《RENESAS：2024车端AI实时分析系统报告（英文版）（25页）.pdf》由会员分享，可在线阅读，更多相关《RENESAS：2024车端AI实时分析系统报告（英文版）（25页）.pdf（25页珍藏版）》请在三个皮匠报告上搜索。

1、 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALREAL-TIME ANALYTICS FOR ENDPOINT AI IN AUTOMOTIVE SYSTEMS JUNE 2024KURT DEKOSKIBUSINESS DEVELOPMENT AND PRODUCT MANAGEMENT FOR AUTOMOTIVE SWSEPBDRENESAS ELECTRONICS CORPORATION 2024 Renesas Electronics Corporation.All righ

2、ts reserved.RENESAS CONFIDENTIALPage 2ML Model DevelopmentExplainability+Hardware AnalyticsSolutions+Reference DesignsA platform powering Edge AIon Renesas processorsfrom 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 3REALITY AI PLATFORM COMBINES ADVANCED SIGNAL P

3、ROCESSING AND MACHINE LEARNING ON MCU/MPU EDGE NODESAdvanced Signal ProcessingAutomatically searches a wide range of signal-processing transforms to create a custom,optimized feature transformMCU/MPU Edge NodesRuns on almost every MCU and MPU core available from Renesas,with new ones added constantl

4、y.Reality AI also supports Renesas motor control boards.SCALABLE FROM 16-BIT to 64-BIT CORESArtificial Intelligence and Anomaly DetectionAutomatically generates machine learning models,explanatory visualizations and hardware design analytics 2024 Renesas Electronics Corporation.All rights reserved.R

5、ENESAS CONFIDENTIALPage 4AI/ML TECHNOLOGY Changing Industry Needs Driving AutomotiveElectrificationBenefits of AI and ML Reduced human error,especially in processes that involve analyzing numerical or signal data.AI wont make oversights due to fatigue or error as a human might.Increase operational e

6、fficiency by automating more functions and identifying and consolidating redundant processes.AutonomousDriving Increased driving safety Decreased reliance on driverSoftware Defined VehicleDigitization and AI Software-first development structure Sensor fusion Reality AI provides the smallest models i

7、n the market in terms of RAM,Flash and processing cycles Reality AI models are supported for Renesas MCUs 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 5AUTONOMOUS MEGA TREND INCREASED LEVEL OF AUTONOMY,REDUCED LEVEL OF DRIVER ENGAGEMENT INCREASED LEVEL OF FEATURE

8、S TO MIMIC DRIVER PERCEPTIONAI software:AI for self-driving cars using sensor-fed machine learning algorithmsAV hardware:Sensors provide the key data inputs to the cars AI systemsLevel 2 AutonomyPartial AutomationVehicle has combined automated functions like acceleration and steering,but the driver

9、must remain engaged with the driving task and monitor the environment at all times.Level 3 AutonomyConditional AutomationDriver is a necessity but is not required to monitor the environment.The driver must be ready to take control of the vehicle at all times with notice.Level 4 AutonomyHigh Automati

10、onThe vehicle is capable of performing all driving functions under certain conditions.The driver may have the option to control the vehicle.Level 5 AutonomyFull AutomationThe vehicle is capable of performing all driving functions under all conditions.The driver may have the option to control the veh

11、icle.Levels of vehicular autonomy defined by the Society of Automotive Engineers(SAE)2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 6REALITY AI SOFTWARE AND SOLUTIONS ADDRESS COMMON ENGINEERING FRUSTRATIONS WITH AI/MLEngineers frustrations with AI/MLHow Reality AI

12、makes it betterMostly use AI for difficult problems where traditional methods arent working and there are no obvious solutionsBuilt specifically for non-visual sensing based on advanced signal processing math and edge deployment on Renesas MCUsInstrumentation and hardware issues add complexity,most

13、optimize for costAnalytics to support hardware design(not just algorithms and model building)Can not deploy“black box”solutionsExplainability based on time-frequency characteristics with full transparency 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 7REALITY AI P

14、RODUCT PORTFOLIOOFFERS BOTH SOFTWARE AND SOLUTIONSSoftware for model creation and hardware optimization on Renesas MCUs,MPUs and motor control kitsComplete framework for specific use cases,including hardware,firmware,software and ML reference designsReality AI SoftwareReality AI ToolsBuild models,vi

15、sualize explanations,optimize hardware buildRealityCheckMotorFor embedded predictive maintenance and control feedbackRealityCheckADFor cloud-based anomaly detection in factory and on-lineReality AI SolutionsRealityCheckHVACComplete framework for smart,self-diagnosing HVACAutomotive SWSComplete frame

16、work for audio-based ADAS sensing for vehiclesArm CoreRenesas Core 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 8Signal ClassificationAnomaly DetectionRegression(continuous value prediction)REALITY AI PRODUCT PORTFOLIO OVERVIEWGetting Started PackageDemo VideosSo

