Engineered System Design and Integration A semantic domain for modeling cyber-physical systems

Transcription

1 Engineered System Design and Integration A semantic domain for modeling cyber-physical systems Pieter J. Mosterman The importance of computation Together with theory and experimentation, computational science now constitutes the third pillar of scientific inquiry, 2013 The MathWorks, Inc. 1 2 The importance of computation Computation as main feature differentiator As new funding becomes available, the following four areas should receive disproportionally larger increases [ ] NIT Systems Connected with the Physical World (which are also called embedded, engineered, or cyber-physical systems) [ ] Software: The NITRD Subcommittee should facilitate efforts by leaders from academia, industry, and government to identify critical issues in software design and development 3 4

2 Computation as main feature differentiator Computation as main feature differentiator 5 6 System integration Agenda System Integration Timing Concurrency Interfaces Shared resources Engineering complex systems Cyber-Physical Systems Modeling paradigms Semantic domains Verification for synthesis Conclusions 7 8

3 System properties System properties Performance Performance Security Size Security Size Safety Power Safety Power Reliability QoS Reliability QoS Cost Cost Known minimum Known maximum Known minimum Known maximum 9 10 System properties System properties Security Performance Size Security Performance Size Safety Power Safety Power Reliability QoS Reliability QoS Cost Cost Known minimum Known maximum Known minimum Known maximum Required 11 12

4 ,QÃFDVHÃ Separation of concerns Divide and conquer A canonical development structure 5(4Ã 7(67Ã UXOH Divide and conquer 'LYLGHÃDQGÃUXOH 6HSDUDWHÃFRQFHUQV 63(&Ã 'LYLGHÃDQGÃUXOH 63(&Ã 63(&Ã 63(&Ã '(6,*1 Having divided to conquer, we must reunite to rule. -- Michael Jackson 0DS V W L Q D U W V Q R & H U X W F H L W K F U $ H U X W F H L W K F U $ 63(&Ã double y; y = k * u;,03/(0(17,03/ã &RPSRVH REM OLE,QWHJUDWH,03/Ã 7(67Ã Michael Jackson, Some Complexities in Computer- Based Systems and Their Implications for System Development, International Conference on Computer Systems and Software Engineering, pp , Tel-Aviv, Israel, May 1990, 16

5 Continuous refinement of abstractions Rise time, overshoot, stability margin algorithm + transfer function Separation of concerns abstraction Stability, minimum phase, corner frequency timed + sample time Output times, latency, communicated values tasked + scheduler Logic values implementation + indexing 18 Abstraction Semantics of a swap operation Axiomatic Good for properties tmp = a; a = b; b = tmp; Denotational Good for specifications {a=a0 b=b0} swap(a,b) {a=b0 b=a0} Operational semantics Details of execution 'HVFULEHÃDÃVHULHVÃRIÃVWDWHÃFKDQJHVÃ ÃLPSHUDWLYH swap: tmp = a; a = b; b = tmp; Operational Good for implementations swap: initial state new state a a0 b b0 tmp tmp0 a a0 b b0 tmp a0 a b0 b b0 tmp a0 tmp = a; a = b; b = tmp; a b0 b a0 tmp a0 swap: tmp = a; a = b; b = tmp; 19 Taken from Frédéric Boulanger and Cécile Hardebolle, Execution of models with heterogeneous semantics 20

