A method for diagnosing rotor water-spraying faults in cross-domain vehicles

Abstract

The present invention provides a method for diagnosing rotor water-spraying faults of a cross-domain aircraft, which relates to the field of fault diagnosis technology, including: constructing a multi-source heterogeneous data set; preprocessing and standardizing the data; converting a one-dimensional sensor signal into a two-dimensional image representation with three-channel characteristics; first inputting the two-dimensional image representation into a residual network, and then sequentially performing dilated convolution and multi-scale feature extraction, attention mechanism and weighted fusion, and bottom-up feature fusion to obtain a feature representation of the fault data; constructing a few-sample dual-graph collaborative propagation network, and training the few-sample dual-graph collaborative propagation network in combination with a dynamic early stopping method based on graph collaboration improvement; identifying and diagnosing the fault type or state. The technical solution of the present invention overcomes the problem in the prior art that it is not possible to perform real-time and accurate fault detection and diagnosis on cross-domain aircraft in the absence of labeled data.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method for diagnosing rotor water-spraying faults of a cross-domain vehicle. Background Art

[0002] Cross-domain vehicles, capable of operating both underwater and in the air, are widely used in mission scenarios such as ocean exploration, underwater rescue, and environmental monitoring. During the medium transition process, especially when exiting the water, the rotors inevitably interact with the water surface, causing "rotor water-kicking." This phenomenon severely impacts the vehicle's performance and stability. Rotor water-kicking has become a key technical bottleneck in cross-domain vehicles' ability to perform cross-medium missions.

[0003] Mechanically, rotor blades are susceptible to wear, deformation, and even fracture due to water hammer and cavitation. In severe cases, this can disrupt thrust output and affect the vehicle's normal operation. At the powertrain level, the high drag at the interface between water and air and the resulting discontinuity in thrust can lead to excessive motor loads, resulting in overheating or protective system shutdowns. Inadequate sealing can even cause water intrusion into the system, compromising the safety of the entire power module. Furthermore, attitude control instability is a significant concern. During underwater exit, uneven thrust distribution and fluid reaction can easily cause pitch, yaw, or roll instability in the rotor, leading to loss of control during exit. Fluid resonance further exacerbates this risk. When the rotor's operating frequency approaches the natural frequency of the vehicle's structure, resonance can induce severe vibrations, compromising the equipment's structural strength and lifespan, and even causing system failure.

[0004] Due to the various potential failure risks mentioned above during the rotor-water launch process, it is particularly necessary to establish an efficient fault diagnosis method and system. On the one hand, the fault diagnosis system can realize real-time monitoring of the status of the rotor and power system. By collecting multiple signals such as rotor vibration, thrust, and attitude, it can promptly identify faults such as mechanical damage to the rotor and motor overload, thereby preventing further expansion of the fault. On the other hand, the fault diagnosis system can provide feedback and optimize the attitude control of the aircraft during the water launch process. By early detection of abnormal phenomena such as thrust imbalance or turbulent interference, it provides data support for the improvement of the control algorithm, thereby improving the stability and safety of the aircraft during cross-domain transitions. In addition, for fatigue damage to the rotor structure and fluid resonance phenomena, the fault diagnosis system can predict the trend of fault occurrence in advance, extend the service life of the aircraft equipment, and reduce maintenance costs.

[0005] In summary, the problem of rotor water splashing in cross-domain vehicles not only affects the stability of mission execution but can also lead to systemic failures and safety hazards. Therefore, developing a method for diagnosing rotor water splashing failures in cross-domain vehicles is a key approach to improving the performance and reliability of cross-domain vehicles. The establishment of such a system will provide cross-domain vehicles with real-time and accurate fault detection and diagnosis capabilities, further ensuring their stable operation in complex mission environments and laying a solid foundation for the future development and application of cross-domain vehicle technology.

[0006] Therefore, there is a need for a cross-domain vehicle rotor water-splashing fault diagnosis method that can detect and diagnose cross-domain vehicle faults in real time and accurately in the absence of labeled data. Summary of the Invention

[0007] The main purpose of the present invention is to provide a method for diagnosing rotor water-spraying faults of a cross-domain aircraft, so as to solve the problem in the prior art that it is impossible to perform fault detection and diagnosis on a cross-domain aircraft in real time and accurately in the absence of labeled data.

[0008] To achieve the above objectives, the present invention provides a method for diagnosing rotor water-spraying faults of a cross-domain vehicle, which specifically includes the following steps:

[0009] S1, obtains sensor data and constructs a multi-source heterogeneous dataset.

[0010] S2, preprocesses and standardizes the data in multi-source heterogeneous datasets.

[0011] S3, performing multi-order differential analysis on the data processed in step S2, and constructing a statistical pattern based on the improved Markov method to convert the one-dimensional sensor signal into a two-dimensional image representation with three-channel characteristics.

[0012] In S4, the two-dimensional image representation is first input into the residual network, and then dilated convolution and multi-scale feature extraction, attention mechanism and weighted fusion, and bottom-up feature fusion are performed in sequence to obtain the feature representation of the fault data.

[0013] S5, using the representation after feature extraction, constructs a few-shot dual-image collaborative propagation network and trains the few-shot dual-image collaborative propagation network.

[0014] S6, after the data to be detected has passed through steps S1 to S4, it is input into the trained few-sample dual-graph collaborative propagation network to identify and diagnose the fault type or state.

[0015] Furthermore, step S3 specifically includes the following steps:

[0016] S3.1, assuming that the original signal after step S2 is x={x1,x2,…,x N}, the first-order difference is Δ 1 x t′ , the second-order difference is Δ 2 x t′ ,but:

[0017] Δ 1 x t′ =x t′ -x t′-1 ,t=2,…,N(1);

[0018] Δ 2 x t′ =Δ 1 x t′ -Δ 1 x t′-1 ,t=3,…,N(2);

[0019] Among them, x t′ Represents the t′th data in the original signal, and N represents a positive integer.

