Insulated bearing fault diagnosis method based on residual error space-time information fusion model

Through the residual space-time information fusion model and bidirectional WDRU neural network empowered by graph knowledge, combined with the scavenger algorithm to optimize hyperparameters, the diagnosis problem of electrical corrosion failure of insulated bearings of wind turbines is solved, and more efficient fault identification and classification is achieved Effect.

CN119089268AActive Publication Date: 2024-12-06QINGDAO FAMBITION HEAVY MASCH CO LTD

Patent Information

Application Number
CN202411190075.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-06
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

The prior art is difficult to accurately identify the electrical corrosion fault of the insulated bearing of wind turbines under high-power frequency conversion control, and the deep learning model has problems of low efficiency and poor accuracy in hyperparameter adjustment and timing information extraction.

Method used

Using the residual space-time information fusion model based on graph knowledge empowerment, a new graph structure and a bidirectional WDRU neural network are constructed, combined with the spatial information fusion module and the time information fusion module, and the scavenger algorithm is used to optimize hyperparameters to realize insulated bearings. Intelligent diagnosis of vibration signals.

Benefits of technology

It improves the accuracy and generalization ability of insulated bearing fault diagnosis, can more effectively dig time sequence information, reduces the randomness and uncertainty of the model, and improves the recognition accuracy and classification performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089268A_ABST
    Figure CN119089268A_ABST
Patent Text Reader

Abstract

The invention discloses an insulated bearing fault diagnosis method based on a residual space-time information fusion model, and the method comprises the steps: collecting the original vibration signals of six channels of an insulated bearing of a doubly-fed wind generator through two three-way vibration sensors, so as to capture the information of the insulated bearing in different health states; dividing the original vibration signal into 80% of training set data and 20% of test set data; constructing a graph knowledge enabled residual space-time information fusion model; hyper-parameters of the graph knowledge enabled residual space-time information fusion model are optimized by adopting a scavenger algorithm; training the model by taking the training set data as the input of a residual space-time information fusion model endowed by graph knowledge and taking the fault category as the output; and inputting test set data into a trained graph knowledge enabled residual space-time information fusion model, and performing fault classification on fused feature data through an automatic space-time feature residual learning process to realize intelligent diagnosis of the doubly-fed wind generator insulated bearing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of bearing fault diagnosis, and relates to an insulating bearing fault diagnosis method based on a residual space-time information fusion model. Background Art

[0002] Wind turbines are known as the "heart" of wind turbines, and insulating bearings are crucial components of the generator's supporting transmission structure. However, since high-power variable-frequency controlled generators will break through the oil film to generate shaft current discharge, the bearing fatigue wear process will be drastically shortened. According to statistics, high-power variable-frequency motor bearings all have this failure form. In addition, the operating conditions of wind turbines are complex and changeable, making them prone to failure. Once the bearing fails, it will affect the safe operation of the entire wind turbine system and even cause huge economic losses. Therefore, the wind turbine bearing fault diagnosis method is studied to timely monitor and diagnose bearing faults, and to formulate inspection and maintenance strategies to ensure the safe and stable operation of wind turbines.

[0003] In recent years, deep learning can fit actual physical models with powerful nonlinear models, explore nonlinear relationships between variables, and achieve end-to-end fault diagnosis. For example, convolutional neural networks rely on unique parameter sharing and local connection mechanisms, which can achieve important translation invariance in Euclidean space. However, most scenes in the real world have clear irregular topological structures, also known as non-Euclidean data, which makes traditional neural network models for Euclidean data difficult to handle. Graph neural networks can collect information from the neighbors of nodes and use it to update their own implicit states, thereby realizing the propagation of high-order structural information, which can usually significantly improve the expressive power of the model. However, due to the over-correlation phenomenon between different feature dimensions of traditional graph network models, it is easy to encounter the problem of high information redundancy of neighbor gain and low content of important information.

