Research On Health Monitoring Of Wind Turbine Tower Flange Joints Based On Piezoelectric Impedance And Convolutional Neural Network

The flange joint of wind turbine tower is critical for the tower’s safety and stability.The loosening of joint bolts can be easily caused by material factors such as initial defects and material aging,load factors such as aerodynamic load and fatigue load,and environmental factors such as extreme temperature and humidity.The loosening will result in connection failure and then promoting the collapse of the structure.Furthermore,because the flange joints on wind turbine towers are internal flanges with connecting bolts inside,the damage is hidden and the health condition cannot be determined from the outside.However,at the moment,health monitoring research is primarily focused on wind turbines and tower foundations,with little attention paid to the flange joint of tower tubes.Regarding the issue above,a new method of health monitoring of wind turbine tower tube node based on piezoelectric impedance technique and convolutional neural network is proposed in this paper.Firstly,a theoretical analysis of piezoelectric coupled electrical impedance is used to verify the viability of the piezoelectric impedance technique for health monitoring of flange joints;Secondly,finite element numerical simulation is used to verify the effectiveness of the piezoelectric impedance technique for flange joint health monitoring;Finally,model testing is used to assess the damage occurrence and location of the damaged bolt,and a convolutional neural network is used to identify nine damage degrees quantitatively.The following are the specific research contents:(1)The theory of piezoelectric coupled electrical impedance is analyzed.The changes in the PZT electrical signal(electrical impedance or electrical admittance)before and after the node loosening can be compared to achieve bolt loosening monitoring inside the node.(2)Finite element numerical simulation software ABAQUS 2018 is used to model the flange joints,high-strength bolts and piezoelectric ceramic sensor PZT-5.Data on conductance is collected,and curves are created.Under five bolt loading scenarios,changes in conductance curves show the occurrence of various degrees of damage.(3)The scale model of tower cylinder flange joint is designed and made.To simulate varying degrees of damage state of tower cylinder flange joints,the PZT sheet is placed over the flange plate and different torques are applied to the bolt.According to the conductance curve of PZT sheet,the occurrence of node damage is judged,and the position of damaged bolt is located according to the conductance RMSD value.(4)An Alex Net convolutional neural network is built using the Keras deep learning framework with the normalized original conductance signal matrix as the input data set and the binary label corresponding to the damage degree as the output data set.The training set is chosen at random from 70% of the input data set,whereas the test set is chosen at random from 30%.Convolution is calculated using six one-dimensional convolution layers,while redundancy information is optimized using three one-dimensional pooling layers.The degree of bolt loosening is properly identified using the trained convolutional neural network.The results show that the proposed method can effectively monitor the occurrence of early minor damage,the position of the damaged bolt and the specific degree of damage of the bolt inside the tower flange joint.Convolutional neural network has strong robustness and can reduce the interference of external factors such as noise to a certain extent.It is more suitable for engineering practice.The proposed method has a certain reference value for the monitoring of bolt looseness in flange joints,and provides a new idea for the health monitoring of wind turbine tower structure.