Feature Classification Method of Frequency Cepstrum Coefficient Based On Weighted Extreme Gradient Boosting

Mingsi Qi,Rui Zhou,Yongsheng Yang,Qianqian Zhang

doi:10.1109/access.2021.3079286

Mingsi Qi, Rui Zhou + Show 2 more

Open Access

https://doi.org/10.1109/access.2021.3079286

Copy DOI

Abstract

Cepstrum coefficient and decision tree classification provide a feasible method for mechanical fault diagnosis. However, this method has the problems of low accuracy of feature extraction and low reliability of feature classification. The efficiency and classification results of this method are not ideal. Therefore, this paper proposes a feature classification method of frequency cepstrum coefficients based on weighted extreme gradient boosting. The proposed method is mainly divided into two parts: feature extraction and feature classification. In the part of feature extraction, Considering the limitation of Mel frequency scale in MFCC, an adaptive frequency cepstrum coefficient (AFCC) method is proposed, which can adaptively obtain the frequency scale of the corresponding signal, thus improving the feature accuracy extracted by MFCC method. In the aspect of feature classification, XGboost algorithm has strong learning ability, which easily leads to over fitting and affects the classification accuracy, A weighted extreme gradient boosting (WXGB) method is proposed, and an improved distance evaluation method (IDE) is introduced to modify the feature weights learned by XGboost, so as to improve the prediction accuracy of the classification model. In order to verify the effectiveness of the method, the vibration signals of rolling bearings with different fault types are taken as samples to carry out feature extraction and classification. The results of feature extraction and classification of different methods are compared, and the effectiveness of the proposed feature classification method based on weighted extremum gradient boosting frequency cepstrum coefficient in rolling bearing fault diagnosis is verified.

Highlights

Rotating machinery structure is a common mechanical equipment structure in modern industrial production
Among a large number of feature extraction methods, a feature extraction method based on speech vibration signal, Mel frequency cepstrum coefficient (MFCC), is noticed [10]
adaptive frequency cepstrum coefficient (AFCC) method is proposed based on the principle of MFCC method, which can be applied to the time-frequency feature extraction of vibration signal of rotating machinery

Summary

INTRODUCTION

Rotating machinery structure is a common mechanical equipment structure in modern industrial production. Among a large number of feature extraction methods, a feature extraction method based on speech vibration signal, Mel frequency cepstrum coefficient (MFCC), is noticed [10]. In the field of fault diagnosis, Jiang et al proposed the method of feature extraction and classification of rolling bearing vibration signals by MFCC method [13]. In order to make full use of MFCC method to extract vibration signal features, we need to improve its Mel frequency scale [17]. AFCC method is proposed based on the principle of MFCC method, which can be applied to the time-frequency feature extraction of vibration signal of rotating machinery. The time-frequency feature extraction of rolling bearing vibration signal can be completed adaptively by the improved method.

AN ADAPTIVE FREQUENCY CEPSTRUM COEFFICIENT IS PROPOSED

Findings

CONCLUSION