Download as PDF, PPTX
![Cost of Gearbox Failures
u Romax Study estimated cost of
planetary bearing failures
>350k[1]
u In 2014 Siemens wrote down
€223M to replace bearings in
fleet <2 yrs. old[2].
u Controlling wind turbines with
data-driven software could,
models show, increase energy
production by at least 10% and
gains of 14-16% are possible[3]
u The average gearbox failure rate
over 10 years is estimated at
5%[4].
1 0.75 0.5 0.25 0 2 4 6 8
Electrical Systems
Electronic Control
Sensors
Hydraulic Systems
Yaw System
Rotor Brake
Mechanical
Brake
Rotor Hub
Gearbox
Generator
Supporting Structure/Housing
Drive Train](https://image.slidesharecdn.com/aweawebinar2016v10-160818162832/75/AWEA-Cognitive-Analytics-for-Predictive-Futures-5-2048.jpg)
Uploaded bySparkCognition
AWEA Cognitive Analytics for Predictive Futures
The document discusses the rising maintenance costs of wind turbines and the financial risks associated with aging equipment, highlighting the importance of predictive analytics. It emphasizes a trial using data-driven software to predict gearbox failures with high accuracy, achieving zero false positives and negatives. The piece also outlines the broader implications of cognitive analytics in improving operational efficiency and reducing costs in energy sector applications.
More Related Content
Similar to AWEA Cognitive Analytics for Predictive Futures
AWEA Cognitive Analytics for Predictive Futures
- 1.
- 2.
- 3.
- 4.
- 5. Cost of Gearbox Failures u Romax Study estimated cost of planetary bearing failures >350k[1] uIn 2014 Siemens wrote down €223M to replace bearings in fleet <2 yrs. old[2]. u Controlling wind turbines with data-driven software could, models show, increase energy production by at least 10% and gains of 14-16% are possible[3] u The average gearbox failure rate over 10 years is estimated at 5%[4]. 1 0.75 0.5 0.25 0 2 4 6 8 Electrical Systems Electronic Control Sensors Hydraulic Systems Yaw System Rotor Brake Mechanical Brake Rotor Hub Gearbox Generator Supporting Structure/Housing Drive Train
- 6.
- 7. About u Develops, Owns, and Operates Power Generation and Energy Storage Units in US and Europe u North America’s largest independent wind power generation company uCurrently operating over 4MW of wind Headquarters Regional Office Wind Project Natural Gas Solar Project Storage
- 8. Gearbox Monitoring Application Trial u Desired Results u Predict gearbox failures with 30- 60 day advanced notice uZero or minimal false positives u “Dummy Light” output u Data Provided u 4 years of historical data from site of ~100 turbines u 27 data variables at 10 minute resolution, no vibration variables collected u Major component failure logs
- 9. Generated Prediction Signatures for all Catastrophic Gearbox Failures Risk Index for Gearbox Health Trial Outcomes • Impending failure (red alert) prediction for catastrophic failure > 1 month • Advanced degradation warning (amber alert) for failures is > 2 months •We had zero false positives, that is no alerts were raised which did not have a failure follow • We had zero false negatives, that is no failures were missed For 100 Turbines 67 352 0 4 0 6 0Days of Warning 50 0 1000 67 Days 35 Days
- 10.
- 11. Some secondary observations u Other failures (likely blade), have very short failure signatures u Failure prediction made in days compared to longer signatures for gearbox failures uSeasonality of Gearbox failures u Catastrophic gearbox failures show high correlation to seasons u Most failures in second half of the year: Q1 – 1, Q2 – 1, Q3 – 9, Q4 – 7 u Other failures relatively independent of seasons u Catastrophic failures distributed more uniformly: Q1 – 3, Q2 – 4, Q3 – 5, Q4 – 5
- 12. Next Steps u Expand to 5 sites u Pending results of expanded sites, committed to enterprise wide roll-out uExplore predictive models for other major components
- 13.
