An Event-Classification Neural Network Approach for Rapid Railroad Bridge Impact Detection

Area of Science:

• Engineering
• Transportation Science
• Artificial Intelligence

Background:

• Railroad bridges are crucial for freight transport, but low-clearance structures face risks from over-height vehicle impacts.
• Current impact detection methods often rely on costly wired sensors and basic thresholding, which can misidentify events like train crossings.
• Accurate detection of impacts is essential for railroad bridge safety, maintenance, and operational continuity.

Purpose of the Study:

• To develop an accurate and efficient machine learning-based system for detecting over-height vehicle impacts on railroad bridges.
• To overcome limitations of traditional wired sensors and threshold-based detection methods.
• To propose a framework for real-time, on-site (edge) event classification.

Main Methods:

• A machine learning approach, specifically a neural network, was developed for impact detection.
• Event-triggered wireless sensors were utilized to collect data from instrumented railroad bridges.
• Key features were extracted from collected event responses to train the neural network model.

Main Results:

• The machine learning model achieved an average classification accuracy of 98.67% through cross-validation.
• The system demonstrated a minimal false positive rate, effectively distinguishing impacts from other events.
• A framework for edge classification was successfully proposed and demonstrated on an edge device.

Conclusions:

• The developed machine learning approach using wireless sensors provides highly accurate detection of over-height vehicle impacts on railroad bridges.
• This method offers a more reliable and potentially cost-effective alternative to traditional detection systems.
• The proposed edge classification framework enables real-time monitoring and rapid response for enhanced railroad infrastructure safety.