Presented by: 
Dr. Stella Kapodistria (TU/e)
Tue 9 May, 12:00 pm - 1:00 pm
67-442 (Priestly Building)

We are interested in the performance of a wind turbine for maintenance and power pricing management purposes.  For maintenance purposes, we will show how to explore the Supervisory Control and Data Acquisition (SCADA) data, available with every wind turbine, in order to build condition based maintenance (CBM) approaches for the main components of the system. More concretely, in this talk, we explore concepts from statistics and connect them to stochastic processes; We use statistical concepts stemming from Statistical Process Control (SPC) and we connect them to CBM and first passage times. To this purpose, we use as a paradigm, mainly for illustration and simplicity purposes, the connection between the Shewhart control chart with the On-Off stochastic process. Such an approach will turn out to produce an astonishingly large number of warning signals that constitute the opportunities of preventive maintenance. We investigate two approaches one based on a Bayesian extension of the Bellman equations and a greedy decision making approach based on machine learning that can be used so as to identify the moment in which it is cost-optimal to perform maintenance.

For the power pricing management purposes, it is necessary to develop models that accurately forecast the power output of a wind turbine. As a first step and following the guidelines of the existing literature we used the SCADA data to model the wind turbine power curve (WTPC). We explored various parametric and non-parametric modelling techniques for the modelling of the WTPC. All of these techniques seem to have an intrinsic limitation in terms of accuracy, making the corresponding models inappropriate for short-term forecasting. To avoid this conundrum, we show that adding a properly scaled autoregressive moving-average (ARMA) modelling layer increases short term prediction performance while keeping the long term prediction capabilities of WTPC models.

The first part on the maintenance is joint on going work with Alessandro Di Bucchianico (TU/e), Paulo Serra (TU/e), and Bert Zwart (CWI), the second part on the power pricing management is joint on going work with Sandor Kolumban (TU/e) and Nazanin Nooraee (TU/e).


Stella Kapodistria is an assistant professor in the section Stochastics of the Department of Mathematics and Computer Science at Eindhoven University of Technology (TU/e), since 2014. She received her PhD from the University of Athens (cum laude) in 2009. Stella was lecturer at the University of the Aegean (2009-2011), postdoc at TU/e (2011-2013), and visiting assistant professor at Groningen University (2013-2014). Her research interests are in data-driven stochastic modelling, performance evaluation, queueing, and stochastic optimization. She is a member of the NETWORKS Zwaartekracht project, and the Data Science Flagship between Philips and TU/e. Since 2013, she has been involved in the development of prognostic and diagnostic algorithms for the DAISY and the DAISY4OFFSHORE projects on wind turbines. She serves on the editorial board of the Probability in the Engineering and Informational Sciences international journal and is a guest editor of the Annals of Operation Research international journal. Stella is the recipient of a TKI WoZ grant (2014).


Further details about SMOR seminars are here: