Tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning

Abstract

The present invention discloses a tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning. The method comprises: arranging strain sensors on the wind turbine tower, collecting strain distribution data caused by simulated damage; reconstructing the three-dimensional displacement distribution under simulated damage by inverse finite element deformation reconstruction algorithm; converting to two-dimensional displacement distribution and inputting into a convolutional neural network model for training, obtaining a trained convolutional neural network model; collecting strain distribution data caused by actual damage to the wind turbine tower, and reconstructing it into a two-dimensional displacement distribution; inputting into the trained convolutional neural network model, outputting the damage position and damage degree information of the wind turbine tower, and realizing structural monitoring of the wind turbine tower. The present invention can reconstruct the displacement distribution of the overall structure of the wind turbine tower by monitoring the strain information of some position points on the wind turbine tower, and can identify whether the tower is damaged and indicate the damage position and degree information, providing guarantee for the safe service of the wind turbine set.

Description

Technical Field

[0001] The present invention relates to a method for monitoring a wind turbine tower structure, and in particular to a method for monitoring a wind turbine tower structure based on inverse finite element deformation reconstruction and deep learning. Background Art

[0002] In existing large and medium-sized wind turbines, taller towers are often used to support the wind turbine impeller and nacelle in order to obtain more wind energy. During the service life of wind turbines, wind turbine towers are often exposed to extreme wind environments, and large deflection deformation and repeated stress cycles can cause damage to the tower. The tower is an important load-bearing component of the wind turbine. The performance of the tower directly affects the stability and reliability of the wind turbine operation. Its damage may lead to catastrophic damage to the entire wind turbine.

[0003] At present, the structural damage monitoring of wind turbine tower mainly adopts the method based on modal characteristics. The principle is that structural damage will cause changes in physical performance parameters (such as mass, damping and stiffness, etc.), thereby changing the modal characteristic parameters of the system (such as natural frequency, damping ratio and modal vibration shape, etc.). The specific implementation is: through a number of vibration sensors arranged on the wind turbine tower, the dynamic response of the tower is measured to obtain its vibration response data, and then the structural modal parameters such as the natural frequency, damping ratio and modal vibration shape of the tower are extracted from it, and the change of modal parameters is judged to determine whether the wind turbine tower is damaged. However, the structural health monitoring of wind turbine towers based on modal characteristics has the following shortcomings: (1) In order to ensure the accurate identification of the tower modal parameters, a large number of fixed vibration sensors need to be installed along the height direction of the tower. The sensors work in a vibration environment for a long time, and the looseness caused by vibration will lead to reduced reliability; (2) The modal parameters are the overall characteristics of the structure. Extracting and analyzing the changes in the modal parameters of the tower from the vibration response can only determine whether the wind turbine tower is damaged, but cannot indicate the specific damage area and damage degree, making it difficult to carry out subsequent positioning and maintenance work; (3) The damage detection method based on modal characteristics requires the modal parameters of the wind turbine tower in a healthy state as a reference. For wind turbines that have been in service, their modal parameters in a healthy state may be missing, resulting in the inability to effectively detect existing damage. Therefore, in order to ensure the safe operation of the wind turbine tower during service, a wind turbine tower damage monitoring method with high reliability and the ability to accurately detect and indicate the damage location and damage degree without relying on the parameters in its healthy state is needed. Summary of the invention

[0004] In order to solve the problems existing in the background technology, the present invention provides a tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning. The present invention can identify the damage location and damage degree of the tower to be tested.

[0005] The technical solution adopted by the present invention is:

[0006] The tower structure monitoring method of the present invention comprises the following steps:

[0007] S1. Several strain sensors are evenly spaced on the inner wall of the wind turbine tower, and strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower are collected through each strain sensor.

[0008] S2. For the strain distribution data caused by a degree of simulated damage at a position of the wind tower collected by each strain sensor, the overall three-dimensional displacement distribution of the wind tower structure under a degree of simulated damage at a position of the wind tower is reconstructed through an inverse finite element deformation reconstruction algorithm; each strain sensor collects the strain distribution data caused by simulated damage at different positions and degrees of the wind tower to obtain the overall three-dimensional displacement distribution of its own wind tower structure.

[0009] S3. Construct a convolutional neural network model, convert the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into an overall two-dimensional displacement distribution, input the overall two-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into the convolutional neural network model for training, and obtain a trained convolutional neural network model.

