Assessing the performance of industrial assets usually requires exploring and combining sensor data, event logs, asset characteristics and domain expert knowledge. Therefore, this process is time and resource consuming. Extrapolating the performances solely from the event logs could lead to more optimal/pro-active planning of maintenance activities. In [1], we have shown that event logs could be numerically encoded into event profiles accurately representing asset event behavior. Therefore, it is possible to extract the event profile of a new operational cycle and link it with the similar event profiles of past operational cycles for which the performance is known. It offers a gain of time and resources when exploring the performance of new operational cycles. We propose a methodology to label asset performances solely based on the event logs, using a standard (numerical) classifiers. The performance of a new asset operational cycle can then be assessed with negligible computational time. The methodology is validated on real-life data from a photovoltaic plant.
[1]
Magnus Sahlgren,et al.
An Introduction to Random Indexing
,
2005
.
[2]
Shanlin Yang,et al.
A new electricity price prediction strategy using mutual information-based SVM-RFE classification
,
2017
.
[3]
Alberto Sillitti,et al.
Failure prediction based on log files using Random Indexing and Support Vector Machines
,
2013,
J. Syst. Softw..
[4]
Elena Tsiporkova,et al.
Annotating the Performance of Industrial Assets via Relevancy Estimation of Event Logs
,
2018,
2018 17th IEEE International Conference on Machine Learning and Applications (ICMLA).
[5]
Xiangliang Zhang,et al.
An up-to-date comparison of state-of-the-art classification algorithms
,
2017,
Expert Syst. Appl..
[6]
Mukesh A. Zaveri,et al.
AUTOMATIC TEXT CLASSIFICATION: A TECHNICAL REVIEW
,
2011
.