| In the actual operation of wind turbines,there are situations where the operating limits go beyond the limit value alarms.Most of these alarms are usually caused by external(related to surrounding conditions)factors and do not cause damage to the WTGs,but there are still a small number of alarms caused by internal faults in the WTGs.If human is not carried out as early as possible,it can lead to damage of other pieces and can cause incalculable losses by forcing the WTGs to shut down.Among the over-limit alarm limits of wind turbines,the over-limit alarm of gearbox limits accounts for most of them.In order to reduce the waste of human and material useful things caused by the limit going beyond the limit value alarm,from the perspective of the wind turbine gearbox limit going beyond limit alarm,this paper mainly completes the following work by judging the fault category through the features of related operation parameters.First,the gearbox fault data of doubly-fed wind turbines in actual wind farms are collected and pre-processed,including the division of wind speed and power,the identification and rejection of abnormal operation data,the addition and filling of missing data,and the early judgment of relationship parameters under the alarm limit fault tree.The similar gearbox fault of doubly-fed wind turbines is then built.Secondly,the doubly-fed wind turbine fault model is implemented by the RBF neural network algorithm,which is a forward-looking network with excellent performance and can rapidly approximate any nonlinear function and map the nonlinear relationship between data.The RBF neural network is used to train the actual operating fault data to increase the number of available samples for known faults,in order to address the problems of low number of individual actual faults,many invalid fault data,and turbines not operating in the effective wind speed range for a long time.Then,a gearbox fault diagnosis algorithm based on improved fuzzy c-mean clustering algorithm is proposed,and the algorithm is validated by simulation with fault data from a gearbox fault simulation testbed.The features of the operational data under different faults are extracted and analyzed using the local preservation projection algorithm.According to the different features of different faults,the study uses unsupervised clustering to differentiate the judgment,and also uses the affiliation degree to represent the classification of the data to be measured,so as to prevent the occurrence of unknown faults in the unit that lead to forced classification.In addition to this,for the problems of unreasonable clustering centroid setting and slow convergence of the clustering method,improvement methods such as setting the density index of subtractive clustering calculation to optimize the clustering centroid and introducing penalty factors are proposed.Finally,the timeliness as well as the feasibility of the fault diagnosis model is verified by an example of an over-limit alarm fault in the gearbox inlet oil pressure signal of a unit in an actual wind farm.The example collected a real fault situation of high oil pressure at the gearbox inlet of wind farm unit No.2 in a wind farm in Chongqing leading to an over-limit alarm,and the operating data including two days before the fault signal over-limit alarm were collected.Improved fuzzy c-mean clustering was used to judge the fault process of the data,and the experimental results showed the feasibility of the fault diagnosis model. |
