RISK-BASED MAINTENANCE OPTIMIZATION OF OFFSHORE WIND SUBSTRUCTURES

Transcription

1 RISK-BASED MAINTENANCE OPTIMIZATION OF Source: Source: Source: Prof. Philippe Rigo ANAST University of Liège

2 Introduction Pablo is a Maritime Engineer specialized in Offshore Renewable Energy and Advanced Design of Offshore Structures Now?... PhD in Risk-Based Maintenance of Offshore Wind Substructures

3 Context: Offshore Wind Far away from shore Complex O&M tasks Reduce LCOE Information available.inspections Monitoring Source: EYReport. Offshore wind in Europe Walking the tightrope to success Source: CrownState Report. A Guide to UK Offshore Wind Operations and Maintenance Foundation Maintenance Source: strain-gauges-as-installed-at-a-belwind-and-b- Northwind Source:

4 Aim: Decision Support Taking the right decision under uncertainty Repair?

5 Uncertainties Modeling Uncertainties?... Physical Model Statistical Measurement Source: Structural Reliability: Bayesian Inference KNOWLEDGE KNOWLEDGE KNOWLEDGE Design Production Operation Decommissioning

6 Deterioration Model - Fatigue Why fatigue? Combined action of wind & waves ~10 8 cycles/lifetime Fracture Mechanics Calibration Fracture Mechanics Similar Reliability SN Curves

7 Probability of ure Updating Reliability - Inspections Optimization: RISK = Probability * Consequence Inspection technique: Ultrasonic Eddy current Visual Inspection result: No detected Detected & repair Detected & no-repair

8 Utilizing Monitoring Data Optimization: RISK = Probability * Consequence Dynamic Bayesian Network (DBN) Monitoring M0 M0 M0 Stress Range S1 S2 Sn Crack size A0 A1 A2 An Inspection I1 I2 In

9 Cost Optimization Optimization: RISK = Probability * Consequence Expected Costs ( ) Inspection cost Repair cost Maintenance Efforts Failure Event Failure cost Optimal Maintenance efforts

10 Decision Problem (II) Pre-posterior Decision Analysis inspection no inspection detection no detection repair no repair repair no repair repair =3.8e21 branches 1 second per branch = 1.24e14 years inspection no inspection detection no detection repair no repair repair no repair no repair

11 Simplification to Decision Problem Heuristic Rule: Constant intervals of time Year 0 Year 20 Every 6 years Y6 Y12 Y18 Every 5 years Y5 Y10 Y15 Y20 More simplifications Perfect inspections Repair if detected

12 Dynamic Approach for Maintenance Planning MARKOV Models Healthy No repair No repair State 0 State 1 State 2 State n Repair Repair Failure Inspection/Monitoring Improves the belief state Partially Observable Markov Decision Processes (POMDP) Point-based algorithms Reduces CPU time significantly 60-states POMDP including 3 combined actions Only 0.32 seconds of CPU time

13 System Effects for Planning Where to perform inspections? Past Components analyzed separately Future Considering Dependencies Shared epistemic uncertainty Similar manufacturing Similar loading Lower costs can be attained Source: Source:

14 Impact: Life extension Fixed-Offshore Wind Farms Year 0 Year 20 Lifetime Reassessment Life extension Utilize gathered data Inspections / SCADA Source:

15 Impact: Desing Optimization Floating Offshore Wind Farms Year 0 Year 20 Probabilistic Design Reduce Safety Design Factor Utilize information Optimize resources Source: 80%99s-first-wind-farm-to-feature-floating-wind-turbines&Itemid=257

16 Conclusion Decision support under uncertainty Utilizing available DATA Engineering: Optimization of the resources! *Contact me for more info