Autonomous Driving Sensor Redundancy for Functional Safety ASIL-D

Overview of Technical Issues:

The sensor redundancy architecture has insufficient capability to detect and isolate individual sensor failures before they compromise safety-critical driving decisions required for ASIL-D functional safety compliance; when sensors degrade or provide conflicting data, the fusion algorithm cannot sufficiently discriminate faulty inputs, creating a harmful effect where unreliable perception data transmits to vehicle control systems, risking safety integrity violations under ISO 26262; the goal is to achieve proven fault detection and fail-safe operation meeting ASIL-D decomposition requirements.

Solution directions generated for this problem

Problem Direction 1 :

ImproveFault detection response time
VS
ConstraintSystem computational complexity

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
Glycemic urgency assessment and alerts interface
Innovative Solution Refine solution

Offline Pre-Computed Fault Signature Library for Real-Time Sensor Validation

Pre-compute fault signatures offline to enable fast runtime lookup
How to solve :
  • During vehicle calibration phase, generate comprehensive fault signature library mapping sensor output patterns to degradation modes (bias drift ±5%, noise increase >20%, intermittent dropout >50ms) using physics-based models and fleet data
  • store signatures as indexed lookup tables (hash tables with O(1) access time) in ECU non-volatile memory (2-4 MB footprint)
  • runtime fault detection performs fast pattern matching against pre-computed signatures using Hamming distance or correlation coefficients, completing checks in <30ms per sensor with <5% CPU overhead
Expected Effect : Detection latency <100ms; CPU overhead <8% vs 40-60% baseline; diagnostic coverage >92%
Risk Control :
  • signature library completeness gaps
  • memory storage capacity limits
  • signature drift over vehicle lifetime

Problem Direction 2 :

ImproveDiagnostic coverage rate
VS
ConstraintSystem computational complexity

Inspiration 1 : Cross-domain reference

Application Principle: #24 Intermediary
Cross-domain applicability Assess applicability
Using dotplots for comparing and finding patterns in sequences of data points
Innovative Solution Refine solution

Proxy-indicator diagnostic layer for lightweight fault coverage expansion

Deploy proxy indicators as intermediary layer
How to solve :
  • Implement lightweight proxy-indicator monitoring layer between raw sensor data and fusion algorithm — track signal variance, cross-sensor correlation coefficients, temporal consistency metrics, and rate-of-change bounds instead of modeling every degradation mode explicitly
  • Pre-compute proxy threshold lookup tables offline during calibration phase: establish normal operating envelopes for variance (σ²<0.05 nominal), correlation (ρ>0.85 between redundant channels), temporal gradient (Δ/Δt within ±15% historical baseline) — store in 2KB ECU memory for O(1) runtime queries
  • Execute parallel proxy checks in 8-12ms per sensor modality using simple threshold comparisons — flag bias drift via moving-average deviation >3σ, noise increase via spectral power ratio >1.4×baseline, intermittent failures via correlation drop events — achieving >95% degradation mode coverage with <8% computational overhead
Expected Effect : Diagnostic coverage 65%→96%; CPU load +7% vs +45% baseline; detection latency <15ms
Risk Control :
  • proxy-indicator calibration drift over vehicle lifetime
  • false-positive rate in edge environmental conditions
  • lookup table memory allocation conflicts

Problem Direction 3 :

ImproveSensor fault isolation capability
VS
ConstraintSensor redundancy implementation cost

Inspiration 1 : Cross-domain reference

Application Principle: #26 Copying
Cross-domain applicability Assess applicability
Methods, systems and apparatus to dynamically facilitate boundaryless, high availability system management
Innovative Solution Refine solution

Physics-based virtual sensor model for analytical redundancy fault isolation

Deploy virtual sensor via vehicle dynamics model
How to solve :
  • Implement physics-based vehicle dynamics model (bicycle model + tire slip estimation) running parallel to physical sensors, generating virtual position/velocity estimates at 20ms cycle time
  • Cross-validate physical sensor outputs (camera, radar, lidar) against model predictions using chi-squared residual test with adaptive threshold (±2σ dynamic envelope)
  • Isolate faulty sensor when residual exceeds threshold for ≥3 consecutive cycles (60ms), triggering safe state transition without requiring 2-3 additional hardware channels per modality
Expected Effect : Fault isolation capability ≥95%, hardware cost +$0, detection latency <80ms
Risk Control :
  • Model parameter drift under tire wear
  • computational overhead 15-20% CPU load
  • false positives during extreme maneuvers