[0020] S3.2, divide the original signal x into M bins according to the value range. Each bin corresponds to a state, where m = 1, 2, ..., M. The bin division is completed by formula (3):

[0021] S m =[Q m-1 ,Q m ),

[0022] Among them, S m Indicates the range of the mth quantile bin, Q m is the mth quantile, Indicates that x has The quantile at which the data is less than or equal to this value.

[0023] Each value x in the signal is binned according to the quantile t′ Mapped to the corresponding number s t :

[0024] s t =m,x t′ ∈S m (4). S3.3, the state transition matrix P represents the transition from state S i Transfer to state S j The probability P ij :

[0025]

[0026] Among them, n ij Indicates state S i To state S jThe number of transitions, J is the total number of states.

[0027] The state transition matrix P is constructed as I, where the pixel value I ij Represents the state transition probability:

[0028] I ih =P ih , i,j=1,2,…,J (6);

[0029] Where I is a single-channel Markov transition field.

[0030] S3.4, for the original signal x, first-order difference Δ 1 x t′ , second-order difference Δ 2 x t′ Execute the Markov method separately to obtain three single-channel transfer field images, and merge the three single-channel transfer field images into a three-channel image I 3C :

[0031] I 3C =Concat(I (0) ,I (1) ,I (2) ) (7);

[0032] Among them, I (0) is the transfer field of the original signal, I (1) is the first-order difference transfer field, I (2) is the transfer field of the second-order difference, and Concat represents the channel merging operation.

[0033] S3.5, scaling the pixel value range of the image, scaling the pixel values ​​of the image to [0, 1], and normalizing according to the mean and standard deviation.

[0034] Furthermore, step S4 specifically includes the following steps:

[0035] S4.1, input three-channel two-dimensional image After ResNet50 extracts multi-scale features, we get the feature map F f , where f = 2, 3, 4, 5.

[0036] S4.2, for each feature map F f , where f = 2, 3, 4, 5. Through the dilated convolution operation with different dilation rates k, three feature maps of different scales are obtained, and then the feature maps are spliced ​​along the channel dimension to form a multi-scale fusion feature map.

[0037] S4.3, the multi-scale fusion feature map is globally average pooled, and after generating the attention weight vector through the fully connected layer, the attention enhanced feature map is generated by channel weighted fusion.

[0038] S4.4, the deepest attention-enhanced feature map is used as the top-level feature map, and the shallow features are upsampled and fused in turn to form the final multi-scale fused feature set, which is finally converted into a feature vector through a global average pooling operation.

[0039] Furthermore, step S5 specifically includes the following steps:

[0040] S5.1, divide the fault feature data into 80% training set and 20% test set, extract the support set S and query set Q from the training set, where the support set contains labeled samples and unlabeled samples; for each labeled sample, the label of the sample is encoded by one-hot encoding. Indicates that Splice and generate the node feature vector V of the labeled sample node :

[0041]

[0042] Where C and d represent the vector dimensions.

[0043] For unlabeled samples, the label vector is initialized to a zero vector Node feature V′ of unlabeled samples node Expressed as:

[0044]

[0045] Among them, Concat represents the concatenation operation.

[0046] S5.2, Node Feature Matrix in the Instance Graph It is composed of the node features of the support set and the query set, where N ins is the total number of samples in the support set and query set; the node features of the labeled samples in the support set are given by The node features of the unlabeled samples in the support set are composed of The label vector is set to zero vector; the node features of the query set samples are composed of Composition; all node features are combined into the instance graph node feature matrix

[0047] S5.3, use K-means clustering algorithm to Divided into N dis distribution clusters and determine the center point C of the distribution cluster k , concatenate the center point with the initialization label vector as the node feature matrix in the distribution graph Represents the distribution information of support set and query set samples in the feature space.

[0048] S5.4, Instance Graph Edge Weight Matrix Initialized by feature similarity between node features, specifically expressed as:

[0049]

[0050] in, and They represent the features of the i-th and j-th nodes in the instance graph respectively, and σ′ is the scale parameter that controls the feature similarity.

[0051] Distribution graph edge weight matrix Initialized by feature similarity between distribution centers:

[0052]

[0053] in, and Represent the features of the kth and lth nodes in the distribution graph respectively.

[0054] Furthermore, step S5 further includes:

[0055] S5.5, Input instance graph node features and instance graph edge weights The instance graph node features are enhanced through the dual-graph collaborative Transformer method. The process of enhancing the instance graph node features is the same as that of enhancing the distribution graph node features. The specific steps of enhancing the instance graph node features are as follows:

[0056] S5.5.1, the instance graph node features Mapped to instance graph query vector Q respectively ins , key vector K ins Sum value vector V ins , the calculation formula is:

[0057]

[0058] in, is the mapping weight matrix, h is the number of heads of multi-head attention, d a The dimensions are for a single head.

[0059] S5.5.2, use the instance graph mask matrix and instance graph edge weight matrix to calculate the attention weights between nodes.

[0060] S5.5.3, the instance graph attention weight α ins and the instance graph value vector Vins Multiply them together to get the instance graph attention output X att,ins , and mapped back to the original feature dimension through linear transformation:

[0061]

[0062] in, is the weight matrix of the output mapping.

[0063] S5.5.4, output the instance graph attention X att,ins Instance graph node features Perform residual connection and layer normalization to obtain the instance graph node feature X after the first step of enhancement res1,ins :

[0064]

[0065] Among them, LayerNorm is the layer normalization operation.

