Show simple item record

dc.rights.licenseRestricted to current Rensselaer faculty, staff and students. Access inquiries may be directed to the Rensselaer Libraries.
dc.contributorBennett, Kristin P.
dc.contributorDrineas, Petros
dc.contributorHendler, James A.
dc.contributor.authorIqbal, Md Ridwan Al
dc.date.accessioned2021-11-03T08:31:51Z
dc.date.available2021-11-03T08:31:51Z
dc.date.created2016-02-09T09:08:05Z
dc.date.issued2015-12
dc.identifier.urihttps://hdl.handle.net/20.500.13015/1586
dc.descriptionDecember 2015
dc.descriptionSchool of Science
dc.description.abstractSupervisory control and data acquisition (SCADA) is a system that automatically collects data from an array of sensors. We propose to use SCADA sensor data from wind turbines and detect potential faults using machine learning techniques. However, fault detection from sensor readings with supervised learning is confronted with several challenges. There is a high amount of variability due to external conditions which reduces learning accuracy. There is also no prior label that identifies which particular turbine has entered a faulty state at a particular time. Another important challenge is the fact that the sensor data is a time series that requires specialized algorithms.
dc.description.abstractExtensive experiments have been performed to compare and contrast all these different techniques. Important features identified by our method have also been analyzed. We provide evidence of multiple root causes for a particular fault and also identify several key sensors that can indicate abnormality. The results support the performance and robustness of our proposed methods and the identified indicators provide a valuable insight for further investigation.
dc.description.abstractWe have proposed two types of normalization to reduce variability. We also propose different constructed features that can utilize the temporal information in the data. We have also proposed a data transformation technique that allows standard classification algorithms to be used on the data with uncertain labels. This novel approach uses a soft decreasing function to assign weights to the instances based on the time interval from future failure. The method then uses classification algorithms such as random forest and SVM, that can train on weighted instances. We have compared this soft label approach with a hard label approach that assigns a specific label to the instances without uncertainty.
dc.description.abstractWind farms are a popular renewable energy source that are costlier to operate than fossil fuels. Automatic fault detection can improve wind turbine fault tolerance and detect faults before they can cause damage and also reduce maintenance cost.
dc.language.isoENG
dc.publisherRensselaer Polytechnic Institute, Troy, NY
dc.relation.ispartofRensselaer Theses and Dissertations Online Collection
dc.subjectComputer science
dc.titleAutomated wind turbine fault detection from SCADA sensor data with machine learning methods
dc.typeElectronic thesis
dc.typeThesis
dc.digitool.pid176919
dc.digitool.pid176920
dc.digitool.pid176921
dc.rights.holderThis electronic version is a licensed copy owned by Rensselaer Polytechnic Institute, Troy, NY. Copyright of original work retained by author.
dc.description.degreeMS
dc.relation.departmentDept. of Computer Science


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record