝑃𝑃0 𝑃𝑃2 𝜃𝜃1

𝑃𝑃1
0.2

0
𝜃𝜃0
-0.2 𝑃𝑃0
-0.4

-0.6

-0.8
𝑃𝑃𝑖𝑖
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

11/18
▣ Stability condition by using PQLF (Luca Zaccarian et. al., 2005, Nesic et. Al., 2011)
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑷𝑷(𝒙𝒙)𝑥𝑥 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴 ℱ𝑖𝑖 ≔ 𝑥𝑥 𝑥𝑥 𝑇𝑇 𝑀𝑀𝑖𝑖 𝑥𝑥 ≥ 0
𝑉𝑉̇ < 0, 𝑥𝑥 ∈ ℱ 𝑥𝑥 + = 𝐴𝐴𝑟𝑟 𝑥𝑥 𝒥𝒥 ≔ 𝑥𝑥|𝑥𝑥 𝑇𝑇 𝑀𝑀0 𝑥𝑥 ≤ 0
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 ≤ 0, 𝑥𝑥 ∈ 𝒥𝒥 𝐴𝐴 =
𝐴𝐴𝑝𝑝
𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝
𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝑨𝑨𝒄𝒄 𝑀𝑀𝑖𝑖 = Θ𝑖𝑖−1 Θ𝑇𝑇𝑖𝑖 + Θ𝑖𝑖 Θ𝑇𝑇𝑖𝑖−1 , 𝑖𝑖 = 1, … , 𝑁𝑁
𝐼𝐼 0 𝑀𝑀0 = Θ𝑁𝑁 Θ𝑇𝑇0 + Θ0 Θ𝑇𝑇𝑁𝑁
⇔ 𝑉𝑉̇ = 𝑥𝑥̇ 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 + 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥̇ < 0, ∀𝑥𝑥 ∈ ℱ𝑖𝑖
𝐴𝐴𝑟𝑟 =
0 0
𝑇𝑇 Θ𝑖𝑖 =
cos 𝜃𝜃𝑖𝑖
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇 sin 𝜃𝜃𝑖𝑖
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 = 𝑥𝑥 + 𝑇𝑇 𝑃𝑃𝑁𝑁 𝑥𝑥 + − 𝑥𝑥 𝑇𝑇 𝑃𝑃0 𝑥𝑥 ≤ 0, ∀𝑥𝑥 ∈ 𝒥𝒥
𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑗𝑗 𝑥𝑥, ∀𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗
𝜃𝜃𝑁𝑁 𝜃𝜃𝑁𝑁−1
=0 degrees
1

0.8 𝑃𝑃𝑁𝑁 𝑃𝑃𝑖𝑖

0.6
𝜃𝜃2
0.4

𝑃𝑃0 𝑃𝑃2 𝜃𝜃1

𝑃𝑃1
0.2

0
𝜃𝜃0
-0.2 𝑃𝑃0
-0.4

-0.6

-0.8
𝑃𝑃𝑖𝑖
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

11/18
▣ Stability condition by using PQLF (Luca Zaccarian et. al., 2005, Nesic et. Al., 2011)
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑷𝑷(𝒙𝒙)𝑥𝑥 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴 ℱ𝑖𝑖 ≔ 𝑥𝑥 𝑥𝑥 𝑇𝑇 𝑀𝑀𝑖𝑖 𝑥𝑥 ≥ 0
𝑉𝑉̇ < 0, 𝑥𝑥 ∈ ℱ 𝑥𝑥 + = 𝐴𝐴𝑟𝑟 𝑥𝑥 𝒥𝒥 ≔ 𝑥𝑥|𝑥𝑥 𝑇𝑇 𝑀𝑀0 𝑥𝑥 ≤ 0
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 ≤ 0, 𝑥𝑥 ∈ 𝒥𝒥 𝐴𝐴 =
𝐴𝐴𝑝𝑝
𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝
𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝑨𝑨𝒄𝒄 𝑀𝑀𝑖𝑖 = Θ𝑖𝑖−1 Θ𝑇𝑇𝑖𝑖 + Θ𝑖𝑖 Θ𝑇𝑇𝑖𝑖−1 , 𝑖𝑖 = 1, … , 𝑁𝑁
𝐼𝐼 0 𝑀𝑀0 = Θ𝑁𝑁 Θ𝑇𝑇0 + Θ0 Θ𝑇𝑇𝑁𝑁
⇔ 𝑉𝑉̇ = 𝑥𝑥̇ 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 + 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥̇ < 0, ∀𝑥𝑥 ∈ ℱ𝑖𝑖
𝐴𝐴𝑟𝑟 =
0 0
𝑇𝑇 Θ𝑖𝑖 =
cos 𝜃𝜃𝑖𝑖
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇 sin 𝜃𝜃𝑖𝑖
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 = 𝑥𝑥 + 𝑇𝑇 𝑃𝑃𝑁𝑁 𝑥𝑥 + − 𝑥𝑥 𝑇𝑇 𝑃𝑃0 𝑥𝑥 ≤ 0, ∀𝑥𝑥 ∈ 𝒥𝒥
𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑗𝑗 𝑥𝑥, ∀𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗
𝜃𝜃𝑁𝑁 𝜃𝜃𝑁𝑁−1
=0 degrees
1