[0010] S4. After the wind turbine tower is actually damaged, the strain distribution data caused by the actual damage on the wind turbine tower is collected through each strain sensor on the wind turbine tower, and the overall three-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is reconstructed through the inverse finite element algorithm and converted into an overall two-dimensional displacement distribution; the overall two-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is input into the trained convolutional neural network model, and the trained convolutional neural network model outputs the damage location and damage degree information of the wind turbine tower to realize the structural monitoring of the wind turbine tower. Damage monitoring can be performed regularly during actual monitoring, and a damage degree of zero indicates no damage.

[0011] In the step S1, a number of strain sensors are evenly spaced on the inner wall of the wind turbine tower. Specifically, the inner wall of the wind turbine tower is evenly divided into a grid, and a strain sensor is arranged at each node position on the grid.

[0012] In the step S1, strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower are collected through various strain sensors. Specifically, a number of magnets with masses ranging from small to large are selected, and magnets with masses ranging from small to large are attached to one of the squares on the grid evenly divided by the inner wall of the wind turbine tower. One magnet is attached each time, so that simulated damage of different degrees is generated at the position of the square. The mass of each magnet corresponds to a degree of simulated damage. The larger the mass of the magnet, the heavier the degree of simulated damage, thereby obtaining a corresponding relationship between the mass of each magnet and the degree of simulated damage; each time a magnet is attached, strain distribution data of the wind turbine tower is collected through various strain sensors. Only one magnet is attached to a square for simulation each time.

[0013] The strain sensor adopts strain gauge or fiber grating and the like.

[0014] In the step S3, the constructed convolutional neural network model includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a second pooling layer and a fully connected layer connected in sequence.

[0015] In the step S3, the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage is converted into an overall two-dimensional displacement distribution, specifically, the three-dimensional structure of the wind turbine tower is unfolded into a two-dimensional rectangular plane along any generatrix of the tower cylindrical surface, and then the three-dimensional structural coordinates of the wind turbine tower are converted into two-dimensional structural coordinates.

[0016] In the step S3, the overall two-dimensional displacement distribution of the wind tower structure at different positions of the wind tower and under different degrees of simulated damage is input into the convolutional neural network model for training. For the overall two-dimensional displacement distribution of the wind tower structure under a degree of simulated damage at a position of the wind tower, the overall two-dimensional displacement distribution includes the actual damage position and actual damage degree of the simulated damage of the wind tower. After processing, the convolutional neural network model outputs the predicted damage degree and predicted damage position of the simulated damage of the wind tower, and the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind tower output by the convolutional neural network model and the actual damage position and actual damage degree of the simulated damage of the wind tower are calculated respectively.

[0017] The training of the convolutional neural network model is completed until the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind turbine tower output by the convolutional neural network model obtained through multiple iterative calculations and the actual damage position and actual damage degree of the simulated damage of the wind turbine tower are less than the preset error threshold.

[0018] The beneficial effects of the present invention are:

[0019] The present invention does not need to identify the modal information of the wind turbine tower, does not need information such as the material properties of the wind turbine tower, avoids the problem of being unable to compare modal parameters due to missing data in a healthy state, and simultaneously reconstructs the displacement distribution of the overall structure of the wind turbine tower by monitoring the strain information of some position points on the wind turbine tower, identifies whether the tower is damaged and indicates the damage location and degree information, thereby achieving real-time positioning and degree monitoring of wind turbine tower damage, and providing guarantee for the safe service of wind turbine sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of the steps of the method of the present invention;

[0021] Figure 2 It is a framework diagram of the convolutional neural network model of the present invention;

[0022] Figure 3 It is a two-dimensional unfolding schematic diagram of a three-dimensional wind turbine tower according to the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, the tower structure monitoring method of the present invention comprises the following steps:

[0025] S1. Several strain sensors are evenly spaced on the inner wall of the wind turbine tower, and strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower are collected through each strain sensor.

[0026] In step S1, a number of strain sensors are evenly arranged on the inner wall of the wind turbine tower, specifically, the inner wall of the wind turbine tower is evenly divided into a grid, and a strain sensor is arranged at each node position on the grid. The strain sensor is a strain gauge or a fiber grating.

