Algorithms for detection of rail wheel squeal

Trackside systems for automatic monitoring of noise from train passbys are becoming more common. Typically these will record an audio file for each passby, and download this file for spectral and other analysis. Automatic detection of the presence and level of wheel squeal from these files provides significant additional information for both operators and environmental authorities. Recently in NSW, two groups have independently developed algorithms for detecting and quantifying wheel squeal. Both are based on a spectral analysis, but details of the procedures differ. Outputs include the maximum level, SEL, duration and spectrum of squeal, and in one case also of flanging noise. This paper compares the procedures and outputs of the two algorithms, using a set of recorded audio files from train passbys. Results indicate the potential of detection based on pattern-recognition techniques in this and similar applications, and also point to some issues associated with their implementation.