Ensemble deep contractive auto-encoders for intelligent fault diagnosis of machines under noisy environment
Introduction
Fault diagnosis of machine is of utmost importance, because early detection of emerging fault significantly improves operational continuity and equipment safety [1]. Among the existing methods, intelligent diagnosis methods do not need to model the physical systems and is very suitable for modern complex equipment [2]. Therefore, it has attracted widespread research interest recently.
Deep auto-encoder (DAE) is a promising intelligent diagnosis algorithm due to its ability to learn salient feature representations automatically [3], [4]. Such ability overcomes the weakness of traditional manual feature engineering that it is ad-hoc, labor-intensive, and may not generalize to multiple contexts [5]. Benefiting from the above advantage, intelligent diagnosis methods based on DAE have achieved great success [6]. Standard DAE was applied to the fault diagnosis of roller bearing [7]. DAE with a new maximum correntropy cost function was applied to fault diagnosis of gearbox [8]. As a variant of DAE, sparse deep auto-encoder (SDAE) encourages sparsity of feature representation. An extension SDAE was proposed for induction motor fault diagnosis [9]. However, in real engineering applications, environment noise corrupts signals [10]. DAE and SDAE cannot learn invariant features from such corrupted data, leading to seriously degraded diagnosis accuracies. Deep denoising auto-encoder (DDAE) provides an access to deal with the noisy data, which is achieved by adding noise with specified intensity to training samples [11]. But in real industrial applications, noise intensity usually varies and is unknown, which limits the application of DDAE. Therefore, it is still urgent to develop powerful intelligent diagnosis methods that have better adaptability to noise.
Deep contractive auto-encoder (DCAE) is a variant of DAE [12]. It explicitly encourages invariance and robustness of feature representation by penalizing the sensitivity to input [13]. DCAE provides an effective way to handle noisy data without knowing noise intensity. However, due to single network structure and difficulty in selecting activation functions, individual DCAE model has low generalization performances in different contexts. To alleviate this problem, ensemble learning provides an approach, which uses some certain strategies to combine the results from multiple individual learners and improves the generalization performances. To the best of our knowledge, aiming at fault diagnosis, only Shao et al. [14] combined ensemble learning with SDAE model. Although their method yielded better diagnosis accuracies than other diagnosis methods, they did not consider the influence of noise. Additionally, their method has several obvious drawbacks (see Section 2.1). Considering that the research of DAE with ensemble learning is still in its infancy, it is meaningful to develop ensemble DAE by taking advantage of both modules.
In this research, a novel ensemble deep contractive auto-encoders (EDCAE) is proposed for intelligent fault diagnosis of machines under noisy environment. It includes three steps: (1) Fifteen different activation functions are adopted to design fifteen DCAE models. These member models learn invariant feature representation from the noisy data. (2) Fisher discriminant analysis selects the low-dimensional features with the maximum class separability and softmax classifier outputs fifteen classification results. (3) A new ensemble strategy is developed to combine the fifteen classification results. Additionally, theory analysis about error bound of the ensemble strategy is conducted to prove its effectiveness. The proposed method is evaluated on three cases including bearing, gear box and Self-priming centrifugal pump. Experiments are conducted under seven different signal-to-noise-ratios (SNR). Results demonstrate that EDCAE is superior to individual DCAE model, DAE, SDAE, DDAE and several ensemble methods.
The rest of this paper is organized as follows. Related works are described in Section 2. In Section 3, the proposed EDCAE is detailed. In Section 4, the experiments and comparisons are conducted. Finally, conclusions and future work are drawn.
Section snippets
Related works
This section provides a brief introduction of related works, including literature review in Section 2.1 and principle of DCAE in Section 2.2.
The proposed EDCAE for intelligent fault diagnosis
This section details the proposed EDCAE method for intelligent fault diagnosis of machines. It contains four parts: (1) DCAE models design; (2) feature learning, selection and identification; (3) combination strategy design; (4) overall procedure of the proposed method.
Experimental verification
Section 4.1 presents the experiment scheme design. Based on the scheme, the proposed method is verified on three case studies, including bearing fault diagnosis, gear box fault diagnosis and self-priming centrifugal pump fault diagnosis.
Conclusion and future work
This paper proposes a new EDCAE method for intelligent fault diagnosis of machine under noisy environment. The main contributions are designing fifteen DCAE models with different characteristics, fusing DCAE model with FDA for feature learning and selection, developing a new combination strategy. Additionally, theory analysis about error bound is implemented to prove the effectiveness of EDCAE. EDCAE is evaluated on three case studies, including motor bearing, gear box and self-priming
CRediT authorship contribution statement
Yuyan Zhang: Methodology, Software, Writing - original draft, Validation, Data curation. Xinyu Li: Formal analysis, Writing - review & editing. Liang Gao: Conceptualization, Supervision, Resources, Funding acquisition. Wen Chen: Visualization, Investigation. Peigen Li: Project administration.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by National Key Research and Development Project [Grant Number 2019YFB1704603], National Natural Science Foundation of China [Grant Number 51775216], Natural Science Foundation of Hubei Province [Grant Number 2018CFA078], and Program for HUST Academic Frontier Youth Team [Grant Number 2017QYTD04].
References (35)
- et al.
HSAE: A hessian regularized sparse auto-encoders
Neurocomputing
(2016) - et al.
Learning transferable features in deep convolutional neural networks for diagnosing unseen machine conditions
ISA Trans.
(2019) - et al.
A novel deep autoencoder feature learning method for rotating machinery fault diagnosis
Mech. Syst. Signal Process.
(2017) - et al.
A sparse auto-encoder-based deep neural network approach for induction motor faults classification
Measurement
(2016) - et al.
A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load
Mech. Syst. Signal Process.
(2018) - et al.
A novel method for intelligent fault diagnosis of rolling bearings using ensemble deep auto-encoders
Mech. Syst. Signal Process.
(2018) - et al.
Deep neural networks: A promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data
Mech. Syst. Signal Process.
(2016) - et al.
A neural network constructed by deep learning technique and its application to intelligent fault diagnosis of machines
Neurocomputing
(2018) - et al.
A new subset based deep feature learning method for intelligent fault diagnosis of bearing
Expert Syst. Appl.
(2018) - et al.
Fault diagnosis of rotary machinery components using a stacked denoising autoencoder-based health state identification
Signal Process.
(2017)
An enhancement deep feature fusion method for rotating machinery fault diagnosis
Knowl.-Based Syst.
A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load
Mech. Syst. Signal Process.
A multiobjective evolutionary algorithm based on decomposition for hybrid flowshop green scheduling problem
Comput. Ind. Eng.
Bearing fault diagnosis based on wavelet transform and fuzzy inference
Mech. Syst. Signal Process.
Imbalanced data fault diagnosis of rotating machinery using synthetic oversampling and feature learning
J. Manuf. Syst.
Deep transfer network with joint distribution adaptation: A new intelligent fault diagnosis framework for industry application
ISA Trans.
Comparison of random forest, artificial neural networks and support vector machine for intelligent diagnosis of rotating machinery
Trans. Inst. Meas. Control
Cited by (53)
A novel digital twin-driven approach based on physical-virtual data fusion for gearbox fault diagnosis
2023, Reliability Engineering and System SafetyModeling automatic pavement crack object detection and pixel-level segmentation
2023, Automation in ConstructionA novel domain adversarial time-varying conditions intervened neural network for drill bit wear monitoring of the jumbo drill under variable working conditions
2023, Measurement: Journal of the International Measurement Confederation