[0004] At the same time, considering that the damage mechanism of the insulated bearings of megawatt-class wind turbines with high-power frequency conversion control is unclear under the interaction of shaft current and fatigue wear, the identification of their electrocorrosion faults has become a technical problem that needs to be overcome urgently. Considering that the characteristics of insulated bearing electrocorrosion faults are affected by factors such as mechanical loads, bearing structural parameters, and electrical performance parameters, the insulated bearing data has strong complexity, nonlinearity, and time-varying properties. However, the existing model has weak time series information extraction capabilities and is difficult to fully mine the characteristic information of electrocorrosion faults.

[0005] Considering that the deep learning model is composed of multiple neurons, the parameter space of the model is usually very large, containing a large number of weights and biases. In particular, the hyperparameters of the deep learning model usually require multiple attempts and repeated experiments to adjust the model parameters through manual trial and error, which makes it have strong randomness and uncertainty, resulting in low efficiency in hyperparameter adjustment of the deep learning model and poor recognition accuracy and generalization performance of the model. Summary of the invention

[0006] In order to solve the above technical problems, the purpose of the present invention is to provide an insulating bearing fault diagnosis method based on a residual space-time information fusion model.

[0007] The present invention provides an insulating bearing fault diagnosis method based on a residual space-time information fusion model, comprising:

[0008] Step 1: Use two three-way vibration sensors to collect the original vibration signals of six channels of the insulating bearing of the doubly-fed wind turbine generator to capture the information of the insulating bearing in different health states;

[0009] Step 2: Divide the original vibration signal into 80% training set data and 20% test set data;

[0010] Step 3: Construct a residual space-time information fusion model empowered by graph knowledge;

[0011] Step 4: Use the scavenger algorithm to optimize the hyperparameters of the residual space-time information fusion model empowered by graph knowledge;

[0012] Step 5: Use the training set data as the input of the residual space-time information fusion model empowered by graph knowledge, and the fault category as the output to perform model training;

[0013] Step 6: Input the test set data into the trained graph knowledge-enabled residual space-time information fusion model. After the automatic space-time feature residual learning process, the fused feature data is classified into fault categories to achieve intelligent diagnosis of the insulating bearings of the doubly-fed wind turbine.

[0014] The insulated bearing fault diagnosis method based on the residual space-time information fusion model of the present invention has at least the following beneficial effects:

[0015] Considering the diagnosis problem of electrical corrosion faults in insulated bearings of megawatt-class wind turbines controlled by high-power frequency conversion, this paper develops a graph knowledge-enabled residual space-time information fusion model in a unique way, which aims to solve the problem that electrical corrosion faults caused by shaft current damage of insulated bearings cannot be accurately identified in actual engineering. The main conclusions are as follows:

[0016] (1) Inspired by the biological neuron structure, a new graph structure is innovatively designed, and the maximum mutual information coefficient formula is derived. A new adjacency matrix is ​​constructed and the validity of the matrix is ​​proved. Based on this graph structure, a spatial information fusion module is designed, which can calculate the non-Euclidean distance relationship between multi-channel insulating bearing vibration signals, making the residual space-time information fusion model framework have strong reasoning and expression capabilities. It provides a powerful modeling tool for the processing and analysis of unstructured data, and also provides a new perspective for the design of learning algorithms such as graph neural networks.

[0017] (2) Considering the strong time-varying nature of the time series characteristics of the wind turbine insulation bearing data, a bidirectional WDRU neural network is proposed, and the back propagation formula is updated. It is cleverly applied to the proposed time information fusion module to deeply mine the time series information of the insulation bearing. The features extracted by the spatial information fusion module and the time information fusion module are fused to achieve information complementarity.

