Feature-level fusion based on wavelet transform and artificial neural network for fault diagnosis of planetary gearbox using acoustic and vibration signals

Abstract

In this article, an intelligent system based on an artificial neural networks (ANN) classifier is proposed for fault diagnosis and classification of planetary gearboxes based on fusing acoustic and vibration data at the feature level. First, the acoustic and vibration signals of the planetary gearbox were collected simultaneously in four gearbox conditions: (1) healthy; (2) worn tooth on planet gear; (3) cracked tooth on ring gear; and (4) broken tooth on ring gear. Then, the time domain signals were transformed to the time-frequency domain by wavelet transform. Thirty statistical features were then extracted from each signal and used as feature vectors to an ANN classifier. The primary classification of the faults was undertaken based on the extracted features from each sensor. The classification accuracy of acoustic and vibration data was about 88.4% and 86.9%, respectively. The final classification accuracy using fused features was 98.6%, indicating the superiority of the proposed method for fault diagnosis of a planetary gearbox. The 10% accuracy increase gained through using the data fusion method can significantly enhance the quality and accuracy of fault diagnosis and, as a result, condition monitoring of the machinery.

Spark plug fault recognition based on sensor fusion and classifier combination using Dempster–Shafer evidence theory

“…In another research, a number of the features have been employed for fault diagnosis of low speed bearing by Widodo et al [29]. Ebrahimi and Mollazade [30] and Khazaee et al [31] have used some of these features for fault diagnosis of tractor starter motor and planetary gearbox, respectively. Devasenapati et al [32] have employed a number of the features for misfire identification in a petrol engine.…”

An intelligent device for diagnosing avian diseases: Newcastle, infectious bronchitis, avian influenza

“…The outputs of any signal processors might not be used as inputs of classifier because the processed signals contain a large number of groups of raw data, therefore some statistical functions should be extracted to define a signal state and to prepare information. Table 1 illustrates all 25 statistical features used in this study to extract statistical information from Frequency domain signals, approximation and details coefficient of DWT (Khazaee et al, 2013;Lei et al, 2008). In this table, x(n) is a time domain signal for n data points (n = 1, 2, .…”

An Intelligent Procedure for the Detection and Classification of Chickens Infected by Clostridium Perfringens Based on their Vocalization

“…Twenty-three features were extracted from the raw data in the present study (Table 2). These indices were the feature functions that have been used by other researchers in the field of data mining and can be used to obtain the information required for data mining (Khazaee et al, 2013;Lei et al, 2008). In this table, x(n) is signal time-series and N is the number of data points.…”