6 Semantics of a swap operation Semantics of a swap operation tmp = a; a = b; b = tmp; tmp = a; a = b; b = tmp; Denotational semantics Results of execution Describe the path from initial to final state Ãdeclarative swap: initial state new state Axiomatic semantics Properties of execution state Describe the change of properties of the state Ãdeclarative {a=a0 b=b0} swap(a,b) {a=b0 b=a0} a a0 b b0 tmp tmp0 swap(a,b) a b0 b a0 tmp a0 Denotational Operational a a0 a b0 b b0 b a0 swap(a,b) Axiomatic Denotational Operational Taken from Frédéric Boulanger and Cécile Hardebolle, Execution of models with heterogeneous semantics 21 Taken from Frédéric Boulanger and Cécile Hardebolle, Execution of models with heterogeneous semantics 22 Separation of concerns Rise time, overshoot, stability margin Separation of concerns Rise time, overshoot, stability margin? Settled value, linearization algorithm + transfer function algorithm + transfer function abstraction Stability, minimum phase, corner frequency timed + sample time Output times, latency, communicated values abstraction Stability, minimum phase, corner frequency Frequency spectrum timed + sample time Output times, latency, communicated values Execution times approximation tasked + scheduler tasked + scheduler Logic values Logic values Memory implementation + indexing implementation + indexing 23 24

7 Composition for system integration? Whoever ate your sandwich does not like bread crusts. -- Sherlock Hemlock C1 a b C2 Divide and rule C1 a b a b C2 C1 a b a b C2? 26 A model-reference adaptive ler Examine real-time behavior reference adapt operator time 150 ms Does not fit in a single process; multi-tasking! Zhi Han and Pieter J. Mosterman, "Detecting Data Store Access Conflict in Simulink by Solving Boolean Satisfiability Problems," in 2010 American Control Conference (ACC 10), pp , Baltimore, MD, June 30-July 2,

8 Multi-tasking Separation of tasks adapt reference adapt operator reference operator 150 ms Shared memory 150 ms Run the adapt task in a separate process Execution fits in our time budget! Execution fits in our time budget! adapt adapt reference operator reference operator 150 ms 150 ms Exploit concurrency in execution resources But what if my adapt task runs longer on another processor? 31 32

9 Variable is read before it is written Incorrect timing? adapt adapt adapt reference operator reference operator reference operator 150 ms 150 ms 150 ms Where does that value for come from? So, uses the previous adapt parameters how much impact could it possibly have? Incorrect timing? Double buffering adapt adapt adapt adapt reference operator reference operator reference operator reference operator 150 ms 150 ms 150 ms 150 ms So, uses the previous adapt parameters how much impact could it possibly have? The same variable has two memory locations In one period, read from one memory location and write to the other 35 36

10 Consistent value independent of architecture Towers of Hanoi adapt adapt reference operator reference operator 150 ms 150 ms Use previous value, even if the adapt computation may be shorter - at times - on different architectures No Determinism: more surprises! no more surprises! Really, though?! 37 as a SCADA system A SCADA system to sort blocks 40

11 A multirate distributed architecture A multirate distributed architecture nozzle z -1 stereo analysis z -1 supervisory slider detection logic pump nozzle z -1 stereo analysis z -1 supervisory slider detection logic pump slider ECU base ECU slider ECU base ECU found picked, placed, nozzle_force NCAP nozzle_force nozzle_mode, position, left_video, right_video NCAP pressure slider force nozzle_mode, slider_force, pressure wireless network found, picked, placed, block_service, position z -1 z -1 z -1 position left_video right_video left right NCAP camera camera found picked, placed, nozzle_force NCAP nozzle_force nozzle_mode, position, left_video, right_video NCAP pressure slider force nozzle_mode, slider_force, pressure wireless network found, picked, placed, block_service, position z -1 z -1 z -1 position left_video right_video left right NCAP camera camera nozzle motor pump actuator slider motor position sensor scene nozzle motor pump actuator slider motor position sensor scene z -1 5 ms delay z ms delay 41 z -1 5 ms delay z ms delay 42 Computation as main feature differentiator System integration System Integration Timing Concurrency Interfaces Shared resources 43 44

12 Computation as main feature differentiator System integration System integration System integration 47 48

13 System integration Agenda Engineering complex systems Cyber-Physical Systems Modeling paradigms Semantic domains Verification for synthesis Conclusions

14

15 Network Physics Network Network Physics Information Networked embedded systems Information Electronics Network Network Physics Physics Information 59 60