Problem Direction 4 :

ImproveFault detection response time
VS
ConstraintSensor redundancy implementation cost

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
Prefetch kernels on data-parallel processors
Innovative Solution Refine solution

Pre-computed fault signature library for rapid sensor validation

Offline pre-compute sensor fault signatures to enable fast runtime matching
How to solve :
  • During vehicle calibration phase, generate fault signature lookup tables mapping sensor output patterns to fault types across 500+ driving scenarios, storing in 2MB ECU flash memory
  • At runtime, compare live sensor data against pre-computed signatures using hash-based pattern matching in <50ms, eliminating need for redundant sensor cross-validation
  • Implement geometric consistency zones pre-calculated from overlapping sensor fields-of-view, enabling single-sensor fault isolation through spatial plausibility checks without additional hardware
Expected Effect : Detection latency <80ms; hardware cost +$0; diagnostic coverage 92%
Risk Control :
  • signature library completeness gaps
  • hash collision false positives
  • field-of-view calibration drift

Problem Direction 5 :

ImproveDiagnostic coverage rate
VS
ConstraintSensor redundancy implementation cost

Inspiration 1 : Cross-domain reference

Application Principle: #26 Copying
Cross-domain applicability Assess applicability
Chroma Quantization in Video Coding
Innovative Solution Refine solution

Analytical redundancy via physics-based virtual sensor models for ASIL-D fault coverage

Deploy virtual sensors using physics models to replicate real sensor behavior without hardware
How to solve :
  • Implement vehicle dynamics Kalman filters to generate virtual camera/radar/lidar outputs based on IMU, wheel speed, and steering angle — cross-validate physical sensor data against model predictions to detect bias drift, noise increase, and intermittent failures
  • Pre-calibrate sensor behavior lookup tables offline using fleet data across 50+ driving scenarios (highway, urban, weather conditions) — store expected output ranges (±5% tolerance for radar range, ±0.3m for camera lane position) in 2MB ECU memory for real-time comparison within 40ms
  • Execute residual-based fault detection by computing deviation between physical sensor output and virtual model prediction every 50ms — flag fault when residual exceeds 3-sigma threshold for >100ms consecutive window, achieving >95% coverage for systematic degradation modes
Expected Effect : Diagnostic coverage 65%→96%; hardware cost +$0; detection latency <80ms; computational overhead +12%
Risk Control :
  • Model accuracy degradation in edge cases
  • calibration dataset representativeness insufficient
  • false positive rate in novel scenarios

Problem Direction 6 :

ImproveSystem computational complexity
VS
ConstraintMust not deteriorate

Inspiration 1 : Cross-domain reference

Application Principle: #10 Preliminary action
Cross-domain applicability Assess applicability
Methods for selecting motion vector prediction values ​​and devices for using them.
Innovative Solution Refine solution

Pre-calibrated fault signature library for real-time sensor diagnostics

Offline pre-compute sensor fault signatures during calibration phase
How to solve :
  • During vehicle calibration, execute comprehensive fault injection testing across all sensor modalities (camera, radar, lidar) under 200+ driving scenarios — record fault signatures (bias drift patterns, noise spectral fingerprints, intermittent failure timing profiles) into a compressed lookup table stored in ECU flash memory (≤8MB footprint)
  • At runtime, perform fast pattern matching using hash-based indexing — compare live sensor data streams against pre-stored signatures within 15-25ms per sensor using simple Euclidean distance metrics, achieving <100ms total detection latency
  • Implement hierarchical signature matching — Level-1 checks (out-of-range, frozen values) execute every 20ms with <5% CPU load, Level-2 checks (drift, noise increase) trigger only when Level-1 anomalies detected, maintaining average computational overhead <18% while achieving >95% diagnostic coverage
Expected Effect : Detection latency <100ms; diagnostic coverage >95%; CPU overhead reduced from 40-60% to <18%; no additional hardware cost
Risk Control :
  • signature library completeness insufficient for edge-case faults
  • flash memory wear from frequent table updates
  • hash collision causing false negatives in pattern matching
Patsnap Eureka Solution