Formal Model-Based Design & Manufacture: A Template for Managing Complexity in Large-Scale Cyber-Physical Systems
Paul Eremenko fmr. Deputy Director/Acting Director Tactical Technology Office Briefing prepared for the Conference on Systems Engineering Research March 21, 2013
The views expressed are those of the author and do not reflect the official policy or position of the Department of Defense or the U.S. Government.
The six frigates (1794) “… the sum of $688,888… to provide, equip and employ, four ships to carry forty guns each, and two ships to carry thirty-six guns each….” --An Act to Provide a Naval Armament, March 27, 1794
2
USS Philadelphia, Tripoli Harbor, February 16, 1804
3
B-52 Stratofortress (1946) "It is desired that the requirements set forth be considered as a goal and that the proposal be for an interim airplane to approximate all requirements, except that emphasis must be placed on meeting the high speed requirement... It is the intent that design proposals should present the best possible over-all airplane..." --Directive letter inviting design proposals for the B-52 bomber, February 13, 1946
F-111 Aardvark (1961)
5
F-35 Joint Strike Fighter (2001)
6
Kolmogorov complexity (sort of)
Length of Technical Specification (pages)
1000
F-22
100
B-52
10
1
Frigate
0.1 1750
1800
1850
1900
1950
2000
2050
Year of Entry into Service Data compiled by Mark Nowack, DARPA/TTO
7
Software complexity 1.E+08 108 F-22A
107
F-35
F/A-18E
1.E+07
F/A-18C F/A-18A
6
10 1.E+06
Shuttle
Source Lines of Code (SLOC)
F/A-18G
105 1.E+05
EA-6B
F-14A
F-14D
ISS AV-8B AH-1W F-14B
SIRTF DS1 Pathfinder
Cassini
MRO
MER
A-6E
A-7E
104 1.E+04
Apollo
Galileo
Viking Voyager
103 1.E+03 Key: • Manned Space Vehicle • Robotic Space Vehicle Manned Combat Aircraft
1.E+02 102 Mariner-6
1.E+01 101 1.E+00 1
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year of Entry into Service Dvorak, D. ed, NASA Study on Flight Software Complexity, Jet Propulsion Laboratory, California Institute of Technology, 5 March2009 Borden, D., Software Acquisition Process Improvement, NAVAIR, undated Agle, D.C., Where Hunters Growl, Air & Space magazine, March 2011
8
Structural & software complexity
Data compiled by Mark Nowack, DARPA/TTO
9
Cost growth $1 quintillion
$1 quadrillion Entire GNP to buy one airplane.
$1 trillion Entire Defense budget to buy one airplane.
$1 billion F-14
F-15 B-52
$1 million
F-35 F-18 F-16 A-10
F-18
P-61 P-39 P-51 Standard E-1
SPAD Morse JN-4A Wright Model A
DH-4
$1 thousand
1900
1950
Source: Norm Augustine, Augustine’s Laws, 6th Edition, AIAA Press, 1997.
2000 2050 Year of Entry into Service
2100
2150 10
Evidence for a causal relationship with complexity
11
Modern systems engineering SWaP used as a proxy metric for cost, and disincentivizes abstraction in design
System decomposed based on arbitrary cleavage lines . . .
MIL-STD-499A (1969) systems engineering process: as employed today
Conventional V&V techniques do not scale to highly complex or adaptable systems–with large or infinite numbers of possible states/configurations
Re-Design System Functional Specification
Cost Optimization
...
SWaP Optimization
SWaP Optimization
System Layout
Power
Data & Control
Thermal Mgmt
. . . and detailed design occurs within these functional stovepipes SWaP = Size, Weight, and Power V&V = Verification & Validation
...
Verification & Validation
Subsystem Design
Component Design
Subsystem Testing
Resulting architectures are fragile point designs
Component Testing
Unmodeled and undesired interactions lead to emergent behaviors during integration Desirable interactions (data, power, forces & torques) Undesirable interactions (thermal, vibrations, EMI)
12
Tools have made it better… Dassault Falcon 7X Two-fold schedule compression for new business jets through faithful application of a digital master model with QA/QC feedback by tail number
Image courtesy of Dassault Systemes
Lockheed Martin F-35 Shimming and ‘drill and fill’ approach significantly worsens production learning effects, leading to delays and cost growth* * GAO-10-382:
Joint Strike Fighter – Additional Costs and Delays Risk Not Meeting Warfighter Requirements on Time, Mar 2010
13 Image courtesy of Lockheed Martin
13
… but the fundamental design flow hasn’t changed! Engineering Change Requests (ECRs) per Month of Program Life Mariner Spacecraft (1960s)
From Project Inception through Midcourse Maneuver, vol. 1 of Mariner Mars 1964 Project Report: Mission and Spacecraft Development, Technical Report No. 32-740, 1 March 1965, JPLA 828, p. 32, fig. 20.
