A method and system for intelligently monitoring the state of a power transmission system of a wind power installation vessel

Abstract

The present invention belongs to the field of wind power installation vessel power monitoring, and specifically relates to a method and system for intelligent monitoring of the power transmission system state of a wind power installation vessel, comprising a signal acquisition module, an information processing module and a prediction output module, and utilizing an improved adaptive threshold function to reduce noise on the signal obtained by the sensor, thereby obtaining a clearer fault signal, and combining a CNN fault diagnosis method to realize real-time fault diagnosis of a wind power installation vessel propeller, a crane power system and a pile leg gear reduction system.

Description

Technical Field

[0001] The present invention belongs to the field of wind power installation vessel power monitoring, and in particular relates to a method and system for intelligently monitoring the state of a power transmission system of a wind power installation vessel. Background Art

[0002] The thrusters, crane, and pile legs are essential components of a wind turbine installation vessel. By driving water, the thrusters control the vessel's forward motion and steering. The crane facilitates the lifting and assembly of the vessel. By extending the pile legs and inserting the pile shoes into the seabed, the vessel maintains its position and performs its work. The thruster system, pile leg gear reduction system, and crane power system are critical to a wind turbine installation vessel's powertrain. Failures in these systems can lead to vessel imbalance, equipment failure, and other issues, threatening the vessel's safety. Summary of the Invention

[0003] To address the above issues, this application uses adaptive wavelet threshold noise reduction to remove environmental noise from vibration sensor, temperature sensor, and torque sensor signals, obtaining clearer fault signals. The processed fault signals are then input into a CNN for training and fault classification, enabling real-time fault diagnosis of key components of wind turbine installation vessels. The technical solution is:

[0004] A method for intelligently monitoring the status of a power transmission system of a wind power installation vessel, wherein a propeller is provided with a temperature sensor, a vibration sensor, and a torque sensor, and a pile leg gear reduction system and a crane power system are provided with vibration sensors, comprising the following steps:

[0005] S1. Collect the vibration signal, temperature signal and torque signal containing noise, and use express, p is the collected sensor signal, ;

[0006] In the formula is the effective signal of the sensor, is the noise signal of the sensor;

[0007] S2. Decompose the signal using a wavelet basis; decompose the original time domain signals of the vibration sensor, temperature sensor, and torque sensor to obtain the low-frequency coefficients and high-frequency coefficients of each layer;

[0008] S3. Filtering high frequency coefficients using an improved adaptive threshold function;

[0009] S4. Reconstruct the signal through inverse transformation to obtain the denoised vibration signal, temperature signal, and torque signal;

[0010] S5. Input the denoised signal from step S4 into the CNN model. For a set of samples, each sensor signal is input as a row, forming a two-dimensional input layer as the input of the CNN model.

[0011] S6. After the CNN training and verification are completed, the training parameters are saved to obtain the CNN diagnostic model.

[0012] Preferably, in step S2, an orthogonal dbN wavelet basis is selected, and the original time domain signals of the vibration sensor, temperature sensor, and torque sensor are decomposed by discrete wavelet transform DWT to obtain low-frequency coefficients and high-frequency coefficients of each layer. The wavelet decomposition formula is as follows (based on the prior art):

[0013] ;

[0014] ;

[0015] Where, Indicates the j +1 layer of decomposition of low-frequency coefficients, Indicates the j The low-frequency coefficients of the layer, is the filter coefficient of the low-pass filter; Indicates the j +1 layer of decomposition of high frequency coefficients, is the filter coefficient of the high-pass filter.

[0016] Preferably, in step S3, the formula of the improved adaptive threshold function is:

[0017] ;

[0018] In the formula is the high frequency coefficient of the j+1th layer, is the adaptive threshold of the j+1th layer, is the trend parameter of the j+1th layer, is the high-frequency coefficient of the j+1th layer after processing.

[0019] Preferably, the rigrsure method is used to calculate the adaptive threshold of the jth layer based on the unbiased likelihood estimation principle. , the trend parameter h can be determined by the following formula:

[0020] ;

[0021] Where, is the trend parameter of the jth layer, is the high frequency coefficient energy of the jth layer, It is the sum of the energy of high-frequency coefficients of each layer. The signal measured by each sensor is denoised separately using an adaptive threshold function.

