Fault Diagnosis Of Wind Turbine Gearbox Based On Multi-Modal Signal Fusion

As one of the most important components in the transmission system of wind turbine,smooth and safe operation of the gearbox is the premise of the wind turbine and the realization of kinetic energy conversion.However,the existing fault diagnosis method based on single-sensor signal cannot fully and reliably describe the status of the gearbox,and it is difficult to provide enough fault information to achieve high precision and efficiency fault diagnosis.Meanwhile,the existing methods based on multi-sensor signals fusion cannot share the common feature space between different sensor signals well,lack effective modal characterization mechanism.It also cannot realize the feature fusion of different signals,so the key information in multi-sensor signals cannot be fully utilized.This paper proposed a fault diagnosis method of wind turbine gearbox,introduces multi-modal learning and multi-task learning strategies,makes use of electromechanical coupling characteristics of wind turbine,combines vibration signal and current signal of generator motor to analyze two modals,and captures the fault diagnosis method of wind turbine gearbox based on multi-mode signal fusion.An effective mode fusion mechanism was established for vibration and current signals,the time-space correlation between the two modes was fully utilized,and the multi-task learning strategy was combined to improve the accuracy of the fault diagnosis model and realize efficient classification and recognition of gearbox faults.The main contents of this paper are as follows:Firstly,the main fault types and internal fault mechanisms of gears and bearings in wind turbine gear box are analyzed,and the fault principle of gear box is summarized.At the same time,the principle of gearbox fault diagnosis based on vibration and current signal is analyzed and summarized.A gearbox fault simulation experiment was designed to simulate the signals generated by gears and bearings under different health conditions,and a gearbox fault diagnosis multi-mode fault classification data set was established.Secondly,aiming at the problem that existing gear box fault diagnosis methods based on multi-sensor signal fusion lack effective signal characterization and fusion mechanism,a multi-modal multi-task with attention fusion network(MTAFN)is proposed,which uses continuous wavelet transform(CWT)to transform vibration and current signals into time-frequency wavelet scale spectra.And it is used as the input of feature extraction network.The extracted different modal features are input into the designed weighted attention fusion module,weighted fusion is carried out according to the proportion of effective features contained in each mode,and the fusion features with dynamic weights are input into the designed multi-task classification module.At the same time,the fault classification is carried out by combining the parameter information of single mode network and fusion mode network.The effectiveness of the proposed method is verified by public data set and self-built test bench data set,and compared with the fault diagnosis method based on a single mode signal.Thirdly,in view of the difficulty in extracting fault features from original signals and the lack of effective deep feature learning mechanism in existing models,a multi-task and multi-modal signal fusion network(M2FN)was proposed by introducing depth measurement learning strategy,aiming at better learning the similarity features of modes and the difference features of different fault types.This method uses a multi-branch convolutional neural network to extract the features of signals of different modes,uses the designed multi-task module with a depth measure learning strategy to carry out deep feature learning of fusion features,shares the parameters of classification tasks and measurement tasks,and learns the similarity and reciprocity characteristics between modes to enhance the inter-class separability and intra-class aggregation of samples.The collaborative enhancement between different kinds of tasks is realized and the accuracy of fault diagnosis is improved.The proposed method is compared with the single mode signal method and the latest fault diagnosis method,and the effectiveness of the proposed method is verified.