Wind turbine bearing fault diagnosis method and diagnosis system

Abstract

The present application discloses a wind turbine bearing fault diagnosis method and diagnosis system, which includes the following steps: acquiring bearing vibration data and establishing a data set; dividing the data set into a source domain and a target domain to obtain source domain sample data and target domain sample data; reconstructing the source domain sample data in phase space based on a chaos theory algorithm to generate a chaotic phase diagram; using the chaotic phase diagram as the input of a neural network to extract features of time and space scales; performing pre-training under multiple working conditions based on bearing samples to obtain a pre-trained model, and performing parameter migration on the pre-trained model based on a transfer learning strategy to obtain a CNN-DLSTM model after parameter migration; analyzing bearing fault information and matching fault type labels to obtain fault diagnosis results; converting strong nonlinear vibration signals into chaotic phase diagrams, and using deep learning methods to perform feature extraction and fault classification on the phase diagrams; and improving fault diagnosis accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a wind turbine bearing fault diagnosis method and system. Background Art

[0002] Wind energy is a key renewable energy source, and wind turbines have been widely installed worldwide in recent years. However, because most wind turbines are located in harsh environments such as mountaintops, fields, and oceans, their critical components—the bearings—are highly susceptible to failure. Bearing failures not only affect the normal operation of wind turbines but can also cause severe damage to the equipment, substantial economic losses, and even lead to safety accidents. Therefore, reliable bearing fault diagnosis is crucial for the operation and maintenance of wind turbines and holds high research value.

[0003] In recent years, with the rapid development of artificial intelligence (AI), deep learning technology has been widely used in various fields. In fault diagnosis, deep learning-based models can automatically extract features from vibration signals through training, significantly reducing labor costs compared to traditional methods.

[0004] However, many existing deep learning fault diagnosis models are based on sufficient bearing fault data. However, in real-world applications, since bearings operate normally most of the time, available fault data is very limited, leading to sample imbalance. Furthermore, wind turbine bearings are complex structures and often operate under variable and noisy conditions. Traditional signal decomposition methods often overlook key information, resulting in poor feature extraction results.

[0005] In summary, the current problems in the research on wind turbine bearing fault diagnosis are:

[0006] (1) Wind turbines operate normally most of the time, with only a few failures. This results in most data samples being collected under normal operating conditions. This phenomenon ultimately leads to the sample imbalance problem.

[0007] (2) Since current wind turbines are highly integrated with their operation in harsh environments, the original collected vibration signals have strong nonlinearity. Traditional signal decomposition methods often have poor results, resulting in poor fault diagnosis results. Summary of the Invention

[0008] The purpose of the present invention is to provide a wind turbine bearing fault diagnosis method, comprising the following steps:

[0009] S1, obtaining bearing vibration data, analyzing the bearing vibration data to determine different working conditions of the bearing, and establishing a data set;

[0010] S2, divide the data set into source domain and target domain, and obtain source domain sample data and target domain sample data;

[0011] S3, based on the chaos theory algorithm, reconstructs the phase space of the source domain sample data, transforms the one-dimensional time series into a chaotic series, and generates a chaotic phase diagram;

[0012] S4, taking the chaotic phase image as the input of the neural network, extracting the characteristics of time and space scales based on the neural network;

[0013] S5: Construct bearing samples based on the features of each size. Pre-train the bearing samples under multiple working conditions (such as different loads and speeds) to obtain a pre-trained model. Parameters of the pre-trained model are transferred based on the transfer learning strategy to obtain a CNN-DLSTM model after parameter transfer.

[0014] S6, based on the CNN-DLSTM model, analyzes the bearing fault information and matches the fault type label to obtain the fault diagnosis result.

[0015] Furthermore, the different operating condition information in step S1 includes one or more of different speeds, different loads, different fault locations or different fault depths.

[0016] Furthermore, in step S3, the one-dimensional time series is embedded into the m-dimensional space using the following formula to obtain the phase point Y:

[0017]

[0018] Where, embedding dimension m and delay time τ;

[0019] Secondly, the Lyapunov exponent method is used to determine the chaotic characteristics of the vibration signal. The formula is as follows:

[0020]

[0021] represents the maximum Lyapunov exponent; Represents the total time span, that is, from the initial time t0 to the end time t M the length of time; is the distance between two adjacent points in the phase space reconstruction, is the distance between two points at the next moment, and M is the total number of iterations.