[0022] Preferably, in step S4, the same wavelet basis function as the forward transform is used to reconstruct the low-frequency coefficients and high-frequency coefficients back to the original signal through inverse wavelet transform, and the signal measured by each sensor is individually subjected to adaptive threshold function noise reduction. The wavelet reconstruction formula is as follows (based on the prior art):

[0023] ;

[0024] Where, It is j +1 layer of low-frequency coefficients, is the filter coefficient corresponding to the mirror filter of the low-pass filter, It is j +1 layer high frequency coefficients, is the filter coefficient corresponding to the mirror filter of the high-pass filter, is the low-frequency coefficient of the j-th layer after reconstruction.

[0025] Preferably, in step S5, the sampling frequency of each sensor is f Hz, for a set of samples, each denoised sensor signal is input as a row, which can form a two-dimensional input layer, that is, i×f The initial two-dimensional signal :

[0026] ;

[0027] The convolution layer in the CNN model uses multiple output channels, and the number of convolution kernels is the number of output channels. The convolution operation method is as follows (based on existing technology):

[0028] ;

[0029] in, is the output feature map c channels, For the c The weight of the convolution kernel corresponding to the output channel, is the bias, activation function, is the input feature map;

[0030] The size of the convolution kernel is 1 row L c Column, that is, the convolution kernel covers 1 row in the vertical direction and L in the horizontal direction c Column, the stride of the convolution kernel is 1 row L s Column, the convolution operation moves L in the horizontal direction each time sThe convolution operation is performed on each row separately.

[0031] Preferably, after each convolutional layer or fully connected layer, before using the activation function, a batch normalization layer operation is performed. The batch normalization formula is as follows (based on the prior art):

[0032] ;

[0033] in, is the variance optimization parameter, is the mean optimization parameter, Is the input data The mean of is the variance of the input data, is a constant.

[0034] Preferably, a pooling layer is used to reduce the dimension of the input data. The pooling method is as follows (according to the prior art):

[0035] ;

[0036] in, Represents the first Rank Elements of the column, Represents the element in the mth row and vth column of the output feature map after maximum pooling, Represents the element window in the input feature map corresponding to the element in the mth row and vth column of the output feature map;

[0037] The window size of the pooling layer is 1 row B c Column, the stride of the pooling layer is 1 row B s Column, the pooling operation moves horizontally each time B s Columns, move row by row in the vertical direction, that is, perform pooling operation independently on each row.

[0038] An intelligent monitoring system for the power transmission system status of a wind power installation vessel, comprising a signal acquisition module, an information processing module and a prediction output module;

[0039] Signal acquisition module: The temperature sensor collects the temperature signal on the thruster; the vibration sensor collects the vibration signal of the thruster, pile leg gear reduction system and crane power system; the torque sensor collects the torque signal on the thruster;

[0040] Information processing module: Uses wavelet basis to perform signal decomposition, decomposing the original time-domain signals of the vibration sensor, temperature sensor, and torque sensor to obtain low-frequency and high-frequency coefficients of each layer; uses an improved adaptive threshold function to filter the high-frequency coefficients; reconstructs the signal through inverse transformation to obtain the denoised vibration signal, temperature signal, and torque signal; after CNN training and verification, saves the training parameters to obtain the CNN diagnostic model;

[0041] Prediction output module: Use the trained CNN diagnosis model to output the fault type and location.

[0042] After the fully connected layer, CNN uses the Sigmoid activation function to generate probability values. This is combined with an improved loss function for model training. During training, model parameters are optimized by minimizing the loss function. The improved loss function simultaneously considers the importance of various fault types, the difficulty of prediction, and the numerical stability of the model during training.

[0043] ;

[0044] in, is the average loss in a batch, z represents the number of samples, Indicates the number of fault categories, is the fault weight, For samples z In category The true label (0 or 1) on For samples z In category The predicted probability of To predict the difficulty penalty coefficient, is the minimum penalty limit value, is the maximum penalty limit value.

[0045] The torque sensor on the first shaft of the thruster includes a conductive slip ring, an elastic beam, and a resistance strain gauge. The elastic beam is located on the upper part of the first shaft and above the oil suction port. There are p elastic beams, and resistance strain gauges are attached to both sides of each elastic beam. The resistance strain gauges and the conductive slip ring form a circuit. When the shaft rotates under a certain torque, one side of the elastic beam is the compression side and the other side is the tension side. The resistance strain gauge is compressed or stretched, resulting in a change in resistance. By applying a known torque and recording the signal, a torque-signal relationship curve equation is established, and the p groups of signals are converted into actual torque values.

