Change-point detection in wind turbine SCADA data for robust condition monitoring with normal behaviour models

Abstract

Abstract. Analysis of data from wind turbine supervisory control and data acquisition (SCADA) systems has attracted considerable research interest in recent years. Its predominant application is to monitor turbine condition without the need for additional sensing equipment. Most approaches apply semi-supervised anomaly detection methods, also called normal behaviour models, that require clean training data sets to establish healthy component baseline models. In practice, however, the presence of change points induced by malfunctions or maintenance actions poses a major challenge. Even though this problem is well described in literature, this contribution is the first to systematically evaluate and address the issue. A total of 600 signals from 33 turbines are analysed over an operational period of more than 2 years. During this time one-third of the signals were affected by change points, which highlights the necessity of an automated detection method. Kernel-based change-point detection methods have shown promising results in similar settings. We, therefore, introduce an appropriate SCADA data preprocessing procedure to ensure their feasibility and conduct comprehensive comparisons across several hyperparameter choices. The results show that the combination of Laplace kernels with a newly introduced bandwidth and regularisation-penalty selection heuristic robustly outperforms existing methods. More than 90 % of the signals were classified correctly regarding the presence or absence of change points, resulting in an F1 score of 0.86. For an automated change-point-free sequence selection, the most severe 60 % of all change points (CPs) could be automatically removed with a precision of more than 0.96 and therefore without any significant loss of training data. These results indicate that the algorithm can be a meaningful step towards automated SCADA data preprocessing, which is key for data-driven methods to reach their full potential. The algorithm is open source and its implementation in Python is publicly available.

A Selective Review on Information Criteria in Multiple Change Point Detection

“…In addition, it offers valuable insights for routine inspection and maintenance planning. The SCADA data of wind turbines is taken from [71]. It consists of 11 proposed SCADA signals of a wind turbine.…”

“…The sample rate of the original data is in a typical 10 min resolution, and then the signals are averaged each day after proposing. Since this dataset has been discussed by [71] in detail, we only present a brief investigation of it.…”

Graph machine learning for predicting wake interaction losses based on SCADA data

“…In step 1, all shutdown and curtailment periods, which were flagged by the project partner, were taken out. Furthermore, power values less than 10 % of rated power, as suggested by [19], as well as values falling outside the cut-in and cut-out wind speed, measured on the nacelle, range were removed. Based on IEC 61400-12 [20] wind speeds larger than 1.5 times the wind speed at 85 % of rated power were filtered out.…”

Enabling Co-Innovation for A Successful Digital Transformation in Wind Energy Using the WeDoWind Ecosystem and A Fault Detection Case Study

“…Then, a kernel change-point detection algorithm is applied in order to screen the prepared signals and flag changes induced by irregular variations of the underlying physical system. The methodology is described in greater detail in [60]. Note that the method works offline and is therefore not suitable to predict failures, but to detect them in existing training data sets.…”

“…Application of the algorithm to the data from WT07 with the settings suggested in [60] resulted in the detection of two CPs (compare Figure 9). The first one coincides with the reported damage of generator bearings and is detected in the temperature measurement of generator bearing 2, therefore providing additional information as to which of the two bearings was presumably affected.…”