Adaptive feature extraction and SVM classification for real-time fault diagnosis of drivetrain gearboxes

Abstract

Abstract-Drivetrain gearboxes play an important role in many modern industrial applications. This paper presents a novel method consisting of adaptive feature extraction and support vector machine (SVM)-based classification for condition monitoring and fault diagnosis of drivetrain gearboxes operating in variable-speed conditions. An adaptive signal resampling algorithm, a frequency tracker, and a feature generation algorithm are integrated in the proposed method for effective extraction of the features of gearbox faults from the stator current signal of the AC electric machine connected to the gearbox. A radial basis function kernel-SVM classifier is designed to identify the fault in the gearbox according to the fault features extracted. Experimental studies are performed for a drivetrain gearbox with a gear crack fault connected with a permanent magnet synchronous machine. Results show that the fault can be effectively identified by the proposed method.

A GA-SVM hybrid classifier for multiclass fault identification of drivetrain gearboxes

“…If the drivetrain gearbox is connected to an AC electric machine, the characteristic frequencies of the gearbox vibration will modulate the stator currents of the AC machine and generate sidebands across the dominant components in the frequency spectra of the currents [8]. As demonstrated in [5], [8], [9], the faults in a gearbox that change the characteristic frequencies of the gearbox vibration will subsequently change the sideband distribution in the frequency spectra of the current signals of the AC machine. For a drivetrain consisting of a multistage gearbox connected with a PMSM, the characteristic frequencies of the faults in a gearbox vibration signal include the shaft rotation-related frequencies ( 1, 2, … , , is the number of gears in the gearbox), and gear meshing-related frequencies…”

“…In order to identify the faults in the gearbox operating in nonstationary conditions, an adaptive feature extraction algorithm was developed in [5] to extract the fault signatures from the nonstationary PMSM stator current signals for pattern recognition in following steps. The proposed feature extraction algorithm consists of three main steps, as shown in Fig.…”

“…The proposed feature extraction algorithm consists of three main steps, as shown in Fig. 1, where , and are the rotating frequencies of gearbox shafts, and the details can be found in [5], where PSD stands for power spectrum density.…”

“…In [5] a current-based method was proposed to identify the fault types for drivetrain gearboxes running at variable speeds. That method used an adaptive feature extraction algorithm to effectively extract the features of gearbox faults from the stator current signal of the permanent-magnet synchronous machine (PMSM) connected to the gearbox.…”

“…To design an effective SVM classifier, the values of the parameters of the SVM have to be chosen properly. In the previous work [5], the best parameters of the SVM classifier were found by a trial-and-error method, which cannot guarantee the parameters obtained to be optimal. An automatic parameter optimization method is therefore required to obtain the optimal design of the SVM.…”