[0022] Furthermore, in step S5, a deep long short-term memory network DLTSM is generated by superimposing multiple layers of LSTM. LSTM determines whether the information in the past memory unit needs to be retained based on the input unit X and output unit h at the previous moment. The output formula is as follows:

[0023] ;

[0024] Where, represents the output gate activation value, which is used to control the information flow from the memory unit to the hidden state; σ represents the Sigmoid activation function, which compresses the output to the [0,1] interval; w L4 Represents the weight matrix of the output gate, which is used to linearly transform the concatenated vector [h t-1 ,x t ], h t-1 represents the hidden state of the previous time step, i.e. short-term memory, x t represents the input vector of the current time step, b L4 Represents the bias term of the output gate, adjusting the activation threshold.

[0025] Furthermore, the input unit controls whether to use the storage unit value at time t to update the storage unit value of the next update unit state, and the formula is:

[0026]

[0027]

[0028] ;

[0029] Where i t The activation value of the forget gate at the current time step t, in the range [0,1], controls the previous memory unit C t-1 Information is discarded; w L2 Represents the weight matrix of the forget gate; represents the bias term of the forget gate, Represents the candidate memory unit of the current time step t, ranging from [-1, 1], storing new information, tanh is the hyperbolic tangent activation function, compressing the value to the interval [-1, 1], w L3 represents the weight matrix of the candidate memory unit, Represents the bias term of the candidate memory unit.

[0030] Furthermore, the output unit outputs the next hidden state value containing the previous input information to update the hidden state. The expression is:

[0031]

[0032]

[0033] ;

[0034] Where C t Represents the state of the memory unit at the current time step t, storing long-term information; Ct-1 represents the state of the memory cell at the previous time step, f t Represents the activation value of the forget gate, ranging from [0,1], generated by the Sigmoid function, which determines how much information of the previous memory unit is retained, h t Represents the hidden state of the current time step t, which is output by the network and passed to the next time step.

[0035] Furthermore, the transfer learning strategy in step S5 includes one of layer-by-layer unfreezing, bottom-up unfreezing, or top-down unfreezing.

[0036] The present invention also provides a wind turbine bearing fault diagnosis system, comprising a processor, a memory, and at least one program, wherein the program is stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the wind turbine bearing fault diagnosis method as described in any one of the above items.

[0037] The present invention further provides a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute to implement any one of the above-mentioned wind turbine bearing fault diagnosis methods.

[0038] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art:

[0039] The present invention converts strong nonlinear vibration signals into chaotic phase diagrams, and then uses deep learning methods to extract features and classify faults from the phase diagrams; finally, transfer learning is used to realize fault diagnosis of wind turbine bearings under multiple working conditions, thereby improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram showing a flow chart of a wind turbine bearing fault diagnosis method provided by an embodiment of the present invention;

[0041] Figure 2 shows the chaotic phase diagram obtained by phase space reconstruction provided by this embodiment;

[0042] Figure 3 The confusion matrix of the diagnosis accuracy provided by this embodiment is shown. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0044] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0045] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0046] like Figure 1-Figure 3 As shown, an embodiment of the present invention provides a method for diagnosing bearing faults of a wind turbine generator set, comprising the following steps:

[0047] S1, obtaining bearing vibration data, analyzing the bearing vibration data to determine the different working conditions of the bearing, and establishing a data set, as shown in Table 1;

[0048]

[0049] Table 1

[0050] S2, divide the data set into source domain and target domain, and obtain source domain sample data and target domain sample data;

[0051] S3, based on the chaos theory algorithm, reconstructs the phase space of the source domain sample data, transforms the one-dimensional time series into a chaotic series, and generates a chaotic phase diagram;

[0052] S4, taking the chaotic phase image as the input of the neural network, extracting the characteristics of time and space scales based on the neural network;

[0053] Specifically, the convolutional neural network is an "end-to-end" data processing method that extracts topological features from the data structure by convolution and pooling the input sample data layer by layer. At the same time, the convolutional neural network has the advantages of sparse connection and parameter sharing, which can reduce the number of model training parameters, avoid algorithm overfitting, and reduce data dimension. Therefore, the present invention chooses to use a convolutional neural network to perform feature extraction on the chaotic phase diagram generated in step S3. The neural network will extract local texture, edge and periodic pattern features at the spatial scale, as well as transient response and long-term evolution features at the time scale. These features can fully characterize the chaotic characteristics of the vibration signal and are used to accurately identify the fault type. Through the above-mentioned feature extraction process, early weak faults and significant damage can be captured simultaneously, thereby improving the accuracy of later fault diagnosis.

