A new automatic convolutional neural network based on deep reinforcement learning for fault diagnosis

Abstract

Convolutional neural network (CNN) has achieved remarkable applications in fault diagnosis. However, the tuning aiming at obtaining the well-trained CNN model is mainly manual search. Tuning requires considerable experiences on the knowledge on CNN training and fault diagnosis, and is always time consuming and labor intensive, making the automatic hyper parameter optimization (HPO) of CNN models essential. To solve this problem, this paper proposes a novel automatic CNN (ACNN) for fault diagnosis, which can automatically tune its three key hyper parameters, namely, learning rate, batch size, and L2-regulation. First, a new deep reinforcement learning (DRL) is developed, and it constructs an agent aiming at controlling these three hyper parameters along with the training of CNN models online. Second, a new structure of DRL is designed by combining deep deterministic policy gradient and long short-term memory, which takes the training loss of CNN models as its input and can output the adjustment on these three hyper parameters. Third, a new training method for ACNN is designed to enhance its stability. Two famous bearing datasets are selected to evaluate the performance of ACNN. It is compared with four commonly used HPO methods, namely, random search, Bayesian optimization, tree Parzen estimator, and sequential model-based algorithm configuration. ACNN is also compared with other published machine learning (ML) and deep learning (DL) methods. The results show that ACNN outperforms these HPO and ML/DL methods, validating its potential in fault diagnosis.

Graphical abstract

Keywords

deep reinforcement learning / hyper parameter optimization / convolutional neural network / fault diagnosis

References

[1] Zhang X , Huang T , Wu B , Hu Y M , Huang S , Zhou Q , Zhang X . Multi-model ensemble deep learning method for intelligent fault diagnosis with high-dimensional samples. Frontiers of Mechanical Engineering, 2021, 16( 2): 340– 352

[2] Chen X F , Wang S B , Qiao B J , Chen Q . Basic research on machinery fault diagnostics: past, present, and future trends. Frontiers of Mechanical Engineering, 2018, 13( 2): 264– 291

[3] Lei Y G , Yang B , Jiang X W , Jia F , Li N P , Nandi A K . Applications of machine learning to machine fault diagnosis: a review and roadmap. Mechanical Systems and Signal Processing, 2020, 138 : 106587

[4] Nath A G , Udmale S S , Singh S K . Role of artificial intelligence in rotor fault diagnosis: a comprehensive review. Artificial Intelligence Review, 2021, 54 : 2609– 2668

[5] Wang J L , Xu C Q , Dai L , Zhang J , Zhong R Y . An unequal deep learning approach for 3-D point cloud segmentation. IEEE Transactions on Industrial Informatics, 2021, 17( 12): 7913– 7922

[6] Chen Z Y , Mauricio A , Li W H , Gryllias K . A deep learning method for bearing fault diagnosis based on cyclic spectral coherence and convolutional neural networks. Mechanical Systems and Signal Processing, 2020, 140 : 106683

[7] Wolpert D H , Macready W G . No free lunch theorems for optimization. IEEE Transactions on Evolutionary Computation, 1997, 1( 1): 67– 82

[8] Wolpert D H . The supervised learning no-free-lunch theorems. In: Roy R, Köppen M, Ovaska S, Furuhashi T, Hoffmann F, eds. Soft Computing and Industry. London: Springer, 2002, 25– 42

[9] Hutter F Kotthoff L Vanschoren J. Automated Machine Learning: Methods, Systems, Challenges. Cham: Springer, 2019

[10] Wen L , Li X Y , Gao L . A new reinforcement learning based learning rate scheduler for convolutional neural network in fault classification. IEEE Transactions on Industrial Electronics, 2021, 68( 12): 12890– 12900

[11] Wen L , Ye X C , Gao L . A new automatic machine learning based hyperparameter optimization for workpiece quality prediction. Measurement and Control, 2020, 53( 7−8): 1088– 1098

[12] Feurer M Eggensperger K Falkner S Lindauer M Hutter F. Practical automated machine learning for the AutoML challenge 2018. In: Proceedings of International Workshop on Automatic Machine Learning at ICML. 2018, 1189‒ 1232

[13] He F X , Liu T L , Tao D C . Control batch size and learning rate to generalize well: theoretical and empirical evidence. In: Proceedings of the 33rd Conference on Neural Information Processing Systems (NeurIPS). Vancouver, 2019, 1143– 1152

[14] Li Y Z, Wei C, Ma T Y. Towards explaining the regularization effect of initial large learning rate in training neural networks. In: Proceedings of the 33rd International Conference on Neural Information Processing Systems. Vancouver, 2019, 11674–11685

