current approaches to system design and …...ge’s product offerings go deep and wide. from...

$: Current Approaches to System Design and …...GE’s product offerings go deep and wide. From Diagnostic imaging products across the standard modalities and new emerging mo\ alities$
Current Approaches to System Design and Modelling at GE Healthcare

Healthcare Information Day June Technical OMG Meeting

Chris UngerGE Healthcare

Page 2OMG MBSE and Healthcare Day June 2014

Introduction to GEHC


Broad solutions for healthcare

Medical diagnostics

Electronic medical records

Diagnostic imaging &surgery

technologies

Discovery tools

Integrated admin. & clinical

Clinicalproducts

Picture Archiving System (PACS)

Protein & cell sciences

Clinical tissue

biomarkers

Broad-based Technologies

LifeSciences

InformationTechnology

Presenter

Presentation Notes

GE’s product offerings go deep and wide. From Diagnostic imaging products across the standard modalities and new emerging modalities like Molecular Imaging, to interventional, surgical guidance, bio pharmaceutical production and integrated IT solutions


1

The Challenge… Energy Conversion & Detection

10-1510-1010-5 10-201051010

1MJ 400 J 2 mJ

Hummer @ 50 mph

Climbing 1 Step Pin Dropping

MR Min SignalMR 3T Gradient Coil

30 MJ 10-18 J1 kJ

Performix Pro e- Noise

8 fJ40 kJ

Presenter

Presentation Notes

The Hercules tube delivers more peak power than any other tube. With the move to volume scanning and fast gantry rotations, it is imperative to deliver high instantaneous power rather than high continuous power. To do this, Hercules has a large diameter anode. The Hercules tube will still be able to support fast gantry rotations since it has a larger bearing cartidge to support this larger target. With the smaller bearings used by our competitors, they will not be able to easily increase their anode size and so will not be able to produce enough dose for good image quality at the faster gantry rotations. Our competitors have chosen to deliver high average power, which is appropriate to single slice or luggage scanning systems. Each imaging modality requires different performance characteristics from the X-ray source. Historically, CT systems have required x-ray sources to produce a medium to high power exposure (24 – 60 kW) that lasts from 40 to 100 seconds and then to repeat it every 5 to 10 minutes. As the modality moves towards more cardiac imaging and volume scanning, the requirement for the tube will switch to very high instantaneous power (100 kW) and for relatively short exposure times (5-20 seconds). The systems will also scan faster thus resulting in a greater “g” loading on the tube. As a result the tube design will evolve in such a way that it can effectively deal with the high instantaneous power input to the track while as the same time keeping the mass of the tube unit within acceptable limits. Performix Power: Balanced bearing design, cg of anode located exactly midway between front and back races. Large Anode: increasing the focal spot radius allows higher instantaneous power on the focal spot without damaging the track. Smallest focal spot in the industry at these power levels, improves resolution of the images. e- collector: collects more than 50% of the off focal radiation which means less skin dose to the patient. What is the vision for the future tube? Vascular: smaller, lighter, higher power for faster acquisitions, better positioning, more robotics. CT: faster acquisitions, higher resolution…higher peak power on smaller focal spot sizes. Patient throughput will be limited to 10/hour, but the acquisition will take 5-10 seconds to freeze motion or the (arterial) contrast bolus. Future Tube: Duplex bearing design: 3X improvement in load carrying capacity means faster gantry rotations. Compact Design: new materials will allow for changes in the overall design of the tube thus resulting in a smaller size for a given power capacity. Higher Resolution: Changes in the focal spot shape and control will allow for improved instantaneous power capability and resolution.