[0046] Compared with the prior art, this application has the following beneficial effects:

[0047] This application proposes an intelligent monitoring method for the power transmission system status of a wind turbine installation vessel. Through a monitoring sensor signal noise reduction and CNN fault diagnosis method, an improved adaptive threshold function is used to reduce the noise of the signal obtained by the sensor, thereby obtaining a clearer fault signal. Combined with the CNN fault diagnosis method, real-time fault diagnosis of the key power transmission system of the wind turbine installation vessel is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 An internal view of a ship propulsion unit installed for wind power;

[0049] Figure 2 A partial enlarged view of the torque sensor for the propeller installed on a wind turbine ship;

[0050] Figure 3 Cross-section of a ship propeller torque sensor installed for wind power;

[0051] Figure 4 Installation diagram of ship thruster flowmeter location for wind power;

[0052] Figure 5 Schematic diagram of the drive part of the pile leg for wind power installation;

[0053] Figure 6 Isometric drawing of ship crane power system for wind power installation;

[0054] Figure 7 Left view of the wind power installation ship crane power system;

[0055] Figure 8 Main view of the power system of the ship crane installed for wind power;

[0056] Figure 9 Flowchart of sensor signal noise reduction and CNN fault diagnosis method;

[0057] Figure 10 This is a diagram of the sensor real-time operating status monitoring system.

[0058] In the figure, 00 torque sensor, 001 conductive slip ring, 002 elastic beam, 003 resistance strain gauge, 1 propeller housing, 2 blade, 3 first mechanical seal, 4 oil suction pipe, 5 third oil inlet pipe, 6 first bevel gear, 7 first temperature sensor, 8 second temperature sensor, 9 third temperature sensor, 10 fourth temperature sensor, 11 fifth temperature sensor, 12 first thrust spherical roller bearing, 13 second thrust spherical roller bearing, 14 first spherical roller bearing, 15 second bevel gear, 16 second shaft, 17 second spherical roller bearing, 18 propeller X-axis vibration sensor, 19 propeller Y-axis vibration sensor, 20 second mechanical seal, 21 sixth temperature sensor, 22 propeller Z-axis vibration sensor, 23 third spherical roller bearing, 24 first shaft, 25 third thrust spherical roller bearing, 26 seventh temperature sensor, 27 sleeve, 28 first oil inlet pipeline, 29 second oil inlet pipeline, 30 first flowmeter, 31 second flowmeter, 32 motor, 33 pile leg motor Z-direction vibration sensor, 34 first coupling, 35 pile leg gear reducer, 36 pile leg gear reducer X-direction vibration sensor, 37 pile leg gear reducer Y-direction vibration sensor, 38 pile leg gear reducer Z-direction vibration sensor, 39 second coupling, 40 pile leg gear, 41 winch, 42 first drive system, 43 second drive system, 44 crane power system motor Z-direction vibration sensor, 45 third drive system, 46 fourth drive system, 47 planetary reducer X-direction vibration sensor, 48 planetary reducer Y-direction vibration sensor, 49 planetary reducer Z-direction vibration sensor, 50 drive pinion, 51 ring gear, 52 crane power system motor, 53 crane power system coupling, 54 planetary reducer. DETAILED DESCRIPTION

[0059] To facilitate understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The accompanying drawings provide embodiments of the present application. However, the present application may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the present application more thorough and comprehensive.

[0060] A method for intelligently monitoring the status of a power transmission system of a wind power installation vessel, characterized in that a propeller is provided with a temperature sensor, a vibration sensor, and a torque sensor, and a pile leg gear reduction system and a crane power system are provided with vibration sensors, comprising the following steps:

[0061] S1. Collect the vibration signal, temperature signal and torque signal containing noise, and use express, p is the collected sensor signal, ;

[0062] In the formula is the effective signal of the sensor, is the noise signal of the sensor;

[0063] S2. Decompose the signal using a wavelet basis; decompose the original time domain signals of the vibration sensor, temperature sensor, and torque sensor to obtain the low-frequency coefficients and high-frequency coefficients of each layer;

[0064] S3. Filtering high frequency coefficients using an improved adaptive threshold function;

[0065] S4. Reconstruct the signal through inverse transformation to obtain the denoised vibration signal, temperature signal, and torque signal;

[0066] S5. Input the denoised signal from step S4 into the CNN model. For a set of samples, each sensor signal is input as a row, forming a two-dimensional input layer as the input of the CNN model.

[0067] S6. After CNN training and verification are complete, the trained parameters are saved to obtain a CNN diagnostic model. After CNN training and verification are complete, the trained parameters are saved to obtain a CNN diagnostic model. By inputting test data into the CNN model, real-time fault diagnosis can be performed to promptly detect bearing faults, gear faults, blade faults, and so on.

