《智能传感器装调与测试（双语）》课件-27.多传感器融合预期功能安全（SOTIF）

多传感器融合预期功能安全（SOTIF）Multi-SensorFusionSafetyoftheIntendedFunctionality(SOTIF)SOTIF框架流程图FlowchartoftheSOTIFframework多传感器融合预期功能安全（SOIF），尤其是国际标准ISO21448在多传感器融合系统中的应用。Theapplicationofmulti-sensorfusionSOIF,especiallytheinternationalstandardISO21448,inmulti-sensorfusionsystems场景识别

Scenariorecognition功能不足触发条件分析

Analysisoftriggeringconditionsforfunctionalinsufficiency风险控制策略设计Designofriskcontrolstrategy验证与确认

VerificationandconfirmationPART01SOTIF基础认知FundamentalsofSOTIFSOTIF定义与融合风险SOTIFDefinitionandFusionRisksSOTIF定义SOTIFdefinitionSOTIF，即预期功能安全，由ISO21448提出，旨在解决系统功能符合设计但在特定场景下因感知或决策偏差导致的安全风险，即‘没坏但用错’的情况。SOTIFwasproposedbyISO21448toaddresssafetyriskscausedbyperceptionordecisiondeviationinspecificscenarios.Specifically,systemfunctionsmeetdesignstandards,butarenotbeingusedfortheirintendedpurposes.与功能安全的区别Differencesfromfunctionalsafety与ISO26262的功能安全不同，SOTIF专注于解决功能设计缺陷，而非硬件故障。功能安全关注硬件的可靠性，而SOTIF关注系统在复杂环境下的可靠性。UnlikefunctionalsafetyinISO26262,SOTIFfocusesonaddressingfunctionaldesigndefectsratherthanhardwarefaults.Whilefunctionalsafetyfocusesonhardwarereliability,SOTIFfocusesonsystemreliabilityincomplexenvironments.融合系统风险Fusionsystemrisks多传感器融合系统易受环境干扰，如暴雨、强光等，导致数据融合算法无法处理复杂场景，可能引发误识别，进而导致误刹车或不刹车，增加碰撞风险。Themulti-sensorfusionsystemiseasilydisturbedbytheenvironmentsuchasheavyrainorstronglight.Thiscancompromisethedatafusionalgorithmincomplexscenarios,leadingtofalsepositivesandsubsequentunnecessarybrakingorfailuretobrake,whichincreasethecollisionrisk.PART02ISO21448四阶段落地ImplementationofISO21448inFourStages四阶段流程全景概览PanoramicOverviewoftheFour-StageProcess四阶段流程Four-stageprocessISO21448标准在多传感器融合系统中的落地分为四个阶段：危险场景定义、功能设计与验证、监控与预警、数据迭代优化。这四个阶段相互衔接，形成一个完整的闭环流程。TheISO21448standardisimplementedinamulti-sensorfusionsysteminfourstages:dangerousscenariodefinition,functionaldesignandverification,monitoringandearlywarning,anddataiterationoptimization.Thesefourstagesareinterconnectedtoformacompleteclosed-loopprocess.多传感器融合画面Multi-sensorfusionscreen多传感器融合系统依赖激光雷达、毫米波雷达、摄像头等多设备的数据协同，一旦某类传感器受环境干扰（如强光、暴雨），或数据融合算法对复杂场景（如行人横穿+车辆加塞）处理不当，就可能出现“感知偏差”，导致车辆误刹车或不刹车。Themulti-sensorfusionsystemreliesonthedatacollaborationofmultipledevicessuchasLiDAR,millimeter-waveradar,andcameras.Onceacertaintypeofsensorisdisturbedbytheenvironment(suchasstronglightorheavyrain),orthedatafusionalgorithmimproperlyhandlescomplexscenarios(suchaspedestriancrossingandvehiclecuttingin),perceptiondeviationmayoccur,resultinginunnecessarybrakingorfailuretobrake.各阶段核心动作拆解BreakdownofCoreActionsinEachStage场景定义阶段Scenariodefinitionstage定义阶段需建立覆盖气候、交通、传感器的三维场景库并排序风险，明确哪些场景可能导致安全风险。Inthedefinitionstage,itisnecessarytoestablishathree-dimensionalscenariolibrarycoveringclimate,traffic,andsensors,rankrisks,andclarifywhichscenariosmayleadtosafetyrisks.天气、交通情况场景仿真验证

