TJ 15 2021 1 156 161
TJ 15 2021 1 156 161
TJ 15 2021 1 156 161
https://doi.org/10.31803/tg-20210205105921
Abstract: In this technical paper, we address the issue of predicting cash dispenser (addressed as ‘Device’ henceforth) failure by harnessing the power of humungous data from
service history, logs, metrics, transactions, and plausible environmental factors. This study helps increase device availability, enhanced customer experience, manage risk &
compliance and revenue growth. It also helps reduce maintenance cost, travel cost, labour cost, downtime, repair duration and increase meantime between failures (MTBF) of
individual components. This study uses a cognitive prioritization model which entails the following at its core; a) Machine Learning engineered features with highest influence on
machine failure, b) Observation Windows, Transition Windows and Prediction Windows to accommodate various business processes and service planning delivery windows, and
c) A forward-looking evaluation of emerging patterns to determine failure prediction score that is prioritized by business impact, for a predefined time window in the future. The
model not only predicts failure score for the devices to be serviced, but it also reduces the service miss impact for the prediction windows.
Keywords: Cash Dispenser Failure Prediction; Cognitive Prioritization Model; Feature Engineering; Machine Learning; Predictive Maintenance (PdM)
on generic approaches around predictive maintenance. Some 3.1 Exploratory Data Analysis (EDA)
of them focus on challenges and reliability aspect of PdM in
their study [6], while others focus only on the cost and An extensive EDA is performed during this study to
economic aspect [2, 15, 19]. While there are studies done for understand data distribution before applying any model.
Industrial and IoT devices [10, 11, 12, 14, 16], there are only Univariate, Bivariate & multivariate analysis is performed to
a handful studies available for cash dispenser devices [21] understand distribution, association & correlation of these
wherein only event/error logs are considered as variables to variables better. The following Figs. 4, 5, 6 & 7 are some of
perform Time Series Classification. Strong indicators of the visuals (illustrative not exhaustive) depicting the same.
failures could be gleaned out from service history, metrics,
transactions, etc. which is missing from the existing work in
this field. The focus of this study is to harness the power of
data points from service history, logs, metrics, transactions,
and plausible environmental factors to get a holistic view
around the device and component failure.
3 PROPOSED METHOD
Predictions are generated at module or component level 3.4 MODEL TRAINING & EVALUATION
and therefore, features are also engineered for each module.
As described earlier, there are two types of features used in Data for CCDM module suffers from imbalance of
this study; static & dynamic. Static features are time windows classes; ‘failure’ & ‘Operational’. Therefore, sampling
independent and dynamic features respect time windows techniques; upsampling, downsampling, SMOTE are
mentioned beforehand. employed to create a dataset with balanced classes. Inbuilt
Inventory details like ‘OEM’, ‘Age’, ‘TAR Action’, weightage techniques provided in R libraries are also
‘TAR problem’ etc. are static. Only active inventory leveraged to overcome class imbalance issue. Cross
(Devices that are not yet deinstalled) is considered for this validation and bootstrapping strategies are used in splitting
study. ‘Last Failure’, ‘Uptime’, ‘Last Visit’ etc. are computed train and test datasets to tackle overfitting issues. Numerous
using Observation Window and most recently closed ticket. experiments are carried out for each module and algorithm
Transaction features ‘Operation Count’, ‘Served Count’, combination by tuning various hyperparameter nobs to arrive
‘Reversed’ etc. are created by performing aggregation. Error at the optimal parameter values resulting in best performance
Figure 12 Recall and Precision for AQ (CCDM) module using Random Forest
Notice
8 REFERENCES
Figure 13 Predictions vs Reactive Tickets for Prediction Rounds [1] Ran, Y., Zhou, X., Lin, P., Wen, Y., & Deng, R. (2019). A
Survey of Predictive Maintenance: Systems, Purposes and
Further, such analysis shows that there is an increase in Approaches. Retrieved from: https://arxiv.org/abs/1912.07383
device availability by 2% if this solution is implemented for [2] S. Spiegel, F. Mueller, D. Weismann, J. Bird (2018). Cost-
predictive maintenance. Meantime between failures of Sensitive Learning for Predictive Maintenance. Retrieved
individual components is almost doubled. The solution from: https://arxiv.org/abs/1809.10979
[3] Tuggener, L., Amirian, M., Rombach, K., Lörwald, S., Varlet,
identifies failures in advance leading to the optimization of
A., Westermann, C., & Stadelmann, T. (2019). Automated
logistics and associated functions. Downtime is also lowered Machine Learning in Practice: State of the Art and Recent
by a significant factor. First touch resolution is increased Results. https://doi.org/10.1109/SDS.2019.00-11
within a week of ticket logging. Repair duration, labor cost [4] Miller, K. & Dubrawski, A. (2020). System-Level Predictive
and travel costs are lowered noticeably too. Retail banks are Maintenance: Review of Research Literature and Gap
able to attain spillover benefits from optimized cash in transit Analysis. Retrieved from:
activities, as they relate to second line maintenance. https://arxiv.org/pdf/2005.05239.pdf
[5] Bousdekis, A., Lepenioti, K., Apostolou, D., & Mentzas, G.