[0066] S5.5.5, input X res1,ins To the two-layer feedforward network FFN, and perform residual connection and layer normalization again, and finally obtain the instance graph enhanced feature

[0067]

[0068] Furthermore, step S5.5.2 specifically includes the following steps:

[0069] First, the query vector Q of the instance graph node i is ins,i and the key vector K of instance graph node j ins,j The dot product between them calculates the raw attention score:

[0070]

[0071] Among them, α raw,ij is the attention score between instance graph node i and node j, is the scaling factor, It's K ins,j The transpose of

[0072] Then, the edge weight matrix of node i and node j in the instance graph is combined and instance graph instance graph mask matrix M ins,ij , shield and adjust the attention score to get the final attention weight α ins,ij :

[0073]

[0074] Among them, Softmax is the activation function, and the instance image mask matrix M ins,ij Defined as:

[0075]

[0076] Furthermore, step S5 also includes: S5.6, using the collaborative attention mechanism to perform cross-graph interaction.

[0077] The specific steps include:

[0078] S5.6.1, Enhance the features of the instance graph Mapping to cross-graph interaction attention query vector Q ins2dis , the distribution map is enhanced Mapped to the cross-graph interaction attention key vector K dis2ins and the cross-graph interaction attention value vector V dis2ins , the specific calculation is:

[0079]

[0080] in, is the linear mapping matrix of cross-graph interactive attention, h is the number of heads of multi-head attention, d a is the characteristic dimension of a single head.

[0081] S5.6.2, by Q ins2dis and K dis2ins Calculate cross-graph attention weight α ins←dis :

[0082]

[0083] Among them, α ins←dis,ij represents the cross-graph attention weight between the i-th node in the instance graph and the j-th node in the distribution graph, is the scaling factor, Q ins2dis,i Enhance features for instance graphs The query vector of the i-th node, K dis2ins,j is the distribution graph node feature The key vector of the j-th node, K dis2ins,j The transpose of .

[0084] S5.6.3, combine the cross-graph attention weights with the value vector V of the distribution graph dis2ins Multiply them together to get the cross-graph fusion feature V of the instance graph node ins←dis :

[0085]

[0086] in, It represents the representation of the i-th node in the instance graph after fusion of distribution graph features.

[0087] S5.6.4, cross-graph fusion features V of instance graph nodes ins←dis Enhanced features with instance graphs Perform residual connection and layer normalization to obtain instance graph cross-graph interaction features

[0088]

[0089] Furthermore, step S5 also includes: S5.7, fusing graph node features.

[0090] The specific steps include:

[0091] S5.7.1, Enhance the features of the instance graph Cross-graph interaction features with instance graphs Splice and get the intermediate feature representation Z ins :

[0092]

[0093] Concat(·) means to and Splice along the feature dimension to get Z ins .

[0094] S5.7.2, by gating coefficient g ins Perform weighted fusion on the enhanced features and cross-graph fusion features to obtain the final node features of the instance graph

[0095]

[0096] in, is the weight matrix of fusion gate; is the bias vector; σ(·) represents the Sigmoid activation function, which ensures that the fusion coefficient is between [0,1].

[0097]

[0098] where ⊙ represents element-wise multiplication.

[0099] Furthermore, step S5 further includes: S5.8, updating the graph edge weights;

[0100] The specific steps include:

[0101] S5.8.1, calculate the square of the difference between the characteristics of node i and node j:

[0102]

[0103] in, Represents the square of the feature difference between the i-th node and the j-th node.

[0104] S5.8.2, As input, the edge weight score h is obtained through two layers of MLP mapping ins,ij , the calculation formula is:

[0105]

[0106] Among them, W e1 and W e2 is the weight matrix of MLP; b e1 and b e2 is the bias vector; ReLU(·) represents the activation function.

[0107] S5.8.3, the edge weight fraction h ins,ij Through exponential mapping, the edge weight is ensured to be positive, which is specifically expressed as:

[0108]

[0109] in, Represents the edge weight after the lth round of update.

[0110] Furthermore, step S5 further includes: S5.9, training the few-shot dual-graph collaborative propagation network with dual-graph collaborative iteration based on improved dynamic early stopping;

[0111] The specific steps include:

[0112] S5.9.1, for node i that supports centralized labeling, the predicted distribution is calculated using the Gaussian kernel function between nodes and normalized by Softmax, which is expressed as follows:

[0113]

[0114] Where t represents the number of current iterations, is the predicted label distribution of node i; is the edge weight between nodes i and j in the instance graph; φ(y j ) is the one-hot encoding representation of the true label of node j; s (·) is the Softmax function.

[0115] S5.9.2, Instance Graph Loss Obtained by the Euclidean distance between the predicted distribution and the true label:

[0116]

[0117] Among them, y i is the true label of node i, S label Represents the set of all marked nodes in the support set.

[0118] S5.9.3, integrate the losses of instance graph and distribution graph, add regularization terms to constrain model parameters W and edge weights E, and the total loss function L (t+1) Defined as:

[0119]

[0120] Among them, λ ins and ξ dis are the weight parameters for instance graph and distribution graph loss, λ reg is the weight coefficient of the regularization term, is the loss of the distribution map; is the instance graph edge weight matrix at iteration t+1; is the edge weight matrix of the distribution graph at the t+1 iteration.

[0121] S5.9.4, rate of change of loss function ΔL (t+1) Defined as:

[0122]

[0123] The cooperativity distance measures the difference between node features in the instance graph and the distribution graph. Defined as:

[0124]

[0125] in, is the instance graph key vector at iteration t+1, It represents the feature representation of the distribution center of the same cluster as the instance node i after iteration.

[0126] S5.9.5: Within the set observation window Ω, if any of the following conditions is met, early stopping is triggered:

[0127] ΔL (t+1) <∈ loss or

[0128] Among them, ∈ loss is the loss change rate threshold, ∈ coh is the cooperativity distance threshold.

[0129] After the dynamic early stop is triggered, the final features V of nodes i and j are used final,i and V final,iPerform label propagation and fault classification; the label of unlabeled node i is weighted by the Gaussian similarity with the labeled nodes in the support set:

[0130]

[0131] Among them, I(y j =c) represents the true label y of node j j Whether it belongs to category c; The Gaussian kernel function measures the feature similarity between nodes; is the predicted label of node i, and arg max is a function that finds the parameter that maximizes the function value.