GEHC Approach to New Product IntroductionTradition NPI process

How Modelling fits in

Recent additions• Formal Reliability process & team• Formal Usability process• Agile methodology (for SW)• Design for Producibility• Design for Six Sigma (revitalization)

Traditional artifactsRequirements = DOORs/Trace

(text based)Systems diagrams in “Visio”

(FBD, state machines, activity diagrams, …)

“Quantitative” performance simulations

Program Kickoff

System Req’tsFreeze

Hardware Freeze

Verification Complete

Pilot Release

Full Production Customer Satisfaction

Challenges• Lack of customer focus• Scope creep• Late integration issues• Lack of model integration• Poor requirements leveling

(capturing design as reqts)

SW: UML models

HW: Performance Models• EE: Cadence/Mentor (Chip->Board)• ME: Thermal, Structural,

Acoustic/Vibration, Life• Reliability allocations and models• Should cost modelling

Systems• Physics (IQ)

Systems• Behavioral• Customer FoM model

MFG: Capacity/Cost Models• Scrap/Cost models• Capacity/workflow models


Examples of Modelling


Method Latin Hypercube Sampling

Monte Carlo Factorial DOEFull/Fractional

Example

Cost Lowest Variable / Higher Highest (per space explored)

Where used Sparsely filling a large design space

Exploring a broad design space

Optimizing response near a design point

Why used Finds response function

Finding unexpecteddesign optima

Finds local response function

When used Medium priorsSemi-expensive sims

Low prior knowledgeInexpensive simulation

High prior knowledgeExpensive simulation

Design Space Exploration

Variable A

Varia

ble

B

Variable A

Varia

ble

B

Variable A

Varia

ble

BX X

X XX

XXX

XX

X

XX

XX

Presenter

Presentation Notes

Latin Hypercube and Monte Carlo are space filling, LHS is more cost effective but requires planning all your runs ahead of time, while MC will become better the longer it runs. Often you will want to study a particular sensitivity. Then the best is to employ a factorial DOE. Consider moving from one scheme to the other depending on what is best suited at this moment in your design process.


Robust Design using “Space Filling” computer experiments

Robustness: move design to center of feasible range

Optimality: move design along Pareto Optimal Edge to

maximize a third Figure of Merit

Needs: Efficient Simulation, Automated Parameterization, Great Visualization tools

Presenter

Presentation Notes

The design shown is just one of the clusters in the Holy Grail. The study here is about investigating that one cluster. DVT analysis of Ys (and Xs) reveals powerfully how a chosen design point is centered (or not) within feasible limits Moving the design as shown above is a graphical method to obtain a robust design Plot Xs vs Xs, Xs vs Ys, Ys vs Ys If you can generate simulation results at reasonable cost, you really should try this graphical design approach. It is also very educational, as you visualize quickly all kinds of correlations and dependencies. Smart cathode is a very successful design that is part of our High Definition tubes in CT.


Computed TomographyModerately complex system with complex behavior

- ~5,000 parts- ~5M lines of code- Triple nested control loops

- Axial, Cradle, mA/kV

First GEHC project using MBSE- <10 engineers using the tool- 3 year process- Principal engineer leads the

effort- Used several consultants to

review and optimize the process

- Focused on a few applications and a few critical components


Computed Tomography

MBSE techniques are used to perform behavioral analysis of key system features and functions.

- discover and verify system requirements

- identify and detail subsystem functions and interfaces

- seed FMEA analysis

- develop system test scenarios

act [Activity View] Position PatientWhiteBoxView [MoveToScan]

«Block» ScannerDesktop

notifyMoveToScanActive

notifyMoveToScanInactive

MTS_OFF

«Block» GantryUI

stopMoveIndicatorOff

moveToScanIndicatorOn

cradleOutIndicatorOn

cradleInIndicatorOn

moveToScanIndicatorOff

cradleInIndicatorOff

cradleOutIndicatorOff

stopMoveIndicatorOn

monitorMoveToScan

Technologist

monitorStopMove

Technologist

«Block» ScannerControl

[else]

[cradle_in_required == true]

CHECK_TOL

checkRequiredCradleDirection

[req_cradle_distance > auto_cradle_confirm_dist]

CHECK_TOL

[else]

[else]

calcReqCradleMoveDistance

[auto_longitudinal_advance == false]

retMoveCradleToStartPos«MessageAction»

[req_cradle_distance > cradle_tolerance_mm]

monitorMoveToScanStatus

[move_to_scan == pressed]MTS_CradleIn

[cradle_in == pressed]

