combining ai and cae in order to save lives
TRANSCRIPT
Agenda
Who is MDGo?
Real-time trauma analysis
Utilizing CAE for AI training
How can MDGo improve CAE accuracy?
Who isMDGo?
MDGo is all about providing customer-oriented holistic support in the case of a car crash
MedicineAISafety
Ongoing collaboration and integrationwith the Israeli medical community
In-house CAEBiomechanical research Employing the top-experts
of AI technology in theIsraeli start-up ecosystem
Vehicle damage & medical information does not exist in real time accident events
According to the EU commission, 44% of car crash fatalities could have been saved if treated according to their specific injury
A long, complicated and vague journey back to normality
Crash
RecoveryDamage
What we want to know
No significant injuries
Serious-Severe Chest Injuries• Skeletal injury: flail chest with 3-5 flail ribs.• Soft tissue injury:
• Hemothorax or pneumothorax.• Lung contusion.
Minor Neck injuriesSoft-tissue injury
Minor lacerations
From Sensor to injury
“Even when the Delta V can be calculated exactly, the amount of force applied to an occupant’s head is literally anyone’s guess”
Measure changes in energy
Medical reports
regression model
𝑃𝑃 𝐴𝐴𝐴𝐴𝐴𝐴 ≥ 3 =1
1 + 𝑒𝑒𝛼𝛼−𝛽𝛽𝚫𝚫𝒗𝒗Probability for severe injury
“ΔV is not a conclusive predictor for cervical spine injury in real-life motor vehicle accidents.”
“Delta-V estimates for cars only were greatly improved but still understated by 16% on average.”
taking the leap
What if we can measure the human rather than the vehicle?
In-cabin sensing Existing vehicle sensors• Monitor the occupants• Not (yet) widespread• Does not directly measure forces
• Monitor the vehicle• Common• Available through OEMs and after market devices
Image by Eyesight Image by NXP semiconductors
Utilizing AI to unlock the data
Crash pulse Neck force Head acceleration
The key – measurable in real-life Human organ measurements – not available in real-life
Crash test predictions
Target driverRib Lateral Acceleration
Target driverHead Lateral Acceleration
Bullet driverAxial Force on Femur
Bullet driverChest Displacement
Getting the dataA deep-learning algorithm is only good as it’s training data
OpenCrash-test Databases
In-houseCAE
In-houseCrash-testing
DataAmplification
In-house
Vice versa – using MDGo for CAE
SolverLoad case
Crash pulse
Dummy pulses Riskfunctions Dummy injury prediction
MDGo’s AI algorithm
Human injury prediction
Human-body pulses
How can MDGo improve CAE
D e l t a
virtual-sensors + medical report = We know what happened in the event
Ground truthoutcomes
CAE,Crash-tests
MDGoVersion-0
Injury riskprediction 0
Injury riskprediction 1
Injury riskprediction 2
MDGoVersion-2
MDGoVersion-1
Trauma unit m
edical reports
Israel test siteA unique ecosystem which allows a complete loop with the medical community
Hyundai-MG EVC Chung visit to MDGo
27 hospitals
National trauma registrarNational EMS
250,000 connected vehicles
Conclusion
Bridge the bio-fidelity gap
Validate CAE with real-life crashes
Improves with number of connected vehicles
Can be tailored to specific vehicle models
CAE is used as a training source for AI algorithm The virtual sensors method, coupled with medical validation, can be used to improve CAE accuracy
(almost) Unlimited flexibility in load case creation
Results are as good as the FE models (number and quality)