Abstract
Predictive maintenance (PdM) is an advanced technique to predict the time to failure (TTF) of a system. PdM collects sensor data on the health of a system, processes the information using data analytics, and then establishes data-driven models that can forecast system failure. Deep neural networks are increasingly being used as these data-driven models owing to their high predictive accuracy and efficiency. However, deep neural networks are often criticized as being “black boxes,” which owing to their multi-layered and non-linear structure provide little insight into the underlying physics of the system being monitored and that are nontransparent and untraceable in their predictions. In order to address this issue, the layer-wise relevance propagation (LRP) technique is applied to analyze a long short-term memory (LSTM) recurrent neural network (RNN) model. The proposed method is demonstrated and validated for a bearing health monitoring study based on vibration data. The obtained LRP results provide insights into how the model “learns” from the input data and demonstrate the distribution of contribution/relevance to the neural network classification in the input space. In addition, comparisons are made with gradient-based sensitivity analysis to show the power of LRP in interpreting RNN models. The LRP is proved to have promising potential in interpreting deep neural network models and improving model accuracy and efficiency for PdM.
Similar content being viewed by others
Data Availability
The raw data used in case study was from NASA Ames Prognostics Data Repository (http://ti.arc.nasa.gov/project/prognostic-data-repository).
References
Monostori L, Kádár B, Bauernhansl T, Monostori L, Kádár B, Bauernhansl T, Kondoh S, Kumara S, Reinhart G, Sauer O, Schuh G, Sihn W, Ueda K (2016) Cyber-physical systems in manufacturing. CIRP Ann 65:621–641. https://doi.org/10.1016/j.cirp.2016.06.005
Menard S (2011) Applied logistic regression analysis. Appl Logist Regres Anal. https://doi.org/10.4135/9781412983433
Wang G, Luo Z, Qin X, Leng Y, Wang T (2008) Fault identification and classification of rolling element bearing based on time-varying autoregressive spectrum. Mech Syst Signal Process 22:934–947. https://doi.org/10.1016/j.ymssp.2007.10.008
Salem O, Guerassimov A, Mehaoua A (2014) Anomaly detection in medical wireless sensor networks using SVM and linear regression models. Int J E-Health Med Commun 5:20–45. https://doi.org/10.4018/ijehmc.2014010102
Yan HC, Zhou JH, Pang CK (2015) Gamma process with recursive MLE for wear PDF prediction in precognitive maintenance under aperiodic monitoring. Mechatronics 31:68–77. https://doi.org/10.1016/j.mechatronics.2015.05.009
Xu W, Wang W (2012) An adaptive gamma process based model for residual useful life prediction. Proc IEEE 2012 Progn Syst Heal Manag Conf PHM-2012 3–6. https://doi.org/10.1109/PHM.2012.6228785
Wei Q, Xu D (2014) Remaining useful life estimation based on gamma process considered with measurement error. ICRMS 2014 - Proc 2014 10th Int Conf Reliab Maintainab Saf More Reliab Prod More Secur Life 645–649. https://doi.org/10.1109/ICRMS.2014.7107275
Zhou D, Yu Z, Zhang H, Weng S (2016) A novel grey prognostic model based on Markov process and grey incidence analysis for energy conversion equipment degradation. Energy 109:420–429. https://doi.org/10.1016/j.energy.2016.05.008
Ertunc HM, Loparo KA, Ocak H (2001) Tool wear condition monitoring in drilling operations using hidden Markov models (HMMs). Int J Mach Tools Manuf 41:1363–1384. https://doi.org/10.1016/S0890-6955(00)00112-7
Ocak H, Loparo KA, Discenzo FM (2007) Online tracking of bearing wear using wavelet packet decomposition and probabilistic modeling: a method for bearing prognostics. J Sound Vib 302:951–961. https://doi.org/10.1016/j.jsv.2007.01.001
Guo L (2017) A recurrent neural network based health indicator for remaining useful life prediction of bearings. Bol Tec Bull 55:585–590. https://doi.org/10.1016/j.neucom.2017.02.045
Huang CG, Huang HZ, Li YF (2019) A bidirectional LSTM prognostics method under multiple operational conditions. IEEE Trans Ind Electron 66:8792–8802. https://doi.org/10.1109/TIE.2019.2891463
Gugulothu N (2017) Predicting remaining useful life using time series embeddings based on recurrent neural networks. CEUR Workshop Proc 2657:1–9. https://doi.org/10.1145/nnnnnnn.nnnnnnn
Montavon G, Samek W, Müller KR (2018) Methods for interpreting and understanding deep neural networks. Digit Signal Process A Rev J 73:1–15. https://doi.org/10.1016/j.dsp.2017.10.011
Böhle M, Eitel F, Weygandt M, Ritter K (2019) Layer-wise relevance propagation for explaining deep neural network decisions in MRI-based Alzheimer’s disease classification. Front Aging Neurosci 11:10. https://doi.org/10.3389/fnagi.2019.00194
