ENERGY Show me the money: making your SCADA data work for you SCADA-Based Condition Monitoring Philippe Brodeur, Eng., M. Eng. 02 February 2017 2017 CanWEA O&M Summit 1 DNV GL 2017 02 February 2017 SAFER, SMARTER, GREENER
Presentation outline Pro-active maintenance Case study: potential O&M gains $$$ SCADA-based condition monitoring (SCM) vs Traditional CMS Background on SCM: DNV GL s development over the years SCM How it works General SCM methodology Examples 2
Pro-active maintenance Why should you do it Historically, major efforts have been put on pre-construction phase of wind projects, but as the industry gets more mature, companies realize the improvements that can be made during the O&M period. Failure of major components Inevitable Expensive repairs: parts, labor, crane costs Associated downtime may be significant! Implementation of pro-active maintenance strategy may allow to: Minimize downtime Group repairs to reduce crane mob-demob costs Reduce incident s part costs Extend life of components 3
Show Me the Money: Potential O&M Cost Savings $$$ Project with 50 x 2 MW turbines Hypothesis: $60/MWh ; 75% of failures identified with SCM 20 year failure estimates based on DNV GL cost and failure model Typical average return to service from DNV GL benchmarking data Event Gearbox full replacement Failure Count Normal Return to Service (days) 27 46 Potential Lost Revenue Savings [k$] 1,034 Potential Replacement Cost Savings [k$] 4,435 Generator replacement 11 31 281 851 Main bearing replacement 20 20 318 Total: 1,633 Big total: 1,413 6,699 8.3M$ 4
Show Me the Money: Potential O&M Cost Savings $$$ Project with 50 x 2 MW turbines Hypothesis: $60/MWh ; 75% of failures identified with SCM 20 year failure estimates based on DNV GL cost and failure model Typical average return to service from DNV GL benchmarking data Event Gearbox full replacement Generator replacement Main bearing replacement Failure Count Normal Return Potential Lost to Service Revenue Savings (days) SCM costs [k$] for 27 46 11 31 20 20 20 years: < 300 k$ 1,034 281 318 Total: 1,633 Big total: Potential Replacement Cost Savings [k$] 4,435 851 1,413 6,699 8.3M$ 5
Traditional CMS vs SCADA-based condition monitoring (SCM) Both aim at detecting deteriorating component health and identify the increased risks of failures Condition monitoring system (CMS) Temperature sensors and vibration sensors, high frequency signals Typically an add-on to your turbine 6
Traditional CMS vs SCADA-based condition monitoring Both aim at detecting deteriorating component health and identify the increased risks of failures Condition monitoring system (CMS) Temperature sensors and vibration sensors, high frequency signals Typically an add-on to your turbine, not installed by default SCADA-based condition monitoring (SCM) Uses 10-minute SCADA data already collected at the turbines Unlike traditional condition monitoring, SCM does not require expensive retrofits of additional sensors or equipment You can decide at any given time in the project s life to apply SCM analysis, as long as you have the SCADA data 7
DNV GL s development of SCM over the years 2011: Start of development, evaluation of different approaches 2012: Validation study, Physical Model approach Blind test 2013 to 2015: General use of the SCM through projects, building experience and fine tuning the model 2016: Development through a GIP (Global Innovation Project) Now: Online SCM available! 2017: Currently 14 sites with ongoing Online SCM Just starting: new blind test to update the one done in 2012 8
DNV GL s SCM blind test results Site Location Operational Data Set Years Predicted failures A Italy 4.8 7 B Ireland 6.0 7 C Ireland 6.5 1 D UK 7.0 5 E UK 2.5 7 Actual Failures True Detections False Detections Score True / False 8 7 0 88% / 0% 8 6 1 75% / 13% 4 1 0 25% / 0% 6 5 0 83% / 0% 10 5 2 50% / 20% 9
SCM overview: how it works Different SCM methods exist DNV GL s SCM applies a physics-of-failure methodology to monitor component health using the known relationships between data channels 10
General SCM Methodology 11 Data collection Model training Periodic comparison of new data to the model Interpretation More than just an algorithm! Reporting, typically followed by turbine inspection
Example Result 1: Predict Gearbox Failure Both charts show different signals on the same turbine: Modelled Temperature Model Inputs Failed Component Advance notice Gearbox Rotor Side Bearing Generator Speed Power Nacelle Temperature Gearbox 9 months T ACTUAL T MODELLED Modelled Temperature Model Inputs Failed Component Advance notice Gearbox Generator Side Bearing Generator Speed Power Nacelle Temperature Gearbox 7 months T ACTUAL T MODELLED 12
Example Result 2: Predict Main Bearing Failure Modelled Temperature Model Inputs Failed Component Advance notice Main Bearing Rotor Speed Power Nacelle Temperature Main Bearing 4 or 7 months T ACTUAL T MODELLED 13
Q&A Acknowledgements Sally Starnes Thomas vandelft Michael Wilkinson Philippe Brodeur Philippe.brodeur@dnvgl.com +1 (514) 272-2175 ext. 248 www.dnvgl.com SAFER, SMARTER, GREENER 14