561
3.2 Unsupervised learning
One of the many possible applications of unsupervised learning is in the field of outlier
detection. It allows the user to identify outliers automatically within a series of data, that is,
points that have different characteristics from the rest of the distribution. The two algorithms
mainly used for this are One Class Support Vector Machine (OCSVM) and isolation forests.
The first one is a variation of the classical SVMs where instead of finding the best separating
hyperplane between different categories, it finds a hyperplane that includes most of the data,
effectively defining what is normal data. On the other hand, isolation forest randomly split
the data in two containers, and repeat the process until all data points are in different containers. By repeating the separating process multiple times, the algorithm can identify outliers by
finding the points that in average takes the least splits to be singled out.
These algorithms are currently used with data generated from each sensor individually, e.g.
rolling average etc, in order to detect faulty behavior. It was found that sensors that produce
more outliers then average are almost always in a faulty condition. This might sound trivial
for some faults where for example the sensors simply do not transmit any value anymore, but
prove to be a very valuable tool for situations where sensors are simply off by a few percent
or vary abnormally. When the integrity of each sensor is confirmed, combinations of sensors
are then tested to detect many more faults, like transmission, engine power or battery problems. The advantage of these algorithms is that no labeling effort needs to be put in and that
they are more likely to detect never-before-seen failures.
4 CONCLUSION
This article presented the results of the application of machine learning technique to predictive maintenance of mobile equipment for underground mines. It showed that the combination of real-time data and Machine Learning helps create large savings for underground mine
operations by detecting multiple different failures in advance. Even though these results are
very promising, they represent only one of the possible fields of application of AI in mining. The authors think that other applications will also soon start creating value, for instance
in the area of short interval control and real-time planning optimization. Early results have
been achieved in these fields promising great improvements in day-to-day operations.
REFERENCES
Beghi, A., L. Cecchinato, C. Corazzol, M. Rampazzo, F. Simmini, and G.A. Susto. 2014. “A One-Class
SVM Based Tool for Machine Learning Novelty Detection in HVAC Chiller Systems.” IFAC Proceedings Volumes (IFAC-PapersOnline) 19: 1953–58. https://doi.org/10.3182/20140824-6-ZA-1003.02382.
Hashemian, H.M., and Wendell C. Bean. 2011. “State-of-the-Art Predictive Maintenance Techniques.”
IEEE Transactions on Instrumentation and Measurement 60 (10): 3480–92. https://doi.org/10.1109/
TIM.2009.2036347.
Laukka, Arto, Juhamatti Saari, Jari Ruuska, Esko Juuso, and Sulo Lahdelma. 2016. “Condition-Based
Monitoring for Underground Mobile Machines.” International Journal of Industrial and Systems
Engineering 23 (1): 74–89.
Li, Hongfei, Dhaivat Parikh, Qing He, Buyue Qian, Zhiguo Li, Dongping Fang, and Arun Hampapur.
2014. “Improving Rail Network Velocity: A Machine Learning Approach to Predictive Maintenance.”
Transportation Research Part C: Emerging Technologies 45: 17–26. https://doi.org/10.1016/j.
trc.2014.04.013.
Mkemai, Rashid. 2011. “Maintenance Procedures and Practices for Underground Mobile Mining
Equipment,” 1–76.
Peng, Ying, Ming Dong, and Ming Jian Zuo. 2010. “Current Status of Machine Prognostics in
Condition-Based Maintenance: A Review.” International Journal of Advanced Manufacturing
Technology 50 (1–4): 297–313. https://doi.org/10.1007/s00170-009-2482-0.
Schwabacher, Mark, and Kai Goebel. 2007. “A Survey of Artificial Intelligence for Prognostics.” AAAI
Fall Symposium, 107–14. http://www.aaai.org/Papers/Symposia/Fall/2007/FS-07-02/FS07-02-016.pdf.
3.2 Unsupervised learning
One of the many possible applications of unsupervised learning is in the field of outlier
detection. It allows the user to identify outliers automatically within a series of data, that is,
points that have different characteristics from the rest of the distribution. The two algorithms
mainly used for this are One Class Support Vector Machine (OCSVM) and isolation forests.
The first one is a variation of the classical SVMs where instead of finding the best separating
hyperplane between different categories, it finds a hyperplane that includes most of the data,
effectively defining what is normal data. On the other hand, isolation forest randomly split
the data in two containers, and repeat the process until all data points are in different containers. By repeating the separating process multiple times, the algorithm can identify outliers by
finding the points that in average takes the least splits to be singled out.
These algorithms are currently used with data generated from each sensor individually, e.g.
rolling average etc, in order to detect faulty behavior. It was found that sensors that produce
more outliers then average are almost always in a faulty condition. This might sound trivial
for some faults where for example the sensors simply do not transmit any value anymore, but
prove to be a very valuable tool for situations where sensors are simply off by a few percent
or vary abnormally. When the integrity of each sensor is confirmed, combinations of sensors
are then tested to detect many more faults, like transmission, engine power or battery problems. The advantage of these algorithms is that no labeling effort needs to be put in and that
they are more likely to detect never-before-seen failures.
4 CONCLUSION
This article presented the results of the application of machine learning technique to predictive maintenance of mobile equipment for underground mines. It showed that the combination of real-time data and Machine Learning helps create large savings for underground mine
operations by detecting multiple different failures in advance. Even though these results are
very promising, they represent only one of the possible fields of application of AI in mining. The authors think that other applications will also soon start creating value, for instance
in the area of short interval control and real-time planning optimization. Early results have
been achieved in these fields promising great improvements in day-to-day operations.
REFERENCES
Beghi, A., L. Cecchinato, C. Corazzol, M. Rampazzo, F. Simmini, and G.A. Susto. 2014. “A One-Class
SVM Based Tool for Machine Learning Novelty Detection in HVAC Chiller Systems.” IFAC Proceedings Volumes (IFAC-PapersOnline) 19: 1953–58. https://doi.org/10.3182/20140824-6-ZA-1003.02382.
Hashemian, H.M., and Wendell C. Bean. 2011. “State-of-the-Art Predictive Maintenance Techniques.”
IEEE Transactions on Instrumentation and Measurement 60 (10): 3480–92. https://doi.org/10.1109/
TIM.2009.2036347.
Laukka, Arto, Juhamatti Saari, Jari Ruuska, Esko Juuso, and Sulo Lahdelma. 2016. “Condition-Based
Monitoring for Underground Mobile Machines.” International Journal of Industrial and Systems
Engineering 23 (1): 74–89.
Li, Hongfei, Dhaivat Parikh, Qing He, Buyue Qian, Zhiguo Li, Dongping Fang, and Arun Hampapur.
2014. “Improving Rail Network Velocity: A Machine Learning Approach to Predictive Maintenance.”
Transportation Research Part C: Emerging Technologies 45: 17–26. https://doi.org/10.1016/j.
trc.2014.04.013.
Mkemai, Rashid. 2011. “Maintenance Procedures and Practices for Underground Mobile Mining
Equipment,” 1–76.
Peng, Ying, Ming Dong, and Ming Jian Zuo. 2010. “Current Status of Machine Prognostics in
Condition-Based Maintenance: A Review.” International Journal of Advanced Manufacturing
Technology 50 (1–4): 297–313. https://doi.org/10.1007/s00170-009-2482-0.
Schwabacher, Mark, and Kai Goebel. 2007. “A Survey of Artificial Intelligence for Prognostics.” AAAI
Fall Symposium, 107–14. http://www.aaai.org/Papers/Symposia/Fall/2007/FS-07-02/FS07-02-016.pdf.