Generative Adversarial Learning Enhanced Fault Diagnosis for Planetary Gearbox under Varying Working Conditions

Abstract

Planetary gearbox is a critical component for many mechanical systems. It is essential to monitor the planetary gearbox health and performance in order to maintain the whole machine works well. The methodology of mechanical fault diagnosis is increasingly intelligent with the extensive application of deep learning. However, the cross-domain issue caused by varying working conditions becomes an enormous encumbrance to fault diagnosis based on deep learning. In this paper, in order to fully excavate potentialities of deep neural network architectures, a novel generative adversarial learning method was introduced for a completely new fault diagnosis based on a deep convolution neural network. In addition, the intelligent fault diagnostic scheme for planetary gearbox under varying speed conditions was developed. After that, some experiments on measured vibration signals of planetary gearbox were conducted to verify the validity and efficiency of the fault diagnostic scheme. The results showed that the proposed method enhanced the capability of the intelligent diagnosis for planetary gear faults under varying speed conditions. Empirical Mode Decomposition (EEMD), machine learning algorithms of Kernel Principle Component Analysis (KPCA) and Learning Vector Quantization (LVQ) [6]. Li used the signal processing method based on Adaptive Multi-scale Morphological Filter (AMMF) and Modified Hierarchical Permutation Entropy (MHPE) to extract fault features in PG vibration signals and then diagnosed the fault by Binary Tree Support Vector Machine (BT-SVM) [7]. Lei also advanced a health condition identification method for multi-stage PG, adopting multi-class relevant vector machine as a classifier and introduced Accumulative Amplitudes of Carrier Orders (AACO) and Energy Ratio-based Difference Spectra (ERDS) as fault features which improved diagnosis performance and robustness [8]. These intelligent methods had certain limitations and uncertainties. Firstly, in the pre-processing stage of vibration signal, the expertise knowledge was highly required, such as signal processing and data statistics. On the other hand, the intrinsic features for the faults of planetary gearbox were not found in these existing literatures, which were independent of the fluctuation of working conditions. Thus, these led to the problem of cross-domain for intelligent fault diagnosis of PG.In recent years, Deep Learning (DL) has been attracting growing attention from various fields. The deep model hierarchical structures make DL capable of learning intrinsic representations of raw complex data, which provide the foundation for its popular application in visual recognition and natural language processing [9][10][11][12][13]. Researchers in the field of mechanical fault diagnosis have taken advantage of the deep learning ability to realize more adaptive feature learning from vibration signals [14]. Some researchers applied various sorts of deep learning models as an upgraded classifier with manually extracted features [15][16][17][18][19][20][21][22][2...

A review on deep learning in planetary gearbox health state recognition: methods, applications, and dataset publication

“…The input of the model was raw vibration signals, and the category error and the authenticity label error were utilized as the loss to train the generator and the discriminator. In addition to the label information, Wen et al [141] added the condition information to the CNN-based generator, and the generator was trained by minimizing the log-likelihoods of the correct source, the fault class and the working condition. The model was optimized using the planetary gearbox signals of different health states under several rotating speeds, and one dataset including all health states under another condition was adopted to test the model.…”

Research on gearbox compound fault diagnosis method and system development based on entire gearbox health maintenance

“…12 In recent years, artificial intelligence diagnosis technology 15,16 has developed vigorously, such as support vector machine 7,15 and deep learning. 13,17 Besides, higher-order statistics 18,19 is a powerful tool for gearbox fault diagnosis.…”

Parallel adversarial feature learning and enhancement of feature discriminability for fault diagnosis of a planetary gearbox under time-varying speed conditions

“…Intelligent fault diagnosis provides an approach to avoiding subjective factors due to the merit of adaptive learning mechanisms, strong fault tolerance, and high non-linear regression ability [5]. In recent years, some typical strategies have been successfully applied to intelligent fault identification of rotating machinery, such as support vector machines (SVMs) [6][7][8], artificial neural networks (ANNs) [9], convolutional neural networks (CNNs) [10][11][12], recurrent neural networks (RNNs) [13], deep belief networks (DBNs) [14], deep auto-encoders (DAEs) [15,16], and generative adversarial networks (GANs) [17][18][19][20], which inspire that of planetary gearboxes.…”

“…In general, many intelligent diagnosis methods mainly including machine learning (ML) algorithms such as the SVM and ANN, and deep learning (DL) algorithms such as the RNN, DBN, DAE, and GAN have been successfully utilized for fault diagnosis [5,21]. However, it was also pointed out in [5,21] that intelligent methods based on ML have some limitations and uncertainties. On the one hand, signal pre-processing and feature extraction are dependent highly on expert knowledge.…”