Smartive trabaja en varios proyectos de alargamiento de vida, entre los que destacan, por ejemplo, “Turbines for Life” de Acciona Energía, WindEx deNaturgy, o alargamiento de vida de Innogy. En todos ellos, la compañía trabaja con datos SCADA, partiendo de datos diezminutales. A partir de los datos SCADA se pueden realizar múltiples análisis aportando información muy útil tanto para el operador cómo para el mantenedor de parques eólicos. El primer análisis, SmartAudit, consiste en una auditoria sobre las variables. Cabe comentar que las variables SCADA son complicadas de tratar, especialmente en parques antiguos donde las comunicaciones pueden ser complicadas. Los datos son filtrados y ajustados, interpolando o extrapolando los huecos de datos. Una vez se disponen de datos completos se analizan alarmas y eventos. Un buen análisis de eventos es fundamental para conseguir buenos niveles de precisión. El primer algoritmo que se utiliza es un algoritmo de selección de variables, que permite identificar las variables más significativas para la detección de un evento. Obtenidas las variables relevantes se procede a la realización de los modelos de normalidad de variables clave. Para dichos modelos se pueden usar las variables más relevantes. Las desviaciones de los modelos de normalidad dan pistas del estado de salud de la máquina. En la Figura 1 se muestra el comportamiento de los modelos. A partir de los modelos de normalidad se pueden sacar variables enriquecidas que ayudarán al diagnóstico final. Con los datos también se realizan técnicas de Clustering y Mapas autoorientados (SOM en sus siglas en Inglés, Self OrientedMaps). El SOM+ Clustering permite distinguir entre puntos de operación sanos y en avería. En la Figura 2 se Smartive is working on several such projects, including, for example, “Turbines for Life” from Acciona Energía, Naturgy’s WindEx R&D project and the Innogy lifetime extension programme. In every case, the company works with SCADA data, based on figures taken every ten minutes. Multiple analyses can be made based on SCADA data, providing both the wind farmoperator and the maintenance teamwith very useful information. The first analysis, SmartAudit, comprises an audit of the variables. It is worth nothing that SCADA variables are complicated to process, particularly in old farms where communications can be difficult. The data is filtered and adjusted, interpolating or extrapolating data gaps. Once the full data is available, alarms and events are analysed. A proper analysis of events is essential in order to achieve good levels of accuracy. The first algorithmused is a variables selection algorithm that is able to identify the most significant variables for detecting an event. Having obtained the relevant variables, normality models of key variables are then performed, using the most relevant variables. Deviations to the normality models give hints as to the state of health of the turbine. Figure 1 illustrates the model behaviour. Based on the normality models, enhanced variables can be extracted that will assist the final diagnosis. Self-organisingmaps (SOM) and clustering techniques are also performed with this data. SOM+ clustering are able to distinguish between healthy and defective operating points. Figure 2 shows an example of clusters, which illustrates the different operating zones of the turbine: healthy (blue and green); defective (red); and at risk (orange). The last algorithm used involves the classifiers. Based on an analysis of variables and metavariables (normality models, clusters), the classifiers give a failure probability over a given period of 90 days. Figure 3 shows how the failure probability evolves over time for a machine experiencing a gearbox breakdown. Depending on the sensitivity of the algorithm, the alarm limit is adjusted: in this specific case, the limit is 50%.When the failure probability exceeds this limit, an alarm goes off in the SCADA. This alarmmeans that there is a very high risk that the component will break within 90 days. This helps the operator and the maintenance team anticipate the preventive maintenance and avoid lost availability. The accuracy of these algorithms is highly dependent on the data.With quality data and training time, the algorithms achieve accurate indicators of above 90%, as shown in Figure 4. By increasing the accuracy of the algorithms it is possible to enhance the data and add expert knowledge. One new trend on which Smartive is working concerns model hybridisation which involves hybridising models based on data analysis, current MANTENIMIENTO PREDICTIVO DE AEROGENERADORES A PARTIR DE DATOS SCADA El mantenimiento predictivo en aerogeneradores es un valor en alza. La extensión de vida es una tendencia clara del sector en la que ya están trabajando operadores como Acciona, Naturgy e Innogy, y ello requiere de un mantenimiento exquisito. Más aún, la aparición de contratos que exigen al mantenedor en torno a un 98% de disponibilidad, incluso en algunos casos llegando al 100% de disponibilidad, imponen requisitos aúnmás exigentes. PREDICTIVE MAINTENANCE FOR WIND TURBINES BASED ON SCADA DATA Predictive maintenance inwind turbines is increasing in value. Lifetime extension is a clear trend in the sector onwhich operators such as Acciona, Naturgy and Innogy are already working, andwhich requires excellent levels of maintenance. Moreover, the emergence of contracts that required the maintainer to offer around 98%, and sometimes up to 100%, availability impose yet more demanding requirements. Figura 1. | Figure 1. Eólica | Wind Power FuturEnergy | Junio June 2019 www.futurenergyweb.es 21
RkJQdWJsaXNoZXIy Njg1MjYx