16 Cyber-physical systems Cyber-physical systems Information Information Electronics Electronics Network Network Physics Physics Cyber-physical systems Design of heterogeneous systems Cyber-physical systems Information Executable models Quick feedback on design options Automate design tasks Automate synthesis tasks Information Shared feature functionality Feature interaction Electronics Computational semantics Electronics Post deployment integration Emerging behavior Network Physics Network Physics 63 64

17 Design of heterogeneous systems Heterogeneity in computational solutions Executable models Quick feedback on design options Automate design tasks Automate synthesis tasks Computational semantics Execution engine Combines many formalisms Execution engine Information ODE State transition Discrete time Control flow Electronics Discrete event DAE Network Discrete event State transition Physics DAE State transition Heterogeneity in computational solutions Agenda Modeling domains Disciplines Engineering complex systems ODE Physical environment Cyber-Physical Systems Simulink Discrete time Simulink Discrete event SimEvents Transition system Electrical hardware Digital hardware Embedded software Modeling paradigms Semantic domains Verification for synthesis Conclusions Stateflow Analog/RF hardware DAE Simscape SimElectronics SimMechanics SimHydraulics SimDriveline Control flow MATLAB Mechanical hardware Communications 67 68

18 Towers of Hanoi as a Cyber-Physical System?

19 A Cyber-Physical System! Modeling paradigms Signal processing Control Supervisory and sequence Feedback Network, communication Physics, plant Justyna Zander and Pieter J. Mosterman, Technical Engine for Democratizing Modeling, Simulation, and Prediction, in Proceedings of the Winter Simulation Conference, Berlin, Germany, December, 2012 Pieter J. Mosterman, Justyna Zander, and Zhi Han, The Towers of Hanoi as a Cyber-Physical System Education Case Study, in proceedings of the First Workshop on CPS Education, Philadelphia, PA, April 8, Modeling the signal processing Modeling the signal processing Algorithmic Assignments Destructive state access Untimed Data centric Algorithmic Assignments Destructive state access Untimed Data centric 75 76

20 Modeling the supervisory and sequence Discrete state based Discrete events cause transitions between states Conditions to guard the transition Untimed Control centric Modeling the supervisory and sequence Discrete state based Discrete events cause transitions between states Conditions to guard the transition Untimed Control centric Modeling the feedback Modeling the feedback Sampled discrete time Fixed sample time Periodic Data centric Sampled discrete time Fixed sample time Periodic Data centric 79 80

21 Modeling network traffic Modeling network traffic Entity flow through a graph Attributes Source Destination Service time Priority Discrete events Preemption Data centric Aperiodic Often stochastic Entity flow through a graph Attributes Source Destination Service time Priority Discrete events Preemption Data centric Aperiodic Often stochastic Modeling the plant physics Modeling the plant physics Domain-specific modeling Simscape Electrical Pneumatic Thermal Differential equation based Noncausal, energy-based, modeling Domain-specific modeling Simscape Electrical Pneumatic Thermal Differential equation based Noncausal, energy-based, modeling 83 84

22 The elements of a model Agenda Concrete syntax Syntax Concrete syntax Abstract syntax Syntactic mapping Semantics Semantic domain Semantic mapping u 2 D c Syntactic mapping y :Const value=2 Abstract syntax :Signal name=c vals=[ ] :Signal name=y vals=[ ] :Delay :Signal Engineering complex systems Cyber-Physical Systems Modeling paradigms Semantic domains Verification for synthesis Conclusions name=u Semantic domain vals=[ ] v(i+1) = u(i) for all i > 0 u(0) = 2 Semantic mapping Developed at the Computer Automated MultiparadigmModeling (CAMPaM) workshop, McGill Bellairs Research Institute A cyber-physical architecture All part of the networked embedded system Signal processing Data intensive algorithms Control Frequent periodic events Network, communication Frequent aperiodic events Physics, plant Continuous with sporadic events ordered 87 88

23 All part of the networked embedded system All part of the networked embedded system ordered ordered synchronous All part of the networked embedded system All part of the networked embedded system ordered ordered synchronous synchronous aperiodic aperiodic periodic 91 92