[0068] The noise reduction and CNN fault diagnosis process described above is stored in an information processor. Noisy sensor signals from the thruster, leg gear reduction system, and crane power system are fed into the information processor, generating real-time fault output. This method enables real-time monitoring of key components while a wind turbine installation vessel is in operation, facilitating timely adjustments.

[0069] In step S2, an orthogonal dbN wavelet basis is selected and the original time domain signals of the vibration sensor, temperature sensor, and torque sensor are decomposed by discrete wavelet transform (DWT) to obtain low-frequency coefficients and high-frequency coefficients of each layer. The wavelet decomposition formula is as follows (based on existing technology):

[0070] ;

[0071] ;

[0072] Where, Indicates the j +1 layer of decomposition of low-frequency coefficients, Indicates the j The low-frequency coefficients of the layer, is the filter coefficient of the low-pass filter; Indicates the j +1 layer of decomposition of high frequency coefficients, is the filter coefficient of the high-pass filter.

[0073] In step S3, the formula of the improved adaptive threshold function is:

[0074] ;

[0075] In the formula is the high frequency coefficient of the j+1th layer, is the adaptive threshold of the j+1th layer, is the trend parameter of the j+1th layer, is the high-frequency coefficient of the j+1th layer after processing.

[0076] The rigrsure method is used to calculate the adaptive threshold of the jth layer based on the unbiased likelihood estimation principle. , the trend parameter h can be determined by the following formula:

[0077] ;

[0078] Where, is the trend parameter of the jth layer, is the high frequency coefficient energy of the jth layer, It is the sum of the energy of high-frequency coefficients of each layer. The signal measured by each sensor is denoised separately using an adaptive threshold function.

[0079] In step S4, the same wavelet basis function as the forward transform is used to reconstruct the low-frequency coefficients and high-frequency coefficients back to the original signal through inverse wavelet transform. The signal measured by each sensor is individually subjected to adaptive threshold function noise reduction. The wavelet reconstruction formula is as follows (based on the prior art):

[0080] ;

[0081] Where, It is j +1 layer of low-frequency coefficients, is the filter coefficient corresponding to the mirror filter of the low-pass filter, It is j +1 layer high frequency coefficients, is the filter coefficient corresponding to the mirror filter of the high-pass filter, is the low-frequency coefficient of the j-th layer after reconstruction.

[0082] In step S5, the sampling frequency of each sensor is f Hz, for a set of samples, each denoised sensor signal is input as a row, which can form a two-dimensional input layer, that is, i×f The initial two-dimensional signal :

[0083] ;

[0084] The convolution layer in the CNN model uses multiple output channels, and the number of convolution kernels is the number of output channels. The convolution operation method is as follows (based on existing technology):

[0085] ;

[0086] in, is the output feature map c channels, For the c The weight of the convolution kernel corresponding to the output channel, is the bias, activation function, is the input feature map;

[0087] The size of the convolution kernel is 1 row L c Column, that is, the convolution kernel covers 1 row in the vertical direction and L in the horizontal direction c Column, the stride of the convolution kernel is 1 row L s Column, the convolution operation moves L in the horizontal direction each time s The convolution operation is performed on each row separately.

[0088] After each convolutional layer or fully connected layer, before using the activation function, a batch normalization layer operation is performed. The batch normalization formula is as follows (based on existing technology):

[0089] ;

[0090] in, is the variance optimization parameter, is the mean optimization parameter, Is the input data The mean of is the variance of the input data, is a constant.

[0091] The pooling layer is used to reduce the dimensionality of the input data. The pooling method is as follows (based on existing technology):

[0092] ;

[0093] in, Represents the first Rank Elements of the column, Represents the element in the mth row and vth column of the output feature map after maximum pooling, Represents the element window in the input feature map corresponding to the element in the mth row and vth column of the output feature map;

[0094] The window size of the pooling layer is 1 row B c Column, the stride of the pooling layer is 1 row B s Column, the pooling operation moves horizontally each time B s Columns, move row by row in the vertical direction, that is, perform pooling operation independently on each row.