[0018] (3) In the insulated bearing fault classification stage, the number of hidden layer neurons, learning rate, regularization coefficient and training times of the residual space-time information fusion model enabled by graph knowledge are optimized by utilizing the characteristics of fast convergence speed and strong optimization ability of the scavenger algorithm. This makes the residual space-time information fusion model enabled by graph knowledge have better classification performance in dealing with the classification problem of time series. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of an insulating bearing fault diagnosis method based on a residual space-time information fusion model of the present invention;

[0020] Figure 2 is a schematic diagram of a weight reduction recurrent unit;

[0021] Figure 3 It is a schematic diagram of a bidirectional WDRU neural network. DETAILED DESCRIPTION

[0022] like Figure 1 As shown, the present invention provides an insulating bearing fault diagnosis method based on a residual space-time information fusion model, comprising:

[0023] Step 1: Use two three-way vibration sensors to collect the original vibration signals of the six channels of the insulating bearing of the doubly-fed wind turbine generator to capture the information of the insulating bearing in different health states.

[0024] Step 2: Divide the original vibration signal into 80% training set data and 20% test set data.

[0025] Step 3: Construct a residual spatio-temporal information fusion model empowered by graph knowledge.

[0026] The graph knowledge-enabled residual space-time information fusion model includes: a spatial information fusion module, a temporal information fusion module and a self-attention module; the original vibration signals of 6 channels are respectively input into the spatial information fusion module and the temporal information fusion module to extract local features and global features, the local features and the global features are fused in the self-attention module to obtain fused features, the fused features are input into the fully connected layer to construct a mapping relationship between features and fault modes and output the diagnosis results.

[0027] In specific implementation, the spatial information fusion module includes: a first hidden layer and a second hidden layer. The first hidden layer includes: a first graph convolution layer and a first activation layer, and the first graph convolution layer includes 64 graph convolution kernels; the second hidden layer includes: a second graph convolution layer, a batch normalization layer, a second activation layer and a discard layer, and the second graph convolution layer includes 128 graph convolution kernels; each graph convolution kernel performs weighted summation on a central vertex and its surrounding 23 adjacent nodes; the original vibration signals of the six channels are sequentially input into the first graph convolution layer, the first activation layer, the second graph convolution layer, the batch normalization layer, the second activation layer and the discard layer, and the local features are output.

[0028] Graph convolutional networks can merge spatial information through graph convolution. In particular, graph neural networks have shown excellent capabilities in modeling the features and structural information of graphs. However, when graph neural networks capture more information from high-order neighbors, there is a new phenomenon called overcorrelation. Overcorrelation causes the captured representation to contain a large amount of redundant information and causes the loss of some useful information, thereby reducing the expressive power of the model. Therefore, in order to solve the technical challenges of the above-mentioned graph network model, and at the same time to fully mine the wind turbine insulation bearing data with non-Euclidean spatial relationships generated based on three-way vibration sensors. The present invention introduces a graph convolutional network based on the spatial domain to describe the non-Euclidean spatial relationship between two six directions based on three-way vibration sensors. First, the Euclidean data is converted into graph data considering the nearest neighbor relationship, and a spatial information fusion module is obtained through graph convolution. The operation process of this module is to learn the interaction between attribute features by performing affine transformation on the attribute information, and then iteratively aggregate the features of neighbor vertices to update the features of the current vertex. With each additional convolution operation of the graph, the central vertex can fuse more information from external vertices.

[0029] Calculating the correlation matrix is ​​a prerequisite for constructing the adjacency matrix. The adjacency matrix is ​​used to store information in the set of edges between nodes in the graph data, and is a two-dimensional array. In the present invention, the adjacency matrix of the graph convolution layer in the first hidden layer and the second hidden layer is constructed, and the output of the graph convolution kernel is obtained, which specifically includes the following steps:

[0030] (1) Calculate the maximum mutual information coefficient of the original vibration signal sequence of adjacent channels:

[0031]

[0032]

[0033]

[0034]