𝑃𝑃𝑁𝑁
𝑥𝑥 𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑖𝑖 + 𝑃𝑃𝑖𝑖 𝐴𝐴 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 𝑥𝑥 < 0, 0.8
𝑃𝑃𝑖𝑖
⇔ 𝑇𝑇 𝑇𝑇 0.6
𝜃𝜃2
𝑥𝑥 𝐴𝐴𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 𝑥𝑥 ≤ 0, 0.4
𝑃𝑃2
𝑃𝑃0 𝜃𝜃1
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, 𝑃𝑃1
0.2

0
𝜃𝜃0
-0.2 𝑃𝑃0
-0.4

-0.6

-0.8
𝑃𝑃𝑖𝑖
-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Θ𝑖𝑖⊥ : the basis of the subspace 𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗

11/18
▣ Stability condition by using PQLF (Luca Zaccarian et. al., 2005, Nesic et. Al., 2011)
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑷𝑷(𝒙𝒙)𝑥𝑥 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴 ℱ𝑖𝑖 ≔ 𝑥𝑥 𝑥𝑥 𝑇𝑇 𝑀𝑀𝑖𝑖 𝑥𝑥 ≥ 0
𝑉𝑉̇ < 0, 𝑥𝑥 ∈ ℱ 𝑥𝑥 + = 𝐴𝐴𝑟𝑟 𝑥𝑥 𝒥𝒥 ≔ 𝑥𝑥|𝑥𝑥 𝑇𝑇 𝑀𝑀0 𝑥𝑥 ≤ 0
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 ≤ 0, 𝑥𝑥 ∈ 𝒥𝒥 𝐴𝐴 =
𝐴𝐴𝑝𝑝
𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝
𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝑨𝑨𝒄𝒄 𝑀𝑀𝑖𝑖 = Θ𝑖𝑖−1 Θ𝑇𝑇𝑖𝑖 + Θ𝑖𝑖 Θ𝑇𝑇𝑖𝑖−1 , 𝑖𝑖 = 1, … , 𝑁𝑁
𝐼𝐼 0 𝑀𝑀0 = Θ𝑁𝑁 Θ𝑇𝑇0 + Θ0 Θ𝑇𝑇𝑁𝑁
⇔ 𝑉𝑉̇ = 𝑥𝑥̇ 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 + 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥̇ < 0, ∀𝑥𝑥 ∈ ℱ𝑖𝑖
𝐴𝐴𝑟𝑟 =
0 0
𝑇𝑇 Θ𝑖𝑖 =
cos 𝜃𝜃𝑖𝑖
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇 sin 𝜃𝜃𝑖𝑖
𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 = 𝑥𝑥 + 𝑇𝑇 𝑃𝑃𝑁𝑁 𝑥𝑥 + − 𝑥𝑥 𝑇𝑇 𝑃𝑃0 𝑥𝑥 ≤ 0, ∀𝑥𝑥 ∈ 𝒥𝒥
𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑗𝑗 𝑥𝑥, ∀𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗
𝜃𝜃𝑁𝑁 𝜃𝜃𝑁𝑁−1
=0 degrees
1

𝑃𝑃𝑁𝑁
𝑥𝑥 𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑖𝑖 + 𝑃𝑃𝑖𝑖 𝐴𝐴 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 𝑥𝑥 < 0, 0.8
𝑃𝑃𝑖𝑖
⇔ 𝑇𝑇 𝑇𝑇 0.6
𝜃𝜃2
𝑥𝑥 𝐴𝐴𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 𝑥𝑥 ≤ 0, 0.4
𝑃𝑃2
𝑃𝑃0 𝜃𝜃1
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, 𝑃𝑃1
0.2

0
𝜃𝜃0
-0.2 𝑃𝑃0
𝐴𝐴𝑇𝑇 𝑃𝑃𝑖𝑖 + 𝑃𝑃𝑖𝑖 𝐴𝐴 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 < 0, LMI-based stability condition -0.4

⇔ 𝑇𝑇
𝐴𝐴𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0,
-0.6

-0.8
𝑃𝑃𝑖𝑖
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, -1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Θ𝑖𝑖⊥ : the basis of the subspace 𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗

11/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇

12/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑉𝑉̇ = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 𝑥𝑥 < 0 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇
⇔ 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 < 0 BMIs

12/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑉𝑉̇ = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 𝑥𝑥 < 0 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇
⇔ 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 < 0 BMIs
Define

⇔ 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 + 𝑍𝑍1𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑍𝑍1 < 0 𝑃𝑃 =