[0132] The present invention has the following beneficial effects:

[0133] (1) The present invention introduces a dual-graph co-evolutionary few-shot learning framework to achieve efficient feature propagation between the support set and the query set, effectively addressing cross-domain and data scarcity issues, and improving the generalization ability of the diagnostic model in new domains and few-shot conditions.

[0134] (2) Compared with traditional diagnostic methods based on single sensor signals or simple features, this invention uniformly preprocesses multi-source sensor data and performs differential-Markov mapping, converting time series signals into multi-channel two-dimensional feature representation images. This process maximizes the mining of time series characteristics and statistical distribution information in the data, providing more discriminative input features for subsequent deep networks, thereby effectively improving the accuracy and robustness of fault identification.

[0135] (3) In the feature extraction stage, the present invention adopts a multi-level feature pyramid network that integrates the residual structure and the attention mechanism for feature extraction, which greatly improves the flexibility and depth of the model in extracting key features, can better capture subtle changes in timing signals, and suppress irrelevant or redundant features, thereby improving the accuracy and efficiency of fault diagnosis.

[0136] (4) In practical applications, traditional graph updating methods often rely on a large amount of experimental data to determine a fixed number of iterations, which can easily lead to model overfitting, decreased generalization ability, or insufficient graph structure updates, making it difficult to achieve optimal diagnostic performance. The present invention designs a dynamic early stopping strategy based on graph synergy improvement. This strategy adaptively determines the timing of iteration cessation by taking into account the changing trend of the loss function and the synergy of dual-graph features. When the performance improvement of the graph co-evolution process is no longer significant, the system automatically terminates the iteration to ensure that the training process converges to the optimal state, reduce the waste of computing resources, further improve the training efficiency and the stability of the diagnostic model, and show significant advantages under complex cross-domain and small sample conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0137] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0138] Figure 1 A flow chart of a cross-domain vehicle rotor water-spraying fault diagnosis method according to the present invention is shown.

[0139] Figure 2 A flow chart showing the application of a cross-domain aircraft rotor water pumping fault diagnosis method of the present invention to a cross-domain aircraft rotor water pumping fault diagnosis system is shown. DETAILED DESCRIPTION

[0140] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0141] like Figure 1 A cross-domain vehicle rotor water-splashing fault diagnosis method is shown, which specifically includes the following steps:

[0142] S1, obtains sensor data and constructs a multi-source heterogeneous dataset.

[0143] S2, preprocesses and standardizes the data in multi-source heterogeneous datasets.

[0144] S3, using an improved multi-order difference Markov mapping method to perform feature representation conversion. Multi-order difference analysis is performed on the data processed in step S2, and statistical pattern construction is performed based on the improved Markov method to convert the one-dimensional sensor signal into a two-dimensional image representation with three-channel characteristics.

[0145] In S4, the two-dimensional image representation is first input into the residual network, and then dilated convolution and multi-scale feature extraction, attention mechanism and weighted fusion, and bottom-up feature fusion are performed in sequence to obtain the feature representation of the fault data.

[0146] S5, using the extracted feature representation, constructs a few-shot dual-graph collaborative propagation network and trains the network. Using the extracted feature representation, fault identification and classification are achieved by constructing a few-shot dual-graph collaborative propagation network. The feature representation of the data is divided into a training set and a test set, with the training set set set at 80% and the test set set at 20%. The training set data is input into the network, and a support set and query set are constructed for the labeled and unlabeled data. A dual-graph structure of instance graph and distribution graph is constructed, and graph propagation is performed based on the dual-graph collaborative evolution mechanism. This achieves cross-domain fault diagnosis decision generalization under few-shot conditions. The model parameters are trained and the graph structure is iteratively optimized using a dynamic early stopping strategy based on graph collaborative improvement, ensuring convergence of the training process and superior classification performance.

[0147] S6: After the data to be tested has been processed through steps S1 to S4, it is input into the trained few-shot dual-image collaborative propagation network to identify and diagnose the fault type or state. After freezing the model parameters, the data to be tested undergoes the same preprocessing and feature extraction process and is then input into the trained few-shot dual-image collaborative propagation network to identify and diagnose the fault type or state.

[0148] Specifically, the sensor data in step S1 includes key signal data obtained from an inertial measurement unit (IMU), accelerometers, strain gauge sensors, current and voltage sensors, rotational speed sensors, and thrust sensors. By deploying multiple sensors, including IMUs, accelerometers, strain gauge sensors, current and voltage sensors, rotational speed sensors, and thrust sensors, key signal data related to the vehicle's attitude, acceleration, structural strain, power state, rotor speed, and thrust are acquired during operation. The data collected by these sensors is then aggregated to construct a multi-source, heterogeneous dataset.

[0149] Specifically, step S2 includes the following steps:

[0150] S2.1, clean the data.

[0151] S2.2, use median filtering to smooth the cleaned data. Based on the sliding window size k, sort the data in the window and take the median to replace the window center value.

[0152] S2.3, normalize and standardize the data, and linearly map the data to the [0,1] interval

[0153] The normalized data are converted to a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0154] S2.4, to address the problems of inconsistent sampling frequencies and asynchronous time bases of multi-source sensors, a timing alignment method is used to unify the time scale of the data, and the sensor data is resampled based on the unified time base.

[0155] Specifically, step S2.1 includes the following steps:

[0156] S2.1.1, calculate the median of each sensor data as the central tendency of the data.

[0157] S2.1.2, calculate the absolute deviation of each data point from the median and find the median of the absolute deviations, i.e., the MAD value.

[0158] S2.1.3, set the threshold to 2.5 times the MAD. If the absolute deviation of a data point from the median exceeds the threshold, the data point is judged as an outlier and the outlier is removed or replaced.