[0035] Among them, X and Y represent the original vibration signal sequences of the two channels respectively, and M IC (X, Y) is the maximum mutual information coefficient between the X sequence and the Y sequence, I(X; Y) represents the mutual information between the X sequence and the Y sequence, H(X) represents the entropy of the X sequence, and H(Y) represents the entropy of the Y sequence; p(x, y) represents the joint probability of the elements in the X sequence and the Y sequence, p(x) represents the probability of an element appearing in the X sequence, and p(y) represents the probability of an element appearing in the Y sequence; p(x i ) represents the probability of the i-th element in the X sequence, p(y i ) represents the probability of the i-th element in the Y sequence appearing, I(x i ) represents the entropy of the i-th element in the X sequence, I(y i ) represents the entropy of the i-th element in the Y sequence, and n represents the number of elements contained in the sequence;

[0036] In order to better quantify the value range of the mutual information coefficient between the original vibration signal sequences of any two channels, when the original vibration signal sequences of the two channels are the same, the maximum self-information between the two channels is set to 1, so that the relationship between the two channels in the adjacency matrix is ​​in the range of [0,1].

[0037] When the X sequence is the same as the Y sequence, it is a self-information relationship. Now we prove it as follows:

[0038] p(x,y)=p(x)=p(y)

[0039]

[0040] therefore:

[0041]

[0042] (2) Use the maximum mutual information coefficient to construct the adjacency matrix G of the graph convolution layer in the first hidden layer and the second hidden layer as follows:

[0043]

[0044] Among them, Ax, Ay, Az, Bx, By and Bz represent the original vibration signal sequences of 6 channels respectively;

[0045] (3) Using Laplace to calculate the normalized adjacency matrix

[0046]

[0047]

[0048] Where E is the identity matrix, yes The degree matrix of

[0049] (4) Get the output of the l+1th layer graph convolution kernel:

[0050]

[0051] Where: H l is the output of the l-th layer graph convolution kernel; f(·) is the nonlinear activation function; W l+1 is the trainable weight matrix of the l+1th layer.

[0052] The gated recurrent unit is a variant of the recurrent neural network, which can better capture the dependencies with large time step distances in the time series. Inspired by the improved idea of ​​the gated recurrent unit, the weight variables of the LSTM model are uniformly controlled by a gate structure weight, and the weight reduction recurrent unit (WDRU) is proposed. It can significantly reduce the number of weights that need to be optimized in the neural network, thereby ensuring that the model training time for data is reduced without reducing the classification accuracy of the model.

[0053] In specific implementation, the time information fusion module includes: an input layer, a third hidden layer and an output layer. Figure 2 As shown, the third hidden layer includes a bidirectional WDRU neural network. The bidirectional WDRU neural network performs sequential and reverse propagation to extract global features, which can better capture the bidirectional dependencies in time series data. The forward WDRU processes the input sequence from the beginning to the end, while the backward WDRU processes it in the opposite direction. The WDRU neural network includes multiple weight reduction recurrent units. The schematic diagram of the weight reduction recurrent unit is shown in Figure 3 The main design idea is to reduce the gate structure to a shared gate, which effectively simplifies the framework structure and improves the model training speed.

[0054] The calculation process of the bidirectional WDRU neural network is as follows:

[0055] (1) Calculate the information status of the shared gate:

[0056] f t =W d ·[h t-1 ,xt ]+b d

[0057] D t =σ(f t )=σ(W d ·[h t-1 ,x t ]+b d )=σ(W d ·h t-1 +W d ·x t +b d )

[0058] Q t =tan(f t )=tan(W d ·[h t-1 ,x t ]+b d )=tan(W d ·h t-1 +W d ·x t +b d )

[0059] Among them, f t represents the intermediate variable of the weight reduction recurrent unit, W d is the weight parameter of the weight reduction recurrent unit, h t-1 represents the output of the previous weight reduction recurrent unit, x t represents the original vibration signal of a channel, b d represents the bias parameter of the weight reduction recurrent unit; D t represents a shared gate, Q t Indicates the information status of the shared door;

