A DIAGNOSTIC MAINTENANCE SYSTEM FOR COMMERICIAL AND NAVAL VESSELS JANE CULLUM

Transcription

1 A DIAGNOSTIC MAINTENANCE SYSTEM FOR COMMERICIAL AND NAVAL VESSELS JANE CULLUM SUPERVISORS: Associate Professor Jonathan Binns, Professor Kiril Tenekedjiev, Dr. Rouzbeh Abbassi, Dr. Vikram Garaniya, Michael Lonsdale

2 COMMERCIAL AND NAVAL VESSEL MAINTENANCE: State-of-the-art 2 1. Periodic planned maintenance and RCM ar e not optimal but work 2. Limited data and knowledge of how to interpret it 3. No need for innovation? HMAS SIRIUS HMAS CAPTAIN SIRIUS COOK AND GRAVING HMAS DOCK, MELBOURNE NSW, SOUTH 2014 CHINA SEA, Applications?

3 3 CHALLENGES: CONDITION BASED AND PREDICTIVE MAINTENANCE Hardware and Infrastructure Mobile asset, maritime environment Useful data Quantity Interpretation

4 4 CHALLENGES: DATA INTERPRETATION Meaningful interpretation of data Idenitfying maintenance tasks Expert Experience - Manual Reliability Centred Maintenance - Manual Diagnostic System Automatic (can also be part of RCM)

5 HMAS WALLER SYDNEY HARBOUR, NSW 5 GOALS? Improve availability and reduce overall maintenance cost Improve maintenance scheduling speed and consistency

6 6 DIAGNOSTIC MAINTENANCE SCHEDULING Diagnose machine health, risk of failure Schedule maintenance if and when required PREDICTIONS PM Interval PM Interval PM Interval PM Interval PM Interval PM Interval System Interval A System Interval B System Interval C REQUIREMENTS Interval A Interval B Interval C Schedule maintenance only when required

7 DIAGNOSTIC MAINTENANCE SYSTEM FOR A COMMERCIAL OR NAVAL VESSEL COMPONENT 7 1. Risk Assessment - Condition Monitoring and Machine Learning COMPONENT APPLICATION FRAMEWORK 3. Performance Measurement - Availability and Overall Maintenance Cost 2. Maintenance Scheduling - Decision Theory NUMBER 2 GENERAL SERVICE PUMP COMPONENT APPLICATION VALUE = TRANSLATE + SCALE + FORECAST Is it BETTER THAN periodic PM?

8 VALUE IN TRANSLATION FOR COMMERCIAL OR NAVAL VESSEL APPLICATIONS 8 HTA ELWING HTA WAREE 1. Create system at component level 2. Tune and re-use for similar components on same or different vessels eg. Estimate system reduces maintenance cost of pump by 10%below current PM: Per Pump : ~$80 AUD per year Total for HTAs, 6 pumps : ~$500 AUD per year Total RAN Fleet 49 ships, boats, submarines, 10 pumps per vessel: ~$40,600 AUD per year

9 VALUE IN SCALE FOR COMMERCIAL OR NAVAL VESSEL APPLICATIONS 9 Fleet 1. Create systems at component level for Vessel Vessel Vessel high priority components Sub-system 1 Sub-system 2 2. Integrate systems to create higher levels using RCM or alternatives Component 1 Add individual component savings Component 2

10 10 VALUE IN FORECASTING Reliability Reliability of Component vs. Time Time Corrosion Wear Fatigue Each set of data points can be generated using system at time t, where R = 1 F (mode(t)), also recommends an action and therefore maintenance cost 10 of 18

11 COMPLETED WORK TO MARCH DATA COLLECTION Designed ten experiments, procured and installed hardware, completed experimental data collection and processing Designed CM data collection process, procured and installed hardware, completed 65%of data collection Wrote scripts for data processing (experimental and CM) Compiled equipment and maintenance data to date for Number 2 General Service Pump Completed survey of Chief Engineer METHODOLOGY Identified novelty and strengths of methodology using literature review process Developed new decision modelling theory in conjunction with supervisor (focus of second paper) Designed and wrote scripts for methodology WRITTEN COMMUNICATION OF RESEARCH Literature review paper published in Ocean Engineering Journal Internal Serco Hub article on research Completed second paper draft currently under review by supervisor Drafted four chapters of Thesis 12