- 14. Category Key Features Business Intelligence (BI) • Centralizedanalysis •Uniform data collection • Average visualizations Rules Based Modeling • Fixed rules must account for all types of transactions in all types of conditions; lead to rule proliferation and management challenges • May be good measures for some simple situations, but average (or even sub-par) measures for others Statistical Analysis • Identifies deviations from “normal” • More a platform for model building and data scientists than an alert generating solution • Not automated to account for changing conditions Physics Based Modeling • Asset-type specific • Model building Is a very hands-on process involving laboratory experiments • Domain experts apply these physical models universally to assets Common Approaches
- 15. Enables machines to penetrate the complexity of data to identify associations Presents powerful techniques to handle unstructured data Continuously learns not only from previous insights, but also for new data entering the system Provides NLP supportto enable human to machine and machine to machine communication Does not require rules, instead relies on hypothesis generation using multiple data sets which might not always appear connected or relevant Benefits of Cognitive Analytics NLP: Natural Language Processing Cognitive Analytics is inspired by the way the human brain operates: Processes Information Draws Conclusions Codifies Instincts & Experience into Learning
- 16.
- 17.
- 18.
- 19.
- 20. Unsupervised vs. Supervised Learning Unsupervised Supervised Index DateTime Asset ID Value 2 5-Apr-10 7:01 750 89 93 22-Mar-13 8:19 904 79 27 20-Oct-14 8:26 545 74 5 10-Jul-12 7:38 552 86 68 15-Sep-11 8:13 942 74 29 1-Jun-11 8:44 900 72 91 20-Jul-11 7:14 587 50 54 12-Jul-10 7:36 765 95 20 5-Sep-14 8:25 813 39 44 30-Jun-11 7:07 983 71 100 5-Oct-12 7:35 802 34 66 12-Mar-10 7:39 726 47 45 6-May-11 7:30 973 98 84 10-Dec-12 7:17 504 68 43 9-Jul-14 8:07 567 74 Action Taken Component Repair Blade Unknown Blade Repair Gearbox Replaced Gearbox Replaced Gearbox NTF Generator Good Generator NTF Blade Repair Generator NTF Gearbox NTF Blade Repair Gearbox Unknown Gearbox Repair Blade Repair Gearbox
- 21.
- 22. Our Algorithms Artemis • Proprietary regularizationtoolfor feature selection • Automated class balancing • Automated model selection • Automated checks on overfitting • Turn-key solutions with health index for industrial use cases Iris • Proprietary clustering algorithm • Optimal clustering of data leading to state generation • Semantic indexing of states • Classification from indexed states • Turn-key solutions with health index for industrial use cases Pythia • Proprietary regularization tool for feature selection • Genetic algorithms for optimizing neural networks
- 23.
- 24.
- 25.
- 27.
- 28.
- 29. Objectives Monitor Critical Assets during startups and coast-downs Predict Remaining Useful Life Analyze failures, alert on impending failures, optimize design Client Asset Big Utility Turbine Generator Big Utility Wind Turbine On-shore driller Electrical Submersible Pump • Data collection from multiple assets • Detects failures, graduating to predictions • Self-learning system with access to in-context advisory powered by IBM Watson • RUL (Remaining Useful Life) prediction and anomaly detection • Automated model building, selection & management • Insights through deeper-order analyses • Failure identification and classification • Automated failure alerting • Critical variable identification • Design and process optimization to reduce specific failures Solution Feature Business Impact • Estimated increase in productivity of 25% –30% • 50X ROI • Estimated savings of ~40% in O&M budgets • ~$2MM per year for 100 MW power generation plant (wind), 40X ROI • 3X increase in life of ESP through proper monitoring and design • Savings of up to $150,000 per asset per year, 50X ROI Other Energy Sector Applications
- 30. Use Case -Improve Safety and Reduce Remediation Cost Through Intelligent Prognostics u SparkCognition has developed an IBM Watson “Advisory” application for Asset Maintenance u SparkCognition’s powered by IBM Watson will allow Directors of Maintenance and technicians to: § Conduct machine to human dialogue to troubleshoot fault codes § Predict impending failures and faults § Identify the right fault codes and troubleshooting tips using natural language queries § Find solutions to problems and advise technicians § Optimize work flow and deliver relevant documentation for a faster turnaround
- 31.
- 32.