MTS_CradleOut

[cradle_out == pressed]

CHECK_TOL CHECK_TOL

monitorStopMoveStatus

setMoveToScanStatus«MessageAction»

setStopMoveStatus«MessageAction»

retStopTableMotionStatus«MessageAction»

MTS_OFF

[else]

«Block» Table

moveCradleToStartPos

stopTableMotion [stop_move_pressed == true]




MTS_OFF

«Block» GantryUI




cradleInIndicatorOn




stopMoveIndicatorOn

monitorMoveToScan

Technologist

monitorStopMove

Technologist


[else]


CHECK_TOL



CHECK_TOL

[else]

[else]








MTS_CradleOut


CHECK_TOL CHECK_TOL





MTS_OFF

[else]

«Block» Table






MTS_OFF

«Block» GantryUI




cradleInIndicatorOn




stopMoveIndicatorOn

monitorMoveToScan

Technologist

monitorStopMove

Technologist


[else]


CHECK_TOL



CHECK_TOL

[else]

[else]








MTS_CradleOut


CHECK_TOL CHECK_TOL





MTS_OFF

[else]

«Block» Table






MTS_OFF

«Block» GantryUI




cradleInIndicatorOn




stopMoveIndicatorOn

monitorMoveToScan

Technologist

monitorStopMove

Technologist


[else]


CHECK_TOL



CHECK_TOL

[else]

[else]








MTS_CradleOut


CHECK_TOL CHECK_TOL





MTS_OFF

[else]

«Block» Table






MTS_OFF

«Block» GantryUI




cradleInIndicatorOn




stopMoveIndicatorOn

monitorMoveToScan

Technologist

monitorStopMove

Technologist


[else]


CHECK_TOL



CHECK_TOL

[else]

[else]








MTS_CradleOut


CHECK_TOL CHECK_TOL





MTS_OFF

[else]

«Block» Table



sd [Package] RevolutionSystemWB_UcSD [MTS_CradleInNormal]

parallelparallelparallelparallel

:GantryUI

cradleInIndicatorOff()

moveToScanIndicatorOff()

:ScannerControl

monitorMoveToScanStatus()

reqMoveToScanIndicatorOff()

reqCradleInIndicatorOff()

:Table

moveCradleToStartPos()

reqMoveCradleToStartPos()

retMoveCradleToStartPosStatus()

stm [State] MoveT oScan [StatechartOfMoveToScan]

MoveToScan

state_25

checkRequiredCradleDirection(cradle_in_required, cradle_out_required);cradleInIndicatorOff(); cradleOutIndicatorOff();

WaitFor_Mov eToScan

moveToScanIndicatorOn();moveToScanIndicatorOff();

[cradle_in_required == true]/cradleInIndicatorOn();

[cradle_out_required == true]/cradleOutIndicatorOn();

ManualMoveOut

OUT_AT_POS

OUT_PRESSED

AT_OUT_LIMIT

evCradleOutPressed

evCradleOutReleased/stopTableMotion();

ManualMoveIn

IN_AT_POS

IN_PRESSED

IN_AT_LIMIT

evCradleInPressed

evCradleInReleased/stopTableMotion();

Mov ingToPos

stopMoveIndicatorOn(); moveCradle...

evMov eToScan/cradleInIndicatorOff();cradleOutIndicatorOff();

AUTO

CradleReleased[else]

[cradle_engaged == false]

evStopMove/stopTableMotion();stopMoveIndicatorOff();

evCradleReleasePressed/engageCradleDriv e();

evCradleReleasePressed/disengageCradleDriv e();

AUTO

CheckingDistance

calcReqCradleMoveDistance(req_cradle_distance);

[auto_cradle_adv ance == true] [else]


[cradle_mov e_done == true]/stopMoveIndicatorOff();

tm(t_move_to_scan_tmo)

tm(t_move_to_scan_tmo)

MTS_TIMEOUT

AT_POSITION


Computed TomographyCT Systems is deploying several model based designs directly to software andhardware.

