Research On Acoustic Emission-based Identification Of Damage Degree For Pitch Bearing In Wind Turbine

The variable-pitch bearing is one of the core components in the transmission system of large-scale wind turbines,which operates in harsh environments and experiences complex loading conditions.Once a fault occurs,it can cause the control function of the entire wind turbine to fail,resulting in serious personal injury and economic loss.Therefore,research on the acoustic emission recognition method for the degree of damage to variable-pitch bearing in wind turbines is of great significance.Firstly,this paper conducts a uniaxial tensile acoustic emission monitoring test on the42 Cr Mo alloy steel material used in variable-pitch bearings,and obtains the distribution law of acoustic emission characteristic signals in the crack initiation,propagation,and fracture stages of the standard 42 Cr Mo specimens.Combined with digital image correlation analysis technology,the acoustic emission-strain correlation analysis at different crack evolution stages is achieved,and the evolution trend of the characteristic parameters of damage evolution process is obtained,laying the foundation for the acoustic emission recognition research of the damage to the retaining ring and the bearing cage of variable-pitch bearings.Secondly,based on the actual structural design of wind turbines,a long-term operation indoor experimental platform for variable-pitch bearings is built,which can simulate the working state of low-speed incomplete rotation of variable-pitch bearings under alternating loads such as axial load,radial load,and overturning moment load.On this experimental platform,the long-term fatigue damage test of variable-pitch bearings under the monitoring of acoustic emission technology is carried out,and the acoustic emission data samples of the variable-pitch bearing under simulated operating conditions are collected.Based on the collected acoustic emission data samples,the degree of damage to the variable-pitch bearing is divided by combining the impact-time curve within the bearing rotation period.The kurtosisbased complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)method is proposed for denoising the continuous waveform signals of acoustic emission from variable-pitch bearings,and the time-domain signals of acoustic emission at each stage of damage to the variable-pitch bearing are transformed into image features by Gramian angular field conversion.Finally,a residual neural network(Res Net)model based on convolutional attention mechanism is constructed to classify and recognize different degrees of damage to variablepitch bearings.The Res Net18-CBAM network model is compared with the standard Res Net18 network,standard VGG16 network,and standard Alex Net network model under the same training,validation,and testing conditions.The accuracy and superiority of the Res Net18-CBAM network are validated,providing an effective and reliable evaluation method for online monitoring of variable-pitch bearings in wind turbines,and having practical significance and application value in the field of early fault detection of wind turbine structures.