[0159] Specifically, step S3 includes the following steps:

[0160] S3.1, assuming that the original signal after step S2 is x={x1,x2,…,x N}, calculate the first-order difference Δ 1 x t′ and the second-order difference Δ 2 x t′ ;

[0161] S3.2, divide the original signal x into M bins according to the value range, each bin corresponds to a state, where m = 1, 2, ..., M. The bin division is completed by formula (1):

[0162] S m =[Q m-1 ,Q m ),

[0163] Among them, S m Indicates the range of the mth quantile bin, Q m is the mth quantile, Indicates that x has The quantile at which the data is less than or equal to this value.

[0164] Each value x in the signal is binned according to the quantile t′ Mapped to the corresponding number s t :

[0165] s t =m,x t′ ∈S m (2). S3.3, the state transition matrix P represents the transition from state S i Transfer to state S j The probability P ij :

[0166]

[0167] Among them, n ij Indicates state S i To state S j The number of transitions, J is the total number of states.

[0168] The state transition matrix P is constructed as I, where the pixel value I ij Represents the state transition probability:

[0169] I ij =P ij , i,j=1,2,…,J (4);

[0170] Where I is a single-channel Markov transition field.

[0171] S3.4, for the original signal x, first-order difference Δ 1 x t′ , second-order difference Δ 2 x t′ Execute the Markov method separately to obtain three single-channel transfer field images, and merge the three single-channel transfer field images into a three-channel image I 3C .

[0172] S3.5, scale the pixel value range of the image to [0, 1] and normalize it according to the mean and standard deviation.

[0173] Specifically, step S4 includes the following steps:

[0174] S4.1, input three-channel two-dimensional image ResNet50 extracts multi-scale features. The network first passes through the initial convolution layer and the maximum pooling operation, and then goes through four stages of bottleneck residual blocks to extract features of different scales:

[0175] Perform 7×7 convolution with step size s=2 and padding value p=3 and 3×3 maximum pooling with step size s=2 and padding value p=1 to obtain the initial feature map F1; F1 passes through 3 bottleneck residual blocks to output the first stage feature map F2; F2 passes through 4 bottleneck residual blocks to output the second stage feature map F3; F3 passes through 6 bottleneck residual blocks to output the third stage feature map F4; fourth stage: F4 passes through 3 bottleneck residual blocks to output the fourth stage feature map F5.

[0176] Among them, F2, F3, F4, and F5 correspond to scale features respectively, the resolution decreases step by step, and the number of channels is 256, 512, 1024, and 2048 respectively.

[0177] S4.2, for each feature map F f , where f = 2, 3, 4, 5, through the expansion convolution operation with different expansion rates k, three feature maps T of different scales are obtained f,1 ,T f,2 ,T f,3 , and then splice the feature maps along the channel dimension to form a multi-scale fusion feature map.

[0178] The feature maps T obtained by three dilated convolutions are f,1 ,T f,2 ,T f,3 , spliced ​​along the channel dimension to form a multi-scale fusion feature map U i .

[0179] The global statistical information of the channel is extracted through the global average pooling GAP operation, and the formula is expressed as:

[0180]

[0181] in, Represents the feature map U i The global average pooling result of the c-th channel in; H and W are the feature maps U i The height and width of h and w represent the row and column indices in the feature map, respectively. Represents the eigenvalue of the cth channel at position (h, w);

[0182] S4.3, the multi-scale fusion feature map is globally average pooled, and after generating the attention weight vector through the fully connected layer, the attention enhanced feature map is generated by channel weighted fusion.

[0183] Specifically:

[0184] The vector p after global average pooling i Input into a fully connected layer to learn the dependencies between channels and generate the attention weight vector a i , the calculation formula is:

[0185] a i =σ(W fc ·p i +b fc ) (6);

[0186] Among them, W fc and b fc Represent the weight matrix and bias vector of the fully connected layer respectively; p iis the input vector of global average pooling; σ represents the Sigmoid activation function;

[0187] The attention weight vector a i and multi-scale fusion feature map U i Multiply channel by channel and combine the weighted features of all channels to obtain the attention-enhanced multi-scale feature map A i .

[0188] S4.4, the deepest attention-enhanced feature map is used as the top-level feature map, and the shallow features are upsampled and fused in turn to form the final multi-scale fused feature set, which is finally converted into a feature vector through a global average pooling operation.

[0189] Specifically, feature fusion starts from the deepest attention-enhanced feature map A5, and A5 is used as the top-level output feature map F top , the formula is:

[0190] F top =A5 (7);

[0191] Among them, F top is the top-level output feature map.

[0192] In order to transfer high-level semantic information to the lower layers, upsampling and feature fusion are performed step by step. Starting from A5, each level A a , where a=4, 3, 2 are all fused with the feature map of the previous layer through upsampling operation; the upsampling operation adopts the nearest neighbor interpolation method, and the target feature map after upsampling is represented as U up ; For the a-th layer feature map A a , the calculation formula of feature fusion is:

[0193] F a =Conv2D(A a +Upsample(F a+1 ,s=2)) (8);

[0194] Among them, F a Represents the fusion feature map of layer a; A a is the attention enhancement feature map of layer a; Upsample(F a+1 ,s=2) means that the feature fusion graph F of the previous layer a+1 Through the upsampling operation, the parameter s = 2, that is, each dimension of the input feature map is expanded by 2 times, so that the resolution of the output feature map is twice that of the input, and finally enlarged to the same size as A a The same size, + represents element-by-element addition operation; Conv2D represents a 3×3 convolution operation.

[0195] Through the bottom-up path, A2, A3, A4, and A5 are fused in sequence, and finally the multi-scale fusion feature maps F2, F3, F4, and F are obtained. top , where F2 represents the lowest level fusion feature, and F top is the top-level feature. The final multi-scale fusion feature map set is expressed as:

[0196]

[0197] in, It is the final multi-scale fusion feature set representation.

