Isaac Segovia Ramírez1, Fausto Pedro García Márquez1 and Alba Muñoz del Río1
1 University of Castilla La Mancha, Avenida Camilo José Cela, 0, 13001, Ciudad Real, Spain
Isaac.Segovia@uclm.es, FaustoPedro.Garcia@uclm.es, Alba.Munoz@uclm.es
Abstract. Wind energy needs advanced analytics to obtain reliable information about the data from wind turbines monitoring and increase the reliability of the maintenance. The analysis of alarm and signal datasets acquired from the wind turbines requires novel algorithms. Artificial intelligence algorithms with complex trainings and high computational costs are traditionally employed. This paper proposes motif analysis using an Internet of the Things platform to study large time series data for wind turbine analysis. The approach employs the periods with more influence in the alarm development and applies motif search. It is presented a real case study analysing interest periods with motif methodology and different algorithms and parameters of the operation, with high accuracy in the results.
Keywords: Wind Turbine, Motif, Internet of the Things, Maintenance Management, SCADA.
1. Introduction
Wind energy has one of the most important growth due to current improvements in wind turbines (WTs) maintenance management. Supervisory control and data acquisition (SCADA) system acquires the data from condition monitoring systems (CMS). The SCADA data is divided into signals and alarms. Alarms are warning messages activated when certain conditions are reached. False alarms are triggered although there are not real failures, producing unnecessary maintenance tasks. New algorithms are required to develop reliable false alarm identification. Motif algorithms are able to find patterns in massive time series data, reducing the computational loads and increasing the reliability of the analysis.
2. Approach
This work presents a novel method based on motif search for alarm analysis in WTs. An IoT platform is employed for time series analysis and pattern identification, allowing the selection of four different algorithms for the motif search: Manhattan, Dynamic Time Warping (DTW), Pearson and Euclidean. It is proposed the analysis of the delay between the detected motif and the real alarm activation and the percentile of the results associated with all other results to validate the patterns.
3. Case Study
The alarm selected for this study is the rotor-generator discrepancy. The new correlated dataset is defined by 3 signals and the base motif definition is obtained with statistical analysis. Only two motifs of the results are considered in this study. The objective is to achieve the maximum percentile value (%PM) with the minimum delay in time scale between the detected motif and the real activation.
3.1. Algorithm influence
The first scenario studies the performance of the algorithms with a motif length of 12 hours. The delay between the motif and the real alarm shows reliable results for all the cases and datasets, except Manhattan and Pearson for motif 2. DTW and Euclidean have the highest accuracy for both datasets. Manhattan algorithm also provides reliable results, although this algorithm presents irregularities.
3.2. Length of the motif
This scenario is designed to test the influence of variations in the length definition. The proposed motif lengths are 3, 12, 24 and 48 hours. High motif lengths reduce the accuracy of the results in DTW, Pearson and Euclidean since increased analysis ranges lead to different issues in the motif identification. The 3 hours motif length achieves consistent results about delay for all the algorithms.
4. Conclusions
This work develops a novel approach with an initial filtering process based on the selection of the critical alarm to be applied in the motif search. It is proposed a real WT case study designed with different scenarios formed by variations in the motif definition. Euclidean and Dynamic Time Warping algorithms achieve accurate results, proving that motif search is a reliable method for wind turbine analysis.
Acknowledgements. The work reported herewith has been financially by the Dirección General de Universidades, Investigación e Innovación of Castilla-La Mancha, under Research Grant ProSeaWind project (Ref.: SBPLY/19/180501/000102).