[0060] (2) Update the control parameters of the information state of the weight reduction cycle unit: C t =D t ×C t-1 +(1-D t )×Q t =σ(W d ·[h t-1 ,x t ]+b d )×C t-1 +(1-σ(f t ))×tan(f t )

[0061] Among them, C t-1 The control parameter representing the information state of the previous weight reduction recurrent unit; C tA control parameter representing the information state of the current weight reduction recurrent unit;

[0062] (3) Calculate the output of the weight reduction recurrent unit:

[0063] h t =D t ×tan(C t )=σ(W d ·h t-1 +W d ·x t +b d )×tan(C t )

[0064] (4) Calculate the output of the bidirectional WDRU neural network:

[0065] y t =σ(h t )=σ(D t ×tan(C t ))

[0066] (5) The update formula of the bidirectional WDRU neural network is as follows:

[0067] h t =W Ft ×h Ft +W Bt ×h Bt +b t

[0068] Among them, W Ft and W Bt Represent the weight variables in the forward and backward propagation processes, b t Indicates the offset, h Ft represents the forward information state, h Bt Indicates the reverse information status,

[0069] The back propagation formula of the bidirectional WDRU neural network is as follows:

[0070] (6) The minimum optimization goal of the time information fusion module is:

[0071]

[0072] Among them, E t represents the error value of the tth period, Y t represents the output of the time information fusion module; W y represents the weight parameter of the weight reduction recurrent unit; b y represents the bias parameter of the weight reduction recurrent unit;

[0073] (7) Variables of the output layer of the temporal information fusion module:

[0074]

[0075]

[0076]

[0077]

[0078] Among them, P t represents the output of the current time weight reduction cycle unit; δy t Represents the change in the output of the bidirectional WDRU neural network model; δP t The current time weight reduction cycle unit output change; δW y Indicates the change in the weight parameter of the bidirectional WDRU neural network; δb y Represents the change in bias parameters of the bidirectional WDRU neural network;

[0079] (8) Calculate the weight change during the back propagation process of the hidden layer of the time information fusion module:

[0080]

[0081]

[0082]

[0083]

[0084]

[0085] Among them, δh t δC t δD t δQ t δf t w h Both represent the change in the weight of the network model during the back propagation process, and T represents the period of the time series signal.

[0086] The Dropout mechanism is added to the hidden layer of the temporal information fusion module for module optimization, so that neurons have a certain probability of inactivation, which makes the module have stronger generalization ability and avoids overfitting of the module. At the same time, the activation function in this module selects the ReLU function, so that the constructed graph knowledge-enabled residual space-time information fusion model has strong anti-interference ability and regularity, while avoiding the problem of gradient disappearance.

[0087] Step 4: Use the scavenger algorithm to optimize the hyperparameters of the residual space-time information fusion model enabled by graph knowledge, so as to improve the recognition accuracy and generalization ability of the model. The hyperparameters include: the number of neurons in the first hidden layer, the second hidden layer and the third hidden layer of the residual space-time information fusion model enabled by graph knowledge, the learning rate, the regularization coefficient, the number of iterations and the batch size.

[0088] In the specific implementation, the scavenger algorithm is used to optimize the hyperparameters of the residual space-time information fusion model empowered by graph knowledge, specifically:

[0089] (1) Initializing the graph knowledge enabled residual space-time information fusion model, including setting the initial number of neurons in the hidden layer of the graph knowledge enabled residual space-time information fusion model.

[0090] (2) Set the learning rate, number of iterations, and batch size range of the residual spatiotemporal information fusion model enabled by graph knowledge. Also set the scavenger target tracking speed coefficient, scavenger population size, and maximum number of iterations.

[0091] (3) Calculate the fitness values ​​of all scavenger individuals, sort them according to their fitness, select the individual with the best fitness, and select the mean square error as the fitness function of the scavenger algorithm. MSE is expressed as:

[0092]

[0093] Where n is the number of samples, Y i With Y i * are the true value and predicted value of the i-th sample data respectively.