[0198] The final multi-scale fusion feature Through the global average pooling operation, it is converted into a vector

[0199] Specifically, step S5 includes the following steps:

[0200] S5.1, divide the fault feature data into 80% training set and 20% test set, extract the support set S and query set Q from the training set, where the support set contains labeled samples and unlabeled samples; for each labeled sample, the label of the sample is encoded by one-hot encoding. Indicates that Splice and generate the node feature vector V of the labeled sample node :

[0201]

[0202] Among them, C and d represent the vector dimensions;

[0203] For unlabeled samples, the label vector is initialized to a zero vector Node feature V′ of unlabeled samples node Expressed as:

[0204]

[0205] Among them, Concat represents the concatenation operation;

[0206] S5.2, Node Feature Matrix in the Instance Graph It is composed of the node features of the support set and the query set, where N ins is the total number of samples in the support set and query set; the node features of the labeled samples in the support set are given by The node features of the unlabeled samples in the support set are composed of The label vector is set to zero vector; the node features of the query set samples are composed of Composition; all node features are combined into the instance graph node feature matrix

[0207] S5.3, Node Feature Matrix in Distribution Graph Represents the distribution information of the support set and query set samples in the feature space. The nodes of the distribution graph do not directly correspond to the samples, but are the distribution center points obtained by clustering the sample features. Use the K-means clustering algorithm to Divided into N dis distribution clusters and determine the center point C of the distribution cluster k , concatenate the center point with the initialization label vector as the node feature matrix in the distribution graph Represents the distribution information of support set and query set samples in the feature space;

[0208] S5.4, Instance Graph Edge Weight Matrix Initialized by feature similarity between node features, specifically expressed as:

[0209]

[0210] in, and They represent the features of the i-th and j-th nodes in the instance graph, σ′ is the scale parameter that controls the similarity of features; exp is the natural exponential function;

[0211] Distribution graph edge weight matrix Initialized by feature similarity between distribution centers:

[0212]

[0213] in, and Represent the features of the kth and lth nodes in the distribution graph respectively.

[0214] Specifically, the node calculation process of the distribution graph in step S5.3 is:

[0215] S5.3.1, sample features in the support set and query set As input, the K-means clustering algorithm is used to divide the sample into N dis distribution clusters, the center point of each cluster is C k Defined as:

[0216]

[0217] Among them, S k is the set of samples belonging to the kth cluster, |S k | represents the number of samples in the cluster;

[0218] S5.3.2, the center point C of each clusterk As the node feature of the distribution graph, initialize the label vector to zero vector Splicing to get the distribution map node features

[0219]

[0220] S5.3.3, all distribution center points are spliced ​​together to form the distribution map node feature matrix Among them, N dis is the number of nodes in the distribution graph, that is, the total number of distribution clusters.

[0221] Specifically, step S5 further includes:

[0222] Step S5 further includes:

[0223] S5.5, Input instance graph node features and instance graph edge weights The instance graph node features are enhanced through the dual-graph collaborative Transformer method. The process of enhancing the instance graph node features is the same as that of enhancing the distribution graph node features. The specific steps of enhancing the instance graph node features are as follows:

[0224] S5.5.1, the instance graph node features Mapped to instance graph query vector Q respectively ins , key vector K ins Sum value vector V ins , the calculation formula is:

[0225]

[0226] in, is the mapping weight matrix, h is the number of heads of multi-head attention, d a The dimension of a single head;

[0227] S5.5.2, use the instance graph mask matrix and the instance graph edge weight matrix to calculate the attention weights between nodes;

[0228] S5.5.3, the instance graph attention weight α ins and the instance graph value vector V ins Multiply them together to get the instance graph attention output X att,ins , and mapped back to the original feature dimension through linear transformation:

[0229]

[0230] in, is the weight matrix of the output mapping;

[0231] S5.5.4, output the instance graph attention X att,ins Instance graph node features Perform residual connection and layer normalization to obtain the instance graph node feature X after the first step of enhancement res1,ins :

[0232]

[0233] Among them, LayerNorm is the layer normalization operation;

[0234] S5.5.5, input X res1,ins To the two-layer feedforward network FFN, and perform residual connection and layer normalization again, and finally obtain the instance graph enhanced feature

[0235]

[0236] The enhancement process of the distribution graph is similar to that of the instance graph. The input distribution graph node features and distribution graph edge weights The enhancement of distribution graph node features is achieved through the dual-graph collaborative Transformer method.

[0237] Specifically, step S5.5.2 includes the following steps:

[0238] First, the query vector Q of the instance graph node i is ins,i and the key vector K of instance graph node j ins,j The dot product between them calculates the raw attention score:

[0239]

[0240] Among them, α raw,ij is the attention score between instance graph node i and node j, is a scaling factor used to stabilize the attention distribution, It's K ins,j The transpose of

[0241] Then, the edge weight matrix of node i and node j in the instance graph is combined and instance image mask matrix M ins,ij , shield and adjust the attention score to get the final attention weight α ins,ij :

[0242]

[0243] Among them, Softmax is the activation function, and the instance image mask matrix M ins,ij Defined as:

[0244]

[0245] This mask ensures that when the edge weight between node i and node j is zero, the attention score will be masked to negative infinity, making the Softmax output zero.

[0246] Specifically, step S5 also includes: S5.6, using the collaborative attention mechanism to perform cross-graph interaction; using the collaborative attention mechanism to perform cross-graph interaction between the instance graph and the distribution graph; the cross-graph interaction process of the instance graph and the distribution graph is consistent.