24 All part of the networked embedded system The semantic domain of a dynamic system ordered synchronous aperiodic periodic Points, [ ] On N On R x N Intervals, [ ² ( ², ]) On R Hybrid point/interval On R MATLAB, Stateflow Discrete time Simulink SimEvents Simulink Simulink, Simscape continuous On R x N Simulink, Simscape The semantic domain of a dynamic system The semantic domain of a dynamic system Points, [ ] On N MATLAB, Stateflow Discrete time Simulink Points, [ ] On N MATLAB, Stateflow Discrete time Simulink On R x N Intervals, [ ² ( ², ]) On R Hybrid point/interval On R SimEvents Simulink On R x N Intervals, [ ² ( ², ]) On R Hybrid point/interval On R SimEvents Simulink Simulink, Simscape Simulink, Simscape On R x N On R x N Simulink Simulink 95 96

25 The semantic domain of a dynamic system The semantic domain of a dynamic system Points, [ ] On N IÃ1Ã ÃR MATLAB, Stateflow Discrete time Simulink Points, [ ] On N IÃ1Ã ÃR MATLAB, Stateflow Discrete time Simulink On R x N SimEvents On R x N SimEvents Intervals, [ ² ( ², ]) On R Simulink Intervals, [ ² ( ², ]) On R Simulink Hybrid point/interval On R Hybrid point/interval On R Simulink, Simscape Simulink, Simscape On R x N On R x N Simulink Simulink The semantic domain of a dynamic system A general operational semantic domain Points, [ ] On N On R x N Intervals, [ ² ( ², ]) On R Hybrid point/interval On R MATLAB, Stateflow Discrete time Simulink SimEvents Simulink Simulink, Simscape Points, [ ] On R x N Without losing the analysis ability and efficiency Integer precision Clock calculus Scheduling Static when possible Dynamic when necessary On R x N Simulink

26 Agenda A surface mount device Engineering complex systems Cyber-Physical Systems Modeling paradigms Semantic domains Verification for synthesis Conclusions Newton and Hooke s Laws Differential equations Contact behavior Discontinuous changes Control behavior Sampled data (periodic) F F board R C A surface mount device A surface mount device Newton s Law 1 F = ma a = m F Newton s Law 1 F = ma a = m Viscous friction F R = Rv F F F F board F board R C R C

27 A surface mount device A surface mount device Newton s Law 1 F = ma a = m Viscous friction F R = Rv F Hooke s Law F C x x F ( ) Cx C = 0 c = Contact behavior Discontinuous changes x( t) Rv( t) + if x( t) < 0 F board ( t) = C otherwise 0 F F F board F board R C R C A surface mount device Explicitly modeled execution engine Controller behavior Sampled data F ( k) = u( k) with t = kt s Completely modeled solver and rate transition with the discontinuous world all with two basic sequential blocks

28 Control synthesis using model checking Control synthesis using model checking A counterexample A counterexample Pieter J. Mosterman, Justyna Zander, Grégoire Hamon, and Ben Denckla, "A Computational Model of Time for Stiff Hybrid Systems Applied to Control Synthesis," in Control Engineering Practice,, vol. 20, nr. 1, pp. 2-13, January Pieter J. Mosterman, Justyna Zander, Grégoire Hamon, and Ben Denckla, "A Computational Model of Time for Stiff Hybrid Systems Applied to Control Synthesis," in Control Engineering Practice,, vol. 20, nr. 1, pp. 2-13, January

29 Agenda Conclusions Engineering complex systems Cyber-Physical Systems Modeling paradigms Semantic domains Verification for synthesis Conclusions Engineered system design principles and challenges Separation of concerns Divide and rule Cyber-physical system Post deployment integration of shared functionality Models are critical in design Many different disciplines, problems, and technologies Multiparadigm modeling! Variety of semantic domains A unifying semantic domain on functions over streams Reasoning across modeling paradigms