Learning Decision Rules by Particle Swarm Optimization (PSO) for Wind Turbine Fault Diagnosis

The next generation wind turbine systems become more and more complex, which requires a more accurate fault detection method to ensure their efficiency. On a wind farm, sibling turbines should see similar wind speed if they both work normally. Based on this, we design wind speed difference tests to detect both hard failures and soft failures, including anemometer faults. In such tests, it is crucial to determine the decision boundary optimally to tell apart the abnormal state from the normal state. We propose a Particle Swarm Optimization (PSO) based approach to learn from historical data to decide the location and size of the boundary. This procedure is adaptable to each turbine using SCADA (Supervisory Control And Data Acquisition) data only. Our approach is advantageous in its applicability and data-driven nature to monitor a large wind farm. The test result has verified the effectiveness of our approach, and we have observed the anemometer aging in data.