[0247] Cross-graph interaction of instance graphs specifically includes the following steps:

[0248] S5.6.1, Enhance the features of the instance graph Mapping to cross-graph interaction attention query vector Q ins2dis , the distribution map is enhanced Mapped to the cross-graph interaction attention key vector K dis2ins and the cross-graph interaction attention value vector V dis2ins , the specific calculation is:

[0249]

[0250] in, is the linear mapping matrix of cross-graph interactive attention, h is the number of heads of multi-head attention, d a is the characteristic dimension of a single head;

[0251] S5.6.2, by Q ins2dis and K dis2ins Calculate cross-graph attention weight α ins←dis :

[0252]

[0253] Among them, α ins←dis,ij represents the cross-graph attention weight between the i-th node in the instance graph and the j-th node in the distribution graph, is the scaling factor, Q ins2dis,i Enhance features for instance graphs The query vector of the i-th node, K dis2ins,j Enhance features for distribution maps The key vector of the j-th node, K dis2ins,j The transpose of

[0254] S5.6.3, combine the cross-graph attention weights with the value vector V of the distribution graph dis2ins Multiply them together to get the cross-graph fusion feature V of the instance graph node ins←dis :

[0255]

[0256] in, It represents the representation of the i-th node in the instance graph after fusion of distribution graph features.

[0257] S5.6.4, cross-graph fusion features V of instance graph nodes ins←dis Enhanced features with instance graphs Carry out residual

[0258] Differential connection and layer normalization to obtain cross-graph interaction features of instance graphs

[0259]

[0260] Similarly, for the distribution graph, the node features are represented by the key vector K of the instance graph. ins2dis and the value vector of the instance graph

[0261] V ins2dis The calculation process is the same as above. Finally, the cross-graph interactive features of the distribution graph are for:

[0262]

[0263] in, It is an enhanced feature of the distribution map.

[0264] Specifically, step S5 further includes: S5.7, fusing features of the instance graph and the distribution graph nodes; the process of fusing features of the instance graph and the distribution graph nodes is the same;

[0265] The instance graph node feature fusion specifically includes the following steps:

[0266] S5.7.1, Enhance the features of the instance graph Cross-graph interaction features with instance graphs Splice and get the intermediate feature representation Z of the instance graph ins :

[0267]

[0268] Concat(·) means to and Splice along the feature dimension to get Z ins ;

[0269] S5.7.2, Gating coefficient g by example diagram ins Perform weighted fusion on the instance graph enhancement features and the instance graph cross-graph fusion features to obtain the final node features of the instance graph

[0270]

[0271] in, is the instance graph weight matrix of the fusion gate; is the bias vector; σ(·) represents the Sigmoid activation function, which ensures that the fusion coefficient is between [0,1].

[0272]

[0273] where ⊙ represents element-wise multiplication.

[0274] For the distribution graph node features, weighted fusion is also performed through the fusion gating mechanism. The specific calculation steps and examples are as follows: Figure 1 Finally, the final node features of the distribution graph are obtained

[0275] Specifically, step S5 further includes: S5.8, updating the instance graph edge weights and the distribution graph edge weights; the instance graph edge weight update process and the distribution graph weight update process are consistent;

[0276] The instance graph edge weight update specifically includes the following steps:

[0277] S5.8.1, calculate the square of the characteristic difference of the node:

[0278]

[0279] in, Represents the square of the feature difference between the i-th node and the j-th node in the instance graph;

[0280] S5.8.2, As input, the instance graph edge weight score h is obtained through two layers of MLP mapping ins, ij, the calculation formula is:

[0281]

[0282] Among them, W e1 and W e2 is the weight matrix of MLP; b e1 and b e2 is the bias vector; ReLU(·) represents the activation function to ensure nonlinear mapping;

[0283] S5.8.3, the instance graph edge weight score h ins,ij Through exponential mapping, the edge weights of the instance graph are ensured to be positive, which can be expressed as follows:

[0284]

[0285] in, Represents the edge weight of the instance graph after the lth round of update.

[0286] For the edge weight update of the distribution graph, the operation is exactly the same as that of the instance graph.

[0287] Specifically, step S5 further includes: S5.9, training the few-shot dual-graph collaborative propagation network based on dual-graph collaborative iteration with improved dynamic early stopping;

[0288] The specific steps include:

[0289] S5.9.1, the nodes of the instance graph are label-propagated by similarity with other nodes through edge weights to obtain the predicted distribution. For node i that supports centralized labeling, the predicted distribution is calculated by the Gaussian kernel function between nodes and normalized by Softmax, which is specifically expressed as follows:

[0290]

[0291] Where t represents the number of current iterations, is the predicted label distribution of node i; is the edge weight between nodes i and j in the instance graph at iteration t+1, which is calculated using node features; j ) is the one-hot encoding representation of the true label of node j; s (·) is the Softmax function, which normalizes the label distribution.

[0292] S5.9.2, Instance Graph Loss Obtained by the Euclidean distance between the predicted distribution and the true label:

[0293]

[0294] Among them, y i is the true label of node i, S label Represents the set of all marked nodes in the support set.

[0295] The design method of the loss function of the distribution graph is exactly the same as that of the instance graph loss function.

[0296] S5.9.3, integrate the losses of instance graph and distribution graph, add regularization terms to constrain model parameters W and edge weights E, and the total loss function L (t+1) Defined as:

[0297]

[0298] λ ins and ξ dis are the weight parameters for instance graph and distribution graph loss, λ reg is the weight coefficient of the regularization term, is the loss of the distribution map; is the instance graph edge weight matrix at iteration t+1; is the edge weight matrix of the distribution graph at the t+1 iteration.

[0299] To prevent overfitting and improve graph update efficiency, a dynamic early stopping mechanism based on graph synergy improvement monitors the rate of change of the loss function and the synergy distance between the two graphs during training.

[0300] S5.9.4, rate of change of loss function ΔL (t+1) Defined as:

[0301]

[0302] The cooperativity distance measures the difference between node features in the instance graph and the distribution graph. Defined as:

[0303]

[0304] in, is the instance graph key vector at iteration t+1, It represents the feature representation of the distribution center of the same cluster as the instance node i after iteration.

[0305] S5.9.5: Within the set observation window Ω, if any of the following conditions is met, early stopping is triggered:

[0306] ΔL (t+1) <∈ loss or

[0307] Among them, ∈ loss is the loss change rate threshold, ∈ coh is the cooperativity distance threshold.

