Abnormal noise monitoring of subway vehicles based on combined acoustic features. (August 2022)

• Record Type: Journal Article
• Authors: Yan, Zhaoli; Zhu, Hang; Zhang, Han; Wan, Hongjie; Liu, Bin
• Is Part Of: Applied acoustics. Volume 197(2022)
• Publication Date: 2022-08
• Subjects: Subway vehicles -- Condition monitoring -- Modified time-spectral kurtosis -- MFCC -- Neural network; Acoustical engineering -- Periodicals; Periodicals; 620.2
• DOI: 10.1016/j.apacoust.2022.108951
• Languages: English
• ISSNs: 0003-682X

Abstract

Highlights: A monitoring technology for abnormal noise in subway vehicles is studied. A modified time-spectral kurtosis (MTSK) feature parameter is proposed. Combined features (MTSK + MFCC) are used for classification. Improvement of the proposed method is effective at both low speeds and high speeds. Abstract: Subways play a significant role in urban rail transit. A monitoring technology for abnormal noise in subway vehicles is studied in this paper to ensure safety. Microphones are placed near the bogie components to acquire acoustic signals during the subway operation. A modified time-spectral kurtosis (MTSK) feature parameter is proposed to take advantage of the sensitivity of the kurtosis to the impact signal and is combined with the mel-frequency cepstral coefficients (MFCC), which reflects the acoustic characteristics of the human ear, to build a neural network model. The normal background noise (including the wheel-rail impact signal at the rail joints) and abnormal impact sound in the bogie area of the vehicle are classified. The results are compared with those before the feature improvement. The experimental results show that the recognition performance of abnormal impact sounds is indeed improved using feature vectors, including MTSK and MFCC. The accuracy rate of the classifier constructed in this work is 99.6% at a speed of 20 km/h. It can also achieve 97.2% and 86.1% at 40 km/h and 60 km/h, respectively. The improvement of the proposed method is more effective even at a lower signal-to-noise ratio compared with other features.