Introduction to Cyber-physical Systems (CPSs)
Ricardo Sanfelice Department of Computer Engineering Hybrid Systems Lab University of California, Santa Cruz
Broad Scope of Cyber-physical Systems
Nonsmooth Control Systems
Digital Control
Cyber-physical Systems
Computer Networks
Multi-mode Control
( (
(
(
(
Distributed Control
(
Power Networks
+ _
Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
u
plant
x
controller
Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
u
plant
x
controller logic for decision making
multiple control laws Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
u
plant
x
distributed controller logic for decision making
multiple control laws Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
u
plant
x
interface
interface controller logic for decision making
multiple control laws Ricardo Sanfelice - University of California, Santa Cruz
Broad Scope of Cyber-physical Systems Systems of today feature: !
Heterogeneous components and interfaces (e.g. humans, networks, analog/digital devices).
!
Modular hardware for flexibility and reconfigurability.
!
Distributed coordination and control.
u
plant
x
environment
ZOH
D/A
interface
interface
A/D
perturbations human interaction
controller network
logic for decision making
multiple control laws Ricardo Sanfelice - University of California, Santa Cruz
Modeling Cyber-Physical Systems Frameworks for modeling of CPSs include not an exhaustive list!
!
Discrete-time models
!
Finite-state machines
!
Continuous-time models
!
Impulsive differential equations
!
Measure-driven differential equations
!
Event-driven systems
!
Timed automata
!
Hybrid automata .. .
!
Ricardo Sanfelice - University of California, Santa Cruz
Discrete-time models Given
G
fnctian
a
Xlkti )
=
G
:lRnx1Rm→Rn ( X(
k
.nlk
)
)
KEN :={
)
.
Xlo )
given
ktm
wd
( k)
¥"÷€zif!€gk In .
.
G
If
is
'
"
}
defined
IRNXIRM
on
2nd
defined
is
m
Existence
0,1 ,
1
an
N
soln
of
genie constraints
Add
Nanmiqveress •
Extra
/
inputs
Hu )
:
ED
Nondetoninisne .
G
set
exzple -
vowed
,u)=G(x,u)+ GXtxlkti 81Pa ×+E
(
×
)
Ricardo Sanfelice - University of California, Santa Cruz
Finite-state machines Consider
FI
QEQ
where
q+=8(
9
Qissfniteset
8
,v)
:
theater
AVH @¥@ qtLq
Not
ni
v=O
8iqM={
a
Bf after ••T
Q=lAD}
SCQM
Teflon
Ricardo Sanfelice - University of California, Santa Cruz
Continuous-time models
xit ,
xeitrn
x°=fcx,u)
th
↳
x
In
general •
denx
donx
Cenpleteness refutnts
:
maximal
of
(
f
of
Rn
→
[
C
norm
,
0
,oD
depends
solutions
on
input )
nd
t#
:
Differential
-
inclusion
EE
Fcx
T
,u)
regulation
.
ff
Fan
:=
F
Q
ctfcxtsrs ) .
for
no a
is
mput ne
a
settled
map
Ricardo Sanfelice - University of California, Santa Cruz
Impulsive differential equations i.
fix )
×+=
gk
x¥M
)
EM
×
MCIR
"
xtirn .
-
.
-
.
.
.
.
ifcx
)
t -4
LTIII
, 00
xltithglxttill
tetti
)i=
Ricardo Sanfelice - University of California, Santa Cruz
Measure-driven differential equations
Ricardo Sanfelice - University of California, Santa Cruz
Event-driven systems
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications
DC bus AC bus
Photovoltaic array DC
Feedback Control for Smart Grids !
Hetereogeneous networked power sources, buses, users, and loads
!
Conversion required between different waveforms
AC DC
DC
Diesel generator
~ Turbine
~ Fuel cells AC
!
Dynamic demands and supplies
!
Multiple time scales
DC Storage AC
(e.g., fast and slow switching)
DC
!
Steady-state and averaged models
!
Linear control design
!
Bening conditions
DC loads
... ...
AC loads
Classical approaches:
AC DC
Communication ( ) and Control ( ) bus
~ Electric power grid
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications
DC bus
Power Conversion for Smart Grids
Photovoltaic array DC
!
Renewables provide power with high fluctuation
!
DC/DC conversion is required before injection to the grid
AC DC
Diesel generator
~ Turbine
~ Fuel cells
Adaptive DC/AC conversion
AC DC Storage
vDC
AC DC/AC s4
s1
s3
s2
Hybrid Controller
L, R
DC
iL io so
vc
c
vg
AC loads
DC loads
... ...
!
DC
AC bus
AC
Ricardo Sanfelice - University DC of California, Santa Cruz
CPS Applications
DC bus
Power Conversion for Smart Grids
Photovoltaic array DC
!
Renewables provide power with high fluctuation
!
DC/DC conversion is required before injection to the grid
!
DC
AC bus AC DC
Diesel generator
~ Turbine
~ Fuel cells
Adaptive DC/AC conversion
AC DC Storage
vDC
AC DC/AC s1 s2
Hybrid Controller
High Penetration of Renewables
L, R
iL io so
vc
c
vg
AC loads
DC loads
... ...
s4 s3
DC
AC
Ricardo Sanfelice - University DC of California, Santa Cruz
CPS Applications
DC bus
Power Conversion for Smart Grids
Photovoltaic array DC
!
Renewables provide power with high fluctuation
!
DC/DC conversion is required before injection to the grid
!