Modern Cyber-Electromechanical System (2000s)
Giffin M., de Weck O., et al., Change Propagation Analysis in Complex Technical Systems, J. Mech. Design, 131 (8), Aug. 2009.
14
Adaptive Vehicle Make
15
Approaches for tackling complexity
Simplify
Disaggregate
Modularize
???
16
Design, Integration, and Testing (months)
DARPA goals for AVM 240 220 200 180 160 MIL-STD-499A 140 120 100 80 Automobile Aerospace Vehicle 1960s 60 1960s 40 Integrated Circuit 1960s 20 Intel 8088 Intel 286 0 1E+03 1E+04 1E+05
Long-Range Strike (est.) ~10X Increase in Manageable Complexity
New IC design flow
Aerospace Vehicle 1990s
~5X Reduction in Development Effort
New automotive design flow
Goal
Automobile 1990s
Integrated Circuit Next Gen
Pentium Intel 386
Automobile Next Gen
1E+06 1E+07 1E+08 Complexity [Part Count + Source Lines of Code (SLOC)]
Data compiled by Mark Nowack, DARPA/TTO
Xeon
1E+09
1E+10
17
Existence proof—VLSI design
increasing abstraction
Transistor model Capacity load
Gate level model Capacity load
System-on-chip IP block performance Design Framework Inter IP communication Wire load performance models Abstract IP blocks
Abstract RTL Abstract
Cluster
RTL clusters
SW models
Cluster
Cluster
Feature Size (µm)
Transistors per chip
Speed (Hz)
Development time (mo)
Sources: Singh R., Trends in VLSI Design: Methodologies and CAD Tools, CEERI, Intel, The Evolution of a Revolution, and Sangiovanni-Vinventelli, A., Managing Complexity in IC Design, 2009
Daily engineer output (Trans/day)
18
Existence proof—foundry-style manufacturing The result: Moved from hundreds of chip designers using verticallyintegrated, captive semiconductor facilities to tens of thousands of designers using pure-play semiconductor foundries to create thousands of products.
An approach to VLSI chip design that separates design from manufacturing (Mead & Conway, 1979). Design implementation: Use of simplified device & component models that trade some performance for automation of design.
Design rules that are independent of and scalable with process technologies.
Semiconductor manufacturing facility becomes the semiconductor foundry. Semiconductor product implementation: Chip prototypes are manufactured in silicon foundries using the same
tools, fabrication processes and materials used for high-volume chip manufacturing… no seams.
Continues to enable, cost-effective custom VLSI products: Generating new markets & new companies including Apple, Silicon Graphics, Cadence, Jazz, TSMC, Broadcom, Nvidia and Qualcomm. 19
MultiAttribute
Design Update Feedback
Preference Surfaces
User Req’ment Synthesis
Visualization Metrics
QA/QC
Source: Paul Eremenko, DARPA/TTO
Constraints from Higher Levels of Abstraction
Rev. 10/18/2011 Design Trade Space Visualization
Dynamic Visualization
Structural & Entropy-Based Complexity Metrics Calculation
Design Space Construction(Stati c Models)
Static Constraint Solver
Qualitative/ Relational Models
Reachability Analysis
Lumped Parameter (ODE) Models
Nonlinear Differential Equation (PDE) Models
Controller/ FDIR Synthesis
CAD Geometry/ Grid Synthesis FEA CFD
Probabilistic Model Checker Requirements Verification
Foundry Trade Space Construct.
Foundry Design Instruction Sets
Monte Carlo Dynamic Sim
Probabilistic Certificate of Correctness
BOM
...
