RCCN WeChat QrCode
Mobile WebSiteWind turbines are electrical equipment that converts wind energy into mechanical energy and mechanical energy into electrical energy. A wind power enterprise is the top three large-scale wind power equipment manufacturers, specializing in large-scale wind turbine and key components of the design, manufacture and sale of wind farms and construction, operation and advisory services in the high-altitude fan market has significant advantages , There are hundreds of online operations, excellent product performance and good after-sales service access to a large number of domestic power investors highly recognized in the industry has a good reputation.
Big Data Thinking to Solve the Century Puzzle of Wind Turbine Small Bolts
Small bolt, big problem
The wind turbine components are mainly bolted, each blade has 50 bolts at the root, because the fan of the pitch operation, the natural aging of the bolt parts or leaves are subject to excessive stress and other factors, the root bolt will be broken or even fall off The situation. The breakage of the blade may cause the bolt to fall off into the engine compartment, causing damage to the unit inside the fan cabin, and when a bolt is problematic, it is likely to cause the other bolts to break continuously and eventually cause the blade to fall and even collapse. Whether the rupture of the blade of the wind turbine is entirely dependent on the manual investigation, the wind field is usually set in remote areas such as mountainous areas, grasslands, seaside or offshore, and a wind farm is usually composed of dozens of fans Of the inspection and can not be found in time, often in the event of a serious failure will be found. How can we find a bolt break in time to avoid the subsequent serious failure occurred, the wind power enterprises need to solve the problem.
Headache, headache foot is not desirable
At present, the wind power enterprises mainly use semi-annual manual inspection and troubleshooting routine. However, the wind field is usually set in remote areas, fan maintenance personnel is not easy to arrange, and a wind field is usually composed of dozens of fans, for a fan by the high frequency of manual investigation is very costly and time cost. Theoretically can also increase the sensor for testing, such as bolt preload force sensor, ring washer sensor, etc., by real-time detection of each bolt preload to determine whether there is no bolt break. Because of the high cost, the main application in the nuclear industry, scientific research and other fields, for a single blade there are 50 bolts fixed fan, on the one hand, cost is too low, the other will make the system more complex.
Generally speaking, it is a headache, not only increased the additional human and material resources, has not brought more extended benefits. According to the measurement and monitoring of other sensors, such as fan speed, inclination, wind speed, direction and other parameters, indirectly find the time point of the root bolt break, promptly notify the operator to replace the broken bolt or take other maintenance Measures. This method eliminates the need for additional sensors to detect broken bolts for the first time, and can be further developed to achieve predictive bolt breakage so that the operator can take measures before the bolt breaks to avoid bolt breakage.
Big data thinking and hear
The wind turbine itself has dozens of sensors that can return hundreds of fields, and these sensor data reflect the various states of the fans. For a long time, the wind power customers have accumulated a large number of blades including blade angle, blade pitch rate, wheel Shell speed and generator speed and other data. With the increase of the sensor, the paper analyzes the problem of fault detection of the fan root bolt and reduces the operating cost.
Based on the machine learning, the supervised learning analysis method refers to the selection of relevant variables from a number of peripheral sensor data indicators, the establishment of normal and abnormal operation of the fan model to determine the time of occurrence of bolt fracture, and then through the threshold of continuous learning algorithm, And gradually find the exact time of fracture occurred, and then to achieve the precise detection of bolt fracture. Specific implementation steps are as follows:
1, feature extraction. Selecting and extracting the variables related to the bolt breakage from a large number of sensor indicators, calculating the difference of the distribution of the sensor data before and after the fracture, and selecting the significant items; and analyzing the frequency of the sensor data before and after the break item;
2, the establishment of normal and abnormal models. According to the wind direction, wind speed and other external environmental factors related to the sensor numerical distribution of the working state of the cut, and to detect the bolt before the day off the smaller window data as a confirmation of abnormal data, training different states of the normal / abnormal decision model, confirm the model The detectability of the abnormal state.
3, to determine the time of failure. In the detection of the bolt before the day of the long window, the use of training to determine the model to detect the normal change to the abnormal jump point, that is normal to the abnormal model point.
4, classification algorithm threshold learning. The new exception data interval is obtained by using the jump point captured in 3, and the exception judgment model is re-trained, and the model is optimized. Repeat steps 3 and 4, step-by-step approximation of the occurrence of real, accurate early time points.
5, fault detection. Based on the final analysis results of 4, the bolt fracture is detected and the time and position of the fault are determined again. After accumulating sufficient data and models, the root bolt failure is further detected based on the trend of the sensor changes before the fault.
At present, the difference of the performance of the sensor and the fault characteristics are determined by the supervised learning analysis method based on the machine learning. The analysis method of finding the exact time of the bolt breakage is given, and the analysis of the existing analysis The results are verified and accumulated, and the prediction of leaf bolt breakage is realized step by step in order to make the operator adjust the equipment condition before the breakage, thus reduce the maintenance cost of the operation and maintenance and equipment and improve the production efficiency of the wind field. The use of data analysis based on machine learning monitoring learning analysis method not only applies to wind power enterprises, can also be applied to more similar large-scale machinery production and operation and maintenance environment for large-scale mechanical failure detection provides a solution ideas and methods.
