Obstacle Detection Testing for Track-Guided Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current obstacle detection systems in the railway sector, particularly for automatic rail travel, face challenges in ensuring reliable detection due to the lack of approval for AI applications in safety-critical areas and the difficulty in understanding their functionality, leading to uncertainties in sensor performance and obstacle clearance in real environments.
Innovation Solution
A method is introduced to test obstacle detection systems under realistic conditions by dividing the testing process into two phases: a first phase outside regular operation for hardware functionality checks and a second phase during regular operation, where detected obstacles are not considered, using virtual or real test conditions to ensure safety standards are met, and utilizing redundant systems for reliable operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Difficulty of detecting and measuring
If AI applications are used for obstacle detection in safety-critical railway areas, then detection capability is improved, but reliability and safety approval are worsened due to lack of understanding of AI functionality
Solution Approach 1:
A test object management system acts as an intermediary between the AI obstacle detection system and the testing environment. This system manages virtual and real test objects, controls their placement in the sensor field, and coordinates testing phases, thereby enabling systematic validation of AI detection capabilities while maintaining safety through controlled test conditions and redundant monitoring systems.
Solution Approach 2:
The testing process is segmented into distinct phases: a first test phase conducted outside regular operation for hardware functionality checks, and a second test phase during regular operation for obstacle detection testing. This segmentation allows thorough testing while maintaining safety by isolating critical functions during different operational states.
2Measurement precision
If testing is conducted during regular operation, then real-world detection performance is improved, but safety risks are worsened due to potential undetected obstacles
Solution Approach 1:
Redundant obstacle monitoring systems are deployed beforehand to cushion against potential detection failures during testing. These backup systems continuously monitor for obstacles during the second test phase, providing a safety net that allows real-world testing while mitigating the harmful effects of potential undetected obstacles through pre-established backup monitoring capabilities.
3Adaptability or versatility
If virtual test objects are used, then testing flexibility is improved, but realism of test conditions is worsened
Solution Approach 1:
The system merges virtual test objects with real test objects in a unified test environment. Virtual test objects provide flexible, programmable test scenarios that can be easily configured and modified, while real test objects (such as physical barriers or test vehicles) add physical realism to the testing. The test object management system coordinates both types of objects, allowing the AI detection system to be tested against a combination of virtual and real obstacles, thereby achieving both flexibility and realism.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The invention comprises a method for testing obstacle detection for a danger zone (GFR) located in the direction of travel (FR) in front of a track-guided vehicle (FZ), wherein a sensor device (SN1, SN2) detects objects (OB) located in or on the danger zone (GFR), the detected objects (OB) are recognized by a computer, and the detected objects (OB) are evaluated by a computer to identify obstacles (HD). In a first test phase (EP1), the functionality of hardware comprising the sensor device (SN1, SN2) is verified outside of normal operation. In a second test phase, the obstacle detection (HD) is tested during normal operation of the vehicle (FZ), whereby the obstacles (HD) detected by the obstacle detection test are disregarded during normal operation of the vehicle (FZ).Furthermore, the invention comprises an arrangement for detecting obstacles, a computer program product and a provisioning device.