[0095] An intelligent monitoring system for the power transmission system status of a wind power installation vessel, comprising a signal acquisition module, an information processing module and a prediction output module;

[0096] Signal acquisition module: The temperature sensor collects the temperature signal on the thruster; the vibration sensor collects the vibration signal of the thruster, pile leg gear reduction system and crane power system; the torque sensor collects the torque signal on the thruster;

[0097] Torque detection: The conductive slip ring 001, the elastic beam 002, and the resistance strain gauge 003 constitute the torque sensor 00 on the first shaft 24 of the thruster; the elastic beam 002 is located on the upper part of the first shaft 24 and above the oil suction port. There are four elastic beams 002, and both sides of each elastic beam 002 are affixed with a resistance strain gauge 003. The resistance strain gauge 003 and the conductive slip ring 001 form a circuit. When the first shaft 24 rotates under a certain torque, one side of the elastic beam 002 is the compression side and the other side is the tension side. The resistance strain gauge 003 is compressed or stretched, resulting in a change in resistance. The torque-signal relationship curve equation is established by applying a known torque and recording the signal, and the four groups of signals are converted into actual torque values.

[0098] Information processing module: The propeller information processor, pile leg gear reduction system information processor, and crane power system information processor each use wavelet basis to perform signal decomposition; decompose the original time domain signals of the vibration sensor, temperature sensor, and torque sensor to obtain low-frequency coefficients and high-frequency coefficients of each layer; use an improved adaptive threshold function to filter the high-frequency coefficients; reconstruct the signals through inverse transformation to obtain the denoised vibration signal, temperature signal, and torque signal; after CNN training and verification, save the training parameters to obtain the CNN diagnostic model for fault prediction;

[0099] Prediction output module: Use the trained CNN diagnosis model to output the fault type and fault location.

[0100] After the fully connected layer, CNN uses the Sigmoid activation function to generate probability values. This is combined with an improved loss function for model training. During training, model parameters are optimized by minimizing the loss function. The improved loss function simultaneously considers the importance of various fault types, the difficulty of prediction, and the numerical stability of the model during training.

[0101] ;

[0102] in, is the average loss in a batch, z represents the number of samples, Indicates the number of fault categories, is the fault weight, For samples z In category The true label (0 or 1) on For samples z In category The predicted probability on (calculated by the sigmoid activation function), To predict the difficulty penalty coefficient, is the minimum penalty limit value, is the maximum penalty limit value.

[0103] A propeller state intelligent monitoring structure includes a torque sensor 00, a conductive slip ring 001, an elastic beam 002, a resistance strain gauge 003, a propeller housing 1, a blade 2, a first mechanical seal 3, an oil suction pipe 4, a third oil inlet pipeline 5, a first bevel gear 6, a first temperature sensor 7, a second temperature sensor 8, a third temperature sensor 9, a fourth temperature sensor 10, a fifth temperature sensor 11, a first thrust spherical roller bearing 12, a second thrust spherical roller bearing 13, a first spherical roller bearing 14, a second bevel gear 15, a second shaft 16, a second spherical roller bearing 17, a propeller X-direction vibration sensor 18, a propeller Y-direction vibration sensor 19, a second mechanical seal 20, a sixth temperature sensor 21, a propeller Z-direction vibration sensor 22, a third spherical roller bearing 23, a first shaft 24, a third thrust spherical roller bearing 25, a seventh temperature sensor 26, a sleeve 27, a first oil inlet pipeline 28, a second oil inlet pipeline 29, a first flowmeter 30, and a second flowmeter 31.

[0104] Among them, the elastic beam 002 is located at the upper part of the first shaft 24 and above the oil suction port. Resistance strain gauges 003 are affixed to both sides of each elastic beam 002. The resistance strain gauges 003 and the conductive slip ring 001 form a circuit. The first mechanical seal 3 is fixedly connected to the sleeve 27. The oil suction pipe 4 is connected to the oil suction port of the sleeve 27. The third oil inlet pipe 5 is connected to the oil inlet of the propeller housing 1. The oil inlet is located near the third temperature sensor 9. The first bevel gear 6 is fixedly connected to the first shaft 24. The first bevel gear 6 is meshed with the second bevel gear 15. The first temperature sensor 7, the second temperature sensor 8, the third temperature sensor 9, the fourth temperature sensor 10, the fifth temperature sensor 11, the propeller X-direction vibration sensor 18, the propeller Y-direction vibration sensor 19, and the propeller Z-direction vibration sensor The dynamic sensor 22 and the sixth temperature sensor 21 are fixedly connected to the thruster housing 1, the first thrust spherical roller bearing 12, the second thrust spherical roller bearing 13, the first spherical roller bearing 14, the second bevel gear 15, and the second spherical roller bearing 17 are connected to the second shaft 16, the second mechanical seal 20 is fixedly connected to the thruster housing 1, the third spherical roller bearing 23 and the third thrust spherical roller bearing 25 are connected to the first shaft 24, the sleeve 27 is fixedly connected to the thruster housing 1, the first oil inlet pipe 28 is connected to the oil inlet of the second mechanical seal 20, the second oil inlet pipe 29 is connected to the oil inlet of the second spherical roller bearing 17, the first flowmeter 30 is located at the upper part of the front thruster port of the wind power installation vessel, and the second flowmeter 31 is located at the lower part of the front thruster port of the wind power installation vessel.

