A carregar...
Publicação
Influence of extreme weather conditions on wind turbine failures for enhanced energy production
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.07 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Esta dissertação analisa a relação entre as condições meteorológicas e as falhas em turbinas eólicas, com foco nas substituições de multiplicadores, através do alinhamento de registos de manutenção com dados meteorológicos horários do parque eólico de Lousã II (Portugal) entre 2019 e 2024. O conjunto de dados integrou ordens de trabalho, indicadores mensais de desempenho obtidos a partir de SCADA (velocidade do vento e potência) e variáveis do Meteostat (temperatura, humidade relativa, velocidade do vento e rajadas). Foram identificados oito eventos de substituição de multiplicadores, comparados com períodos de controlo sazonais equivalentes.
A metodologia incluiu a rotulagem supervisionada de variáveis meteorológicas em classes interpretáveis, a aplicação do algoritmo k-means para identificar regimes de funcionamento, e duas abordagens de aprendizagem automática: uma Árvore de Decisão, que gera regras condicionais do tipo se–então, e a Mineração de Regras de Associação (Apriori), que deteta combinações frequentes de condições meteorológicas antes das falhas. Em complemento, foi calculado um indicador simples de perda de energia, comparando a produção real com a teórica, e criado um OilScore para sintetizar resultados laboratoriais sobre a condição dos lubrificantes.
Os resultados mostraram que a maioria das falhas ocorreu em condições meteorológicas comuns, sobretudo em ventos calmos a moderados (velocidade < 15 m/s) e rajadas moderadas (≈15–25 m/s). Rajadas fortes ou extremas (> 35 m/s) estiveram associadas a poucas falhas, em linha com paragens automáticas de segurança. Humidade relativa elevada (≥80%) surgiu como contexto recorrente, em especial nos sistemas de passo e guinada. No paruqe éolico, verificou-se uma associação positiva entre perdas mensais de energia e temperatura, sendo que os meses mais quentes apresentaram maiores desvios face à curva de referência. Nas turbinas com substituições de multiplicadores, picos de perda de energia precederam várias intervenções, sugerindo utilidade para aviso prévio.
A hipótese de que as falhas aumentam em condições extremas não foi confirmada. Os riscos estão antes ligados à exposição prolongada a regimes húmidos e ventos moderados, o que reforça a importância de integrar métricas de perda de energia, dados meteorológicos e condição do óleo em estratégias de manutenção preditiva.
Palavras Chave: turbinas eólicas; caixa multiplicadora; falhas; manutenção
This dissertation investigates the relationship between weather and wind turbine failures, with emphasis on gearbox replacements, using maintenance records aligned with hourly meteorological data from the Lousã II wind farm (Portugal) for 2019–2024. The dataset integrates event-based work orders, monthly performance indicators derived from SCADA wind speed and power, and Meteostat weather variables (temperature, relative humidity, wind speed, and gusts). Eight gearbox replacement events were analysed alongside season-matched control periods. Methods included supervised labelling of weather into bins, k-means clustering to reveal operating regimes, and two machine-learning approaches: a Decision Tree to derive if–then rules, and Association Rule Mining (Apriori) to detect frequent co-occurring weather conditions preceding failures. In parallel, a simple energy-loss metric compared actual to theoretical production, and an OilScore summarised lubricant health. Failures across components occurred mainly under ordinary weather conditions rather than extremes. Counts were highest in calm-to-low wind bins and moderate gusts (≈15–25 m/s), while very few events occurred during strong or extreme gusts—consistent with protective shutdowns. Higher relative humidity (≥80%) was a recurring backdrop, particularly for pitch and yaw systems. Fleet-level analysis revealed a positive link between monthly energy loss and temperature, with warmer months showing greater deviations from the reference curve. For turbines with gearbox replacements, pronounced energy loss spikes often preceded interventions, suggesting operational value for early-warning. In the gearbox-focused analyses, neither the Decision Tree nor the association rules revealed a single dominant trigger. Patterns involving low gusts and high humidity appeared before some events but should be regarded as exploratory due to limited data. The evidence does not support the hypothesis that failures increase during extreme weather. Instead, risks align with prolonged exposure to humid, moderate-wind regimes. The study highlights opportunities for predictive maintenance by combining labelled weather, energy loss metrics, and oil condition data, and points to future work using exposure-based risk models, high-frequency SCADA features, and cross-site validation.
This dissertation investigates the relationship between weather and wind turbine failures, with emphasis on gearbox replacements, using maintenance records aligned with hourly meteorological data from the Lousã II wind farm (Portugal) for 2019–2024. The dataset integrates event-based work orders, monthly performance indicators derived from SCADA wind speed and power, and Meteostat weather variables (temperature, relative humidity, wind speed, and gusts). Eight gearbox replacement events were analysed alongside season-matched control periods. Methods included supervised labelling of weather into bins, k-means clustering to reveal operating regimes, and two machine-learning approaches: a Decision Tree to derive if–then rules, and Association Rule Mining (Apriori) to detect frequent co-occurring weather conditions preceding failures. In parallel, a simple energy-loss metric compared actual to theoretical production, and an OilScore summarised lubricant health. Failures across components occurred mainly under ordinary weather conditions rather than extremes. Counts were highest in calm-to-low wind bins and moderate gusts (≈15–25 m/s), while very few events occurred during strong or extreme gusts—consistent with protective shutdowns. Higher relative humidity (≥80%) was a recurring backdrop, particularly for pitch and yaw systems. Fleet-level analysis revealed a positive link between monthly energy loss and temperature, with warmer months showing greater deviations from the reference curve. For turbines with gearbox replacements, pronounced energy loss spikes often preceded interventions, suggesting operational value for early-warning. In the gearbox-focused analyses, neither the Decision Tree nor the association rules revealed a single dominant trigger. Patterns involving low gusts and high humidity appeared before some events but should be regarded as exploratory due to limited data. The evidence does not support the hypothesis that failures increase during extreme weather. Instead, risks align with prolonged exposure to humid, moderate-wind regimes. The study highlights opportunities for predictive maintenance by combining labelled weather, energy loss metrics, and oil condition data, and points to future work using exposure-based risk models, high-frequency SCADA features, and cross-site validation.
Descrição
Palavras-chave
Turbinas eólicas Caixa multiplicadora Falhas Manutenção