DC
AC bus AC DC
Diesel generator
~ Turbine
~ Fuel cells
Adaptive DC/AC conversion
AC DC Storage
vDC
AC DC/AC s1 s2
Hybrid Controller
High Penetration of Renewables
L, R
iL io so
vc
c
vg
AC loads
DC loads
... ...
s4 s3
DC
AC
Ricardo Sanfelice - University DC of California, Santa Cruz
CPS Applications Control of Reconfigurable Multi-Robot Systems !
!
Groups of heterogeneous networked agents
Adversaries
dynamic communication network
Adversaries can disrupt the network (jamming, destructive actions)
!
Locations and capabilities unknown Static and Mobile Agents
Scenarios of DoD interest: !
Electronic warfare
!
Satellite communications
!
Disaster relief
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications Reconfigurable Satellite Communications !
Dynamic ground-satellite links
!
Dynamic signal-to-noise ratio on communication channels
!
Jamming attacks (MILSATCOM)
dynamic network
ground station
adversary
GEO
satellite LEO
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications Reconfigurable Satellite Communications !
Dynamic ground-satellite links
!
Dynamic signal-to-noise ratio on communication channels
!
Jamming attacks (MILSATCOM)
dynamic network
ground station
adversary
GEO
satellite LEO
Recent initiatives, such as the National Broadband Plan, challenge the traditional FCC approach to allocating spectrum, requesting a new U.S. spectrum policy allowing for dynamic allocation and utilization.
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications Control of Aerial Vehicles with Limited and Faulty Sensors !
Autonomous recovery control
!
Low cost sensing for autonomous navigation
!
Sensor failures affect stability and performance
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications Control of Aerial Vehicles with Limited and Faulty Sensors !
Autonomous recovery control
!
Low cost sensing for autonomous navigation
!
Sensor failures affect stability and performance
Ricardo Sanfelice - University of California, Santa Cruz
CPS Applications Control of Aerial Vehicles with Limited and Faulty Sensors UAVs in the National Air Space ! Autonomous recovery control (NAS): 2012 bill giving FAA three yearsnavigation to “integrate” UAVs into the ! Low cost sensing for autonomous NAS (set policies, standards, etc.) ! Sensor failures affect stability and performance
Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges The U.S. National Academy of Engineering has listed 14 grand challenges that relate environmental, health, and societal issues; these issues will clearly benefit from advances achieved in cyber-physical systems.1 “The control engineering research community can play a leading role in the development of cyber-physical systems.”2
1 2
http://www.engineeringchallenges.org/challenges.aspx The Impact of Control Technology, T. Samad and A.M. Annaswamy (eds.), 2011. Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges3 Compositionality !
Compositionality that cuts across the heterogeneous cyber and physical aspects of CPS is a major scientific challenge.
!
Modeling and predicting performance of composition.
!
An potential solution could be “plug & play” cyber-physical components that integrate seamlessly.
3
Cyber-Physical Systems Summit, 2008. Report of the Steering Committee for Foundations in Innovation for Cyber-Physical Systems, 2013. Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges Distributed Event-based Sensing and Control !
Collecting adequate information and asserting control in a distributed environment.
!
New science and theory is needed regarding how communication can facilitate safe operation, failure interlocks, and adaptation.
!
Integration of information and actions across time (with understanding of uncertainty at different scales) is essential.
!
When do samples need to be collected for distributed decisions? What information needs to be collected? Where should the computations be performed?
Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges Physical and Cyber Constraints !
Current algorithm designs seldom include the computational limitations of the hardware/software on which they are implemented as explicit constraints. Computational capabilities, such as computer architecture and processing power are not typically explicitly considered during design.
!
Current “one-size-fits-all” approach typically rules out viable solutions or leads to very conservative solutions, affecting the system?s performance and robustness.
!
Need to generate tools for systematic analysis and design of computationally aware algorithms for CPSs.
Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges Physical, Human Interfaces and Integration !
A hierarchy of models of physical contact with varying levels of resolution and degrees of freedom (e.g., continuous, discrete, etc).
!
The mathematical properties the models in the hierarchy, the corresponding algorithms with known performance properties, and the couplings of the contact models to the cyber and other physical components of the system.
!
A unified model including all components is needed.
!
New science and theory is needed to define safe hand-offs between human and cyber-based control, cyber-physical interlocks, protocols for mixed initiative inter-operation, and maintaining the operator’s mental model and appropriate skepticism. Ricardo Sanfelice - University of California, Santa Cruz
CPS Challenges Robustness, Adaptation, Reconfiguration !
Contrary to well-established methods of robust control, which can handle modeling uncertainty, we need new notions of robust system design that address uncertainty at the cyber level (computing decision or scheduling choices), and are resilient with respect to massively uncertain/untrusted data and structural/topological uncertainties and reconfiguration.
!
CPS will also need to be reconfigurable and adaptive to overcome faults in both physical and cyber levels.
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
!
Modeling of the physical and the cyber
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
!
Modeling of the physical and the cyber
!
Tools to analyze the behavior of systems with both components
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
!
Modeling of the physical and the cyber
!
Tools to analyze the behavior of systems with both components tools solely for the physical or for the cyber do not apply
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
!
Modeling of the physical and the cyber
!
Tools to analyze the behavior of systems with both components tools solely for the physical or for the cyber do not apply
!
Understand core theoretical concepts needed to study CPS
Ricardo Sanfelice - University of California, Santa Cruz
Looking Ahead !
Systems that combine physical and cyber components, potentially networked and with computations integrated with physical process physical = physical plant/process/network cyber = software/algorithm/computation
!
Modeling of the physical and the cyber
!
Tools to analyze the behavior of systems with both components tools solely for the physical or for the cyber do not apply
!
Understand core theoretical concepts needed to study CPS
!
Apply tools and concepts to a CPS application Ricardo Sanfelice - University of California, Santa Cruz