[0105] Lubrication Process: First lubrication oil circuit: Oil enters from the third oil inlet pipe 5, flows through the oil inlet of the thruster housing 1, and flushes the first thrust spherical roller bearing 12 and the second thrust spherical roller bearing 13. The oil flowing through the second thrust spherical roller bearing 13 flushes the first spherical roller bearing 14. Second lubrication oil circuit: Oil enters from the second oil inlet pipe 29, passes through the second mechanical seal 20, and enters the second spherical roller bearing 17, flushing the second spherical roller bearing 17. Third lubrication oil circuit: Oil flowing out of the flushed bearings, such as the first spherical roller bearing 14 and the second spherical roller bearing 17, enters the thruster housing 1, and gradually flushes upwards to the second bevel gear 15, the first bevel gear 6, the third spherical roller bearing 23, and finally reaches the third thrust spherical roller bearing 25. Oil outlet circuit: Oil is sucked up through the oil suction pipe 4, completing the oil circulation.

[0106] Sealing process: Oil at a set pressure (the oil pressure is set by setting the oil tank liquid level, which needs to be slightly higher than the seawater pressure outside the thruster) enters the first oil inlet 28 and enters the second mechanical seal 20, which not only plays a lubricating role, but also serves as a further sealing role to prevent external seawater from entering the second mechanical seal 20 and the thruster housing 1.

[0107] Temperature Detection: There are seven temperature sensors. The second temperature sensor 8 detects normal oil temperature. The first temperature sensor 7 detects the temperature of the oil flowing through the third spherical roller bearing 23. The third temperature sensor 9 detects the temperature of the oil flowing through the first spherical roller bearing 14. The fourth temperature sensor 10 detects the temperature of the oil flowing through the second thrust spherical roller bearing 13. The fifth temperature sensor 11 detects the temperature of the oil flowing through the first thrust spherical roller bearing 12. The sixth temperature sensor 21 detects the temperature of the oil flowing through the second spherical roller bearing 17. The seventh temperature sensor 26 detects the temperature of the oil flowing through the third thrust spherical roller bearing 25. Due to environmental influences, oil temperature can fluctuate. If the temperature of the oil flowing through a bearing or gear changes abnormally, it indicates that the bearing or gear is overheating, which may indicate wear or aging of the bearing or gear, and the bearing or gear needs to be replaced promptly.

[0108] Vibration Detection: The propeller's X-axis vibration sensor 18, Y-axis vibration sensor 19, and Z-axis vibration sensor 22 monitor propeller vibration. Under normal circumstances, propeller vibration is minimal and acceptable. However, if a propeller malfunctions, the vibration may exceed the normal range, allowing the vibration sensors to detect the fault immediately.

[0109] Functionality Testing: The first and second flow meters 30, 31 are staggered vertically to accurately measure flow signals. Damage or obstruction to the propeller blades can affect the flow of water around them, causing the flow meter reading to drop or fluctuate significantly. If the propeller is operating but the flow meter's measured flow rate differs significantly from the theoretical calculated flow rate, this indicates a propeller blade failure and requires repair.

[0110] Leg gear reduction system:

[0111] It includes a motor 32, a pile leg motor Z-direction vibration sensor 33, a first coupling 34, a pile leg gear reducer 35, a pile leg gear reducer X-direction vibration sensor 36, a pile leg gear reducer Y-direction vibration sensor 37, a pile leg gear reducer Z-direction vibration sensor 38, a second coupling 39, and a pile leg gear 40.

[0112] Among them, the first coupling 34 connects the motor 32 and the pile leg gear reducer shaft 35, the second coupling 39 connects the pile leg gear 40 and the pile leg gear reducer shaft 35, the pile leg gear reducer X-direction vibration sensor 36, the pile leg gear reducer Y-direction vibration sensor 37, and the pile leg gear reducer Z-direction vibration sensor 38 are all installed on the pile leg gear reducer 35 to monitor the vibration of the pile leg gear reducer 35 in real time, and the pile leg motor Z-direction vibration sensor 33 is installed on the motor 32 to monitor the vibration of the motor 32 in real time.

