Combining the Power of Data with Physics to Reduce Maintenance Costs and Increase Energy Production
27 Feb 18
11:30 AM
-
12:00 PM
Tracks:
Asset Management, Emerging Technology
In recent years, two separate technological advances have been utilized to improve wind farm operations and maintenance practices. The first is a physics based approach, which uses physics models of turbine system and components to predict the effect of various operational strategies on the performance of the component or system. The second, more recent approach, is the application of machine learning techniques to wind turbine SCADA data. In this approach, various algorithms are used to develop models of expected turbine behavior based solely on the historical data, with no explicit model of the turbine system or sub-component. Both methods have strengths and weaknesses, and thus there have been successful and less successful implementations of each method. Our approach is to combine the two methods, and to do so in a way that captures the strengths of each method. We will show several examples of how we have successfully combined a physics based approach with a machine learning approach, to achieve outcomes not possible using only a single approach. We will demonstrate application of this hybrid approach for anomaly detection of operational characteristics of a wind turbine and for prediction of component failure. We will also discuss best practices for data storage and event labeling, to maximize the value of the wind farm data collected. We will also show the value proposition of the hybrid approach for predictive analytics for wind farm owners and operators.