A New Method for Pantograph-Catenary Arc Detection Based on Acoustic Signals

Objective To enhance the accuracy and precision of arc detection in urban rail transit and address the limitations of existing arc detection methods such as electrical and optical approaches, it is necessary to develop new pantograph-catenary arc detection approaches.

Method The Mel Frequency Cepstral Coefficient (MFCC) method is adopted to extract sound features, including the preprocessing of acoustic signals, analysis and calculation of spectral power spectrum, Mel filtering, and discrete cosine transform. For different acoustic signals, MFCCs are used as feature vector sets and audio features to distinguish various pantograph-catenary accompanying sounds, such as pantograph-catenary arcing, ambient noise, aerodynamic noise, and current noise. The K-means clustering algorithm is employed for feature recognition and differentiation of audio classification results. A Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) is introduced for modelling and identification of arcing sound features, and its accuracy is verified by comparison with the measured data from a metro line in Shanxi Province, China.

Result & Conclusion  The proposed GMM-HMM exhibits excellent recognition performance for arcing audio. It can serve as a novel acoustic signal-based detection method for pantograph-catenary arcing, thereby making up for the limitations of conventional arcing detection methods.