[0054] S5: Construct bearing samples based on the features of each size. Pre-train the bearing samples under multiple working conditions (such as different loads and speeds) to obtain a pre-trained model. Parameters of the pre-trained model are transferred based on the transfer learning strategy to obtain a CNN-DLSTM model after parameter transfer.

[0055] Specifically, the core idea of ​​transfer learning is to train a model using data from the source domain and fine-tune the pre-trained model based on a small sample of data from the target dataset. This model can then handle the target domain task without requiring a large amount of target data. The source and target problems need to be similar, but their differences cannot be directly represented. This behavior enables transfer learning models to extract specific and common essential features between the source and target problems.

[0056] In fault diagnosis, transfer learning models can analyze and learn various fault characteristics from the source domain dataset, thereby more accurately analyzing similar diagnostic problems in the target domain, ultimately improving the performance and efficiency of the model. Specifically, the prediction layer of the model pre-trained using the dataset under specific working conditions is replaced by a new fully connected layer with randomly generated weights. This fully connected layer is used to fine-tune the identification of bearing fault types under other working conditions.

[0057] S6, based on the CNN-DLSTM model, analyzes the bearing fault information and matches the fault type label to obtain the fault diagnosis result.

[0058] According to an embodiment of the present invention, the different operating condition information in step S1 includes one or more of different speeds, different loads, different fault locations, or different fault depths.

[0059] According to an embodiment of the present invention, in step S3, the one-dimensional time series is embedded into the m-dimensional space using the following formula to obtain the phase point Y:

[0060]

[0061] Where, embedding dimension m and delay time τ;

[0062] Secondly, the Lyapunov exponent method is used to determine the chaotic characteristics of the vibration signal. The formula is as follows:

[0063]

[0064] represents the maximum Lyapunov exponent; Represents the total time span, that is, from the initial time t0 to the end time t M the length of time; is the distance between two adjacent points in the phase space reconstruction, is the distance between two points at the next moment, and M is the total number of iterations.

[0065] Specifically, using chaos theory to reconstruct the phase space of the original one-dimensional time series yields two parameters: the embedding dimension m and the delay time τ. These two parameters are used to construct the phase points. The M-dimensional space corresponds to the reconstructed phase space sequence, not the two-dimensional chaotic sequence.

[0066] The role of phase points: The trajectories of phase points (i.e., attractors) preserve the geometric and topological properties of the original system (such as fractal dimension and Lyapunov exponent), making unobservable multivariate dynamics visible and analyzable.

[0067] The main methods for selecting τ include the autocorrelation function method and the mutual information method. The mutual information method has been widely used because it simultaneously considers both linear and nonlinear correlations between data. Therefore, the present invention uses the mutual information method and the Cao method to calculate the delay time and embedding dimension, respectively.

[0068] Furthermore, the vibration signals of healthy machinery typically exhibit periodicity or stable randomness, while faults can introduce nonlinear perturbations, causing the signal to become chaotic. Analyzing chaotic characteristics primarily reveals the dynamic behavior changes hidden within the vibration signal, allowing for accurate identification of fault conditions.

[0069] According to an embodiment of the present invention, in step S5, a deep long short-term memory network DLTSM is generated by superimposing multiple layers of LSTM. The LSTM determines whether the information in the past memory unit needs to be retained based on the input unit X and output unit h at the previous moment. The output formula is as follows:

[0070] ;

[0071] Where, represents the output gate activation value, which is used to control the information flow from the memory unit to the hidden state; σ represents the Sigmoid activation function, which compresses the output to the [0,1] interval; w L4Represents the weight matrix of the output gate, which is used to linearly transform the concatenated vector [h t-1 ,x t ], h t-1 represents the hidden state of the previous time step, i.e. short-term memory, x t represents the input vector of the current time step, b L4 Represents the bias term of the output gate, adjusting the activation threshold.

[0072] Specifically, as the depth of the neural network increases, the overall performance of the neural network improves as a whole. Therefore, the present invention generates and adopts a deep long short-term memory network (DLTSM) by superimposing multiple layers of LSTM. The same LSTM layer is time-corrected, and the output of the previous LSTM layer will be looped into this layer as input. Different LSTM layers are spatially correlated, and data is output from the previous layer to the next layer. In this way, functions can be built at a lower level and then combined at a higher level. Each LSTM layer contains a large number of neurons that collect long-term dependencies in the signals. In each LSTM layer, neurons also exchange information with each other. DLSTM is more compact than LSTM and has improved generalization ability.