[0113] The motor 32 and the leg gear 40 are located on the same side of the leg gear reducer 35. There are four sensors: the leg motor Z-axis vibration sensor 33, the leg gear reducer X-axis vibration sensor 36, the leg gear reducer Y-axis vibration sensor 37, and the leg gear reducer Z-axis vibration sensor 38.

[0114] Crane power system:

[0115] It includes a winch 41, a first drive system 42, a second drive system 43, a crane power system motor Z-direction vibration sensor 44, a third drive system 45, a fourth drive system 46, a planetary reducer X-direction vibration sensor 47, a planetary reducer Y-direction vibration sensor 48, a planetary reducer Z-direction vibration sensor 49, a drive pinion 50, a ring gear 51, a crane power system motor 52, a crane power system coupling 53, and a planetary reducer 54.

[0116] The four drive systems jointly drive the winches and share the same structure: a crane power system motor 52, a crane power system coupling 53, a planetary reduction gear 54, and a drive pinion 50. Each drive system is equipped with four sensors: a crane power system motor Z-axis vibration sensor 44, a planetary reduction gear X-axis vibration sensor 47, a planetary reduction gear Y-axis vibration sensor 48, and a planetary reduction gear Z-axis vibration sensor 49.

[0117] Each thruster contains seven temperature sensors, three vibration sensors, and one torque sensor: the first temperature sensor 7, the second temperature sensor 8, the third temperature sensor 9, the fourth temperature sensor 10, the fifth temperature sensor 11, the sixth temperature sensor 21, the seventh temperature sensor 26, the thruster X-axis vibration sensor 18, the thruster Y-axis vibration sensor 19, the thruster Z-axis vibration sensor 22, and the torque sensor 00. A wind turbine installation vessel is equipped with multiple thrusters, each with more than one set of sensors. Similar thrusters share a common information processor, so there is also the nth temperature sensor, the thruster X-axis vibration sensor n, the thruster Y-axis vibration sensor n, the thruster Z-axis vibration sensor n, and the nth torque sensor. The noisy signals measured by these sensors are input into the information processor for noise reduction and fault diagnosis. After the diagnosis is complete, the information is transmitted to the console display via bus 1.

[0118] Each leg gear reduction system contains four vibration sensors: a leg motor Z-axis vibration sensor 33, a leg gear reduction box X-axis vibration sensor 36, a leg gear reduction box Y-axis vibration sensor 37, and a leg gear reduction box Z-axis vibration sensor 38. A wind turbine installation vessel typically has four legs, each with multiple leg gear reduction systems. Each leg shares a common information processor, thus including a leg motor Z-axis vibration sensor n, a leg gear reduction box X-axis vibration sensor n, a leg gear reduction box Y-axis vibration sensor n, and a leg gear reduction box Z-axis vibration sensor n. The noisy signals measured by these sensors are input into the information processor for noise reduction and fault diagnosis. Once the diagnosis is complete, the signals are fed to the console display via bus 2.

[0119] Each crane power system includes four drive systems, each of which contains four vibration sensors: crane power system motor Z-axis vibration sensor 44, planetary gearbox X-axis vibration sensor 47, planetary gearbox Y-axis vibration sensor 48, and planetary gearbox Z-axis vibration sensor 49. A wind turbine installation vessel has multiple crane power systems. Similar crane power systems share a common information processor, including crane power system motor Z-axis vibration sensor n, planetary gearbox X-axis vibration sensor n, planetary gearbox Y-axis vibration sensor n, and planetary gearbox Z-axis vibration sensor n. The noisy signals measured by these sensors are fed into the information processor for noise reduction and fault diagnosis. After diagnosis, the signals are transmitted to the console display via bus 3.