𝑆𝑆
𝑁𝑁 𝑇𝑇
𝑁𝑁
�
𝑋𝑋
, 𝑃𝑃 −1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝑀𝑀
�
𝑌𝑌
, 𝑍𝑍1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝐼𝐼
0
, 𝑍𝑍2 =
𝐼𝐼 𝑆𝑆
0 𝑁𝑁 𝑇𝑇

𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 𝐴𝐴𝑝𝑝 𝑹𝑹 𝐼𝐼

𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 = 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑍𝑍1 = >0
𝑆𝑆𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝑅𝑅 + 𝑆𝑆𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 + 𝑁𝑁𝐴𝐴𝑐𝑐 𝑀𝑀𝑇𝑇 𝑆𝑆𝐴𝐴𝑝𝑝 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝐼𝐼 𝑺𝑺

12/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑉𝑉̇ = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 𝑥𝑥 < 0 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇
⇔ 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 < 0 BMIs
Define

⇔ 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 + 𝑍𝑍1𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑍𝑍1 < 0 𝑃𝑃 =

𝑆𝑆
𝑁𝑁 𝑇𝑇
𝑁𝑁
�
𝑋𝑋
, 𝑃𝑃 −1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝑀𝑀
�
𝑌𝑌
, 𝑍𝑍1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝐼𝐼
0
, 𝑍𝑍2 =
𝐼𝐼 𝑆𝑆
0 𝑁𝑁 𝑇𝑇

𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 𝐴𝐴𝑝𝑝 𝑹𝑹 𝐼𝐼

𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 = 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑍𝑍1 = >0
𝑆𝑆𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝑅𝑅 + 𝑆𝑆𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 + 𝑁𝑁𝐴𝐴𝑐𝑐 𝑀𝑀𝑇𝑇 𝑆𝑆𝐴𝐴𝑝𝑝 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝐼𝐼 𝑺𝑺
Variable replacement LMIs
� 𝒄𝒄
𝐴𝐴𝑝𝑝 𝑹𝑹 + 𝐵𝐵𝑝𝑝 𝑪𝑪 𝐴𝐴𝑝𝑝
=
� 𝒄𝒄
𝑨𝑨 𝑺𝑺𝐴𝐴𝑝𝑝 + 𝑩𝑩� 𝒄𝒄 𝐶𝐶𝑝𝑝

12/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑉𝑉̇ = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 𝑥𝑥 < 0 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇
⇔ 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 < 0 BMIs
Define

⇔ 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 + 𝑍𝑍1𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑍𝑍1 < 0 𝑃𝑃 =

𝑆𝑆
𝑁𝑁 𝑇𝑇
𝑁𝑁
�
𝑋𝑋
, 𝑃𝑃 −1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝑀𝑀
�
𝑌𝑌
, 𝑍𝑍1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝐼𝐼
0
, 𝑍𝑍2 =
𝐼𝐼 𝑆𝑆
0 𝑁𝑁 𝑇𝑇

𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 𝐴𝐴𝑝𝑝 𝑹𝑹 𝐼𝐼

𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 = 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑍𝑍1 = >0
𝑆𝑆𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝑅𝑅 + 𝑆𝑆𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 + 𝑁𝑁𝐴𝐴𝑐𝑐 𝑀𝑀𝑇𝑇 𝑆𝑆𝐴𝐴𝑝𝑝 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝐼𝐼 𝑺𝑺
Variable replacement LMIs
� 𝒄𝒄
𝐴𝐴𝑝𝑝 𝑹𝑹 + 𝐵𝐵𝑝𝑝 𝑪𝑪 𝐴𝐴𝑝𝑝
=
� 𝒄𝒄
𝑨𝑨 𝑺𝑺𝐴𝐴𝑝𝑝 + 𝑩𝑩� 𝒄𝒄 𝐶𝐶𝑝𝑝

Stability condition for reset control

𝐴𝐴𝑇𝑇 𝑃𝑃𝑖𝑖 + 𝑃𝑃𝑖𝑖 𝐴𝐴 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 < 0,
𝐴𝐴𝑇𝑇𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0,
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0,

12/18
▣ LMI-based design method for linear controller (Carsten W. Scherer et. al., 1997)
𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃, 𝑃𝑃 > 0. 𝐴𝐴𝑝𝑝 𝐵𝐵𝑝𝑝 𝑪𝑪𝒄𝒄
𝐴𝐴 =
𝑉𝑉̇ = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 𝑥𝑥 < 0 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄
𝑇𝑇
𝑥𝑥 = 𝑥𝑥𝑝𝑝𝑇𝑇 𝑥𝑥𝑐𝑐𝑇𝑇
⇔ 𝑃𝑃𝑃𝑃 + 𝐴𝐴𝑇𝑇 𝑃𝑃 < 0 BMIs
Define

⇔ 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 + 𝑍𝑍1𝑇𝑇 𝐴𝐴𝑇𝑇 𝑃𝑃𝑍𝑍1 < 0 𝑃𝑃 =

