HUIZHI XIE (JOINTLY WITH HONGFEI LI AND YASUO AMEMIYA)

Transcription

1 A statistical framework for hydrant reliability analysis and prediction 1 HUIZHI XIE (JOINTLY WITH HONGFEI LI AND YASUO AMEMIYA)

2 Outline Introduction Statistical modeling and analysis Summary Future research directions 2

3 Background Based on the 2010 summer intern project in IBM Aims to improve the client s physical asset management using IBM s management software and advanced analytics As part of the overall project, our focus is on the management of hydrant 3

4

5 Goal of our analysis Reduce or control the occurrences of hydrant failure for the entire system Give a good forecasted distribution of time to failure for each hydrant from a future time point based on the information up to that time point Implement an analytics driven maintenance planning based on monitored and predicted reliability 5

6 Work order records Work type Preventive maintenance Failure Repair Data Three types of problem are considered here. 6 WONUM FAILURECODEPROBLEMCODEWORKTYPEQNUM STATUS REPORTDATE LONGITUDLATITUDE 'FH2509' 'HYDRANTS' 'CA' 'PM' '2722' 'COMP' 8/10/ : 'FH2390' 'HYDRANTS' 'CA' 'PM' '2755' 'COMP' 8/4/ : 'FH2210' 'HYDRANTS' 'CA' 'PM' '3364' 'COMP' 7/20/ 'FH2246' 'HYDRANTS' 'CA' 'PM' '3381' 'COMP' 7/22/ : 'FH2608' 'HYDRANTS' 'CA' 'PM' '3418' 'COMP' 8/11/2005 6: 'FH2348' 'HYDRANTS' 'CA' 'PM' '3424' 'COMP' 8/1/ : 'FH2470' 'HYDRANTS' 'CA' 'PM' '3426' 'COMP' 8/2/ :

7 Data(Cont d) Detailed information on hydrant Hydrant manufacturer Install date Spatial location 7 EQNUM ASSETNULOCATIONDESCRIPTIOMANUFACINSTALLDATLONGITUDLATITUDEHYDRANTHYDRANTFLOWRATFLOWUNI LIFECYCLHYDRANT_ELEBANDCOLHYDRANTHYDRANTHYDRANT 3028 H01615 Hydrant, CLO /26/ EDDY Potable 3153 GPM Active st 29-Jan H09250 Other - Hydra /18/ EDDY Potable GPM Active White 4th 5-Apr H09090 Hydrant, CLO EDDY Potable GPM Active Ana-1st A H09086 Hydrant, CLO EDDY Potable GPM Active Ana-1st A H09088 Hydrant, CLO EDDY Potable GPM Active Ana-1st A H09570 Hydrant, CLO /2/ EDDY Potable GPM Active Low L H03655 Hydrant, CLO /2/ EDDY Potable 0 GPM Active Red 1st

8 Factors that can affect hydrant reliability Hydrant age Number of previous repairs Hydrant manufacturer Code of the previous problem Spatial location Other factors possibly related to spatial location 8

9 Pressure zone information The whole region is divided into seven pressure zones which have different water pressure. We use pressure zone to cluster the hydrants into different areas.

10 Introduction to survival analysis A branch of statistics dealing with death in biological organisms and failure in mechanical systems; reliability analysis in engineering; duration analysis in economics or sociology Models time to event data Event: death, infection of some disease, job loss, etc. Time origin: e.g. the entry of patient into a medical study Censoring: event not observed due to some reason 10

11 Introduction to survival analysis(cont d) Survival function 11 Hazard function Cox model

12 Some notes on hydrant reliability analysis Failure or event is defined as one of the three types of problems Hydrant is a repairable system; repeated observations on each hydrant Assume both preventive maintenance and repair restore the hydrants to the new state 12 The time point at which a PM occurs is considered as a censoring point for the failure of hydrant

13 Four different cases 13 PM Censored PM Censored Event Failure Event Failure

14 Competing risk Failure types are due to different physical mechanisms with different behaviors 14 When any failure occurs, the hydrant is repaired to make it new with respect to all the three failure types

15 Frailty model Functional form for hydrant i, period j Correlation between recurrent events on the same subject Explain the variation unexplained by the listed factors 15

16 Our model Assign a random variable for each type of failure and assume they are independent. Analyze the three types of failure separately using the frailty model 16 for hydrant i, period j

17 Prediction 17 At any given time in the future, based on the information up to that point, predict the distribution of time to failure from that point on for each existing hydrant Failure probability for each hydrant as a function of time Expected cumulative number of failures of all the existing hydrants as a function of time

18 Illustration of PM scheduling 18

19 Illustration of PM scheduling(cont'd) Strategy Strategy description Expected number of failures in 90 days 1 Hydrant 1(30) Hydrant 2(60) 2 Hydrant 1(60) Hydrant 2(30) 3 Hydrant 1(30,60) Hydrant 2(NO) 4 Hydrant 1(NO) Hydrant 2(30,60)

20 Summary and Future research direction Developed an approach for reliability or failure monitoring/forecasting, incorporating the time and individual specific information, that can issue information directly useful for planning and managing a large physical system, and that can be implemented to be used over time Interval prediction method development, and extensions to cover more general problems 20

21 21 Thank you!