Verificationbysimulatingtheweatherandtrafficconditions各阶段核心动作拆解BreakdownofCoreActionsinEachStage功能设计与验证阶段Functionaldesignandverificationstage验证阶段通过SIL、HIL、封闭道路测试证明指标达成，确保设计的功能能够有效应对风险场景。Intheverificationstage,SIL,HIL,andclosedroadtestsarecarriedouttoprovethattheindicatorshavebeenachieved,soastoensurethatthedesignedfunctionscaneffectivelycopewithriskscenarios.封闭道路场景仿真验证

Verificationbysimulatingclosedroadscenarios各阶段核心动作拆解BreakdownofCoreActionsinEachStage监控预警与设计阶段Monitoring,earlywarning,anddesignstage设计阶段将风险转化为安全指标，完成传感器权重分配、融合策略冗余及降级方案，确保系统在风险场景下的可靠性。Inthedesignstage,risksareconvertedintosafetyindicators,andsensorweightallocation,fusionstrategyredundancy,anddegradationplansarecompletedtoensuresystemreliabilityinriskscenarios.监控预警画面上传系统

Monitoringandearlywarningscreenuploadingsystem各阶段核心动作拆解BreakdownofCoreActionsinEachStage监控与迭代阶段Monitoringanditerationstage监控阶段部署在线性能检测，对误差超阈值场景即时预警；迭代阶段收集实车数据，用数据驱动重新标定权重与场景库，形成闭环。Inthemonitoringstage,onlineperformancedetectionisdeployedtoprovideatimelyearlywarningforscenarioswhereerrorsexceedthresholds.Intheiterationstage,realvehicledataiscollectedtoenabledata-drivenrecalibrationofweightsandscenariolibrariestoformaclosedloop.监控画面实时预警

Real-timeearlywarningofthemonitoringscreenPART03高速暴雨实战案例High-SpeedRainstormPracticalCases超声波雷达UltrasonicRadar激光雷达失效LiDARfailure在高速暴雨场景中，激光雷达点云被雨幕吸收反射率下降60%，目标缺失率上升，导致感知误差显著增加。Inhigh-speedrainstormscenarios,aftertheLiDARpointcloudisabsorbedbytheraincurtain,reflectivitydecreasesby60%andthetargetmissingrateincreases,resultinginasignificantincreaseinperceptionerrors.毫米波雷达失效Millimeter-waveradarfailure毫米波雷达虽可穿透但多径效应使测距误差由±0.3m增至±1.5m，融合系统若继续等权平均，将误判前车距离。Althoughmillimeter-waveradarhasthepenetrationcapability,themultipatheffectincreasestherangingerrorfrom±0.3mto±1.5m.Ifthefusionsystemcontinuescalculatingtheequalweightedaverage,itwillmisjudgethedistancetothevehicleahead.摄像头失效Camerafailure摄像头因雨痕与低对比度导致边框断裂，识别置信度掉至0.3，进一步加剧了融合系统的感知误差。Thecameraframeisbrokenduetorainmarksandlowcontrast,andtherecognitionconfidencedropsto0.3,furtherexacerbatingtheperceptionerrorsofthefusionsystem.ISO21448防护与迭代ISO21448ProtectionandIteration防护措施Protectivemeasures针对暴雨场景，通过动态调整传感器权重，降低激光雷达和摄像头权重，提升毫米波雷达比重，并引入GPS车速进行校验，有效降低误判风险。Forrainstormscenarios,thesystemreducestheriskofmisjudgmenteffectivelybydynamicallyadjustingsensorweights,reducingLiDARandcameraweights,increasingtheweightofmillimeter-waveradar,andintroducingGPStoverifyvehiclespeed.迭代优化Iterationoptimization收集暴雨场景下的实车数据，重新训练权重策略并通过OTA推送，使目标跟踪误差从1.5米降至0.8米，显著提升系统可靠性。Thesystemcollectsrealvehicledatainheavyrainscenarios,retrainsweightstrategies,andpushesthemthroughOTA,sothatthetargettrackingerrorisreducedfrom1.5metersto0.8meters,significantlyimprovingsystemreliability.小结SummarySOTIF（ISO21448）解决多传感器融合系统“没坏但用错”的风险，它通过四个关键环节落地应用，先定义高风险场景明确风险点，再通过功能设计与验证应对风险，接着实时监控预警防止风险发生，最后用数据反馈迭代补全风险漏洞；它和功能安全互