Automated mast data monitoring assistance: miss fewer issues, improve consistency while reducing analyst time & fatigue
11 Sep 18
2:30 PM
-
3:45 PM
Tracks:
Data Science and Digitalization
Organizations often maintain archives of mast data that have been ‘cleaned’ during previous assessments. These historical 'clean' data can be leveraged using statistical and machine learning methods to build automated "intelligent" detectors that can help analysts identify anomalies in mast data monitoring campaigns. We present the evidence of applying these methods to over 25 masts with the results showing that not only do these methods help analysts identify obvious issues (e.g. a frozen sensor) but also enable the detection of easily unnoticed issues such as subtle drifts in sensor readings or early onset of icing. We will also benchmark the detector against a typical mast cleaned by multiple analysts. Applying these methods can lead to fewer missed issues, reduce analyst review time, and benefit from a more consistent and repeatable approach to mast data quality assurance. Finally, such methods can also be deployed for real time monitoring of met masts.