[PPT] - Holistic Control for Wireless Control Systems Yehan Ma CSE 521S PowerPoint Presentation

SLIDE 1

Holistic Control for Wireless Control Systems

Yehan Ma CSE 521S

SLIDE 2

Industrial Process Automation

1

Ø Large Scale Ø Relative low speed Ø Challenging environment

q Physical plant q Network communication

Ø Stability is critical

q Health, Safety, and Environment (HSE)

Ø Automation à Industrial 4.0

Courtesy: Emerson Process Management

SLIDE 3

Controller State Observer Actuators Plant Sensors Reference ( ) u t ˆ( ) u t

( ) y t

ˆ( ) y t

ˆ( ) x t pr pressure va valve ve sensor data Sensor Actuator control command Controller

Cyber-Physical Dependability

Wireless Interference Physical Disturbance

Dependable control requires

control performance
resiliency under interference
energy efficiency of wireless network

Most of today’s industrial wireless networks are for monitoring

2

SLIDE 4

Control Backgrounds

3

𝑦(𝑢) ref ref

Ø Traditional controller design: PID, optimal control law

q Guarantee control performance

Stability

Stability index: Lyapunov function 𝑊 𝑦(𝑢) 𝑊 𝑦(𝑢) ≥ 0, 𝑊̇ 𝑦(𝑢) ≤ 0àSystem is stable

Regulation error

Regulation error index: Integral Absolute Error 𝐽𝐵𝐹 = ∫ |𝑠𝑓𝑔 − 𝑦(𝑢)|𝑒𝑢

67 8

SLIDE 5

Current Industrial Control

Ø Wireless: use redundancy to reduce data loss Ø Control: tolerate data loss and physical disturbances Ø Operate in isolation!

Network Manager

Reconfiguration Signals Performance Measurements

Wireless Sensor Network

Wireless Interference Physical Disturbance

4

SLIDE 6

Holistic Control

Ø Close the loop between control and network Ø Holistic controller manages both the physical plant and network configurations based on states of the plant and the network Ø Improve control performance while reducing energy cost in spite

f cyber and physical interference

5

Controller Plant

Reconfiguration Signals Acks Control Inputs Plant Outputs

Wireless Sensor Network Network Manager Holistic Controller Plant

Reconfiguration Signals Acks

Wireless Sensor Network Network Manager Holistic Controller Plant

Reconfiguration Signals Acks Plant Outputs

Wireless Sensor Network Network Manager

Network State

Holistic Controller Plant

Reconfiguration Signals Acks Control Inputs Plant Outputs

Wireless Sensor Network Network Manager

Network State Network Configuration

SLIDE 7

Holistic Control Framework

12/3/19

6

Controller Physical State Observer Actuators Sensors

t

u

t

y ˆt y ˆt x

Plant

ˆt u

Controller Physical State Observer Actuators Sensors Plant Holistic Controller Physical State Observer Actuators Sensors Plant

R or Tn 3 4

Physical Disturbance Wireless Interference

Network Reconfiguration Signal 2

Ø Holistic controllers

q monitor control performance q compute (1) network configurations and (2) control commands

Ø Network

q transmits control commands q reconfigures itself when needed

Ø Wireless control systems with enhanced resiliency and efficiency!

1

SLIDE 8

Holistic Control Strategies

Ø Reconfigure wireless networks in response to system states

q Adapting number of transmissions q Adapting sampling rates q Self-triggered control q Adapting transmission schedules

7

Y. Ma, D. Gunatilaka, B. Li, H. Gonzalez and C. Lu, Holistic Cyber-Physical Management for Dependable

Wireless Control Systems. ACM Transactions on Cyber-Physical Systems, 3(1), Article No. 3, 2018.

Y. Ma, C. Lu and Y. Wang, Efficient Holistic Control: Self-Awareness across Controllers and Wireless

Networks,ACM Transactions on Cyber-Physical Systems, Special Issue on Self-Awareness in Resource Constrained Cyber-Physical Systems, accepted.

Y. Ma, J. Guo, Y. Wang, A. Chakrabarty, H. Ahn, P

. Orlik and C. Lu, Optimal Dynamic Scheduling of Wireless Networked Control Systems, ACM/IEEE International Conference on Cyber-Physical Systems (ICCPS'19), April 2019.