𝑆𝑆
𝑁𝑁 𝑇𝑇
𝑁𝑁
�
𝑋𝑋
, 𝑃𝑃 −1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝑀𝑀
�
𝑌𝑌
, 𝑍𝑍1 =
𝑅𝑅
𝑀𝑀𝑇𝑇
𝐼𝐼
0
, 𝑍𝑍2 =
𝐼𝐼 𝑆𝑆
0 𝑁𝑁 𝑇𝑇

𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 𝐴𝐴𝑝𝑝 𝑹𝑹 𝐼𝐼

𝑍𝑍1𝑇𝑇 𝑃𝑃𝑃𝑃𝑍𝑍1 = 𝑍𝑍1𝑇𝑇 𝑃𝑃𝑍𝑍1 = >0
𝑆𝑆𝐴𝐴𝑝𝑝 𝑅𝑅 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝑅𝑅 + 𝑆𝑆𝐵𝐵𝑝𝑝 𝐶𝐶𝑐𝑐 𝑀𝑀𝑇𝑇 + 𝑁𝑁𝐴𝐴𝑐𝑐 𝑀𝑀𝑇𝑇 𝑆𝑆𝐴𝐴𝑝𝑝 + 𝑁𝑁𝐵𝐵𝑐𝑐 𝐶𝐶𝑝𝑝 𝐼𝐼 𝑺𝑺
Variable replacement LMIs
� 𝒄𝒄
𝐴𝐴𝑝𝑝 𝑹𝑹 + 𝐵𝐵𝑝𝑝 𝑪𝑪 𝐴𝐴𝑝𝑝
=
� 𝒄𝒄
𝑨𝑨 𝑺𝑺𝐴𝐴𝑝𝑝 + 𝑩𝑩� 𝒄𝒄 𝐶𝐶𝑝𝑝
𝑀𝑀𝑖𝑖 is indefinite matrix
Relaxing these BMIs into LMIs is extremely difficult
Stability condition for reset control
𝑇𝑇 𝑇𝑇 𝑇𝑇
𝐴𝐴𝑇𝑇 𝑃𝑃𝑖𝑖 + 𝑃𝑃𝑖𝑖 𝐴𝐴 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 < 0, ⇔ 𝑍𝑍1𝑖𝑖 𝑃𝑃𝑖𝑖 𝐴𝐴𝑍𝑍1𝑖𝑖 + 𝑍𝑍1𝑖𝑖 𝐴𝐴 𝑃𝑃𝑖𝑖 𝑍𝑍1𝑖𝑖 + 𝜏𝜏𝐹𝐹𝑖𝑖 𝒁𝒁𝑻𝑻𝟏𝟏𝑖𝑖 𝑴𝑴𝒊𝒊 𝒁𝒁𝟏𝟏𝑖𝑖 < 0
𝐴𝐴𝑇𝑇𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0,
𝑇𝑇 𝑇𝑇
𝑍𝑍10 𝑇𝑇
𝐴𝐴𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 𝑍𝑍10 − 𝑍𝑍10 𝑃𝑃0 𝑍𝑍10 + 𝜏𝜏𝑅𝑅 𝒁𝒁𝑻𝑻𝟏𝟏𝟏𝟏 𝑴𝑴𝟎𝟎 𝒁𝒁𝟏𝟏𝟏𝟏 ≤ 0,
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, 𝜣𝜣𝑻𝑻𝒊𝒊⊥ 𝑷𝑷𝒊𝒊 − 𝑷𝑷𝒊𝒊−𝟏𝟏 𝜣𝜣𝒊𝒊⊥ = 𝟎𝟎

12/18
▣ Descriptor system approach
Plant dynamics Controller dynamics
𝑥𝑥̇ 𝑝𝑝 = 𝐴𝐴𝑝𝑝 𝑥𝑥𝑝𝑝 + 𝐵𝐵𝑝𝑝 𝑢𝑢 𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑦𝑦 𝑥𝑥𝑐𝑐+ = 0
𝑦𝑦 = 𝐶𝐶𝑝𝑝 𝑥𝑥𝑝𝑝 𝑢𝑢 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
0 � 𝑢𝑢̇ = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐 − 𝑢𝑢
𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝
0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐 ⇔ 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝐸𝐸 Beautifully arranged for the use of variable replacement technique!
𝐴𝐴

13/18
▣ Descriptor system approach
Plant dynamics Controller dynamics
𝑥𝑥̇ 𝑝𝑝 = 𝐴𝐴𝑝𝑝 𝑥𝑥𝑝𝑝 + 𝐵𝐵𝑝𝑝 𝑢𝑢 𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑦𝑦 𝑥𝑥𝑐𝑐+ = 0
𝑦𝑦 = 𝐶𝐶𝑝𝑝 𝑥𝑥𝑝𝑝 𝑢𝑢 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
0 � 𝑢𝑢̇ = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐 − 𝑢𝑢
𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝
0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐 ⇔ 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝐸𝐸 Beautifully arranged for the use of variable replacement technique!
𝐴𝐴
▣ Stability Analysis
𝑇𝑇 𝑃𝑃 0
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝐸𝐸 𝑃𝑃�𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃�𝑖𝑖 𝐸𝐸𝐸𝐸, 𝑥𝑥 ∈ ℱ𝑖𝑖 𝑃𝑃�𝑖𝑖 = 𝑖𝑖
𝑋𝑋 𝑌𝑌

