TY - GEN
T1 - Supporting maintenance decisions with expert and event data
AU - Kunttu, Susanna
AU - Kortelainen, Helena
N1 - Project code: G2SU00152
PY - 2004
Y1 - 2004
N2 - A successful maintenance program incorporates planning and follow-up processes, including systematic feedback and data collection systems and routines. In the process industry, maintenance data collection systems do not typically contain the methods and analysis tools needed to support the continuous update of the maintenance program. Implementation of the necessary procedures and tools can be successful only if the data collection and updating is simple and automated, and does not appreciably increase the workload of the people responsible for maintenance development. The aim of our study is to find methods for predicting the number of failures and the time to the next failure using expert data, which is updated with the collected event data. In this study, three methods for predicting the number of failures were compared. The event and expert data was collected from a Finnish board mill. Tested predicted methods included the moving average, and models for the Poisson process and power law process. With our data set, moving average delivered as good estimates as the more sophisticated ones. One of the four test cases showed especially large variations in the recorded yearly failure rate - and none of the testing predicting methods delivered reliable estimates in this case. Because maintenance actions are carried out also during other stoppages, the event data proved to be insufficient for time to failure predictions. The results proved that a continuously improving maintenance program should be based, not only on the event data, but also on all other relevant information. This means than data from different sources need to be combined and the quality of the recorded data must be high.
AB - A successful maintenance program incorporates planning and follow-up processes, including systematic feedback and data collection systems and routines. In the process industry, maintenance data collection systems do not typically contain the methods and analysis tools needed to support the continuous update of the maintenance program. Implementation of the necessary procedures and tools can be successful only if the data collection and updating is simple and automated, and does not appreciably increase the workload of the people responsible for maintenance development. The aim of our study is to find methods for predicting the number of failures and the time to the next failure using expert data, which is updated with the collected event data. In this study, three methods for predicting the number of failures were compared. The event and expert data was collected from a Finnish board mill. Tested predicted methods included the moving average, and models for the Poisson process and power law process. With our data set, moving average delivered as good estimates as the more sophisticated ones. One of the four test cases showed especially large variations in the recorded yearly failure rate - and none of the testing predicting methods delivered reliable estimates in this case. Because maintenance actions are carried out also during other stoppages, the event data proved to be insufficient for time to failure predictions. The results proved that a continuously improving maintenance program should be based, not only on the event data, but also on all other relevant information. This means than data from different sources need to be combined and the quality of the recorded data must be high.
KW - maintenance
KW - process models
KW - prediction
U2 - 10.1109/RAMS.2004.1285511
DO - 10.1109/RAMS.2004.1285511
M3 - Conference article in proceedings
SN - 0-7803-8216-1
T3 - Proceedings of the Annual Reliability and Maintainability Symposium
SP - 593
EP - 599
BT - Proceedings of the Annual Reliability and Maintainability Symposium 2004
PB - IEEE Institute of Electrical and Electronic Engineers
T2 - Annual Reliability and Maintainability Symposium 2004
Y2 - 26 January 2004 through 29 January 2004
ER -