SLIDE 9

(1) Adapting #Tx

Ø Adjust the number of transmission (#Tx)

8

−84 −82 −80 −78 −77 −76 −75 −74 −73 −72 0.2 0.4 0.6 0.8 1 PDR Noise Strength (dBm) 1 Tx 2 Tx 3 Tx 4 Tx 5 Tx 6 Tx

Controller Physical State Observer Actuators Sensors

t

u

t

y ˆt y ˆt x

Plant

ˆt u

Controller Physical State Observer Actuators Sensors Plant Holistic Controller Physical State Observer Actuators Sensors Plant

R or Tn 3 4

Physical Disturbance Wireless Interference

#Tx 2 1

SLIDE 10

Measure Control Performance

9

Ø Monitor control performance based on Lyapunov function 𝑊(𝒚:) Ø Trend of 𝑊(𝒚:) à system stability Ø Value of 𝑊(𝒚:) à upper bound of physical states errors ||𝒚: − 𝒚𝒔|| t

( )

t

V x

t

x

SLIDE 11

#Tx Adaptation Algorithm

12/3/19

10

If Increase threshold → #Tx ↑ If Decrease threshold for a time interval → #Tx ↓ 𝑊(𝑢) 𝑢 Increase threshold Decrease threshold

Ø Simplified rate adaptation algorithm

#Tx ↑ #Tx ↓

SLIDE 12

Network Reconfiguration

Ø Asymmetric scheduling

q 1 Tx for sensing flows q adapt #Tx for actuation flows

Ø Piggyback mechanism

q Piggyback #Tx in each packet sent to actuators q Each node checks #Tx and switches schedule in next period

11

SLIDE 13

Network Reconfiguration Cont.

A B C D

F1:AàBàC F2:AàBàD

12

#TX = 1 #TX = 1 #TX = 1 #TX = 3 #TX = 3 #TX = 3 #TX = 2 #TX = 2

SLIDE 14

Wireless Cyber-Physical Simulator

13

Ø Realistic and holistic simulation environment for wireless control systems (open source: http://wcps.cse.wustl.edu)

q Integrate TOSSIM and Simulink

Python interface Routing layer TDMA MAC layer Wireless link model Message pool Packet collector Cross-platform function call Data Block Controller Plant model Reference Routing Scheduling Wireless Signal Wireless Noise

Sensor Data after delay and loss Command Data after delay and loss

Simulink TOSSIM User Inputs

Wireless Network Network Manager Interfacing Block

Routing Scheduling RSSI Noise Sensor Data

Data with SS/UDS decisions

Command Data

Sensor and Command Data Return values of function call

SLIDE 15

Wireless Cyber-Physical Simulator

14

Ø Realistic and holistic simulation environment for wireless control systems.

q Incorporate WirelessHART network protocol stack q Provide Dockerized (container-based) installation q Incorporate holistic control mechanisms

run-time network adaptation
simulate communicational and computational latency

SLIDE 16

Experimental Settings

Ø Physical plant: 5-state linear time-invariant plant model Ø Wireless network: 16-node WirelessHART network Ø Network interference: noise in wireless channels Ø Physical disturbance: sensor bias

15

SLIDE 17

−2 2 Physical states 10

−3

10 10

3

Lyapunov function −78 −75 Wireless noise 1 2 3 4 #TX 20 40 60 80 100 120 140 160 180 200 0.5 1 Actuation PDR Time (s) (a) (d) (e) (b) (c) −2 2 Physical states 10

−3

10 10

3

Lyapunov function −78 −75 Wireless noise 1 2 3 4 #Tx 20 40 60 80 100 120 140 160 180 200 0.5 1 Actuation PDR Time (s) (a) (d) (e) (b) (c)

16

Performance under cyber interference

SLIDE 18

140 160 180 200 220 0.5 1 1.5 2 System Lifetime under network interference (Day) Mean Absolute Error

17

Performance under interferences

140 150 160 170 180 190 200 0.2 0.4 0.6 0.8 1 System Lifetime under physical disturbance (Day) Mean Absolute Error 2 TX 3 TX 4 TX Pure Network Adaptation Holistic Management

Holistic control closes the loop between network and control à

Ø prolong network lifetime Ø maintain resilience to both cyber and physical interferences

SLIDE 19

(2) Rate Adaptation

12/3/19

18

𝑊(𝑢) 𝑢 Increase threshold Decrease threshold

Sampling rate ↑ Sampling rate ↓

Controller Physical State Observer Actuators Sensors

t

u

t

y ˆt y ˆt x

Plant

ˆt u

Controller Physical State Observer Actuators Sensors Plant Holistic Controller Physical State Observer Actuators Sensors Plant

