sameer m. prabhu, ph. d. industry marketing director sameer.prabhu@mathworks · finite-state...
TRANSCRIPT
© 2012 The MathWorks, Inc.
Getting to Smart
Sameer M. Prabhu, Ph. D.
Industry Marketing Director
smart│ systems of systems
getting to smart
Experimentation system that is: – easy to use
– transparent in operation
– and flexible
Finite-state machine execution with interfaces
to C++, ROS, and MATLAB
MATLAB: – quadrotor modeling, control design, trajectory planning, simulation
– interface for experiment definition and execution
Continual development of increasingly complex behaviors
Courtesy: Daniel Mellinger, Vijay Kumar, et al.
GRASP Lab, University of Pennsylvania
With looptune, you can tune the
whole control system in one shot!
smart│ systems of systems
getting to smart – what we are doing
Simulink
Control
Design
With looptune, you can tune the
whole control system in one shot!
smart│ systems of systems
getting to smart – what we are doing
Simulink
Control
Design
development process smart│ Credit: General Motors LLC 2011
Creating and Choosing
Engine
Generator
Electric
Drive Unit
Li-ion
Battery
Control
Strategies
Design Reuse
“We‟re using the same models to control different systems.
It benefits us to use these [models], single sources that
people can understand; it allows us to go faster.”
Greg Hubbard Senior Manager of Hybrid and Electric Drive Controls, GM
smart│ development process Credit: General Motors LLC 2011
Collaborative Team Development
Keynote: The Hope and Reality of Global Model-Based Development
Kent Helfrich, Executive Director, Electronic Controls & Software, General Motors
MathWorks Automotive Conference, May 4, 2011, Novi, MI, USA smart│ development process
smart│ development process
getting to smart
Concurrently tackling
what is possible (the science)
what is feasible and optimal (the engineering)
Leveraging Model-Based Design
across development centers around the world
across engineering teams
across development phases
across vehicles with the same design platform
smart│ development process
Modeling paradigms
Simscape + Physical Modeling
Discrete Events
Model Reference
Variant Modeling
Simulink Projects Local Source
Control
Subversion
(SVN)
Other Source
Control Tools
Connectivity
through
adapter
getting to smart – what we are doing
Courtesy of FLUXNET
smart│ understanding
FEEDING ARTERIES
DRAINING VEINS
NIDUS
Arteriovenous Malformation (AVM) Philips Electronics India Ltd.
getting to smart
Automatic segmentation of image into various
compartments as arteries and veins
Electrical circuit for each segmented vessel of the
compartment using R,L,C
Improved understanding of pressure that causes rupture in
brains
Courtesy: Y.Kiran Kumar
Philips Electronics India Ltd. smart│ understanding
smart│ understanding
• Big images or large
datasets of small
images
• Block processing
• “UseParallel” mode
• Automatic Image
Registration
• align images from
different data sets
getting to smart – what we are doing
smart│ decisions
getting to smart
The domain expert created the tool
Decision-support tool based on data
Performance improves as more data is provided
HKM optimize a steel production process
smart│ organization
Constraints – complex process
– equipment capability
– raw material availability
– environmental regulations
– demands of multiple customers
Previously: – dispatcher scheduled entire plant manually
– required extensive process knowledge - only a few qualified
– significant business risk If dispatcher is absent
HKM algorithms at work
smart│ organization
Develop algorithm – Algorithm development accelerated by a factor of 10
– Optimization time cut from 1 hour to 5 minutes
Deploy algorithm – integrate into HKM‟s existing scheduling application.
getting to smart
Scale to fit: from prototyping ideas to
24/7 number-crunching
“In some cases the automated system
has produced a solution that our
dispatchers did not consider.”
Deploy knowledge to others in the
organization
smart│ organization
Parallel Computing Toolbox and MATLAB
Distributed Computing Server – Multiple cores to computing clusters
– The number of GPU-enabled MATLAB
functions continues to grow in
– GPU-enabled functions can be
deployed using MATLAB Compiler
MATLAB Mobile – iPhone and iPad
– Access „in the field‟
smart│ organization
getting to smart – what we are doing
smart│ component design
The Semtech SX1231 wireless RF transceiver.
Semtech transitioning from analog to digital
To improve performance and
decrease power consumption
Want to evaluate multiple
design ideas during the
study phase
Want to improve the
development workflow,
especially between study
phase and implementation
smart│ component design
getting to smart
Two engineers
System-level model, including channel noise
and demodulation schemes
Digital filters, converters, PLLs
designed to integrate together
Multiple designs explored,
analyzed, and optimized
Automatic HDL generation meant
FPGA prototypes 50% faster
“We used the time we saved to improve the design”
getting to smart
Modeling
different abstractions,
multiple purposes
System simulation and analysis
Component design exploration
System integration
Automatic HDL code generation
Test benches
smart│ component design
The Semtech SX1231 wireless transceiver. The Semtech SX1231 wireless transceiver.
smart│ component design
HDL Coder and HDL Verifier
IMPLEMENTATION & VERIFICATION
DESIGN
FPGA ASIC
VHDL, Verilog
MATLAB & Simulink
HDL Coder
• Prototype and
implement algorithms
on FPGAs or ASICs
• Verify HDL code using
HDL co-simulation or
FPGA-based simulation
• Support for DSP and
Communications
algorithms
getting to smart – what we are doing
HDL Verifier
smart│ people
getting to smart
Combining classroom instruction and hands-on work
From basic concepts of kinematics and control design,
to the implementation and certification of flight code
Attracting, engaging, and retaining students,
and equipping them to succeed in industry
smart│ people
Simulink support for Lego, Arduino, and Beagle Board
Real-Time Windows Target
xPC Target Turnkey
getting to smart – what we are doing
smart│ in many ways
Weinmann MEDUMAT Transport ventilator
The MEDUMAT Transport ventilator.
Image © Weinmann Medical Technology.
Designed for emergency
transport and care damage to lungs or chest wall?
can patient breathe?
using mask or tube?
Multiple operating modes, ranging
from assistive to full control
Must be portable and
exceptionally reliable
Requires certification to ISO/IEC 62304, ISO 10651-3,
DIN EN ISO 13485, DIN EN ISO 14971
smart│ in many ways
Embedded Coder –
production code for control system
and sensor signal processing
Simulink –
plant model includes
hardware components
and a mechanical model
of human lungs
Simulink and Stateflow –
model of controller and
state machines
DSP System Toolbox –
digital filters and algorithms to
calculate gas flow velocity
Simulink Verification
and Validation –
model coverage analysis
for unit tests
Weinmann MEDUMAT Transport ventilator
Embedded Coder –
targets code to
Infineon MCU and
TI DSP
Image © Weinmann Medical Technology.
smart│ in many ways
Weinmann MEDUMAT Transport ventilator
50% faster code development and reviews
100% automatically generated
embedded software
50% less time on internal reviews
Over 40 design alternatives explored
60% reuse of core design in next-generation
ventilator, which uses a different processor
Image © Weinmann
Medical Technology.
smart components
systems
systems of systems
development processes
understanding
decisions
people
organizations
getting to
Do you….
… have a strong foundation of algorithms and components
to build “smart” systems?
… use models to share ideas and verify implementations?
… organize to develop product lines with re-used components?
… coordinate your research and product development teams?
… put algorithms and decision-making tools
in the hands of all the right people?
… educate students with theory, modeling, hands-on practice,
and a systems perspective?
… hire for those skills and build them in your teams?