[0094] The smaller the MSE value, the smaller the gap between the predicted value and the true value of the residual space-time information fusion model enabled by graph knowledge, which means that the position of the scavenger individual is better; the position of the scavenger individual corresponds one-to-one to the hyperparameters of the residual space-time information fusion model enabled by graph knowledge. The position coordinates of the scavenger individual with the optimal fitness are the optimal hyperparameters of the residual space-time information fusion model enabled by graph knowledge.

[0095] (4) Determine whether the scavenger algorithm iteration has ended. When the maximum number of iterations is reached, the algorithm iteration is completed. At this time, the position of the scavenger individual with the best fitness is assigned to the residual space-time information fusion model empowered by graph knowledge, that is, the best hyperparameters found by the scavenger algorithm are assigned to the residual space-time information fusion model empowered by graph knowledge. If the scavenger algorithm iteration has not ended, return to the previous step.

[0096] Step 5: Use the training set data as the input of the residual space-time information fusion model enabled by graph knowledge, and the fault category as the output to perform model training.

[0097] In the specific implementation, the training set data is used as the input of the residual space-time information fusion model enabled by graph knowledge, and the fault category is used as the output to complete the feature extraction and fusion of the original vibration data of the insulating bearing. The loss function of the Adam optimizer and the Softmax classifier is selected. When the set number of iterations is reached, the training is stopped, the fault type is output, and the model training is completed.

[0098] Step 6: Input the test set data into the trained graph knowledge-enabled residual space-time information fusion model. After the automatic space-time feature residual learning process, the fused feature data is classified into fault categories to achieve intelligent diagnosis of the insulating bearings of the doubly-fed wind turbine.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the concept of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An insulating bearing fault diagnosis method based on residual space-time information fusion model, characterized in that: include: Step 1: Use two three-way vibration sensors to collect the original vibration signals of six channels of the insulating bearing of the doubly-fed wind turbine generator to capture the information of the insulating bearing in different health states; Step 2: Divide the original vibration signal into 80% training set data and 20% test set data; Step 3: Construct a residual space-time information fusion model empowered by graph knowledge; Step 4: Use the scavenger algorithm to optimize the hyperparameters of the residual space-time information fusion model empowered by graph knowledge; Step 5: Use the training set data as the input of the residual space-time information fusion model empowered by graph knowledge, and the fault category as the output to perform model training; Step 6: Input the test set data into the trained graph knowledge-enabled residual space-time information fusion model. After the automatic space-time feature residual learning process, the fused feature data is classified into fault categories to achieve intelligent diagnosis of the insulating bearings of the doubly-fed wind turbine.

2. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 1 is characterized in that: The graph knowledge-enabled residual space-time information fusion model includes: a spatial information fusion module, a temporal information fusion module and a self-attention module; the original vibration signals of 6 channels are respectively input into the spatial information fusion module and the temporal information fusion module to extract local features and global features, the local features and the global features are fused in the self-attention module to obtain fused features, the fused features are input into the fully connected layer to construct a mapping relationship between features and fault modes and output the diagnosis results.

3. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 2 is characterized in that: The spatial information fusion module includes: a first hidden layer and a second hidden layer; The first hidden layer includes: the first graph convolution layer and the first activation layer, and the first graph convolution layer includes 64 graph convolution kernels; the second hidden layer includes: the second graph convolution layer, the batch normalization layer, the second activation layer and the discard layer, and the second graph convolution layer includes 128 graph convolution kernels; each graph convolution kernel performs weighted summation on a central vertex and its 23 surrounding adjacent nodes; the original vibration signals of 6 channels are sequentially input into the first graph convolution layer, the first activation layer, the second graph convolution layer, the batch normalization layer, the second activation layer and the discard layer, and the local features are output.

4. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 3 is characterized in that: Constructing the adjacency matrix of the graph convolution layer in the first hidden layer and the second hidden layer and obtaining the output of the graph convolution kernel specifically includes the following steps: (1) Calculate the maximum mutual information coefficient of the original vibration signal sequence of adjacent channels: Among them, X and Y represent the original vibration signal sequences of the two channels respectively, and M IC (X, Y) is the maximum mutual information coefficient between the X sequence and the Y sequence, I(X; Y) represents the mutual information between the X sequence and the Y sequence, H(X) represents the entropy of the X sequence, and H(Y) represents the entropy of the Y sequence; p(x, y) represents the joint probability of the elements in the X sequence and the Y sequence, p(x) represents the probability of an element appearing in the X sequence, and p(y) represents the probability of an element appearing in the Y sequence; p(x i ) represents the probability of the i-th element in the X sequence, p(y i ) represents the probability of the i-th element in the Y sequence appearing, I(x i ) represents the entropy of the i-th element in the X sequence, I(y i ) represents the entropy of the i-th element in the Y sequence, and n represents the number of elements contained in the sequence; (2) Use the maximum mutual information coefficient to construct the adjacency matrix G of the graph convolution layer in the first hidden layer and the second hidden layer as follows: Among them, Ax, Ay, Az, Bx, By and Bz represent the original vibration signal sequences of 6 channels respectively; (3) Using Laplace to calculate the normalized adjacency matrix Where E is the identity matrix, yes The degree matrix of (4) Get the output of the l+1th layer graph convolution kernel: Where: H l is the output of the l-th layer graph convolution kernel; f(·) is the nonlinear activation function; W l+1 is the trainable weight matrix of the l+1th layer.

5. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 2 is characterized in that: The time information fusion module includes: an input layer, a third hidden layer and an output layer, wherein the third hidden layer includes a bidirectional WDRU neural network, and sequential and reverse propagation are performed through the bidirectional WDRU neural network to extract global features; the WDRU neural network includes a plurality of weight reduction recurrent units; The calculation process of the bidirectional WDRU neural network is as follows: (1) Calculate the information status of the shared gate: f t =W d ·[h t-1 ,x t ]+b d D t =σ(f t )=σ(W d ·[h t-1 ,x t ]+b d )=σ(W d ·h t-1 +W d ·x t +b d ) Q t =tan(f t )=tan(W d ·[h t-1 ,x t ]+b d )=tan(W d ·h t-1 +W d ·x t +b d ) Among them, f t represents the intermediate variable of the weight reduction recurrent unit, W d is the weight parameter of the weight reduction recurrent unit, h t-1 represents the output of the previous weight reduction recurrent unit, x t represents the original vibration signal of a channel, b d represents the bias parameter of the weight reduction recurrent unit; D t represents a shared gate, Q t Indicates the information status of the shared door; (2) Update the control parameters of the information state of the weight reduction cycle unit: C t =D t ×C t-1 +(1-D t )×Q t =σ(W d ·[h t-1 ,x t ]+b d )×C t-1 +(1-σ(f t ))×tan(f t ) Among them, C t-1 The control parameter representing the information state of the previous weight reduction recurrent unit; C t A control parameter representing the information state of the current weight reduction recurrent unit; (3) Calculate the output of the weight reduction recurrent unit: h t =D t ×tan(C t )=σ(W d ·h t-1 +W d ·x t +b d )×tan(C t ) (4) Calculate the output of the bidirectional WDRU neural network: y t =σ(h t )=σ(D t ×tan(C t )) (5) The update formula of the bidirectional WDRU neural network is as follows: h t =W Ft ×h Ft +W Bt ×h Bt +b t Among them, W Ft and W Bt Represent the weight variables in the forward and backward propagation processes, b t Indicates the offset, h Ft represents the forward information state, h Bt Indicates the reverse information status, The back propagation formula of the bidirectional WDRU neural network is as follows: (6) The minimum optimization goal of the time information fusion module is: Among them, E t represents the error value of the tth period, Y t represents the output of the time information fusion module; W y represents the weight parameter of the weight reduction recurrent unit; b y represents the bias parameter of the weight reduction recurrent unit; (7) Variables of the output layer of the temporal information fusion module: Among them, P t represents the output of the current time weight reduction cycle unit; δy t Represents the change in the output of the bidirectional WDRU neural network model; δP t The current time weight reduction cycle unit output change; δW y Indicates the change in the weight parameter of the bidirectional WDRU neural network; δb y Represents the change in bias parameters of the bidirectional WDRU neural network; (8) Calculate the weight change during the back propagation process of the hidden layer of the time information fusion module: Among them, δh t δC t δD t δQ t δf t w h Both represent the change in the weight of the network model during the back propagation process, and T represents the period of the time series signal.

6. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 1 is characterized in that: The step 4 uses the scavenger algorithm to optimize the hyperparameters of the residual space-time information fusion model enabled by graph knowledge, so as to improve the recognition accuracy and generalization ability of the model. The hyperparameters include: the number of neurons in the first hidden layer, the second hidden layer and the third hidden layer of the residual space-time information fusion model enabled by graph knowledge, the learning rate, the regularization coefficient, the number of iterations and the batch size.

7. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 6 is characterized in that: In step 4, the scavenger algorithm is used to optimize the hyperparameters of the residual space-time information fusion model enabled by graph knowledge, specifically: (1) Initializing the graph knowledge enabled residual space-time information fusion model, including setting the number of neurons in the initial hidden layer of the graph knowledge enabled residual space-time information fusion model; (2) Set the learning rate, number of iterations, and batch size range of the residual space-time information fusion model enabled by graph knowledge, and also set the scavenger target tracking speed coefficient, scavenger population size, and maximum number of iterations; (3) Calculate the fitness values ​​of all scavenger individuals, sort them according to their fitness, select the individual with the best fitness, and select the mean square error as the fitness function of the scavenger algorithm. MSE is expressed as: Where n is the number of samples, Y i With Y i * are the true value and predicted value of the i-th sample data respectively; The smaller the MSE value, the smaller the difference between the predicted value and the true value of the residual space-time information fusion model enabled by graph knowledge, which means that the position of the scavenger individual is better; the position of the scavenger individual corresponds one-to-one to the hyperparameters of the residual space-time information fusion model enabled by graph knowledge. The position coordinates of the scavenger individual with the optimal fitness are the optimal hyperparameters of the residual space-time information fusion model enabled by graph knowledge. (4) Determine whether the scavenger algorithm iteration has ended. When the maximum number of iterations is reached, the algorithm iteration is completed. At this time, the position of the scavenger individual with the best fitness is assigned to the residual space-time information fusion model empowered by graph knowledge, that is, the best hyperparameters found by the scavenger algorithm are assigned to the residual space-time information fusion model empowered by graph knowledge. If the scavenger algorithm iteration has not ended, return to the previous step.

8. The insulated bearing fault diagnosis method based on the residual space-time information fusion model according to claim 1 is characterized in that: The step 5 is specifically as follows: using the training set data as the input of the residual space-time information fusion model empowered by graph knowledge, and the fault category as the output, completing the feature extraction and fusion of the original vibration data of the insulating bearing, selecting the loss function of the Adam optimizer and the Softmax classifier, stopping the training when the set number of iterations is reached, outputting the fault type, and completing the model training.

Citation Information

Patent Citations

  • Transformer fault diagnosis method and device based on self-attention heterogeneous network

    CN116955951A

  • Rolling bearing fault diagnosis method based on multi-scale residual network and improved GRU

    CN118094371A

  • Enhanced discriminate feature learning deep residual CNN for multi-task rotating machinery fault diagnosis with information fusion

    US20230351177A1

Cited By

  • Fault diagnosis positioning technology

    CN119984386A