Ensuring the reliability of wind power, a strategic renewable source, is a fundamental challenge given the operational repercussions of climate change. In this regard, this study investigates the impact of high ambient temperatures on wind turbine reliability, with specific attention to the occurrence of a heat-induced fault, and proposes a mitigation strategy to reduce related energy losses. A multi-phase methodology was developed, involving the identification of recurring patterns preceding failures through time series clustering of ambient temperature and active power which correspond to different operating conditions. Subsequently, the patterns identified by the clusters are systematically mapped onto the entire historical dataset of the plant by time series classification. Using a similarity-based approach, each temporal window extracted from the historical dataset is assigned to a cluster. Two clustering methods were compared: Hierarchical Clustering (using Dynamic Time Warping, DTW, distance) and HDBSCAN (based on extracted features and UMAP dimensionality reduction). The findings revealed a progressive increase in critical operating hours, correlating with recent climatic trends and decreasing system reliability. To estimate the fault rate, a Bayesian Gamma-Poisson model was implemented, which provided robust predictions, quantified uncertainty, and confirmed the rising trend of faults. Then based on the identified clusters, a derating strategy was developed to proactively reduce power output. The study was conducted on approximately 12 years of data recorded by SCADA system from a wind farm composed of 37-turbine Vestas V90 and located in Sardinia. It demonstrates that rising global temperatures directly impact wind turbine reliability. The proposed methodology, which combines advanced data analytics with Bayesian modeling, provides a robust framework for reliability assessment and fault prediction. Simulations of the derating strategy demonstrated its notable effectiveness, achieving a 51\% reduction in energy losses attributed to the fault during the 2022--2024 period. The results support the adoption of optimized derating strategies as a preventive measure in operations and maintenance (O\&M), thereby enhancing wind farm resilience and ensuring more stable and reliable energy production in an uncertain climatic future. Additionally, a post-commissioning verification confirmed the validity of the original design parameters, though with an indication of increasing stress attributed to evolving environmental conditions.
Garantire l’affidabilità dell’energia eolica, una fonte rinnovabile strategica, rappresenta una sfida fondamentale alla luce delle ripercussioni operative dei cambiamenti climatici. In questo contesto, il presente studio indaga l’impatto delle alte temperature ambientali sull’affidabilità delle turbine eoliche, con particolare attenzione al verificarsi di un guasto indotto dal calore, e propone una strategia di mitigazione volta a ridurre le perdite di energia associate. In questo contesto è stata sviluppata una metodologia in più fasi che prevede l’identificazione di pattern ricorrenti che precedono i guasti mediante il clustering di serie temporali di temperatura ambientale e potenza attiva, rappresentativi di differenti condizioni operative. Successivamente, i pattern identificati dai cluster sono stati sistematicamente mappati sull’intero dataset storico dell’impianto attraverso tecniche di classificazione di serie temporali. Con un approccio basato sulla similarità, ogni finestra temporale estratta dal dataset è stata assegnata a un cluster. Sono stati confrontati due metodi di clustering: il Clustering Gerarchico (basato sulla distanza Dynamic Time Warping, DTW) e l’HDBSCAN (basato su estrazione di feature e riduzione dimensionale tramite UMAP). I risultati hanno evidenziato un incremento progressivo delle ore operative critiche, in correlazione con le recenti tendenze climatiche e con una diminuzione dell’affidabilità complessiva del sistema. Per stimare il tasso di guasto è stato implementato un modello Bayesiano Gamma–Poisson, che ha fornito previsioni robuste, quantificazione dell’incertezza e conferma del trend crescente dei guasti. Sulla base dei cluster individuati, è stata quindi sviluppata una strategia di derating, volta a ridurre proattivamente la potenza erogata. Lo studio, condotto su circa 12 anni di dati registrati dal sistema SCADA di un parco eolico composto da 37 turbine Vestas V90 situato in Sardegna dimostra che la metodologia proposta, fornisce un quadro solido per la valutazione dell’affidabilità e la previsione dei guasti. Allo stesso tempo le simulazioni della strategia di derating hanno mostrato una notevole efficacia, ottenendo una riduzione del 51\% delle perdite di energia attribuibili al guasto nel periodo 2022–2024 risultando un'ottima misura preventiva nelle attività di O\&M per garantire una produzione energetica più stabile e affidabile in un futuro climatico incerto. Inoltre, una verifica post-commissioning ha confermato la validità dei parametri di progetto originari, pur evidenziando un incremento delle sollecitazioni dovuto alle mutate condizioni ambientali.
A methodology for the reliability assessment of wind turbines under climate change
Fares, Rita Pia
2024/2025
Abstract
Ensuring the reliability of wind power, a strategic renewable source, is a fundamental challenge given the operational repercussions of climate change. In this regard, this study investigates the impact of high ambient temperatures on wind turbine reliability, with specific attention to the occurrence of a heat-induced fault, and proposes a mitigation strategy to reduce related energy losses. A multi-phase methodology was developed, involving the identification of recurring patterns preceding failures through time series clustering of ambient temperature and active power which correspond to different operating conditions. Subsequently, the patterns identified by the clusters are systematically mapped onto the entire historical dataset of the plant by time series classification. Using a similarity-based approach, each temporal window extracted from the historical dataset is assigned to a cluster. Two clustering methods were compared: Hierarchical Clustering (using Dynamic Time Warping, DTW, distance) and HDBSCAN (based on extracted features and UMAP dimensionality reduction). The findings revealed a progressive increase in critical operating hours, correlating with recent climatic trends and decreasing system reliability. To estimate the fault rate, a Bayesian Gamma-Poisson model was implemented, which provided robust predictions, quantified uncertainty, and confirmed the rising trend of faults. Then based on the identified clusters, a derating strategy was developed to proactively reduce power output. The study was conducted on approximately 12 years of data recorded by SCADA system from a wind farm composed of 37-turbine Vestas V90 and located in Sardinia. It demonstrates that rising global temperatures directly impact wind turbine reliability. The proposed methodology, which combines advanced data analytics with Bayesian modeling, provides a robust framework for reliability assessment and fault prediction. Simulations of the derating strategy demonstrated its notable effectiveness, achieving a 51\% reduction in energy losses attributed to the fault during the 2022--2024 period. The results support the adoption of optimized derating strategies as a preventive measure in operations and maintenance (O\&M), thereby enhancing wind farm resilience and ensuring more stable and reliable energy production in an uncertain climatic future. Additionally, a post-commissioning verification confirmed the validity of the original design parameters, though with an indication of increasing stress attributed to evolving environmental conditions.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_10_Fares_01.pdf
non accessibile
Descrizione: Thesis PDF
Dimensione
7.24 MB
Formato
Adobe PDF
|
7.24 MB | Adobe PDF | Visualizza/Apri |
|
2025_10_Fares_02.pdf
non accessibile
Descrizione: Executive summary PDF
Dimensione
762.58 kB
Formato
Adobe PDF
|
762.58 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/243596