Early Warning for Wind Turbine Blade Collision Using the Event Camera
Objective In extreme environments,blade collisions with the tower due to complex aerodynamic forces can lead to severe accidents.Early warning systems for blade collisions are crucial for taking proactive measures to prevent accidents and minimize losses.Current methods for detecting blade collisions face two main issues:firstly,installing sensors on the wind turbine's surface is challenging.Secondly,conventional methods use high-speed cameras to track blade motion,but the accuracy of clearance distance measurement depends on the camera's frame rate,and large amounts of data complicate continuous monitoring.Additionally,traditional optical cameras are affected by harsh lighting conditions.In this study,we propose a novel method utilizing an event camera for early warning of wind turbine blade collisions.The event camera's high dynamic range allows monitoring under complex lighting conditions,and its low latency provides high temporal resolution.This reduces errors associated with frame rates compared to optical cameras.The low data volume and low power consumption of the event camera enhance monitoring efficiency.Methods We monitor the rapidly rotating blade tip in challenging lighting conditions using event stream data captured by the event camera.Our method employs polarity to extract blade coordinates from the event stream.We first accumulate the event stream into frames at a predetermined interval based on the monitoring frequency.Next,we analyze the polarity of edge events when the blade tip is at its minimum clearance distance.We then use a neighborhood search method to determine the precise blade tip coordinates from the candidate events.With these coordinates,we compute the distance between the blade tip and the tower's central axis on the event frame and determine the minimum clearance distance of the blade using central perspective projection principles.Results and Discussions A point cloud model of the wind turbine is obtained using a 3D scanner during shutdown.We compare our method with optical camera-based methods under the same conditions.The event camera,with a resolution of 1280 pixel×720 pixel and a delay of 10 μs,and the optical camera,with a resolution of 2448 pixel×2048 pixel and a frame rate of up to 30 frames per second,are both used to monitor blade motion and calculate minimum clearance distance.The raw event stream output from the event camera is shown in Fig.7.Unlike traditional optical cameras,the event camera does not use frames.Fig.8 illustrates the accumulation of event frames at different time intervals.When the time interval is too short,the event frame cannot capture all the information about blade motion.Conversely,when the time interval is too long,the edges of the blades in the event frame become unclear.Fig.9 compares accumulated event frames at 1/30 s intervals with images from an optical camera,showing that event cameras can capture information between the frames of optical cameras.Figs.10 and 11 compare imaging results from optical cameras and event cameras under harsh lighting conditions,demonstrating the advantages of event cameras in environments with strong light.Fig.12 shows the experimental setup for verifying the accuracy of blade tip coordinate extraction.The results in Table 1 confirm that the method proposed in this study accurately extracts blade tip coordinates.Table 2 and Fig.13 present the results of measuring the minimum clearance distance of blades using both optical cameras and event cameras.The comparison reveals that while the event camera accurately measures the minimum clearance distance,the optical camera's measurements have an error of over 1m.This discrepancy can lead to false alarms and serious consequences during blade scanning tower warnings.Conclusions We propose an early warning method for wind turbine blade collisions using an event camera.This method leverages the event camera's ability to capture sparse event stream data with high temporal resolution to monitor blade rotation.Our approach analyzes event polarity to extract candidate events and uses neighborhood search to determine blade tip coordinates,enabling accurate calculation of the minimum clearance distance and early collision warnings.The experiments demonstrate that,compared to the traditional optical camera monitoring approach,the proposed method offers significant advantages such as low latency and reduced data volume.Additionally,it effectively filters out the superfluous information from static background objects and substantially reduces measurement errors in blade clearance distance,which are typically caused by the frame rate limitations of optical cameras during wind turbine operation under harsh lighting conditions.
NETL
NSTL
万方数据