R or Tn 3 4

Wireless Interference

Sampling Rate 2 1

Physical Disturbance

SLIDE 20

(3) Self-triggered Control

12/3/19

19

Ø Event trigger rule

q Stability index is specified by: 𝑇(𝑢) q Trigger when 𝑊 𝑢 ≥ 𝑇 𝑢

Ø Self triggered control

q Predict when the trigger condition will be violated based on model

àInter-transmission time

𝑊(𝑢) 𝑢 𝑇(𝑢)

Controller Physical State Observer Actuators Sensors

t

u

t

y ˆt y ˆt x

Plant

ˆt u

Controller Physical State Observer Actuators Sensors Plant Holistic Controller Physical State Observer Actuators Sensors Plant

R or Tn 3 4

Wireless Interference

Inter-transmission Time 2 1

Physical Disturbance

SLIDE 21

Network Design

12/3/19

20

Ø Glossy flooding

q One to many q Fast (propagation delay < 10 ms in 100-node mesh network)

Ferrari, F., et. al. Efficient network flooding and time synchronization with glossy. In IPSN, 2011.

SLIDE 22

Glossy flooding

12/3/19

21

Ø One to many Ø Constructive interference Ø Radio event driven Ø Fast (propagation delay < 10 ms in 100-node mesh network)

𝑢

+

𝑢

=

𝑢

SLIDE 23

Network Reconfiguration

12/3/19

22

Ø Glossy flooding

q One to many q Fast (propagation delay < 10 ms in 100-node mesh network)

Ø Low-power Wireless Bus (LWB) network protocol

q Maps all communication on fast Glossy floodsà many to many

Ø Advantages of LWB

q Fast q T

pology independent

q Suitable for network-wide adaptation

Ferrari, F., et. al. Efficient network flooding and time synchronization with glossy. In IPSN, 2011. Ferrari, F., et. al Low-power wireless bus. In Sensys, 2012.

SLIDE 24

WCPS-RT for Hybrid Simulation

12/3/19

23

WCPS-RT: Hybrid Simulations Ø Real wireless networks + simulated physical plants Ø Capture wireless dynamics that are hard to simulate accurately Ø Leverage simulation support for controllers and plants.

d12

t

y

ˆt u

Holistic Controller State Observer

,

t

u ˆt y ˆt x

R or Tn Controller Side s11 Socket Socket Actuators Plant Sensors Socket s12 Socket Simulink Desktop Real-Time WSAN d11 Simulink Desktop Real-Time Plant Side WSAN Testbed @ WUSTL d12

ˆ

ˆ ˆ

s11 s12 WSAN d11 WSAN Testbed @ WUSTL d12

t

y

ˆt u

Holistic Controller State Observer

,

t

u ˆt y ˆt x

R or Tn Controller Side s11 Socket Socket Actuators Plant Sensors Socket s12 Socket Simulink Desktop Real-Time WSAN Socket Socket d11 Socket Socket Pi Interfacing Block Serial Serial Serial Serial Interfacing Block Simulink Desktop Real-Time Plant Side

SLIDE 25

Experimental Settings

12/3/19

24

104 102

103 101

105 106 107 108 113 112 114 111

109

115 116 117 118 119

s11 s21

3

110 s31 s41 s51

104 102

103 101

105 106 107 108 113 112 114 111

109

115 116 117 118 119

s11 s21

3

110

s31 s41 s51

140 121 122 123 124 125 126 127 129 128 132

131

130 136 135 134 133 139 120 138 137 141 142 143 144

d21

4

156 155 154 153 152 151 150 148 147 169 146 157 145 159 158 160 163 162 161

164

165 149 168 167 166 170

d11

5

157 167 169

d31 d41 d51

Loop1 WSAN Testbed

u1 x y

Plant1 x1

y u x

Kalman Filter1

x Tn u

Holistic Controller1 Real-Time Sync Real-Time Synchronization

u Tn

simulink to s11@pi314_2

u seq_no Tn

d11@pi536 to simulink Seq_no1 Tn1 rec.

us upi

Converter1

upi us

Converter2

Phy Dis

Dis u u1

Actuator1

x

ant2 x2 ter2 c 17 ink Seq_no2 c. r3 r4 is1 Actuator2

× 5

Ø Physical plant and controller

q Up to five load positioning plants

Ø 3- floor WSAN@WUSTL

q 70 T

elosB motes

q Low-power Wireless Bus network protocol

SLIDE 26

1 0.5 0.25 RA ST 2 3 4

Avg. energy cost (mW)

1 0.5 0.25 RA ST 1 2 3 4

MAE

Normal Condition

12/3/19

25

Control performance metric:

RA and ST have comparative control performance to fixed 1Hz sampling
while consuming 40+% less energy in the network!
ST is more aggressive in energy saving than RA

Ø RA: Rate Adaptation Ø ST: Self-Triggered control

SLIDE 27

12/3/19

26

1 0.5 0.25 RA ST 2 3 4

Avg. energy cost (mW)

1 0.5 0.25 RA ST 10 20 30

MAE

Under Wireless Interference

RA and ST have similar control performance to fixed 1Hz sampling,

achieving resiliency to wireless disturbance!