[0027] In step S1, strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower are collected by various strain sensors. Specifically, several magnets with masses from small to large are selected, and magnets with masses from small to large are attached to one of the squares on the grid evenly divided by the inner wall of the wind turbine tower. One magnet is attached each time, so that simulated damage of different degrees is generated at the position of the square. The mass of each magnet corresponds to a degree of simulated damage. The larger the mass of the magnet, the heavier the degree of simulated damage, thereby obtaining the corresponding relationship between the mass of each magnet and the degree of simulated damage; each time a magnet is attached, strain distribution data of the wind turbine tower is collected by various strain sensors. Only one magnet is attached to a square for simulation each time. The degree of simulated damage is divided into levels according to the magnets with masses from small to large. The degree of simulated damage corresponding to the mass between 0 and the smallest mass is at the first degree level of simulated damage, and the degree of simulated damage corresponding to the mass between the nth magnet and the n-1th magnet is at the nth degree level of simulated damage.

[0028] S2. For the strain distribution data caused by a degree of simulated damage at a position of the wind tower collected by each strain sensor, the overall three-dimensional displacement distribution of the wind tower structure under a degree of simulated damage at a position of the wind tower is reconstructed through an inverse finite element deformation reconstruction algorithm; each strain sensor collects the strain distribution data caused by simulated damage at different positions and degrees of the wind tower to obtain the overall three-dimensional displacement distribution of its own wind tower structure.

[0029] S3. Construct a convolutional neural network model, convert the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into an overall two-dimensional displacement distribution, input the overall two-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into the convolutional neural network model for training, and obtain a trained convolutional neural network model.

[0030] In step S3, the constructed convolutional neural network model includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a second pooling layer and a fully connected layer connected in sequence.

[0031] In step S3, the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage is converted into an overall two-dimensional displacement distribution, specifically, the three-dimensional structure of the wind turbine tower is unfolded into a two-dimensional rectangular plane along any generatrix of the tower cylindrical surface, and then the three-dimensional structural coordinates of the wind turbine tower are converted into two-dimensional structural coordinates.

[0032] In step S3, the overall two-dimensional displacement distribution of the wind turbine tower structure at different positions and under different degrees of simulated damage of the wind turbine tower is input into the convolutional neural network model for training. For the overall two-dimensional displacement distribution of the wind turbine tower structure under a degree of simulated damage at a position of the wind turbine tower, the overall two-dimensional displacement distribution includes the actual damage position and actual damage degree of the simulated damage of the wind turbine tower. After processing, the convolutional neural network model outputs the predicted damage degree and predicted damage position of the simulated damage of the wind turbine tower, and the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind turbine tower output by the convolutional neural network model and the actual damage position and actual damage degree of the simulated damage of the wind turbine tower are calculated respectively.

[0033] The training of the convolutional neural network model is completed until the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind turbine tower output by the convolutional neural network model obtained through multiple iterative calculations and the actual damage position and actual damage degree of the simulated damage of the wind turbine tower are less than the preset error threshold.

[0034] S4. After the wind turbine tower is actually damaged, the strain distribution data caused by the actual damage on the wind turbine tower is collected through each strain sensor on the wind turbine tower, and the overall three-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is reconstructed through the inverse finite element algorithm and converted into an overall two-dimensional displacement distribution; the overall two-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is input into the trained convolutional neural network model, and the trained convolutional neural network model outputs the damage location and damage degree information of the wind turbine tower to realize the structural monitoring of the wind turbine tower. Damage monitoring can be performed regularly during actual monitoring, and a damage degree of zero indicates no damage.

[0035] In summary, the present invention arranges a number of strain sensors on the inner wall of the wind turbine tower, and obtains discrete strain distribution data sets on the tower under different positions and different degrees of damage by simulating damage. The overall displacement distribution of the three-dimensional tower structure is reconstructed through an inverse finite element deformation reconstruction algorithm, and the three-dimensional displacement distribution is expanded into a two-dimensional displacement matrix; the two-dimensional displacement matrix data set is used as input, and the matrix composed of the damage position coordinates and the degree level is used as output to train a convolutional neural network model with a custom structure, and classification model parameters are obtained through multiple iterative calculations; after completing the model training, the discrete strain distribution of the wind turbine tower under unknown damage is reconstructed and expanded to obtain the two-dimensional displacement matrix, which is input into the model to identify the damage position and degree of the tower to be tested.

