Industry 4.0: A wind farm seizes the opportunity of digital innovations

Digital innovations have enabled a wind farm operator to achieve a dual objective:

• Continuous optimisation of performance parameters to produce more electricity.
• Reduction of maintenance costs by predicting failures, determining the residual life of components, optimising intervention schedules and controlling future expenses.

The technologies used:

Firstly, data acquisition is an essential point. Firstly, the sampling frequency must be determined according to the objectives sought and the technical constraints in order to guarantee the relevance of the data. Secondly, local pre-processing may be necessary when sending raw data is not justified. This is more specifically the case for vibration measurements and current harmonic analysis.

Secondly, it is generally necessary to clean up the data, complete the missing data, and synchronise them when the measurement frequencies are different: typically 1 second for SCADA data, about ten minutes for vibration measurements on the slow transmission shaft side and up to 6 hours for weather data. The aim is to ensure the relevance of further processing.  

Thirdly, to make this processing optimal, several models are applied. These are built and validated using different modelling approaches by exploiting historical data combined with maintenance and process expertise:

• A first approach is based on a combination of alarms, operational data and frequencies of occurrence of events measured by the SCADA.
• A second approach uses analytical tools such as decision trees, neural networks, ISHM models, etc.
• A third approach is based on physical laws, including meteorological models used to build wind speed and direction series around the turbine (in historical and forecast mode) and power curve models (capacity of the turbine to recover energy from the wind) and torque models (efficiency of the transformation of mechanical energy into electrical energy).
• A fourth approach, called hybrid, is based on a combination of the three previous ones. These include models derived from the power curve and the torque curve as well as predictive maintenance models.

Fourthly, in order to render the usable information provided by the models, it is necessary to set up display tools such as dashboards, trend curves and data cross-analysis graphs, and also a conversational assistant that answers the user's questions about the state of his machines.

Figure 1: Prognosis of residual life span

Early detection of faults and prognosis

The replacement of large components is costly in terms of intervention and production losses. However, early detection of faults combined with a prognosis of the remaining service life reduces the costs of interventions and allows them to be planned in good time. For example, a fault on the main bearing was detected 1.5 years before it caused damage to the peripheral equipment. This detection was achieved using a hybrid vibration model enhanced by SCADA measurements. The planning of the intervention was optimised thanks to the adaptive prognosis of the residual life, calculated by a hybrid degradation prediction model built on the basis of historical data and on the advice of maintenance experts.

Identifying the causes of performance degradation

On wind turbines, performance degradations have multiple origins, such as nacelle misalignment to the wind, blade erosion, loss of generator efficiency, etc. Identifying the cause of the degradation usually requires specific measurement campaigns and/or costly inspections. With hybrid models, some degradations can be isolated and even evaluated in terms of production loss and necessary correction. This was the case for a wind turbine with a nacelle misalignment. Isolation of the cause and evaluation of the correction required was achieved by combining hybrid power and torque curve models. After correction, the average improvement in output was estimated to be about 15%.