12 REMAINING WORK 11 DATA COLLECTION [September 2018] Complete remaining 2/ 3 of CM 8 fortnightly sessions - 8 hours total time Process remaining CM data Record recommendations of Engineer and preventative maintenance alongside system recommendations METHODOLOGY Tune model Generate recommendations from CM data using tuned model Graph recommendations from methodology, Engineer and PM schedule, calculate availability and maintenance cost of the three policies WRITTEN COMMUNICATION OF RESEARCH Complete second paper draft and submission Complete results paper draft and submission Complete thesis 13

13 12 COMPONENT APPLICATION: NUMBER 2 GENERAL SERVICE PUMP 1. Risk Assessment - Condition Monitoring and Machine Learning a. Data for Algorithm Training and Condition Monitoring 13 b. Machine Learning Examples 23 c. Applying a Machine Learning Algorithm Maintenance Scheduling - Decision Theory a. Maintenance Actions as Lotteries 25 b. Modelling Lottery Prizes: Multi-attribute Utility 26 c. Making a Decision: Maximum Expected Utility Performance Measurement -Availability and Overall Maintenance Cost Availability and Maintenance Cost, Validation 28

14 13 DATA FOR MACHINE LEARNING AND CONDITION MONITORING Two Purposes: 1. From Experiments on Test Pump- Build model 2. From Condition Monitoring on No. 2 General Service Pump Use model to predict condition of No. 2 General service pump CREATE DATASETS DESCRIBING COMMON CENTRIFUGAL PUMP FAULTS: 1. No fault Run pump under normal operational conditions alongside 2. No fault Run pump under normal operational conditions engines running 3. No fault Run pump under normal operational conditions at sea 4. Worn Impeller - Lathe impeller fluid side and polish 5. Worn bearing Measure pump bearing with many running hours 6. Damaged bearing Grind outer race of new bearing flat and polish 7. Unbalanced shaft/ Static Imbalance Lathe off material from one point of shaft 8. Misaligned shaft/ Offset misalginment Misalign pump- motor coupling 9. Loose packing Loosen casing bolts 10. Poor mounting Loosen mounting bolt on pump foot

15 14 DATA FOR MACHINE LEARNING AND CONDITION MONITORING Two Purposes: THE DATASETS (20 min sessions): SAMPLE RATE: 1. From Experiments on Test Pump- Build model 2. From Condition Monitoring on No. 2 General Service Pump Use model to predict condition of No. 2 General service pump 1. Vibration: Dual channel on pump Every 2 minutes 2. Temperature: Thermal imaging camera Per Minute 3. Pressure: Suction and discharge gauges Per Minute 4. Motor current: Current clamp on cord Per Minute 5. Packing drip rate: Visual inspection Per Minute 6. Shaft rotation: Tacometer Per experiment

16 ELWING BILGE/ FIRE SYSTEM TEMPORARY CONFIGURATION 15 Operating conditions for all pumps: -0.2 bar Suction 2.1 bar Discharge

17 16 TEST PUMP SETUP NUMBER 2 GENERAL SERVICE PUMP

18 TEST RIG SETUP 17

19 18 DATA COLLECTION - EXPERIMENTAL Two Purposes: 1. From Experiments on Test Pump- Build model 2. From Condition Monitoring on No. 2 General Service Pump Use model to predict condition of No. 2 General service pump 1. TEST PUMP/ AIR CONDITIONING PUMP Conduct TEN EXPERIMENTS of 20 min sessions. PhD Objective Build a model which can detect the following: 1. No fault, no engines, ship alongside 2. No fault, engines running, ship alongside 3. No fault, ship at sea 4. Worn Impeller 5. Loose packing 6. Damaged bearing 7. Worn bearing (Air Conditioning Pump) 8. Unbalanced shaft/static Imbalance 9. Misaligned shaft/offset Misalignment 10. Poor Mounting