Simulation

kV

InverterVoltage

InverterCurrent

Lab HW

kV

InverterVoltage

InverterCurrent

x-Ray Generator KV Control Loop- Control/Plant models designed/analyzed in SIMULINK. - Auto-generated vhdl

Active X-Ray Beam Position Control- Control/Plant models designed/analyzed in SIMULINK. - Auto-generating C++ code

Cardiac Acquisition and Emission Modulation- Feature analysis and simulation performed in

SIMULINK- Auto-generating C++ code

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4-0.2

0

0.2

0.4

0.6

0.8

1

1.2

Time (seconds)

loop e

rror B


Customer Workflow Modeling

ED Renovation

Client Scenario Simulation Results

Current ED old and over-crowded, client planning to dramatically expand / replace

existing capacity in 3 phases while continuing to provide 24/7 emergency care services.

Construction Cost $1.3M

Staffing Costs $2M

Reduced Waiting & LoS +25% vol

Simulation enabled client to “shell” one pod and redesign staffing• Gather the requirements: observational

research, data mining from records• Proprietary GE Tool (capacity vs. staffing,

equipment, layout…)• Review conclusions and recommendations


GEHC Modelling Maturity LevelsHighly Mature• Quantitative

Modelling• Field Strength• Air flow• Noise• Resolution• Structure / vibration• Electronics• …

Needs

• Customer Work Systems

• Disease state models• Interoperability• Outcomes (health, economic)

Developing• Process map/Utilization

• Factory utilization simulations

• Customer workflow productivity

• Customer Task QoS• Tumor Visualization• Artifacts

• Cost• Integrated should cost simulations

• Integrated System Models

• Image quality from customer to components

• Architecture model


Future Challenges


The industry faces many challengesThe medical industry product developers face problems with ……• Extreme time to market pressures

– 1st to market usually gains 80% of that market• Compliance with regulations

– FDA, IEC, ISO, HIPAA, ICD-10, ACA, etc.• Defects are VERY costly to handle

– Want to avoid audit, decrees, warning letters, recalls, etc...• Most products are developed in a geographically distributed way

– Need to communicate and define tasks• Technology is impacting development and delivery

– IoT, product variants, Mobile Medical Apps, complex deployment models, cloud

Source: IBM Rational


Key Industry Challenges for MBSE adoptionWhat are the most critical barriers to faster adoption of MBSE? High barrier

to entry with uncertain payback• ROI – Assured cost, Unquantified return

• Fear of the unknown – no clear success stories with a business case• Many best practices…you pay for the tools and then need to pay for a consultant to

tailor a process• Difficulty to understand how to introduce on an existing product – how to start? (not

going to throw out the existing DOORS requirements database)• Many things don’t scale…need an incredible investment…hard to justify

• Concerns about FDA acceptance• The tools are not validated archival mechanisms, so the archive has to be done in a

document storage tool (in textual requirements)• If we have to capture everything in textual requirements anyway (for audits), what is

the advantage of the model?


Lowering the barrier to entryManagement is confronted with many competing priorities for investment

Recommendation: Develop an implementation use case/cookbook, with a library of testimonials/businesses cases for upper management

Biggest cost is not the tool…need a way to make ‘the pill easier to swallow’• Big bang: full in on one project, with a complete strategy…needs business case for upper

management to justify the investment• Get your feet wet: partial implementation (one feature, one subsystem)…needs cookbook

on how best to integrate a partial MBSE implementation with prior processes and tools

Tool Tailoring

Training

Trial &

ErrorUsage

CostAvoidance

QualityBenefit

VisionReuse

Time

Inve

stm

ent

Ret

urn


MBSE Challenge INCOSE productINCOSE (International Council on Systems Engineering) has working

groups on Biomedical Healthcare and Model Based Systems Engineering

Those WGs have sponsored a Healthcare MBSE challenge group developing a medical pump model

Recommendation: INCOSE publish a reference model

• Demonstrate the value and utility of MBSE for biomedical-healthcare related applications

• Develop frameworks and templates that can be used to accelerate the development and approval of biomedical devices.

• Demonstrate integration of risk management, safety assurance, and other regulatory concerns.