Modeling Languages
Semantic Integration
Component Model Library
DomainSpecific
Context Model Library
Model abstraction
The META-iFAB Integrated Tool Chain
Manufacturability Constraints
Hierarchical abstraction
Electromagnetic Thermal Mechanical Hydraulic Electrical
PLM
Ass’y Selection Process Mapping Machine Selection
Machine/Ass’y Mod Lib
CNC Generator
Process Model Library
20
Formal Model-Based Design
21
Component models As of today: • 131 component classes • 469 component instances • 43 parametric components • 112 ITAR protected models • 357 non-ITAR protected models
Source: Ricardo plc
22
Context models
Example • Probabilistic model of drive train • Extended to add vehicle load and state computation • Terrain context model specified as Markov chain
Gear_ratio
Trans_in_rpm
TransmissionController Trans_rot Gear_ratio_in
1 Throttle
Position
Trans_in_rpm_sensor
VehicleState
Throttle_in Engine_rot
Rot_in
Trans_rot_out
Slope_in Weight
Trans_load
Torque_out
VehicleLoad
DieselEngine Transmission
Contex t
Slope Position
Load_out
Trans_torque_in
Terrain
W VehicleWeight
Vehicle drivetrain
Extensi on
Contex t
Probabilistic Context Model •
Generate discrete time Markov chain (DTMC) models of terrain from digital elevation data
Specify terrain resolution for model and generate histogram (simple)
DTED elevation data
formula slope = s = 0 ? down : s = 1 ? level : s = 2 ? up : 1; module terrain s : [0..2] init [] s=0 -> 0.8 : [] s=1 -> 0.1 : [] s=2 -> 0.0 : endmodule
-0.02
Compute autocorrelation matrix for terrain data to incorporate relationship between adjacent locations (more realistic)
1; (s'=0) + 0.2 : (s'=1) + 0.0 : (s'=2) ; (s'=0) + 0.8 : (s'=1) + 0.1 : (s'=2) ; (s'=0) + 0.2 : (s'=1) + 0.8 : (s'=2) ;
-0.01
0.00
0.01
0.02
Markov chain representation of terrain for probabilistic model checking Customized for required horizontal and vertical resolution
Source: BAE Systems Land & Armaments Division
23
Integration of formal semantics across domains Composition • Continuous Time • Discrete Time • Discrete Event
META Semantic Integration
Simulink/ Stateflow
Embedded Software Modeling
Hybrid Bond Graph
Modelica
TrueTime
• Energy flows • Signal flows • Geometric
Functional Mock-up Unit Equations Modelica-XML
FMU-ME S-function FMU-CS
Formal Verification
Stochastic Co-Simulation
Distributed Simulation
• • • •
• Open Modelica • Delta Theta • Dymola
• • • •
Qualitative reasoning Relational abstraction Model checking Bounded model checking
NS3 OMNET Delta-3D CPN
Source: Vanderbilt ISIS
High Level Architecture Interface (HLA)
24
Hierarchical and model abstraction
Components
Assemblies ~102
Subsystems ~105
# of Design Alternatives
Hierarchal Abstraction
~10
System ~1010
Static Models
Qualitative Models
Relational Models
Linear / ODE
Nonlinear / PDE
Model Abstraction
25
Hierarchical abstraction—assembly level
Source: Vanderbilt ISIS
26
Hierarchical abstraction—subassembly/component level
Source: Vanderbilt ISIS
27
Cloud-hosted commercial tools instantiation
Source: CyDesign Labs
28
Model abstraction for verification Qualitative Reasoning
Static Trade Space Exploration Component Models Modelica State Flow Bond Graphs XML Geometry
Semantic Integration
• • • • •
Embedded Software Synthesis • Auto code generation • Generation of hardwarespecific timing models • Monte Carlo simulation sampling to co-verify • Hybrid model checking under investigation
• • • • • •
Static constraint application Manufacturability constraints Structural complexity metrics Info entropy complexity metrics Identify Pareto-dominant designs 10^10 10^4 designs
• • • • •
Linear Differential Equation Models
Relational Abstraction A
CAD & Partial Differential Equation Models • Generate composed CAD geometry for iFAB • Generate structured & unstructured grids • Provide constraints and input data to PDE solvers • Couple to existing FEA, CFD, EMI, & blast codes • 10 1 design Sources: GATech; Xerox PARC; SRI; Vanderbilt ISIS