[0073] According to an embodiment of the present invention, the input unit controls whether to use the storage unit value at time t to update the storage unit value of the next update unit state, and the formula is:

[0074]

[0075]

[0076] ;

[0077] Where i t The activation value of the forget gate at the current time step t, in the range [0,1], controls the previous memory unit C t-1 Information is discarded; w L2 Represents the weight matrix of the forget gate; represents the bias term of the forget gate, Represents the candidate memory unit of the current time step t, ranging from [-1, 1], storing new information, tanh is the hyperbolic tangent activation function, compressing the value to the interval [-1, 1], w L3 represents the weight matrix of the candidate memory unit, Represents the bias term of the candidate memory unit.

[0078] According to an embodiment of the present invention, the output unit outputs the next hidden state value containing the previous input information to update the hidden state, which is expressed as:

[0079]

[0080]

[0081] ;

[0082] Where C t Represents the state of the memory unit at the current time step t, storing long-term information; C t-1 represents the state of the memory cell at the previous time step, f t Represents the activation value of the forget gate, ranging from [0,1], generated by the Sigmoid function, which determines how much information of the previous memory unit is retained, h t Represents the hidden state of the current time step t, which is output by the network and passed to the next time step.

[0083] According to an embodiment of the present invention, the transfer learning strategy in step S5 includes one of layer-by-layer unfreezing, bottom-up unfreezing, or top-down unfreezing.

[0084] Specifically, layer-by-layer unfreezing: After training begins, a fully connected layer is added and the remaining layers are frozen. After a period of training, the next layer is released and trained again for another period of time. This process is repeated until all layers are trainable.

[0085] Bottom-up unfreezing: Unfreeze the network layers from bottom to top. Initially, all layers are frozen except for the single layer being fine-tuned. Each training process is performed 50 times. After optimizing each layer, the best-performing model is selected for the next round of fine-tuning. All layers are unfrozen in the final round of fine-tuning.

[0086] Top-down thawing: Similar to the bottom-up thawing method. The only difference is that the order of thawing each layer changes from high to low.

[0087] After testing and comparison, we chose bottom-up thawing as the migration strategy to improve the effect of transfer learning.

[0088] In summary, the present invention converts strong nonlinear vibration signals into chaotic phase diagrams, and then uses deep learning methods to extract features and classify faults from the phase diagrams. Finally, transfer learning is used to realize fault diagnosis of wind turbine bearings under multiple working conditions, thereby improving the accuracy of fault diagnosis.

[0089] This embodiment also provides a wind turbine bearing fault diagnosis system, including a processor, a memory and at least one program, the program is stored in the memory and is configured to be executed by the processor, and the program includes instructions for executing any of the above-mentioned wind turbine bearing fault diagnosis methods.

[0090] The present invention also provides a computer-readable storage medium, which stores a computer program. The computer program enables a computer to execute to implement any of the above-mentioned wind turbine bearing fault diagnosis methods.

[0091] Those skilled in the art will appreciate that, for ease of explanation, the following example illustrates a configuration in which one memory and one processor are provided. In an actual terminal or server, multiple processors and memories may exist. A memory may also be referred to as a storage medium or storage device, etc., which is not limited in the present embodiment.

[0092] It should be understood that in the embodiments of the present application, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor may also be a general-purpose microprocessor, a graphics processing unit (GPU), or one or more integrated circuits for executing relevant programs to implement the functions required to be executed in the embodiments of the present application.

[0093] The processor can also be an integrated circuit chip with signal processing capabilities. During the implementation process, the various steps of the present application can be completed by the integrated logic circuit of the hardware in the processor or by instructions in the form of software. The above-mentioned processor can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the art such as random access memory, flash memory and read-only memory, programmable read-only memory or electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor reads the information in the memory and, in combination with its hardware, completes the functions required to be performed by the units included in the method, device and storage medium of the embodiments of the present application.

[0094] It should also be understood that the memory mentioned in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache.

[0095] By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM).

[0096] The memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may be independent and connected to the processor via a bus. The memory may also be integrated with the processor, and the memory may store a program. When the program stored in the memory is executed by the processor, the processor is used to execute the various steps of the determination method in the above-mentioned embodiment of the present application.

[0097] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) is integrated into the processor. It should be noted that the memory described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0098] It should be understood that the term "and / or" in this document simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0099] During implementation, each step of the above method can be completed by an integrated logic circuit of the hardware in the processor or an instruction in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0100] Those skilled in the art will appreciate that the various illustrative logical blocks (ILBs) and steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0101] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer-programmed program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a processor, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer may be a general-purpose computer, a computer network, or other programmable device.