13/18
▣ Descriptor system approach
Plant dynamics Controller dynamics
𝑥𝑥̇ 𝑝𝑝 = 𝐴𝐴𝑝𝑝 𝑥𝑥𝑝𝑝 + 𝐵𝐵𝑝𝑝 𝑢𝑢 𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑦𝑦 𝑥𝑥𝑐𝑐+ = 0
𝑦𝑦 = 𝐶𝐶𝑝𝑝 𝑥𝑥𝑝𝑝 𝑢𝑢 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
0 � 𝑢𝑢̇ = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐 − 𝑢𝑢
𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝
0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐 ⇔ 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝐸𝐸 Beautifully arranged for the use of variable replacement technique!
𝐴𝐴
▣ Stability Analysis
𝑇𝑇 𝑃𝑃 0
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝐸𝐸 𝑃𝑃�𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃�𝑖𝑖 𝐸𝐸𝐸𝐸, 𝑥𝑥 ∈ ℱ𝑖𝑖 𝑃𝑃�𝑖𝑖 = 𝑖𝑖
𝑋𝑋 𝑌𝑌
𝑉𝑉̇ = 𝑥𝑥̇ 𝑇𝑇 𝐸𝐸𝑃𝑃𝑖𝑖 𝑥𝑥 + 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖𝑇𝑇 𝐸𝐸𝑥𝑥̇ < 0, 𝑥𝑥 ∈ ℱ𝑖𝑖
𝑇𝑇 0 𝑃𝑃𝑖𝑖
⇔ 𝑉𝑉̇ = 𝐸𝐸𝑥𝑥𝑥𝑥̇ 𝑃𝑃𝑇𝑇 𝜏𝜏 𝑀𝑀 𝐸𝐸𝑥𝑥𝑥𝑥̇ < 0 By using S-procedure
𝐹𝐹𝑖𝑖 𝑖𝑖
𝑇𝑇 0
𝑖𝑖
=0 =0 𝑇𝑇
𝐸𝐸 𝑥𝑥̇ 𝑃𝑃𝑖𝑖 𝐸𝐸 𝑥𝑥̇ 𝐸𝐸 𝑥𝑥̇ 𝑇𝑇
𝐸𝐸𝑥𝑥 𝐸𝐸 𝑥𝑥̇ 𝑇𝑇
𝐸𝐸𝑥𝑥
⇔ 𝑉𝑉̇ = 𝑇𝑇 + 𝐺𝐺𝑖𝑖 −𝐼𝐼 𝐴𝐴 + 𝐺𝐺𝑖𝑖 −𝐼𝐼 𝐴𝐴 <0
𝑥𝑥 𝑃𝑃𝑖𝑖 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 𝑥𝑥 𝑥𝑥 𝑥𝑥 𝑥𝑥 𝑥𝑥

13/18
▣ Descriptor system approach
Plant dynamics Controller dynamics
𝑥𝑥̇ 𝑝𝑝 = 𝐴𝐴𝑝𝑝 𝑥𝑥𝑝𝑝 + 𝐵𝐵𝑝𝑝 𝑢𝑢 𝑥𝑥̇ 𝑐𝑐 = 𝐴𝐴𝑐𝑐 𝑥𝑥𝑐𝑐 + 𝐵𝐵𝑐𝑐 𝑦𝑦 𝑥𝑥𝑐𝑐+ = 0
𝑦𝑦 = 𝐶𝐶𝑝𝑝 𝑥𝑥𝑝𝑝 𝑢𝑢 = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐
0 � 𝑢𝑢̇ = 𝐶𝐶𝑐𝑐 𝑥𝑥𝑐𝑐 − 𝑢𝑢
𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝
0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐 ⇔ 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝐸𝐸 Beautifully arranged for the use of variable replacement technique!
𝐴𝐴
▣ Stability Analysis
𝑇𝑇 𝑃𝑃 0
𝑉𝑉 = 𝑥𝑥 𝑇𝑇 𝐸𝐸 𝑃𝑃�𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃�𝑖𝑖 𝐸𝐸𝐸𝐸, 𝑥𝑥 ∈ ℱ𝑖𝑖 𝑃𝑃�𝑖𝑖 = 𝑖𝑖
𝑋𝑋 𝑌𝑌
𝑉𝑉̇ = 𝑥𝑥̇ 𝑇𝑇 𝐸𝐸𝑃𝑃𝑖𝑖 𝑥𝑥 + 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖𝑇𝑇 𝐸𝐸𝑥𝑥̇ < 0, 𝑥𝑥 ∈ ℱ𝑖𝑖
𝑇𝑇 0 𝑃𝑃𝑖𝑖
⇔ 𝑉𝑉̇ = 𝐸𝐸𝑥𝑥𝑥𝑥̇ 𝑃𝑃𝑇𝑇 𝜏𝜏 𝑀𝑀 𝐸𝐸𝑥𝑥𝑥𝑥̇ < 0 By using S-procedure
𝐹𝐹𝑖𝑖 𝑖𝑖
𝑇𝑇 0
𝑖𝑖
=0 =0 𝑇𝑇
𝐸𝐸 𝑥𝑥̇ 𝑃𝑃𝑖𝑖 𝐸𝐸 𝑥𝑥̇ 𝐸𝐸 𝑥𝑥̇ 𝑇𝑇
𝐸𝐸𝑥𝑥 𝐸𝐸 𝑥𝑥̇ 𝑇𝑇
𝐸𝐸𝑥𝑥
⇔ 𝑉𝑉̇ = 𝑇𝑇 + 𝐺𝐺𝑖𝑖 −𝐼𝐼 𝐴𝐴 + 𝐺𝐺𝑖𝑖 −𝐼𝐼 𝐴𝐴 <0
𝑥𝑥 𝑃𝑃𝑖𝑖 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 𝑥𝑥 𝑥𝑥 𝑥𝑥 𝑥𝑥 𝑥𝑥