6 CONCLUSION (2019). Decision Making in Predictive Maintenance: Literature
Review and Research Agenda for Industry 4.0. IFAC-
In this work, we have studied how machine learning PapersOnLine, 52(13), 607-612.
heuristic models augmented with sophisticated feature https://doi.org/10.1016/j.ifacol.2019.11.226
engineering mechanism could be used to empower retail [6] Meyer Zu Wickern, V. - Challenges and Reliability of
bankers by predicting component that are mostly likely to fail Predictive Maintenance. (2019). Challenges and Reliability of
Predictive Maintenance.
in the next four weeks. Multidimensional data from disparate
https://doi.org/10.13140/RG.2.2.35379.89129
touch points of devices is employed to facilitate predictive [7] Poór, P. & Basl, J. (2019). Predictive maintenance as an
maintenance in real time. intelligent service in Industry 4.0. Journal of Systems
This study could easily be prolonged to predict failure in Integration, 10(1), 3-10. https://doi.org/10.20470/jsi.v10i1.364
other devices such as POS machines, Self-checkout kiosks, [8] Paolanti, M., Romeo, L., Felicetti, A., Mancini, A., Frontoni,
etc. E., & Loncarski, J. (2018). Machine Learning approach for
Predictive Maintenance in Industry 4.0. The 14th IEEE/ASME
7 WAY FORWARD International Conference on Mechatronic and Embedded
Systems and Applications (MESA), 1-6.
This study has several avenues of enhancement to add https://doi.org/10.1109/MESA.2018.8449150
more capabilities into the overall Predictive maintenance [9] Schmidt, B., Sandberg, U., & Wang, L. (2014). Next
Generation Condition Based Predictive Maintenance. The 6th
solution.
International Swedish Production Symposium,
https://www.researchgate.net/publication/318775487_Next_G
eneration_Condition_Based_Predictive_Maintenance.
[10] Bonci, A., Longhi, S., & Nabissi, G. (2019). Predictive
Maintenance System using motor current signal analysis for
Industrial Robot. The 24th IEEE International Conference on
Emerging Technologies and Factory Automation (ETFA),
1453-1456. https://doi.org/10.1109/ETFA.2019.8869067
[11] Cakir, M., Ali Guvenc, M., & Mistikoglu, S. (2020). The
Experimental Application of Popular Machine Learning
Algorithms on Predictive Maintenance and the Design of IIoT
based Condition Monitoring System. Computers & Industrial
Engineering, 106948. Available online 30 October 2020.
https://doi.org/10.1016/j.cie.2020.106948
[12] Maashi, M., Alwhibi, N., Alamr, F., Alzahrani, R., Alhamid,
A., & Altawallah, N. (2020). Industrial Duct Fan Maintenance
Predictive Approach Based on Random Forest. The 9th
International Conference on Information Technology
Convergence and Services (ITCSE 2020), 177-184.
https://doi.org/10.5121/csit.2020.100516
[13] Raza, A. & Ulansky, V. (2016). Modelling of predictive
maintenance for a periodically inspected system. Procedia
CIRP 59, 95-101. https://doi.org/10.1016/j.procir.2016.09.032
[14] Vibha, K., Pal, H., Bendigeri, M. V., & Panna, N. (2020).
Development of Predictive Maintenance System of Motors.
International Journal of Advanced Science and Technology,
29(7), 2344-2351.
[15] Jing Yu, D. T. & Wing-Keung, A. L. (2020). An Economic
Perspective on Predictive Maintenance of Filtration Units.
Retrieved from: https://arxiv.org/abs/2008.11070
[16] Zheng, H., Paiva, A. R., & Gurciullo, C. S. (2020). Advancing
from Predictive Maintenance to Intelligent Maintenance with
AI and IIoT. Retrieved from: https://aiotworkshop.github.io/
published/AIoT_4_Zheng_TechnicalPaper_KDD2020.pdf
[17] Serradilla, O., Zugasti, E., & Zurutuza, U. (2020). Deep
learning models for predictive maintenance: a survey,
comparison, challenges and prospect. Retrieved form:
https://arxiv.org/pdf/2010.03207.pdf
[18] Satta, R., Cavallari, S., Pomponi, E., Grasselli, D., Picheo, D.,
& Annis, C. (2017). A dissimilarity-based approach to
predictive maintenance with application to HVAC systems.
Retrieved from:
https://arxiv.org/ftp/arxiv/papers/1701/1701.03633.pdf
[19] Alsyouf, I. (2007). The role of maintenance in improving
companies’ productivity and profitability. International
Journal of Production Economics, 105(1), 70-78.
https://doi.org/10.1016/j.ijpe.2004.06.057
[20] Guillaume, A., Vrain, C., & Wael, E. (2020). Time series
classification for predictive maintenance on event logs.
Retrieved from: https://arxiv.org/pdf/1911.10996.pdf
Authors’ contacts: