1 distribution a. approved for public release; distribution unlimited. (approval afrl pa #...

32
1 DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014) Reducing the Wrapping Effect in Flowpipe Construction using Pseudo-Invariants Stanley Bak United States Air Force Research Lab Information Directorate – Rome, NY

Upload: mitchell-dennis

Post on 26-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

1DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

Reducing the Wrapping Effect in Flowpipe Construction using Pseudo-Invariants

Stanley Bak

United States Air Force Research Lab

Information Directorate – Rome, NY

2DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

2

Cyber-Physical Systems

• Include computational (discrete) components, and physical-world (continuous) aspects

• Discrete components are typically modeled usingfinite state machines with switching rules. The physical world is typically described by differential equations.

Autonomous Cars Air Traffic ControlFault-Tolerant

Power Distribution

3DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

3

Outline

• Hybrid Automata as Models for Cyber-Physical Systems

• Flowpipe Construction

• Wrapping-Effect Error

• Pseudo-Invariants

• Improvement Example

4DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

4

Hybrid Automaton

u

continuous dynamics

invariant: hybrid automaton may remain in u as long as x I(u)

)(

)(

xFx

uIx

u

location (discrete state)

u’

reset condition

guard conditionedge

0Xx

initial condition

)(

)(

xFx

uIx

u

)(: 11 eGxe

),( 1 xeRx

),( 4 xeRx

)(: 44 eGxe

)(: 22 eGxe

)(: 33 eGxe ),( 3 xeRx

),( 2 xeRx

• Fusion of discrete dynamics with continuous dynamics

From: “Hybrid System Verification Using Discrete Model Approximations,” Chutinan

5DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

5

Hybrid Automaton Example

• Simple example: on/off heater in a room (x = temperature)

6DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

6

Hybrid Automaton Execution

• Executions of a hybrid automaton capture one valid trajectory in the model. There can be unaccountably many.

7DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

7

Hybrid Automaton Reachability

• Reachability captures all possible behaviors, and is therefore useful for verification.

8DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

8

Outline

• Hybrid Automata as Models for Cyber-Physical Systems

• Flowpipe Construction

• Wrapping-Effect Error

• Pseudo-Invariants

• Improvement Example

9DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

9

Reachability Algorithm

• One way to (overapproximate) reachability is to iteratively compute discrete and continuous successors

10DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

10

Continuous Successors

• Many ways to compute continuous successors (abstraction, symbolic reasoning, flowpipe construction)

• Flowpipe construction methods compute the set of states at snapshots in time

T=0.0

T=0.1

T=0.2

Between 0.0 and 0.1

Between 0.1 and 0.2

11DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

11

Discrete Successors

• Flowpipe methods typically aggregate states across discrete transitions

• Otherwise, a singletracked set canspawn multiple flowpipes insubsequent modes

Image from: “Safety Analysis of Hybrid Systems with SpaceEx,” Frehse et al., http://cmacs.cs.cmu.edu/seminars/slides/frehse.pdf

12DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

12

Demonstration(buck_open.hyc)

Aggregation

13DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

13

No Aggregation

14DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

14

With Aggregation

15DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

15

Outline

• Hybrid Automata as Models for Cyber-Physical Systems

• Flowpipe Construction

• Wrapping-Effect Error

• Pseudo-Invariants

• Improvement Example

16DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

16

Flowpipe Construction Issues

• Iterative flowpipe construction leads to overapproximation error

• Depends on representation

Image from: “Perspectives on Enclosure Methods”, Kulisch et al.

17DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

17

Outline

• Hybrid Automata as Models for Cyber-Physical Systems

• Flowpipe Construction

• Wrapping-Effect Error

• Pseudo-Invariants

• Improvement Example

18DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

18

Pseudo-Invariants

• Key Idea of the Talk: Use discrete successor aggregation to reduce wrapping-effect error

• How? Force discrete-successor aggregation to occur by introducing an artificial invariant (called the pseudo-invariant) and associated transition

– Basically, split one mode into two withidentical dynamics

19DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

19

Splitting a Single Mode

• The new automaton is a bisimulation of the original one

20DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

20

Effect on Reachability

21DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

21

Outline

• Hybrid Automata as Models for Cyber-Physical Systems

• Flowpipe Construction

• Wrapping-Effect Error

• Pseudo-Invariants

• Improvement Example

22DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

22

Van der Pol Dynamics

(Demonstration, ShowReachability.jar)

23DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

23

Van der Pol Simulations

24DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

24

Computing Reachability

• Using Flow*, a reachability tool that uses Taylor Models as a state-set representation, we ran the following two models (initial state: x = 1.0, y=[-0.5, 0.5]):

poly ode 1 { x'= y y'= y - x - x^2 * y }

first { poly ode 1 { x'= y y'= y - x - x^2 * y } inv { x in [0.75, 999] } }

second { poly ode 1 { x'= y y'= y - x - x^2 * y } inv {}}

jumps { first -> second guard { x in [-999,0.75] } reset { } parallelotope aggregation {}}

Error gets too large at t=1.54, tool exits

CompletesSuccessfully!

25DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

25

Flow* Reachability

26DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

26

What Happens Around t=1.54?

27DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

27

With Pseudo-Invariant

28DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

28

Conclusion

• Key Idea of the Talk: Use discrete successor aggregation to reduce wrapping-effect error

• Concerns:

– What if the intersections are not accurate?

– How do you come up with pseudo-invariants?

– What if you choose a “poor” pseudo-invariant?

29DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

29

Bonus

• How do you come up with pseudo-invariants?

• Idea: during reachability, when the wrapping errors appears to be getting large, generate a pseudo-invariant on-the-fly based on the currently-tracked set

(demonstration, van_der_pol.hyc)

30DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

30

Without Pseudo-Invariants

31DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

31

Using Pseudo-Invariants

32DISTRIBUTION A. Approved for public release; Distribution unlimited. (Approval AFRL PA # 88ABW-2014-1552, 09 April 2014)

32

Using Pseudo-Invariants (2)