⇔
0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑇𝑇𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0, 𝑉𝑉 𝑥𝑥 + − 𝑉𝑉 𝑥𝑥 ≤ 0, ∀𝑥𝑥 ∈ 𝒥𝒥
+ 𝐻𝐻𝐻𝐻 𝐺𝐺𝑖𝑖 −𝐼𝐼 𝐴𝐴 <0
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑖𝑖 𝑥𝑥 = 𝑥𝑥 𝑇𝑇 𝑃𝑃𝑗𝑗 𝑥𝑥, ∀𝑥𝑥 ∈ 𝑋𝑋𝑖𝑖 ∩ 𝑋𝑋𝑗𝑗
Theorem 1
13/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝐸𝐸 𝐴𝐴
𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

𝑅𝑅1 𝑅𝑅2 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝

𝐺𝐺𝑖𝑖 𝐴𝐴 = Ω𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0
𝑅𝑅1 𝑅𝑅2 0 𝑪𝑪𝒄𝒄 −𝐼𝐼

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

𝑅𝑅1 𝑅𝑅2 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 Nonlinear terms

𝐺𝐺𝑖𝑖 𝐴𝐴 = Ω𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑅𝑅2 𝐴𝐴𝑐𝑐 , 𝑅𝑅1 𝐵𝐵𝑐𝑐 , 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝑅𝑅1 𝑅𝑅2 0 𝑪𝑪𝒄𝒄 −𝐼𝐼

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

𝑅𝑅1 𝑅𝑅2 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 Nonlinear terms

𝐺𝐺𝑖𝑖 𝐴𝐴 = Ω𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑅𝑅2 𝐴𝐴𝑐𝑐 , 𝑅𝑅1 𝐵𝐵𝑐𝑐 , 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝑅𝑅1 𝑅𝑅2 0 𝑪𝑪𝒄𝒄 −𝐼𝐼
Replacing
𝐴𝐴̂ 𝑐𝑐 = 𝑅𝑅2 𝐴𝐴𝑐𝑐
𝐵𝐵�𝑐𝑐 = 𝑅𝑅1 𝐵𝐵𝑐𝑐
𝐶𝐶̂𝑐𝑐 = 𝑅𝑅2 𝐶𝐶𝑐𝑐

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

𝑅𝑅1 𝑅𝑅2 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 Nonlinear terms

𝐺𝐺𝑖𝑖 𝐴𝐴 = Ω𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑅𝑅2 𝐴𝐴𝑐𝑐 , 𝑅𝑅1 𝐵𝐵𝑐𝑐 , 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝑅𝑅1 𝑅𝑅2 0 𝑪𝑪𝒄𝒄 −𝐼𝐼
Theorem 2
Replacing 0 𝑃𝑃𝑖𝑖 It is now LMIs
𝐴𝐴̂ 𝑐𝑐 = 𝑅𝑅2 𝐴𝐴𝑐𝑐 + 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 Ω𝑖𝑖 <0
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖
𝐵𝐵�𝑐𝑐 = 𝑅𝑅1 𝐵𝐵𝑐𝑐 where 𝐻𝐻𝐻𝐻 𝐴𝐴 ≔ 𝐴𝐴 + 𝐴𝐴𝑇𝑇
𝐶𝐶̂𝑐𝑐 = 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝐴𝐴𝑇𝑇𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0,
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0,

14/18
▣ Design condition of reset control based on the stability analysis

0 𝑃𝑃𝑖𝑖 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 𝑥𝑥𝑝𝑝

+ 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 𝑮𝑮𝒊𝒊 𝑨𝑨 <0 𝐼𝐼 0 0 𝑥𝑥̇ 𝑝𝑝
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 0 𝐼𝐼 0 𝑥𝑥̇ 𝑐𝑐 = 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑥𝑥𝑐𝑐
0 0 0 𝑢𝑢̇ 0 𝑪𝑪𝒄𝒄 −𝐼𝐼 𝑢𝑢
𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝐴𝐴
Choose 𝐺𝐺𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝐸𝐸 𝑥𝑥̇ = 𝐴𝐴𝐴𝐴
𝑅𝑅1 𝑅𝑅2