Qualitative abstraction of dynamics Computationally inexpensive Quickly eliminate undesirable designs State space reachability analysis 10^4 10^3 designs
B
• Models are fully composable • Simulation trace sampling to verify correctness probability • Application of probabilistic model checking under investigation • 10^2 10 designs
• • • • •
Relational abstraction of dynamics Discretization of continuous state space Enables formal model checking State-space reachability analysis 10^3 10^2 designs
29
Verification on a adiabatic quantum computer Number of Qubits
1000
Vesuvius, 512 qubits
Leda, 28 qubits
Vesuvius Ranier
100 Leda Europa
10 Calypso 1 2002
Calypso, 4 qubits
2004
2006
2008
2010
2012
2014
ring
Hz es
8&
Median run time to 99% certainty [microseconds]
1.E+18 1.E+16 1.E+14 1.E+12 1.E+10 1.E+08 1.E+06 1.E+04 1.E+02 1.E+00 0 Source: USC/ISI
64
128
192
256
320
Number of Variables [N]
384
448
512 30
Probabilistic verification through simulation
Source: Vanderbilt ISIS
31
Probabilistic certificates of correctness (PCCs)
Source: Vanderbilt ISIS
32
Design space visualization
Source: Vanderbilt ISIS
33
Geometric composition for gridding/higher-order modeling
Source: Vanderbilt ISIS
34
Model-Based Manufacturing
35
Manufacturing process models As of today: • 7 material shaping processes • 19 general processes • 231 machine instantiations • 64 manual labor units • 3,212 tools
Sources: Penn State ARL; GM Research
36
Design decomposition
Topological Decomposition
Source: Xerox PARC
“Reverse Composition”
37
Foundry configuration tradespace exploration
Source: Penn State ARL
38
Sequencing & scheduling
Source: Penn State ARL
39
Tasking the distributed foundry & feedback to design Joint Manufacturing Technology Center Rock Island Arsenal, IL
• • • • Information Goods Agreements
Source: Penn State ARL
• •
ANALYSIS TIMING Part Decomposition - ~10 min Assembly Analysis - ~120 min Purchased Parts - ~1 min Manufactured Parts • aPriori - ~2 min/part • CNC-Ana - ~35 min Design Configuration - ~10 min Build Schedule Gen - ~5 min 40
Ecosystem
41
Source: Vanderbilt ISIS
Collaboration platform—configuration control
42
Collaboration platform—component model ontology
Source: Vanderbilt ISIS
43
Collaboration platform—immersive multi-user visualization
Sources: Electrotank; Vanderbilt ISIS
44
Critical scale for a model-based product ecosystem AUTOSAR Consortium
Two-Sided Market Model
Boston Fusion ROM Estimate of Investment Scale
Sources: Ferrari, A. An Overview of (Electronic) System Level Design: beyond hardware-software co-design, SFM-06:HV, Univ. of Urbino 2006; Jorge Tierno, Boston Fusion
45
FANG Challenge 1 – Mobility and Drivetrain subsystems
Prize: $1,000,000 Initial roll-out - 1/14/2013 Finalist team selection - 3/17/2013 Registration closes - 4/1/13 Challenge closes - 4/15/2013 Winner announced - 4/22/2013 Build - Summer 2013 (tbd)
As of today: • 1,077 participants • 267 total teams • 18 teams qualified for finals • Largest team size ~ 27
www.vehicleforge.org
46
FANG Challenge 2 – Chassis and Structural subsystems
Prize: $1,000,000
47
FANG Challenge 3 – Full Vehicle Design
Prize: $2,000,000 Winner entered into Marine ACV prototype fly-off
48
Modeling shows promise for 5X time compression Traditional Design Flow
META Design Flow
400 Req/M 400 10 Arch/M 80 400 Spec/M 600 2000 Tests/M 600 Req*/M 800
200 Req/M 200 5 Arch/M 40 200 Spec/M 300 1000 Tests/M 300 Req*/M 400
0
12
24
36
48
60 72 Time (Month)
Requirements Elicitation : METAm-on/off-with-change Concept Exploration : METAm-on/off-with-change Design and Integration : Metam-on/off-with-change Verification : METAm-on/off-with-change Validation : METAm-on/off-with-change Certificate of Completion : METAm-on-with -change
Source: Olivier de Weck, MIT
84
96
0
0
12
24
36
48
60 72 Time (Month)
84
96
Requirements/Month Architectures/Month Specifications/Month Tests/Month Requirements/Month
49
For more information: FANG Challenges: http://www.vehicleforge.org Source Code: http://www.cps-vo.org DARPA PM:
[email protected] Me:
[email protected] Coming soon… special issue of Journal of SE! 50