[15] Zhou P , Peng Z K , Chen S Q , Yang Y , Zhang W M . Non-stationary signal analysis based on general parameterized time-frequency transform and its application in the feature extraction of a rotary machine. Frontiers of Mechanical Engineering, 2018, 13( 2): 292– 300

[16] Wang J L , Xu C Q , Yang Z L , Zhang J , Li X O . Deformable convolutional networks for efficient mixed-type wafer defect pattern recognition. IEEE Transactions on Semiconductor Manufacturing, 2020, 33( 4): 587– 596

[17] Xu G W , Liu M , Jiang Z F , Shen W M , Huang C X . Online fault diagnosis method based on transfer convolutional neural networks. IEEE Transactions on Instrumentation and Measurement, 2020, 69( 2): 509– 520

[18] Li Z X, Zheng T S, Wang Y, Cao Z, Guo Z Q, Fu H Y. A novel method for imbalanced fault diagnosis of rotating machinery based on generative adversarial networks. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 3500417

[19] Chen J B, Huang R Y, Zhao K, Wang W, Liu L C, Li W H. Multiscale convolutional neural network with feature alignment for bearing fault diagnosis. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 3517010

[20] Jiao J Y , Zhao M , Lin J , Liang K X . A comprehensive review on convolutional neural network in machine fault diagnosis. Neurocomputing, 2020, 417 : 36– 63

[21] Yao Y , Zhang S , Yang S X , Gui G . Learning attention representation with a multi-scale CNN for gear fault diagnosis under different working conditions. Sensors, 2020, 20( 4): 1233

[22] Li S , Wang H Q , Song L Y , Wang P X , Cui L L , Lin T J . An adaptive data fusion strategy for fault diagnosis based on the convolutional neural network. Measurement, 2020, 165 : 108122

[23] Kolar D , Lisjak D , Pająk M , Pavković D . Fault diagnosis of rotary machines using deep convolutional neural network with wide three axis vibration signal input. Sensors, 2020, 20( 14): 4017

[24] Wang R X , Jiang H K , Li X Q , Liu S W . A reinforcement neural architecture search method for rolling bearing fault diagnosis. Measurement, 2020, 154 : 107417

[25] Zhang K Y , Chen J L , He S L , Xu E Y , Li F D , Zhou Z T . Differentiable neural architecture search augmented with pruning and multi-objective optimization for time-efficient intelligent fault diagnosis of machinery. Mechanical Systems and Signal Processing, 2021, 158 : 107773

[26] Cabrera D , Guamán A , Zhang S H , Cerrada M , Sánchez R V , Cevallos J , Long J Y , Li C . Bayesian approach and time series dimensionality reduction to LSTM-based model-building for fault diagnosis of a reciprocating compressor. Neurocomputing, 2020, 380 : 51– 66

[27] Li L , Jamieson K , DeSalvo G , Rostamizadeh A , Talwalkar A . Hyperband: a novel bandit-based approach to hyperparameter optimization. The Journal of Machine Learning Research, 2018, 18( 1): 6765– 6816

[28] Li H , Zhang Q , Qin X R , Sun Y T . Raw vibration signal pattern recognition with automatic hyper-parameter-optimized convolutional neural network for bearing fault diagnosis. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science, 2020, 234( 1): 343– 360

[29] Long J Y , Zhang S H , Li C . Evolving deep echo state networks for intelligent fault diagnosis. IEEE Transactions on Industrial Informatics, 2020, 16( 7): 4928– 4937

[30] Han J H , Choi D J , Park S U , Hong S K . Hyperparameter optimization using a genetic algorithm considering verification time in a convolutional neural network. Journal of Electrical Engineering & Technology, 2020, 15( 2): 721– 726

[31] Wei J A , Huang H S , Yao L G , Hu Y , Fan Q S , Huang D . New imbalanced fault diagnosis framework based on cluster-MWMOTE and MFO-optimized LS-SVM using limited and complex bearing data. Engineering Applications of Artificial Intelligence, 2020, 96 : 103966

[32] Hansen S. Using deep Q-learning to control optimization hyperparameters. 2016, arXiv:1602.04062

[33] Zhang Z Z , Chen J L , Chen Z B , Li W P . Asynchronous episodic deep deterministic policy gradient: toward continuous control in computationally complex environments. IEEE Transactions on Cybernetics, 2021, 51( 2): 604– 613

