Sensor Failure Detection via Physical Consistency Checks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sensor failure detection methods require pre-generated models to identify anomalies, limiting their effectiveness in real-time scenarios, especially when partial sensor malfunctions occur due to dirt or dust, which can lead to malfunctioning driving support systems.
Innovation Solution
A failure detection apparatus that acquires sensor data over a past reference period and determines anomalies by checking if characteristics of normal data are present in the detected data, allowing for real-time anomaly detection without pre-generated models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pre-generated models are used for sensor anomaly detection, then detection accuracy for known failure patterns is improved, but the system cannot detect anomalies outside the model's training scenarios and requires complex model generation processes
Solution Approach 1:
The sensor system performs self-diagnosis by automatically analyzing its own output data patterns without requiring external model generation or complex preprocessing. The detection unit examines whether detected objects and their characteristics are consistent with expected sensing results, enabling the system to self-identify anomalies such as dirt accumulation or sensor failures.
Solution Approach 2:
The invention extracts only the essential detection logic from complex models - specifically checking whether detected objects and their characteristics (position, size, movement) are physically consistent with the sensor's field of view and expected behavior. This simplified extraction enables anomaly detection without requiring full model generation.
2Reliability
If complex pre-generated models are used for sensor failure detection, then detection capability for specific failure modes is improved, but real-time detection speed is reduced and the system lacks adaptability to new failure patterns
Solution Approach 1:
The sensor system performs real-time self-monitoring by continuously analyzing its own output data for physical consistency. The detection unit checks whether detected objects' positions, sizes, and movements are consistent with the sensor's field of view and expected physical behavior, enabling immediate anomaly detection without model generation delays.
Solution Approach 2:
The detection method dynamically adapts to different sensor types and failure modes by using general physical consistency checks rather than fixed models. The system can detect various anomalies (dirt accumulation, lens contamination, sensor failures) by evaluating whether detected data patterns are physically plausible, providing both speed and adaptability.
3Adaptability or versatility
If sensor data is analyzed without reference models, then real-time detection and adaptability to new failure patterns are improved, but the ability to distinguish true anomalies from normal variations deteriorates
Solution Approach 1:
Instead of checking whether data matches a pre-generated model, the system inverts the approach by checking whether data violates physical consistency principles. The detection unit identifies anomalies by finding deviations from expected physical relationships (e.g., objects appearing outside the sensor's field of view, impossible movement patterns), which reliably indicates sensor failures without requiring training models.
Data Source
AI summary
A failure detection apparatus (10) acquires, as target data, sensor data output in a past reference period by a sensor (31), such as a millimeter wave radar or LiDAR (Light Detection And Ranging), mounted on a moving body (100). The failure detection apparatus (10) determines whether detected data indicating a characteristic of a detected object indicated by normal data, which is sensor data output when the sensor (31) is normal, is included in the acquired detected data in the past reference period. In this way, the failure detection apparatus (10) determines whether a failure has occurred in the sensor (31).