Track Rail Defect Detection With Two-Stage AI Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated railway track inspection systems face challenges in balancing high-speed data processing with low false positive rates, leading to inefficient and potentially dangerous oversight of critical defects.
Innovation Solution
A two-stage AI model approach using a lightweight YOLO-Tiny model for initial defect detection and a heavyweight YOLOv5 model for refinement, combined with cloud-based verification, to enhance accuracy and reduce false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If a high-speed AI model is used for defect detection, then processing speed is improved, but the false positive rate increases
Solution Approach 1:
The patent divides the defect detection process into two sequential stages: a first stage using a lightweight AI model for rapid initial screening, and a second stage using a heavyweight AI model for refined verification. This segmentation allows the system to achieve both high processing speed (through the lightweight model's fast initial detections) and low false positive rate (through the heavyweight model's accurate verification), directly resolving the contradiction between speed and reliability.
Solution Approach 2:
The lightweight AI model performs preliminary defect detection before the heavyweight model analyzes the data. This preliminary action filters out obvious non-defects early, reducing the workload for the verification stage and enabling the system to maintain high throughput while ensuring accuracy through the subsequent heavyweight model analysis.
2Measurement precision
If an accurate AI model is used for defect detection, then detection accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent segments the detection task into two models with different computational characteristics: a lightweight model optimized for speed that handles initial screening, and a heavyweight model optimized for accuracy that performs verification. This segmentation enables the system to achieve high detection accuracy through the heavyweight model while maintaining high processing speed through the lightweight model's rapid initial filtering.
Solution Approach 2:
The system applies partial action by using the lightweight model to handle the majority of processing tasks (initial detection of most frames), reserving the computationally intensive heavyweight model for only those cases requiring verification. This partial application of the accurate but slow model ensures high accuracy where needed while maintaining overall processing speed.
3Measurement precision
If manual inspection methods are used, then detection accuracy can be maintained, but labor intensity and time consumption increase
Solution Approach 1:
The patent replaces manual inspection with an automated dual-stage AI-based inspection system. The lightweight and heavyweight AI models work together to perform defect detection automatically, eliminating the need for manual inspection while achieving both high detection accuracy and improved productivity through rapid automated processing of track rail images.
Data Source
AI summary
A method and system for detecting defects in track rails is disclosed. A processor receives imaging data of one or more-track rails in real-time using an imaging device coupled to a railway train. A set of image frames of the one or more-track rails are determined for each time instance. A first processed frame is determined from the set of image frames based on detection of at least one first defect from a set of predefined defects in the set of image frames using a first artificial intelligence (AI) model. The first processed frame is processed to determine a second processed frame from the first processed frame based on detection of at least one second defect from the set of predefined defects in the first processed frame using a second AI model.