17、ftware ManualApplication NotesSWS-EVDRealityCheck HVACRealityCheckADNative deployment available now for all Renesas RA,RX,RZ,RL78,RH850,and R-Car families of devicesRealityCheckMotor 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 9REALITY AI ADDRESSES THE FULL AIOT

18、 DESIGN LIFECYCLEOnly 5%of typical project costs are spent on building modelsSoftware helps with the other 95%Instrumentation&hardware development,data collection,validation and internal confidence buildingAlgorithmically-Driven Feature DiscoveryAI ExploreTM(AutoML)Sensor Selection and BOM Optimizat

19、ionData ReadinessEdge AI/TinyMLCode Optimization 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 10RENESAS/REALITY AI FAMILYCustomers Application Software StackApplication Input SignalOutput Application TriggersML/AI Generated CodeAI ExploreTM(AutoML)Sensor Selectio

20、n and BOM OptimizationData ReadinessEdge AI/TinyMLCode OptimizationNative deployment available now for all Renesas RA,RX,RZ,RL78,RH850,and R-Car families of devicesAutoML(no coding)ExplainabilityCost-optimized specificationsMinimum sensor setAutomated ConsistencyQualityCoverageEmbedded code generati

21、on C,C+Ease of deploymentMATLAB compatibility 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 11REALITY AI TOOLSAPPLICATION INPUT SIGNAL100 dB Siren 0m100 dB Siren 15m30+dB shift in overall noise levelFully explainable.“When the resonance peak at 146hz disappears,we

22、 know the fan is blocked”Image 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 12“GET STARTED”WITH REALITY AI PRODUCT PORTFOLIOMost customers start with one of our“Get Started”packages that combine software with enhanced onboarding and services,before switching to a

23、n annual software subscription8-week Trial RunRequirements and data assessmentOnboarding and setupTwo iterations run by Reality AI teamUser training on own data and modelsUp to four weeks of access for evaluation and continued development4-month Prototype PoC8-week Trial Run with access extended to

24、four monthsEnhanced support from Reality AI,including pre-processing,model construction and post-processingConstruction of laptop-based or other non-embedded prototype/demo6-month Edge AI Prototype4-month Prototype PoC,with access extended to six monthsDelivery of Edge AI model compiled for Cortex M

25、-class architecture or LinuxAssistance with hardware and application integration/debugging8-week“Phase 0”Requirements and data assessmentData coverage and collection planInstrumentation/hardware planFeasibility PoC planR&D roadmap with machine learning5-month PoC8-week“Phase 0”Monitoring of client d

26、ata collectionFour iterations run by Reality AI teamUser training on own data and modelsValidation statistics for evaluation of PoC results7-month Working Prototype8-week“Phase 0”Monitoring of client data collection5-month POCConstruction of laptop-based or another non-embedded prototype/demoStarter

27、PoCPrototypeI have dataI dont have data 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 13POWEREFFICIENCYREALITY AI DEVELOPMENT WITH RENESAS AUTOMOTIVE MCUs and SoCsArmCoreRenesas Core32-bit MCU64-bit SoC16-bit MCUSCALABLE WITH SoCHIGH PERFORMANCESUPER LOW POWER32-b

28、it MCURenesas meets the varied requirements of our customers with a rich lineup of automotive MCUs and SoCs Mass ProductionSamplingDevelopmentConceptEV Drive MotorCBMRealityCheckMotorSWS-EVDBMSRealityCheckHVACOD-GRRealityCheck ADEmbeddedAI-enabledSolutionReality AI Tools 2024 Renesas Electronics Cor

29、poration.All rights reserved.RENESAS CONFIDENTIALPage 14AUTOMOTIVE,CONSUMER AND INDUSTRIAL IOT APPLICATIONSReality AI TinyMLmodels run on every Renesas MCUHear approaching emergency vehicles,cars and bicyclesDetect weather,terrain,road conditions and micro-collisionsMake tires that report their stat

30、e of wearHome Appliance electric motors that monitor the health of the systems they drivePredict e-bike chain deteriorationPumps that know they are dry or cavitatingFilters that know how long until they will clogAir conditioners that predict their own maintenance needsRemote livestock behavior monit

31、oringDiagnose the source of power quality issues based on 3-phase ACAutomotive SWSComplete framework for audio-based ADAS sensing for vehiclesRealityCheck HVACComplete framework for smart,self-diagnosing HVAC(now engaging with early adopter customers)RealityCheck MotorFor embedded predictive mainten

32、ance and control feedback 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 15TARGET USE CASE SEEING WITH SOUNDEnvironment participants emit sounds that can often be heard before they are seenAround cornersObstructed viewsAt a distanceVehicle blind spotsWhere other se