20 19 1 TEST PUMP Point Measurement Vibration Vibration Vibration Temperature Temperature Temperature Vibration Temperature Temperature Vibration Location Motor, Vertical Motor, Horizontal Drive-end bearing Pump Casing Motor, Drive End Bearing Casing Coupling Pump Drive End Bearing Casing, Horizontal Shaft Pump, Bearing Casing Pump Casing, Horizontal

21 20 AIR CONDITIONING PUMP Point Measurement Vibration Vibration Vibration Temperature Temperature Temperature Vibration Temperature Temperature Vibration Location Motor, Vertical Motor, Horizontal Drive-end bearing Pump Casing Motor, Drive End Bearing Casing Coupling Pump Drive End Bearing Casing, Horizontal Shaft Pump, Bearing Casing Pump Casing, Horizontal

22 21 DATA FOR MACHINE LEARNING AND CONDITION MONITORING Two Purposes: 1. From Experiments on Test Pump- Build model 2. From Condition Monitoring on No. 2 General Service Pump Use model to predict condition of No. 2 General service pump 2. Number 2 General Service Pump CONDITION MONITORING for one 20 min session, repeat fortnightly for 6 months. PhD Objective - Detect the following using CM measurement: 1. No fault, alongside 2. No fault, engines running 3. No fault, at sea 4. Worn Impeller 5. Loose packing 6. Damaged bearing 7. Worn bearing 8. Unbalanced shaft/static Imbalance 9. Misaligned shaft/offset misalignment 10. Loose mounting

23 22 NUMBER 2 GENERAL SERVICE PUMP Point Measurement Vibration Vibration Vibration Temperature Temperature Temperature Vibration Temperature Temperature Vibration Location Motor, Vertical Motor, Horizontal Drive-end bearing Pump Casing Motor, Drive End Bearing Casing Coupling Pump Drive End Bearing Casing, Horizontal Shaft Pump, Bearing Casing Pump Casing, Horizontal

24 = MACHINE LEARNING CLASSIFICATION 23

25 24 APPLYING A MACHINE LEARNING ALGORITHM CM Vector Machine Learning Algorithm Gr oup Probabilities Input: Set of Measurements from Number 2 General Service Pump: Vibration Temperature Pressure Naive Bayes Algorithm Simple modelling approach Good performance on few data and many features Results: Probability that pump is in each group: OK - No fault Impeller wear Damaged PDE bearing

26 25 MAINTENANCE ACTIONS AND HORSE RACING

27 MAXIMUM EXPECTED UTILITY 27

28 28 3. PERFORMANCE MEASUREMENT Availability vs. PM Overall Maintenance Cost vs. PM Validation against expert recommendations

29 HMAS PERTH AUSTRALIAN MARINE COMPLEX COMMON USER FACILITY, WA, Innovation needed in maintenance of commercial and naval vessels Outlined a diagnostic maintenance system application to a shipboard pump Tuning and validation of system is in progress (TBC September 2018) SUMMARY

30 THANKYOU! ACKNOWLEDGEMENTS The candidate acknowledges the support of the ARC Research Training Centre for Naval Design and Manufacturing (RTCNDM) in this investigation Serco Defence Asia-Pacific and the Condition Monitoring division, Fleet Base East. The RTCNDM is a University-Industry partnership established under the Australian Research Council Industry Transformation grant scheme (ARC IC ). The candidate also acknowledges the support of Serco Defence Asia-Pacific and the Condition Monitoring Division, Fleet Base East in providing guidance and resources for this research.

31 WORN IMPELLER 37 of 18

32 UNBALANCED SHAFT/ STATIC IMBALANCE 38 of 18

33 DAMAGED BEARING 39 of 18

34 MISALIGNED SHAFT/ OFFSET MISALIGNMENT 40 of 18

35 VIBRATION DATA QUALITY 2.5 Misaligned Shaft/Offset Misalignment 2 No fault alongside Amplitude mms Hz 75 Hz Expect higher amplitudes at 25, 50 and 75Hz due to misaligned shaft - Mobius Institute Training Manual (2008) Hz Frequency Hz 41 of 18

36 VIBRATION DATA QUALITY No fault alongside Worn Impeller 3 25 Hz Amplitude mms Hz Expect higher amplitudes at 25 and 520Hz due to worn impeller - Mobius Institute Training Manual (2008) Frequency Hz 42 of 18