Arras L, Montavon G, Müller K-R, Samek W (2018) Explaining recurrent neural network predictions in sentiment analysis. 159–168. https://doi.org/10.18653/v1/w17-5221
Hochreiter S. The vanishing gradient problem during learning recurrent neural nets and problem solutions
Werbos PJ (1990) Backpropagation through time: what it does and how to do it. Proc IEEE 78:1550–1560. https://doi.org/10.1109/5.58337
Hochreiter S (1997) Long short-term memory. 1780:1735–1780
Wu D, Jiang Z, Xie X, Wei X (2020) LSTM learning with Bayesian and Gaussian processing for anomaly detection in industrial IoT. 16:5244–5253
Wang Y, Perry M, Whitlock D, Sutherland JW (2020) Detecting anomalies in time series data from a manufacturing system using recurrent neural networks. J Manuf Syst. https://doi.org/10.1016/j.jmsy.2020.12.007
Li Z, Li J, Wang Y, Wang K (2019) A deep learning approach for anomaly detection based on SAE and LSTM in mechanical equipment. Int J Adv Manuf Technol 103:499–510. https://doi.org/10.1007/s00170-019-03557-w
Dudukcu HV, Taskiran M, Kahraman N (2020) LSTM and WaveNet implementation for predictive maintenance of turbofan engines. 20th IEEE Int Symp Comput Intell Informatics, CINTI 2020 - Proc 151–156. https://doi.org/10.1109/CINTI51262.2020.9305820
Zhang W, Guo W, Liu X, Liu Y, Zhou J, Li B, Lu Q, Yang S (2018) LSTM-based analysis of industrial IoT equipment. IEEE Access 6:23551–23560. https://doi.org/10.1109/ACCESS.2018.2825538
Zhao R, Wang J, Yan R, Mao K (2016) Machine health monitoring with LSTM networks. Proc Int Conf Sens Technol ICST:17–22. https://doi.org/10.1109/ICSensT.2016.7796266
Binder A, Bach S, Montavon G et al (2016) Layer-wise relevance propagation for deep neural network architectures. Lect Notes Electr Eng 376:913–922. https://doi.org/10.1007/978-981-10-0557-2_87
Arras L, Arjona-Medina J, Widrich M, et al (2019) Explaining and interpreting LSTMs. Lect Notes Comput Sci (including Subser Lect Notes Artif Intell Lect Notes Bioinformatics) 11700 LNCS:211–238. https://doi.org/10.1007/978-3-030-28954-6_11
Wu H, Huang A, Sutherland JW (2020) Avoiding environmental consequences of equipment failure via an LSTM-based model for predictive maintenance. Procedia Manuf 43:666–673. https://doi.org/10.1016/j.promfg.2020.02.131
University of Cincinnati (2006). “Bearing data set”, NASA Ames prognostics data repository (http://ti.arc.nasa.gov/project/prognostic-data-repository), NASA Ames Research Center, Moffett Field, CA
Qiu H, Lee J, Lin J, Yu G (2006) Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics. J Sound Vib 289:1066–1090. https://doi.org/10.1016/j.jsv.2005.03.007
Lee J, Wu F, Zhao W, Ghaffari M, Liao L, Siegel D (2014) Prognostics and health management design for rotary machinery systems - reviews, methodology and applications. Mech Syst Signal Process 42:314–334. https://doi.org/10.1016/j.ymssp.2013.06.004
Zurada JM, Malinowski A, Cloete I (1994) Sensitivity analysis for minimization of input data dimension for feedforward neural network. Proc - IEEE Int Symp Circuits Syst 6:447–450. https://doi.org/10.1109/iscas.1994.409622
Caesarendra W, Tjahjowidodo T (2017) A review of feature extraction methods in vibration-based condition monitoring and its application for degradation trend estimation of low-speed slew bearing. Machines 5:21. https://doi.org/10.3390/machines5040021
Acknowledgements
The authors gratefully acknowledge the support of the Wabash Heartland Innovation Network (WHIN) grant at Purdue University.
Funding
This research was supported by the Wabash Heartland Innovation Network (WHIN) grant at Purdue University.
Author information
Authors and Affiliations
Contributions
Haiyue Wu: Conceptualization, investigation, methodology, writing-original draft.
Aihua Huang: Writing-reviewing and editing.
John W. Sutherland: Supervision, writing-reviewing and editing.
Corresponding author
Ethics declarations
Ethics approval
The authors would like to declare that the work described was original which has not been published previously or not under consideration for publication elsewhere.
Consent to participate
Not applicable
Consent for publication
All of the authors have reviewed the final version of the manuscript and approved it for publication.
Conflict of interest
The authors declare no competing interests.
Additional information
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Rights and permissions
About this article
Cite this article
Wu, H., Huang, A. & Sutherland, J.W. Layer-wise relevance propagation for interpreting LSTM-RNN decisions in predictive maintenance. Int J Adv Manuf Technol 118, 963–978 (2022). https://doi.org/10.1007/s00170-021-07911-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00170-021-07911-9