[34] Zhu Z Y , Peng G L , Chen Y H , Gao H J . A convolutional neural network based on a capsule network with strong generalization for bearing fault diagnosis. Neurocomputing, 2019, 323 : 62– 75

[35] Wang Y , Ning D J , Feng S L . A novel capsule network based on wide convolution and multi-scale convolution for fault diagnosis. Applied Sciences, 2020, 10( 10): 3659

[36] Wen L , Li X Y , Gao L . A new two-level hierarchical diagnosis network based on convolutional neural network. IEEE Transactions on Instrumentation and Measurement, 2020, 69( 2): 330– 338

[37] Zhang W , Peng G L , Li C H , Chen Y H , Zhang Z J . A new deep learning model for fault diagnosis with good anti-noise and domain adaptation ability on raw vibration signals. Sensors, 2017, 17( 2): 425

[38] Hoang D T , Kang H J . Rolling element bearing fault diagnosis using convolutional neural network and vibration image. Cognitive Systems Research, 2019, 53 : 42– 50

[39] Jiang G Q , He H B , Yan J , Xie P . Multiscale convolutional neural networks for fault diagnosis of wind turbine gearbox. IEEE Transactions on Industrial Electronics, 2019, 66( 4): 3196– 3207

[40] Qiao H H , Wang T Y , Wang P , Zhang L , Xu M D . An adaptive weighted multiscale convolutional neural network for rotating machinery fault diagnosis under variable operating conditions. IEEE Access: Practical Innovations, Open Solutions, 2019, 7 : 118954– 118964

[41] Song Y , Li Y B , Jia L , Qiu M K . Retraining strategy-based domain adaption network for intelligent fault diagnosis. IEEE Transactions on Industrial Informatics, 2020, 16( 9): 6163– 6171

Nomenclature

Abbreviations
ACNN	Automatic convolutional neural network
ANN	Artificial neural network
AVG	Average prediction accuracy
AWMS-CNN	Adaptive weighted multiscale convolutional neural network
BO	Bayesian optimization
CNN	Convolutional neural network
DAN	Domain adaption network
DDPG	Deep deterministic policy gradient
DL	Deep learning
DRL	Deep reinforcement learning
ELM	Extreme learning machine
FC	Fully connected
GS	Grid search
HCNN	Hierarchical convolutional neural network
HPO	Hyper parameter optimization
ICN	CNN based on a capsule network with an inception block
IF	Inner race fault
LSTM	Long short-term memory
ML	Machine learning
MSCNN	Multiscale convolutional neural network
NAS	Neural architecture search
OF	Outer race fault
RF	Roller fault
RL	Reinforcement learning
RS	Random search
SMAC	Sequential model-based algorithm configuration
SVM	Support vector machine
TPE	Tree Parzen estimator
VI-CNN	CNN based on vibration image
WDCNN	Deep convolutional neural networks with wide first-layer kernels
WMSCCN	Wide convolution and multiscale convolution
Variables
a	Action of DDPG algorithm
at	Current action
bt	Current batch size
bmax, bmin	The upper and lower boundaries for batch size
E_{\mathrm{\Pi }}	Expected value under policy \mathrm{\Pi }
f	Fourier frequency in short-time Fourier transform
lt	Current L2-regulation value
lmax, lmin	The upper and lower boundaries for L2-regulation value
lrmax, lrmin	The upper and lower boundaries for learning rate, respectivley
lrt	Current learning rate
losst	Loss value of the CNN model at time step t
L	Training loss of critic network
M	Number of the sequencing training loss
n	Number of samples in the experience storage D
p	Transition probability function
Q^{\mathrm{\Pi }} (s, a)	Q-value function under policy \mathrm{\Pi }
r	Reward of DDPG algorithm
s	State of DDPG algorithm
st	State at time step t
STFT	Short-time Fourier transform formulation
t	Time step
w(t)	Window function
yt	Actual Q-value
α	Factor to control the degree of soft updating
γ	Discount factor
\mathrm{\Pi }	Policy of the agent to choose the action
{\theta }^{\mu }	Online actor network
{\theta }^{{\mu }^{\prime }}	Target actor network
{\omega }^Q	Online critic network
$$\begin{gathered} {\omega }^{Q^{\prime }} \\ \end{gathered}$$	Target critic network

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China (Grant Nos. 51805192 and U21B2029), the Major Special Science and Technology Project of Hubei Province, China (Grant No. 2020A EA009), and the State Key Laboratory of Digital Manufacturing Equipment and Technology of Huazhong University of Science and Technology, China (Grant No. DMETKF2020029).

RIGHTS & PERMISSIONS

2022 Higher Education Press 2022