ADAS Sensor Fault Diagnosis Using Deep Learning Shared Representation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Advanced Driver Assistance Systems (ADAS) face challenges in efficiently and comprehensively verifying sensor data due to varying environmental conditions and the complexity of sensor failures, which can lead to defects and reduced reliability, especially with the increasing use of multiple sensors in autonomous navigation systems.
Innovation Solution
An apparatus and method utilizing deep learning for fault diagnosis and backup of ADAS sensors, incorporating an individual sensor diagnosis unit, inter-sensor mutual diagnosis unit, and integrated diagnosis unit to evaluate sensor reliability through shared representation and prediction models, reducing the influence of low-reliability sensors on other operational sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple sensors are used to improve ADAS functionality and coverage, then the system can provide more comprehensive environmental recognition, but the complexity of verifying sensor data and diagnosing faults increases significantly
Solution Approach 1:
The patent divides sensor verification into two independent modules: individual sensor diagnosis (evaluating each sensor separately using its own historical data and environmental conditions) and inter-sensor mutual diagnosis (evaluating sensors based on their mutual relationships and shared environmental factors). This segmentation reduces the overall verification complexity by breaking down the complex multi-sensor verification problem into manageable independent sub-problems that can be processed separately and then integrated.
2Reliability
If traditional verification methods are used, then the system structure remains simple, but it is substantially difficult and inefficient to secure all verification cases for various sensor operating environments
Solution Approach 1:
The patent introduces a unified sensor verification system that handles multiple sensor types (cameras, radars, LiDARs, ultrasonic sensors) and various operating environments (different regions, times, weather conditions) through a single integrated framework. The system uses universal verification modules that can adapt to different sensor types and environmental conditions, eliminating the need to design separate verification cases for each scenario while maintaining comprehensive coverage.
Solution Approach 2:
The patent implements feedback mechanisms where sensor verification results are continuously fed back into the system. Individual sensor diagnosis uses historical sensor data and environmental information to evaluate current sensor status, while inter-sensor mutual diagnosis uses verification results from other sensors to cross-validate each sensor's output. This feedback loop enables the system to adapt to various operating environments dynamically without requiring pre-programmed verification cases for every possible scenario.
3Productivity
If sensor verification is performed without considering environmental factors, then the verification process is simpler, but the accuracy of fault diagnosis decreases under varying environmental conditions
Solution Approach 1:
The patent applies local quality by tailoring the verification approach to specific environmental conditions and sensor types. The system adjusts verification parameters and thresholds based on local environmental factors such as weather conditions, time of day, and geographic region. For example, verification criteria for camera sensors differ between daytime and nighttime conditions, and radar verification parameters are adjusted based on weather conditions. This localized approach maintains high verification efficiency while ensuring accurate fault diagnosis across varying environmental conditions.
Data Source
AI summary
An apparatus for fault diagnosis and back-up of advanced driver assistance system sensors based on deep learning, the apparatus including: an individual sensor diagnosis unit configured to quantitatively evaluate a reliability of an output result of each sensor at each moment on the basis of a model for an output of each sensor under a normal operation; an inter-sensor mutual diagnosis unit configured to extract shared representation between the sensors and quantitatively evaluate a normal-operation reliability of the output result of each sensor on the basis of the extracted shared representation; and an integrated diagnosis unit configured to quantitatively evaluate a final reliability of each sensor on the basis of output results of the individual sensor diagnosis unit and the inter-sensor mutual diagnosis unit.