[0120] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligently monitoring the state of a power transmission system of a wind power installation vessel, characterized in that: The thruster is provided with a temperature sensor, a vibration sensor and a torque sensor, and the pile leg gear reduction system and the crane power system are provided with a vibration sensor, including the following steps: s1. Collect the vibration signal, temperature signal and torque signal containing noise, represented by c(p), where p is the collected sensor signal, c(p)=a(p)+b(p)p=0, 1, 2..., N-1; Where a(p) is the effective signal of the sensor, and b(p) is the noise signal of the sensor; S2. Use wavelet basis to perform signal decomposition; decompose the original time domain signals of the vibration sensor, temperature sensor, and torque sensor to obtain low-frequency coefficients and high-frequency coefficients of each layer; S3. Filter the high frequency coefficients using an improved adaptive threshold function; S4. Reconstruct the signal through inverse transformation to obtain the denoised vibration signal, temperature signal, and torque signal; S5. Input the denoised signal in step S4 into the CNN model. For a set of samples, each sensor signal is input as a row to form a two-dimensional input layer as the input of the CNN model. S6. After the CNN training and verification are completed, the training parameters are saved to obtain the CNN diagnosis model; In step S3, the formula of the improved adaptive threshold function is: Where d j+1 [n] is the high frequency coefficient of the j+1th layer, φ j+1 is the adaptive threshold of the j+1th layer, h j+1 is the trend parameter of the j+1th layer, ω j+1 is the high frequency coefficient of the j+1th layer after processing; The rigrsure method is used to calculate the adaptive threshold φ of the jth layer based on the unbiased likelihood estimation principle. j , the trend parameter h can be determined by the following formula: In the formula, h j is the trend parameter of the jth layer, E j is the high frequency coefficient energy of the jth layer, ∑E j It is the sum of the energy of high-frequency coefficients of each layer. The signal measured by each sensor is denoised by an adaptive threshold function separately.

2. The method for intelligently monitoring the state of a power transmission system of a wind power installation vessel according to claim 1, characterized in that: In step S5, the sampling frequency of each sensor is fHz. For a set of samples, each denoised sensor signal is input as a row, which can form a two-dimensional input layer, that is, an initial two-dimensional signal I of i×f. a : The convolutional layer in the CNN model uses multiple output channels; the size of the convolution kernel is 1 row L c Columns, that is, the convolution kernel covers 1 row in the vertical direction and L in the horizontal direction c Column, the stride of the convolution kernel is 1 row L s The convolution operation moves L columns in the horizontal direction each time. s Columns, moving row by row in the vertical direction, so that each row is convolved separately, and a batch normalization layer is performed after each convolutional layer or fully connected layer and before the activation function is used.

3. The method for intelligently monitoring the state of a power transmission system of a wind power installation vessel according to claim 2, characterized in that: Use the pooling layer to reduce the dimension of the input data. The window size of the pooling layer is 1 row B c Column, the stride of the pooling layer is 1 row B s Column, the pooling operation moves B each time in the horizontal direction s Columns, move row by row in the vertical direction, that is, perform pooling operations independently on each row.

4. An intelligent monitoring system for the power transmission system of a wind power installation vessel, used to execute an intelligent monitoring method for the power transmission system of a wind power installation vessel according to any one of claims 1 to 3, characterized in that: It includes a signal acquisition module, an information processing module and a prediction output module; Signal acquisition module: The temperature sensor collects the temperature signal on the thruster; the vibration sensor collects the vibration signal of the thruster, pile leg gear reduction system and crane power system; the torque sensor collects the torque signal on the thruster; Information processing module: Use wavelet basis to decompose the signal, decompose the original time domain signals of vibration sensor, temperature sensor and torque sensor, and obtain low-frequency coefficients and high-frequency coefficients of each layer; use improved adaptive threshold function to filter high-frequency coefficients; reconstruct the signal through inverse transformation to obtain denoised vibration signal, temperature signal and torque signal; after CNN training and verification, save the training parameters to obtain CNN diagnostic model; Prediction output module: Use the trained CNN diagnosis model to output the fault type and location.

5. The intelligent monitoring system for the power transmission system status of a wind power installation vessel according to claim 4, characterized in that: After the fully connected layer, CNN generates probability values ​​through the Sigmoid activation function, and implements model training in combination with the improved loss function. During the training process, the model parameters are optimized by minimizing the loss function. The loss function is calculated as follows: Among them, loss batch is the average loss in a batch, z represents the number of samples, ψ represents the number of fault categories, η ψ is the fault weight, is the true label of sample z in category ψ, is the predicted probability of sample z in category ψ, diff ψ is the prediction difficulty penalty coefficient, θ is the minimum penalty limit value, and ζ is the maximum penalty limit value.

6. The intelligent monitoring system for the power transmission system status of a wind power installation vessel according to claim 4, characterized in that: The torque sensor on the first shaft of the thruster includes a conductive slip ring, an elastic beam, and a resistance strain gauge; the elastic beam is located on the upper part of the first shaft and above the oil suction port. The number of the elastic beams is p. Resistance strain gauges are attached to both sides of each elastic beam. The resistance strain gauge and the conductive slip ring form a circuit. When the first shaft rotates, one side of the elastic beam is the compression side and the other side is the tension side. The resistance strain gauge is compressed or stretched, resulting in a change in resistance. The torque-signal relationship curve equation is established by applying a known torque and recording a signal, and the p group of signals are converted into actual torque values.

Images