Wind farms currently belong mostly to companies/operating companies usually unrelated to wind turbine manufacturers. These organizations are finding a number of problems, delays and losses related to availability, due to corrective maintenance actions performed on the turbines. Highlighted among these issues:
- The information provided by the wind turbine is not sufficient to know the status of its major components, often causing unexpected breakdowns to occur
- The support provided by the manufacturer may not be of appropriate quality or speed from the point of view of the maintenance provider
- When repairs are necessary, replacement components are often difficult to obtain, with long delivery times, causing the turbine to become inoperative during this period
- The wind turbines installed in a park may be from different manufacturers. There is therefore no homogeneity in their specifications, components or maintenance
The objective of this project is the development and validation of a technology for the integrated management of diagnosis and analysis of wind turbine systems. The system, using an intelligent prediction module and set of historical and current data of the monitored parameters, predicts the state of the elements.
Predicting breakdowns at an earlier stage results in a reduction in wind turbine maintenance and operating costs, and of course increases their availability, which is otherwise hampered due to waiting times from the onset of the unexpected failure to the replacement of the damaged component. Another of the important factors of the project is the remote monitoring and automation of monitored components as well as the ample gaining of information regarding the operation.
The system primarily performs an acquisition of analog signals from sensors of different types: accelerometers and tachometers, located in the drive train (low speed shaft, main bearing, gearbox and generator). Subsequently, the information is processed, and the results recorded. In addition, the system allows for autonomous operation, external communications for remote work applications running therein and verification of its status and sensors connected to it.
The system has been developed with a tiered structure, each operating independently. Communication between them is performed by TCP / IP protocol. The description and functions of each level are as follows:
- Physical level: Comprised of signal acquisition equipment and physical parameter sensor gauges used to try to check the status of the monitored components.
- Processing level. On this level the software modules responsible for processing the “raw” vibration signals from the acquisition device are implemented. It uses the LabVIEW Sound and Vibration Analysis Toolkit, extracting characteristic information of sensorised components from each of the signals. Afterwards, the results are dumped into a database for use in the following levels.
- Transport level. Here is found everything related to communication between the data acquisition device and processing modules, and between those and the database that stores the results. The network uses TCP / IP protocol.
- Analysis and diagnosis level: Consisting of three processing modules based on artificial intelligence.
- Prognosis Module: It is responsible for making estimates of the future values of the signal being monitored, in order to predict the behavior of the system, using artificial intelligence techniques to determine the real-time status of each of the elements.
- Detection module: This module’s mission is to determine the current and future states of the analysed components. The method employs a feed-forward neural network with an auto-encoder structure. In this case only assumed normal behaviour data is available, and the goal is to model this behavior and detect data that is out of the usual pattern. Subsequently, we have an estimator of the density function for normal behaviour and for three other states that represent abnormal situations (mild, severe, or very severe failure) to serve as a reference for comparing future system data and for establishing the distribution that it belongs to. Finally, a status determination unit based on the application of a sequential statistical test obtains the probability of where the system will be in each of the statements referred to.
- Diagnosis module: Establishes the nature of the fault or abnormality detected through the study and analysis of the symptoms and signs observed. This is done using knowledge systems based on rules that are allowed to explain the expert knowledge.
METHODOLOGY / PROJECT STAGES:
- Definition of prerequisites and system components.
This stage was ascertained to monitor the elements of the wind turbines, the types of fault to consider, the most suitable location of the sensors, the sensor type and the hardware and software to use.
- Acquisition module design and data storage
- Acquisition module mounting
- Installation of the turbine equipment at the Sotavento Wind Farm.
Installing sensors, acquisition modules and hardware for data capture in wind turbines AE01, AE05, AE09, AE13 and AE20 at Sotavento Wind Park. Field tests to confirm the suitability of the equipment.
- Integration of equipment with the communications network at the wind farm
- Design and installation of the central server
- Registration and data monitoring
- Development of AEROGIDAS intelligent forecasting system
- Field trials, testing and validation of equipment and systems
- INDRA Systems
- Sotavento Experimental Wind Farm Galicia
- UDC (University of Coruña), R & D Laboratory of Artificial Intelligence (LIDIA)