33、nsors cannot seeWhere drivers are not focusedUsing sounds,the AI technology for SWS provides early warning of emergency sirens,on-coming vehicles and other sound sources to augment existing autonamous driving and ADAS systemsAutomotive Application FeaturesEmergency Vehicle Detection+Blind Spot Detec

34、tion+Other Vehicles+Horn Honks+Pedestrians+CyclistsAutonomous Emergency Braking Pedestrian/CyclistCyclist Dooring Prevention+Pavement Conditions/TerrainWhy SWS Contributing factors and causes:Distracted Driving Active and Passive Noise Cancelation SWS-EVD 2024 Renesas Electronics Corporation.All rig

35、hts reserved.RENESAS CONFIDENTIALPage 16TARGET USE CASE SEEING WITH SOUNDDemonstration VideoSirens 1km normal operating conditions and speedsVehicles 30m obstructed view/intersection with moderate noiseBlind spots 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 17 T

36、argeting various noise patterns that indicate differing road surfaces that affect driving habits Dry Road Surface Wet Road Surface(how much)Snowy/Icy Road Surface Gravel or Sand Road Surfaces Cobblestone surfaces Development Level=PoCTARGET USE CASE:CONDITION MONITORINGROAD CONDITIONS AND TERRAINArm

37、CoreRenesas Core 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 18TARGET USE CASEMICRO-COLLISION Detection of small impacts against the vehicle Glass Breakage Scrapes/Scratches Small Dings/DentsDevelopment Level=PoC ArmCoreRenesas Core 2024 Renesas Electronics Corp

38、oration.All rights reserved.RENESAS CONFIDENTIALPage 19TARGET USE CASEAUDIO AND VOICE COMMANDS Presence Detection of each occupant and seating location.Allows for targeted Sound Zone for audio performance Allows for controlled in-cabin communication Anti-spoofing Recorded verse actualIdentify occupa

39、nts for individualized sound zoneClear communication from front to rear occupantsAnti-Spoofing DemoAI/ML trained Voice based User Access ControlAnti SpoofingVoiceArmCoreRenesas Core 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 20TARGET USE CASESPRESENCE DETECTION

40、 AND DRIVER/OCCUPANT BEHAVIOR AND PROFILINGPresence DetectionMethod 1:Sound VideoUse AI/ML to learn GesturesOccupant MovingOccupant BreathingGesture RecognitionPresence Detection and Occupant Behavior and ProfilingMethod 2:Doppler Radar VideoMethod 3:UWBUses doppler radar or UWB to distinguish betwe

41、en humans,pets and inanimate objects(shifting in seat,jostling,breathing,gestures,etc.)AI/ML trained to recognize human soundsOccupant DetectionRenesas Core 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 21TARGET USE CASES:CONDITION MONITORINGELECTRIC MOTOR (Realit

42、yCheck Motor)Enables sensor-less ML models to be deployed directly to the motor control board:Uses electrical information already available on the boardTakes high sample-rate current and voltage as a proxy extra components(like accelerometers)EV Drive Motor Development Level=ConceptDevelopment Level

43、 for Small motors=Pre-ProductionDemo VideoAI/ML Motor Speed/Torque+Sensor-less FOCHigh SpeedLow TorqueLow SpeedHigh ToqueMotor Speed/TorqueAI/ML Motor Mis-alignment+Sensor-less FOCAlignedMis-AlignedMotor Mis-alignmentAI/ML Unbalanced Load Detection+Sensor-less FOCNO ADDITIONAL SENSORS!Leverages Avai

44、lable Electrical&Motor SignalsBalancedUnbalanced12mm M4 screw,11mm off center to create unbalanced conditionUnbalanced DetectionRealityCheckMotor 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 22TARGET USE CASES:CONDITION MONITORINGPREDICTIVE MAINTENANCE,REMAINING

45、USEFUL LIFE AND ANOMALY DETECTIONDemo Video Uses external microphones mounted in wheel wells to monitor the state of wear of tires Uses vibration and motor cues to determine remaining filter life Uses vibration cues to monitor rotating shafts for balance and operating condition and anomaly detection

46、 Utilizing vibration cues for pumps that know they are dry or cavitatingDevelopment Level=Pre-ProductionRenesas Core 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 23TARGET USE CASE:CONDITION MONITORINGRealityCheck HVACUsing machine learning and edge processing,it

47、is now possible to build HVAC units with the native ability to detect a wide range of faults and operating conditionsTarget Conditions for Detection/PredictionBlocked cabin/outdoor airflowCondenser/evaporator blockageRefrigerant charge issues undercharge/overchargeFaulty fan/blowerFaulty compressorF

48、ailing capacitorHeating/cooling capacity reductionFilter life predictionOther conditionsDevelopment Level=ConceptRealityCheckHVAC 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALPage 24AWARD-WINNING,PATENTEDTECHNOLOGY 2024 Renesas Electronics Corporation.All rights reserved.RENESAS CONFIDENTIALTHANKYOU+