Business Intelligence and Data Analytics in Renewable Energy Sector
LRMM_TG5_1
1. Chalmers University of Technology
Maintenanace Management of Wind Turbines
Abstract:
Wind power has been one of the most promising sources of renewable energy in recent times. Unfortunately, reliability of wind turbines
has been lower than expected resulting in larger downtimes and higher maintenance costs. Condition based maintenance presents a
good solution reduce maintenance costs and in long term improve reliability. Various condition monitoring methods like vibration
monitoring, acoustic monitoring, visual inspections, oil analysis etc; have been employed for fault diagnosis and prognosis. However,
these condition monitoring approaches have not been successfully integrated with optimal maintenance decision making.
This poster presents an ANN (Artificial Neural Network) based condition monitoring approach, which uses data from SCADA (supervisory
control and data acquisition) and result for a case study. Furthermore, an outline of a procedure to utilize the condition monitoring
information towards maintenance decision making has been presented.
Department of Energy and Environment
Pramod Bangalore, PhD. student
Division of Electric Power Engineering
Pramod.bangalore@chalmers.se
From Data to Maintenance Decisions
Vibration
Based CMS
Preventive
Maintenance
Forecast of
Operating
Conditions
Intimation for
Inspection
Corrective
Maintenance
Service
Maintenance
Management
Decision
PMSPIC Based Scheduler
Maintenance Activities:
1. Repair
2. Minor Replacement
3. Major Replacement
Assignment of Resources:
1. External Resources
2. Internal Resources
3. Spares
ANN Based
Monitoring
Self Evolving
Approach
Maintenance
Reports
Measurements
Alarms &
Warnings
SCADA
Maintenance
Decision
Support
Issues with Maintenance Management for Wind
1. Low quality of information from maintenance
reports
2. No systematic procedure to relate failure to
SCADA alarms
3. No methodology for proper utilization of
information from CMS towards maintenance
otpimization
4. Mathematical optimization methods used very
seldom for decision support
Advantages of SEMS
1. A systematic procedure to associate each failure
to a SCADA alarm, improving reliability analysis
2. A procedure for utilization of information of CMS
and SCADA for decision support
3. Output from SEMS can be used for financial
planning
4. Opportunistic maintenance decision making
5. Optimal maintenance decision possible with
different objectives depending on requirement
from owner/operator
Jul Aug Sep Oct Nov
0
5
10
15
Gear Bearing Model
MahalanobisDistance
Jul Aug Sep Oct Nov
0
5
10
Date
MahalanobisDistance
Gear Oil Model
Average Mahalanobis Distance
Normal Operation Limit
SCADA Alarms/Warnings
Data Missing
Gearbox breakdown 19. Nov.
Gearbox breakdown 19. Nov.
First alarm - 13. Sept.
First alarm - 05. Okt.
SCADA Based Condition monitoring
Self Evolving Maintenance Scheduler (SEMS) Approach
Non-linear Autoregressive Neural Network
with exogenous input (NARX) used for
modelling SCADA data
Robust filtering algorithm for input data
selection for better cofindence in output
Best structure of ANN decided based on
detailed analysis of different structures
Case Study: Failure detection in gearbox
Statistical threshold calculation based on
model performace used for error detection
Project Supported by: Project Partner:
Features of ANN based Condition Monitoring Approach
Project Sponsored by: