Coupling fault diagnosis of wind turbine gearbox based on multitask parallel convolutional neural networks with overall information

Abstract

With the development of smart grid, capacity of wind power that connects to the grid increases gradually, which makes the continuous and stable operation of wind turbine (WT) critically important. Therefore, by considering gearbox structure and operating condition, a diagnosis approach for coupling faults of WT gearbox is proposed based on multitask parallel convolutional neural network with reinforced input (RI-MPCNN). The overall information array of gearbox that fuses wavelet packet transform of vibration signals, domain knowledge of gearbox components and operating condition s used as RI-MPCNN input. Then, RI-MPCNN that has parallel sub-convolutional neural networks (sub-CNNs) and multiple classifiers realizes the diagnosis of coupling faults of multiple components simultaneously. Meanwhile, a reinforced input is added to each sub-CNN to improve the diagnosis accuracy of each component. It is notable that the proposed approach not only fuses the overall gearbox information at system level, but also realizes fault diagnosis at component level. In the approach evaluation based on two case studies, the proposed approach can improve diagnosis accuracies by about 3 and 20% compared with the existing methods, respectively.