• Capture learnings on how to make the shift from a document-centric to a model-centric systems engineering environment


Concern: Regulatory Acceptance

Reporting of computational modeling studies in medical device submissions

Draft Guidance for Industry and FDA StaffOwner: Tina M. Morrison, Ph.D., [email protected].

Example CDRH Modelling PaperOne concern is that regulations can impede progress toward higher quality processes• Auditors can be unclear on what is

acceptable in a model, and where to poke for quality gaps

• FDA has published a draft guidance on computational (quantitative) modelling for industry and

• Gives guidance on what to include…in general, and for four types of models

• Does not address behavioral/architecture (SysML) models

A consistent approach on how to summarize, review, and document models would ease acceptance

Recommendation: FDA (and industry) publish a guidance on submitting behavioral simulation results


Summary – Benefits to Industry of MBSEImproved Systems Thinking

• Use Case/Performance/Interface Analysis critical for a complete design specification.• Logical model to provide high level of abstraction for ease of understanding, improved

reuse or design sharing

Improved Communication• Visual vs. Textual leads to Clearer, more precise communication & better reviews• Visual designs & models are easier for global teams (less language barrier)

Improved Quality• Verify correctness and completeness of requirements/design – robustness / stress

testing of design rather than simply reviewing in quality• Improved design of test cases, derived from weaknesses exposed in the model

Improved Predictability and Efficiency (Time to Market)• Verify correctness and completeness of requirements/design – robustness / stress

testing of design rather than simply reviewing in quality• Improved leveling of requirements (efficiency in verification and documentation)• Auto code generation (no translation errors in implementation)

Questions?


APPENDIX


Systems Engineering: Turning Needs into Solutions

A Great Integrated Design happens when Systems Engineers Thinkers are effective and enabled

• The product seamlessly integrates into the customer’s workflow and systems, reliably meets all their needs, and delights the customer,

• The key performance parameters (CTQs) ensure robust delivery of clear market differentiation,

• technical scope/program work is clearly tied to market impact, • technical risks are retired early and robustly, • creative ideas come from everyone and designs are optimized across

organizational boundaries, • design decisions are clearly identified and closed in a timely fashion (and

stay closed), • designs integrate easily, • quality problems (when they exist) are found and resolved early and few

design issues escape to the field, and• institutional knowledge is available to everyone when and how they need it.

Presenter

Presentation Notes

Hino details – Q4 2012 HAYST pilot: Getting good results to detect very rare bugs before V&V (Perenna Tube Control Firmware, DV24 Host SW, etc.) SWTC to start evaluation soon. DRBFM: Identified >20 technical risk areas to apply DRBFM for Cj3.0 and >50 for KIZUNA. Full deployment for Single Crystal Probe (Kagura). Focused 1 week DFSS training for 15 new employees & 20 hours DFSS recap course for 15 PE/Arch. DFSS course included Taguchi’s Robust Design. Hino MR’s good practice: Technical risks identification by DRBFM HALT in MKE. (DV24 DTRSW) Manage/evaluate technical risks early in development .


What’s Working? What Can Be Improved?

Threats• Infrastructure Cost (tool, training, time)• Focus on the tool, not the (thought) process• Communication of models to non-modelers• No guide on how to get started

• Small enough to be successful• Large enough to be complete; net impact has a positive

ROI

Weaknesses / Challenges• Many regulations (and increasing)

• Unclear standards (leads to over-documenting)

• Global markets…need local understanding• Challenge of language skill levels• Lots of small competitors

• High Quality Standards (ALARP vs. No Defects)• Still get late quality surprises in new products

• Global markets…need local understanding• Challenge of non-uniform language skill levels

• No standardized systems process• Systems Thinking is hard to teach (or even develop!)

Opportunities• Behavioral modelling

• Higher standard quality (tested models with margin)

• Reusable for shared components• Improved handoffs between functions

Strengths• Excellent, hard working engineers• Global footprint with good reputation• Good quantitative modelling capability• Broad product offering

current approaches to system design and …...ge’s product offerings go deep and wide. from...

Documents