[0102] This embodiment further provides a computer-readable storage medium, which stores a computer program. The computer program enables a computer to execute to implement the above-mentioned wind turbine bearing fault diagnosis method.

[0103] It should be noted that computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via wired (e.g., coaxial cable, optical fiber) or wireless (e.g., infrared, wireless, microwave, etc.) means, or can be transmitted from one website, computer, server or data center to a mobile phone processor via wired means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media. Available media can be magnetic media (e.g., floppy disk, hard disk), optical media (e.g., DVD), or semiconductor media (e.g., solid-state drive), etc.

[0104] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent replacements for some of the technical features therein. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A wind turbine bearing fault diagnosis method, characterized in that: The following steps are involved: S1, obtaining bearing vibration data, analyzing the bearing vibration data to determine different working conditions of the bearing, and establishing a data set; S2, divide the data set into source domain and target domain, and obtain source domain sample data and target domain sample data; S3, based on the chaos theory algorithm, reconstructs the phase space of the source domain sample data, transforms the one-dimensional time series into a chaotic series, and generates a chaotic phase diagram; S4, taking the chaotic phase image as the input of the neural network, extracting the characteristics of time and space scales based on the neural network; S5: Construct bearing samples based on the features of each size, perform pre-training under multiple working conditions based on the bearing samples to obtain a pre-trained model, and perform parameter migration on the pre-trained model based on the transfer learning strategy to obtain a CNN-DLSTM model after parameter migration. S6, analyzes bearing fault information based on the CNN-DLSTM model, matches fault type labels, and obtains fault diagnosis results; In step S5, a deep long short-term memory network DLTSM is generated by superimposing multiple layers of LSTM. LSTM determines whether the information in the past memory unit needs to be retained based on the input unit X and output unit h at the previous moment. The output formula is as follows: ； Where, represents the output gate activation value, which is used to control the information flow from the memory unit to the hidden state; σ represents the Sigmoid activation function, which compresses the output to the [0,1] interval; w L4 Represents the weight matrix of the output gate, which is used to linearly transform the concatenated vector [h t−1 ,x t ], h t−1 represents the hidden state of the previous time step, i.e. short-term memory, x t represents the input vector of the current time step, b L4 Represents the bias term of the output gate, adjusting the activation threshold; The input unit controls whether to use the storage unit value at time t to update the storage unit value of the next update unit state. The formula is: ； ； Where i t The activation value of the forget gate at the current time step t, in the range [0,1], controls the previous memory unit C t-1 Information is discarded; w L2 Represents the weight matrix of the forget gate; represents the bias term of the forget gate, Represents the candidate memory unit of the current time step t, ranging from [-1, 1], storing new information, tanh is the hyperbolic tangent activation function, compressing the value to the interval [-1, 1], w L3 represents the weight matrix of the candidate memory unit, Represents the bias term of the candidate memory unit; The output unit outputs the next hidden state value containing the previous input information to update the hidden state. The expression is: ； ； ； Where C t Represents the state of the memory unit at the current time step t, storing long-term information; C t−1 represents the state of the memory cell at the previous time step, f t Represents the activation value of the forget gate, ranging from [0,1], generated by the Sigmoid function, which determines how much information of the previous memory unit is retained, h t Represents the hidden state of the current time step t, which is output by the network and passed to the next time step.

2. The wind turbine bearing fault diagnosis method according to claim 1, characterized in that: The different operating condition information in step S1 includes one or more of different speeds, different loads, different fault locations or different fault depths.

3. The wind turbine bearing fault diagnosis method according to claim 2, characterized in that: In step S3, the one-dimensional time series is embedded into the m-dimensional space using the following formula to obtain the phase point Y: ； Where, embedding dimension m and delay time τ; Secondly, the Lyapunov exponent method is used to determine the chaotic characteristics of the vibration signal. The formula is as follows: ； represents the maximum Lyapunov exponent; Represents the total time span, that is, from the initial time t0 to the end time t M the length of time; is the distance between two adjacent points in the phase space reconstruction, is the distance between two points at the next moment, and M is the total number of iterations.

4. The wind turbine bearing fault diagnosis method according to claim 1, wherein: The transfer learning strategy in step S5 includes one of layer-by-layer unfreezing, bottom-up unfreezing, or top-down unfreezing.

5. A wind turbine bearing fault diagnosis system, characterized in that: The system comprises a processor, a memory and at least one program, wherein the program is stored in the memory and configured to be executed by the processor, and the program comprises instructions for executing the wind turbine bearing fault diagnosis method according to any one of claims 1 to 4.

Images