Machine learning-based rail corrugation recognition: a metro vehicle response and noise perspective

Area of Science:

• Engineering
• Artificial Intelligence
• Materials Science

Background:

• Rail corrugation is a prevalent issue in metro systems, impacting operational efficiency and maintenance.
• Accurate recognition of rail corrugation's wavelength and amplitude is crucial for timely intervention.

Purpose of the Study:

• To develop an efficient algorithm for recognizing rail corrugation wavelength and amplitude.
• To utilize in-vehicle noise and bogie acceleration data for corrugation analysis.

Main Methods:

• A particle probabilistic neural network (PPNN) algorithm was developed, integrating particle swarm optimization and probabilistic neural networks.
• In-vehicle noise features (sound pressure levels at specific frequencies) were used to identify rail wavelengths (30 and 50 mm).
• Bogie acceleration data, processed via complete ensemble empirical mode decomposition with adaptive noise, was used to identify rail amplitudes (0.1 and 0.2 mm).

Main Results:

• The PPNN algorithm achieved an average accuracy of 96.43% in recognizing rail wavelengths using in-vehicle noise.
• The PPNN algorithm achieved an average accuracy of 95.40% in recognizing rail amplitudes using bogie acceleration energy entropy.
• A stepwise moving window search algorithm was developed for effective feature selection.

Conclusions:

• The developed PPNN algorithm offers a highly accurate and efficient method for detecting rail corrugation parameters.
• This AI-driven approach demonstrates significant potential for improving the monitoring and maintenance of metro infrastructure.
• The study highlights the effectiveness of integrating acoustic and dynamic measurements for diagnosing rail defects.