Multi-Sensor Fault Detection, Identification, Isolation and Health Forecasting for Autonomous Vehicles

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Autonomous driving systems rely on sensor data for navigation and decision-making.
• Sensor failures in autonomous vehicles can lead to critical safety issues and accidents.
• Current systems require improved methods for early fault detection and prediction.

Purpose of the Study:

• To develop a novel architecture for fault detection, isolation, identification, and prediction (FDIDP) in multi-sensor systems for autonomous vehicles.
• To enhance the reliability and safety of autonomous driving by addressing sensor fault vulnerabilities.
• To create a system capable of forecasting potential sensor failures.

Main Methods:

• Utilized real-world autonomous vehicle data combined with artificially injected sensor faults.
• Developed and implemented two distinct deep neural network architectures for fault detection, identification, and isolation.
• Introduced a health index measure derived from the fault detection system's output.
• Trained a health index forecasting network using the developed health index.

Main Results:

• Achieved very impressive performance in detecting, identifying, and isolating multi-sensor faults.
• The proposed architecture demonstrated high accuracy in identifying various types of sensor faults.
• Successfully developed a quantifiable health index for sensor systems.
• The health index forecasting network showed promise in predicting future sensor health.

Conclusions:

• The novel FDIDP architecture significantly improves the ability to manage sensor faults in autonomous vehicles.
• The system offers a robust approach to enhance the safety and reliability of self-driving technology.
• Forecasting sensor health provides a proactive maintenance strategy, reducing the risk of critical failures.