𝑅𝑅1 𝑅𝑅2 𝐴𝐴𝑝𝑝 0 𝐵𝐵𝑝𝑝 Nonlinear terms

𝐺𝐺𝑖𝑖 𝐴𝐴 = Ω𝑖𝑖 = 𝐺𝐺1𝑖𝑖 𝑅𝑅1 𝑅𝑅2 𝑩𝑩𝒄𝒄 𝐶𝐶𝑝𝑝 𝑨𝑨𝒄𝒄 0 𝑅𝑅2 𝐴𝐴𝑐𝑐 , 𝑅𝑅1 𝐵𝐵𝑐𝑐 , 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝑅𝑅1 𝑅𝑅2 0 𝑪𝑪𝒄𝒄 −𝐼𝐼
Theorem 2
Replacing 0 𝑃𝑃𝑖𝑖 It is now LMIs
𝐴𝐴̂ 𝑐𝑐 = 𝑅𝑅2 𝐴𝐴𝑐𝑐 + 𝐻𝐻𝐻𝐻 −𝐺𝐺𝑖𝑖 Ω𝑖𝑖 <0
𝑃𝑃𝑖𝑖𝑇𝑇 𝜏𝜏𝐹𝐹𝑖𝑖 𝑀𝑀𝑖𝑖 After solving LMIs,
𝐵𝐵�𝑐𝑐 = 𝑅𝑅1 𝐵𝐵𝑐𝑐 where 𝐻𝐻𝐻𝐻 𝐴𝐴 ≔ 𝐴𝐴 + 𝐴𝐴𝑇𝑇 𝑅𝑅2−1 𝐴𝐴̂ 𝑐𝑐 = 𝐴𝐴𝑐𝑐
𝐶𝐶̂𝑐𝑐 = 𝑅𝑅2 𝐶𝐶𝑐𝑐
𝐴𝐴𝑇𝑇𝑟𝑟 𝑃𝑃𝑁𝑁 𝐴𝐴𝑟𝑟 − 𝑃𝑃0 + 𝜏𝜏𝑅𝑅 𝑀𝑀0 ≤ 0, 𝑅𝑅1−1 𝐵𝐵�𝑐𝑐 = 𝐵𝐵𝑐𝑐
Θ𝑇𝑇𝑖𝑖⊥ 𝑃𝑃𝑖𝑖 − 𝑃𝑃𝑖𝑖−1 Θ𝑖𝑖⊥ = 0, 𝑅𝑅2−1 𝐶𝐶̂𝑐𝑐 = 𝐶𝐶𝑐𝑐

14/18
▣ Motivation
▣ Background
▣ LMI-based design of Reset Control
▣ Numerical Example
▣ Conclusion

15/18
4th order linear plant 𝑑𝑑
8𝑠𝑠 2 + 18𝑠𝑠 + 32 + 𝑢𝑢
𝑃𝑃 𝑠𝑠 = 4 𝑟𝑟 𝑒𝑒 𝑦𝑦
𝑠𝑠 + 6𝑠𝑠 3 + 14𝑠𝑠 2 + 24𝑠𝑠 ℛ 𝒫𝒫
+ +
−

By solving the optimization problem

min 𝛾𝛾
subject to
LMIs derived for 𝐻𝐻∞ control
𝑦𝑦 2
where, sup ≤ 𝛾𝛾
𝑑𝑑 2

16/18
4th order linear plant 𝑑𝑑
8𝑠𝑠 2 + 18𝑠𝑠 + 32 + 𝑢𝑢
𝑃𝑃 𝑠𝑠 = 4 𝑟𝑟 𝑒𝑒 𝑦𝑦
𝑠𝑠 + 6𝑠𝑠 3 + 14𝑠𝑠 2 + 24𝑠𝑠 ℛ 𝒫𝒫
+ +
−

By solving the optimization problem

min 𝛾𝛾
subject to
LMIs derived for 𝐻𝐻∞ control
𝑦𝑦 2
where, sup ≤ 𝛾𝛾
𝑑𝑑 2

𝐶𝐶𝑐𝑐 = 0 0 0 1

16/18
4th order linear plant 𝑑𝑑
8𝑠𝑠 2 + 18𝑠𝑠 + 32 + 𝑢𝑢
𝑃𝑃 𝑠𝑠 = 4 𝑟𝑟 𝑒𝑒 𝑦𝑦
𝑠𝑠 + 6𝑠𝑠 3 + 14𝑠𝑠 2 + 24𝑠𝑠 ℛ 𝒫𝒫
+ +
−

