A Collaborative Data-driven Methodology for Improving Wind Farm Operations and Maintenance

Description

Majority of worldwide installed wind turbines are aging. As a population of remotely distributed large-size power generation units, their operations and maintenances (O&M) are challenging and costly. Recent advancement of sensing and information technologies (i.e. supervisory control and data acquisition (SCADA) system and unmanned aerial/ground vehicle based imaging system) deployed in wind farms offers an unprecedented opportunity for timely collections of massive volumes of a variety of data. Such large data form a valuable platform for evolving the condition monitoring (CM) and operations & maintenance (O&M) mode from a descriptive and individualbased level to a predictive and population-based scale through employing large data analytics and integrated methods rooted in realism. This project aims at developing a collaborative data-driven methodology for more predictive and dynamic wind farm CM and O&M based on four proposed research tasks:1) develop a novel self-organized deep-learning based modeling method for depicting the heterogeneity of CM signal patterns among wind turbines through learning large volumes of collected SCADA data. Next, a collaborative framework will be developed to derive monitoring criteria with data-driven CM signal models and predict impending failures; 2) develop a two-phase data-driven framework of establishing tailor-made models for individually characterizing wind turbine power generation processes based on the joint learning of their SCADA data and image data. The first phase will develop a computer vision based method to extract data summarizing turbine rotor blade conditions through analyzing their images. Next, a method for modeling their power generation processes by joint learning image analytics data and SCADA data will be explored; 3) develop a comprehensive decision-making model for dynamically determining optimal wind farm O&M schedules based on constraints derived from CM results and developed data-driven power generation models. Solving the proposed decision-making model will be a complex mixed-integer nonlinear programming problem and a novel hierarchical heuristic will be developed; and 4) validate the proposed methods through extensive computational studies and the field testing.