RA and ST incurs higher energy cost due to recovery
but still lower than 1Hz sampling
ST consumes more energy than RA, due to packet loss recovery

Wireless interference generated by Wi-Fi

SLIDE 28

12/3/19

27

1 0.5 0.25 RA ST 2 3 4

MAE

1 0.5 0.25 RA ST 2 2.5 3 3.5 4

Avg. energy cost (mW)

Under Physical Disturbance

RA and ST have similar control performance to fixed 1Hz sampling,

achieving resiliency to physical disturbance!

while reducing energy consumption by 30+%

Physical disturbance generated by adding constant bias to actuators

SLIDE 29

Controller Physical State Observer Actuators Sensors

t

u

t

y ˆt y ˆt x

Plant

ˆt u

Controller Physical State Observer Actuators Sensors Plant Holistic Controller Physical State Observer Actuators Sensors Plant

R or Tn

(4) Dynamic Scheduling Framework

12/3/19

28

Ø Multiple control loops share the same wireless network Ø Holistic controller

q generate actuation commands q predict the physical states q predict link quality q dynamically schedule the network traffics based on predicted link

quality and predicted control performances

3 4 Dynamic Schedule 2 1

Wireless Interference Physical Disturbance

SLIDE 30

Physical State Prediction

12/3/19

29

Controller determines u(k) based on state x(k) and system model Ø packet of loop i at t = k arrives (closed loop): Ø packet of loop i is lost, and 𝑣 @A(𝑙 − 1) is actuated (open loop):

SLIDE 31

Link Quality Prediction

12/3/19

30

Ø Link quality metric: PRR

q Holt’s additive trend prediction method

Ø 1-step PRR prediction under noise −75dBm of one link Ø Sliding window with a window size of 15

SLIDE 32

Control Cost Metric

12/3/19

31

Ø Quadratic cost function of loop i: Ø Over all cost Ø Given schedule s(k), the expectation of 𝐾A(𝑦A(𝑙 + 1)) is:

#Tx 1-PRR PDR

SLIDE 33

Optimal Scheduling

12/3/19

32

Ø The optimal scheduling problem is formulated as: Ø where L is the total number of slots assigned for all actuation flows in each superframe. Ø Nonlinear integer programming problem: NP-hard Ø Linear programming relaxation

q 99.98% of cases yield the optimal solutions

SLIDE 34

Experimental Settings

12/3/19

33

Ø Four double-water-tank systems over IEEE 802.15.4 network Ø Two groups of experiments

q Baseline: fixed (periodic) scheduling q Optimal (OPT) scheduling

SLIDE 35

Constant Background Noise

12/3/19

34

Ø Control performance

SLIDE 36

Constant Background Noise

12/3/19

35

Ø OPT schedule: (Avg. PRR: loop 1-4: 80.73%, 72.82%, 64.14%, 72.40%) Ø Slot allocation in various time intervals

SLIDE 37

Conclusion

Ø Holistic control framework for wireless control system

q Closing the loop between network and control

Ø Holistic control algorithms and protocols

q Online network reconfigurations based on physical states

Ø Resilient to both cyber and physical interference while prolonging network lifetime Ø WCPS(-RT) provides realistic and holistic simulation environment

36

SLIDE 38

Reference

Ø Y. Ma, D. Gunatilaka, B. Li, H. Gonzalez and C. Lu, Holistic Cyber- Physical Management for Dependable Wireless Control Systems. ACM Transactions on Cyber-Physical Systems, 3(1), Article No. 3, 2018. Ø Y. Ma, C. Lu and Y. Wang, Efficient Holistic Control: Self-Awareness across Controllers and Wireless Networks,ACM Transactions on Cyber-Physical Systems, Special Issue on Self-Awareness in Resource Constrained Cyber-Physical Systems, accepted. Ø Y. Ma, J. Guo, Y. Wang, A. Chakrabarty, H. Ahn, P. Orlik and C. Lu, Optimal Dynamic Scheduling of Wireless Networked Control Systems, ACM/IEEE International Conference on Cyber-Physical Systems (ICCPS'19), April 2019.

37