By solving the optimization problem

min 𝛾𝛾
3.5

subject to
3
3

LMIs derived for 𝐻𝐻∞ control

2.5
2

disturbance input
2

1
1.5
𝑦𝑦 2
where, sup ≤ 𝛾𝛾
𝑑𝑑 2 1 0
0 0.05 0.1 0.15 0.2
0.5

0
0 2 4 6 8 10
Time (s)

0 ≤ 𝑡𝑡 < 0.1
𝑑𝑑 = � 10
𝐶𝐶𝑐𝑐 = 0 0 0 1 0 0.1 ≤ 𝑡𝑡

16/18
 0.5
Linear Control
Reset Control
Output

-0.5
0 2 4 6 8 10
Time (s)

0.5

0
Input

-0.5

-1
0 2 4 6 8 10
Time (s)

17/18
 0.5
Linear Control
Reset Control
Output

-0.5
0 2 4 6 8 10
Time (s)

0.5 Table. Comparison of ℒ2 gains

𝑦𝑦 2
0 sup
𝑑𝑑 2
Input

-0.5 Linear control 0.5370

Reset Control 0.3833

-1
0 2 4 6 8 10
28.61% decreased by reset control
Time (s)

17/18
▣ LMI-based design of reset control is addressed

▣ Previous stability condition yields BMI design condition

▣ Formulated closed-loop system as a descriptor hybrid system

▣ Proposed two theorems

▲ The stability analysis of descriptor hybrid system
▲ LMI-based design of reset control system

▣ Numerical example showed the feasibility of the design strategy

18/18
Thank you for listening!
Q&A
▣ A linear matrix inequality (LMI) in the variable 𝑥𝑥 ∈ ℝ𝑛𝑛 has the
form
𝐹𝐹 𝑥𝑥 ≔ 𝐹𝐹0 + 𝑥𝑥1 𝐹𝐹1 + 𝑥𝑥2 𝐹𝐹2 + ⋯ + 𝑥𝑥𝑛𝑛 𝐹𝐹𝑛𝑛 ≽ 0,

where 𝐹𝐹0 ∈ ℝ𝑚𝑚×𝑚𝑚 , … , 𝐹𝐹𝑛𝑛 ∈ ℝ𝑚𝑚×𝑚𝑚 are symmetric matrices.

Example of LMIs (Lyapunov stability)

𝐴𝐴𝑇𝑇 𝑷𝑷 + 𝑷𝑷𝐴𝐴 < 0, 𝑃𝑃 > 0
where, 𝑃𝑃 is matrix variables. (𝐴𝐴 is stable if and only if there exist such 𝑃𝑃.)
𝑝𝑝1 𝑝𝑝2 𝑎𝑎11 𝑎𝑎12
For 𝑃𝑃 = 𝑝𝑝 𝑝𝑝3 , 𝐴𝐴 = 𝑎𝑎21 𝑎𝑎22 ,
2

2𝑎𝑎11 𝑎𝑎12 2𝑎𝑎21 𝑎𝑎11 + 𝑎𝑎22 0 𝑎𝑎21

𝑝𝑝1 + 𝑝𝑝2 + 𝑝𝑝3 <0
𝑎𝑎12 0 𝑎𝑎11 + 𝑎𝑎22 2𝑎𝑎12 𝑎𝑎21 2𝑎𝑎22
1 0 0 1 0 0
𝑝𝑝1 + 𝑝𝑝2 + 𝑝𝑝3 >0
0 0 1 0 0 1
▣ References
▲ Advantages of Reset control
 Zhao, Guanglei, and Jingcheng Wang. "On L2 gain performance improvement of linear systems with Lyapunov-based
reset control." Nonlinear Analysis: Hybrid Systems 21 (2016): 105-117.

▲ Design of reset control based on Frequency domain

 Guo, Yuqian, Youyi Wang, and Lihua Xie. "Frequency-domain properties of reset systems with application in hard-disk-
drive systems." IEEE Transactions on Control Systems Technology17.6 (2009): 1446-1453.
 Van Loon, S. J. L. M., et al. "Frequency-domain tools for stability analysis of reset control systems." Automatica 82
(2017): 101-108.
▲ LMI-based stability analysis of Reset control systems
 Zaccarian, Luca, Dragan Nesic, and Andrew R. Teel. "First order reset elements and the Clegg integrator revisited."
Proceedings of the 2005, American Control Conference, 2005.. IEEE, 2005.
 Aangenent, W. H. T. M., et al. "Performance analysis of reset control systems." International Journal of Robust and
Nonlinear Control 20.11 (2010): 1213-1233.
 Nesic, Dragan, Andrew R. Teel, and Luca Zaccarian. "Stability and performance of SISO control systems with first-order
reset elements." IEEE Transactions on Automatic Control 56.11 (2011): 2567-2582.
▲ LMI-based design of Dynamic output feedback control
 Scherer, Carsten, Pascal Gahinet, and Mahmoud Chilali. "Multiobjective output-feedback control via LMI optimization."
IEEE Transactions on automatic control 42.7 (1997): 896-911.