BwimNet: A Novel Method for Identifying Moving Vehicles Utilizing a Modified Encoder-Decoder Architecture.

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Transportation Infrastructure

Background:

• Traffic load monitoring is crucial for bridge structural health, aiding in overload detection and safety evaluation.
• Bridge Weigh-In-Motion (BWIM) systems typically require additional sensors for vehicle configuration, which can be costly and unstable.
• Existing data-driven methods often struggle with obtaining vehicle weight labels.

Purpose of the Study:

• To develop a novel method for identifying moving vehicle properties (velocity, wheelbase, axle weight) using only bridge dynamic responses.
• To overcome the limitations of conventional BWIM systems, including the need for extra sensors and labeled weight data.
• To enhance the accuracy and reliability of traffic load monitoring in bridges.

Main Methods:

• A modified encoder-decoder architecture with a signal-reconstruction layer was proposed.
• An unsupervised learning approach was employed to extract higher-level features from raw data.
• A numerical bridge model based on vehicle-bridge coupling vibration theory was established for validation.

Main Results:

• The proposed method successfully identified vehicle properties without additional sensors or vehicle weight labels.
• Numerical simulations demonstrated the approach's applicability and effectiveness.
• Parametric studies confirmed improved stability and reliability, even with significant data noise.

Conclusions:

• The novel encoder-decoder approach offers a promising solution for accurate and cost-effective traffic load monitoring in bridges.
• This method eliminates the dependency on fragile and expensive auxiliary sensors.
• The technique provides a robust and reliable means for identifying moving vehicle characteristics from bridge responses.