[0036] The present invention can reconstruct the displacement distribution of the overall structure of the wind turbine tower by monitoring the strain information of some points on the wind turbine tower, and can identify whether the tower is damaged and indicate the location and degree of damage, thereby providing guarantee for the safe service of the wind turbine set.

Claims

1. A tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning, characterized in that: The steps include: S1. Arrange a number of strain sensors evenly and spaced apart on the inner wall of the wind turbine tower, and collect strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower through each strain sensor; S2. Reconstruct the overall three-dimensional displacement distribution of the wind tower structure under a degree of simulated damage at a position of the wind tower by using an inverse finite element deformation reconstruction algorithm for strain distribution data caused by a degree of simulated damage at a position of the wind tower collected by each strain sensor; each strain sensor collects strain distribution data caused by simulated damage at different positions and degrees of the wind tower to obtain the overall three-dimensional displacement distribution of its own wind tower structure; S3, constructing a convolutional neural network model, converting the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into an overall two-dimensional displacement distribution, inputting the overall two-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage into the convolutional neural network model for training, and obtaining a trained convolutional neural network model; S4. After actual damage occurs to the wind turbine tower, strain distribution data caused by the actual damage on the wind turbine tower is collected through various strain sensors on the wind turbine tower, and the overall three-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is reconstructed through an inverse finite element algorithm and converted into an overall two-dimensional displacement distribution; the overall two-dimensional displacement distribution of the wind turbine tower structure under the actual damage of the wind turbine tower is input into the trained convolutional neural network model, and the trained convolutional neural network model outputs the damage location and damage degree information of the wind turbine tower, thereby realizing structural monitoring of the wind turbine tower; In the step S3, the overall two-dimensional displacement distribution of the wind turbine tower structure at different positions and under different degrees of simulated damage of the wind turbine tower is input into the convolutional neural network model for training. For the overall two-dimensional displacement distribution of the wind turbine tower structure under a degree of simulated damage at a position of the wind turbine tower, the overall two-dimensional displacement distribution includes the actual damage position and actual damage degree of the simulated damage of the wind turbine tower. After processing, the convolutional neural network model outputs the predicted damage degree and predicted damage position of the simulated damage of the wind turbine tower, and the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind turbine tower output by the convolutional neural network model and the actual damage position and actual damage degree of the simulated damage of the wind turbine tower are calculated respectively. The training of the convolutional neural network model is completed until the errors between the predicted damage position and predicted damage degree of the simulated damage of the wind turbine tower output by the convolutional neural network model obtained through multiple iterative calculations and the actual damage position and actual damage degree of the simulated damage of the wind turbine tower are less than the preset error threshold.

2. The tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning according to claim 1 is characterized in that: In the step S1, a number of strain sensors are evenly spaced on the inner wall of the wind turbine tower. Specifically, the inner wall of the wind turbine tower is evenly divided into a grid, and a strain sensor is arranged at each node position on the grid.

3. The tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning according to claim 2 is characterized in that: In the step S1, strain distribution data caused by simulated damage at different positions and degrees of the wind turbine tower are collected through various strain sensors. Specifically, a number of magnets with increasing mass are selected, and magnets with increasing mass are attached to one of the squares on the grid evenly divided by the inner wall of the wind turbine tower. One magnet is attached each time, so that simulated damage of different degrees is generated at the position of the square. The mass of each magnet corresponds to a degree of simulated damage. The larger the mass of the magnet, the heavier the degree of simulated damage, thereby obtaining a corresponding relationship between the mass of each magnet and the degree of simulated damage; each time a magnet is attached, the strain distribution data of the wind turbine tower is collected through various strain sensors.

4. The tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning according to claim 1 is characterized in that: The strain sensor adopts a strain gauge or a fiber grating.

5. The tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning according to claim 1 is characterized in that: In the step S3, the constructed convolutional neural network model includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a second pooling layer and a fully connected layer connected in sequence.

6. The tower structure monitoring method based on inverse finite element deformation reconstruction and deep learning according to claim 1 is characterized in that: In the step S3, the overall three-dimensional displacement distribution of the wind turbine tower structure at different positions of the wind turbine tower and under different degrees of simulated damage is converted into an overall two-dimensional displacement distribution, specifically, the three-dimensional structure of the wind turbine tower is unfolded into a two-dimensional rectangular plane along any generatrix of the tower cylindrical surface, and then the three-dimensional structural coordinates of the wind turbine tower are converted into two-dimensional structural coordinates.

Images