[0308] After the dynamic early stop is triggered, the final features V of nodes i and j are used final,i and V final,j Perform label propagation and fault classification; the label of unlabeled node i is weighted by the Gaussian similarity with the labeled nodes in the support set:

[0309]

[0310] Among them, I(y j =c) represents the true label y of node j j Whether it belongs to category c; The Gaussian kernel function measures the feature similarity between nodes; is the predicted label of node i, and arg max is a function that finds the parameter that maximizes the function value.

[0311] Specifically, in step S6, after freezing the model parameters, the data to be detected is input into the trained few-sample dual-graph collaborative propagation network after the same preprocessing and feature extraction process to identify and diagnose the fault type or state.

[0312] Step S6 also includes the following steps:

[0313] S6.1, fix all parameters in the trained few-shot dual-graph cooperative propagation network to ensure that the network is no longer updated during the inference phase.

[0314] S6.2, preprocess the input data to be tested according to the same process as the training phase, including: data cleaning and standardization: eliminating invalid data and unifying the data format; normalization and alignment: normalizing the data collected by different sensors and performing time series alignment; feature conversion: converting the time series signal into a three-channel two-dimensional image representation through multi-order difference and Markov mapping methods.

[0315] In S6.3, the preprocessed two-dimensional image is input into the feature extraction module that combines the trained residual network with the multi-scale attention mechanism to generate multi-level fusion features.

[0316] In S6.4, the extracted feature representation is used as the node feature and input into the frozen few-shot dual-graph collaborative propagation network. The feature propagation mechanism of the instance graph and the distribution graph is combined to identify the fault type or state.

[0317] S6.5, in the few-shot dual-graph collaborative propagation network, the fault label of the node is calculated through label propagation and Gaussian kernel similarity mechanism, and the final diagnosis result of the fault type is output.

[0318] S6.6, deploy the frozen model to the cross-domain vehicle system to provide real-time fault diagnosis capabilities. According to actual application requirements, integrate it into computing devices for online detection and diagnosis.

[0319] In the design of the cross-domain vehicle, a variety of sensors are deployed to comprehensively monitor various rotor water-spraying faults and provide stable operation guarantees. Strain gauge sensors are installed at the root of the rotor blades. They are primarily used to capture the instantaneous impact force and stress changes during rotor startup, detecting deformation or fatigue damage on the blades, and thus ensuring the mechanical health of the rotor. A thrust sensor is located adjacent to the rotor support structure. This sensor monitors the output of the rotor thrust in real time, determines whether the thrust is uniform, and quickly identifies problems such as insufficient thrust or abnormal thrust fluctuations. A matching speed sensor is installed at the shaft end of the rotor motor. It is responsible for monitoring the speed changes during rotor startup and operation, and promptly detecting problems such as unstable speed or abnormal output of the power system.

[0320] At the same time, current and voltage sensors are deployed at the power module and motor input terminals of the power system to monitor the fluctuations in current and voltage during the operation of the motor and to diagnose whether the power system has current overload or power anomalies. In order to capture the overall vibration of the aircraft, acceleration sensors are deployed in the front and rear areas of the fuselage. This deployment can effectively identify abnormal vibration and impact signals caused by rotor startup, fluid disturbance or structural failure. In addition, in order to monitor the attitude changes of the aircraft during the process of leaving the water, an inertial detection unit is installed at the center of gravity of the aircraft to provide real-time attitude data such as pitch angle, roll angle and yaw angle, to assist the attitude control system in attitude adjustment, and ensure that the aircraft can smoothly complete the transition from underwater to air.

[0321] Through the precise deployment of these sensors, the vehicle can comprehensively collect key information such as the rotor mechanical status, power system output, attitude changes and structural vibration, thereby providing strong data support for the fault diagnosis of rotor water-splashing.

[0322] like Figure 2 As shown, the present invention first uses a sensor module to collect data, and the sensor module includes: an inertial detection unit, a strain gauge sensor, a current and voltage sensor, a thrust sensor, a speed sensor and an acceleration sensor; then the collected data is cleaned, denoised, normalized and standardized by a data processing module, and data alignment and data synchronization are performed; then the data is subjected to feature extraction and fault diagnosis by a calculation processing module, and then the fault diagnosis result is sent to a human-computer interaction module, and the human-computer interaction module includes: a monitoring screen, a diagnostic parameter adjustment panel, and an alarm system. The human-computer interaction module is used to generate a fault report; finally, the generated fault report is sent to an execution control module, and the attitude controller, power controller and safety protection unit in the execution control module are used to control the cross-domain vehicle according to the generated fault report.

[0323] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the scope of protection of the present invention.

Claims

1. A method for diagnosing rotor water-spraying faults of a cross-domain vehicle, characterized in that: The specific steps include: S1, acquires sensor data and constructs a multi-source heterogeneous dataset; S2, preprocesses and standardizes data from multi-source heterogeneous datasets; S3, performing multi-order difference analysis on the data processed in step S2, and constructing a statistical model based on an improved Markov method to convert the one-dimensional sensor signal into a two-dimensional image representation with three-channel characteristics; In S4, the two-dimensional image representation is first input into the residual network, followed by dilated convolution and multi-scale feature extraction, attention mechanism and weighted fusion, and bottom-up feature fusion to obtain the feature representation of the fault data; S5, using the representation after feature extraction, constructs a few-shot dual-graph collaborative propagation network, and combines it with a dynamic early stopping method based on graph synergy improvement to train the few-shot dual-graph collaborative propagation network; S6, after the data to be tested has passed through steps S1 to S4, it is input into the trained few-sample dual-graph collaborative propagation network to identify and diagnose the fault type or state; Step S3 specifically includes the following steps: S3.1, assuming that the original signal after step S2 is x={x1,x2,…,x N }, the first-order difference is Δ 1 x t′ , the second-order difference is Δ 2 x t′ ,but: Δ 1 x t′ =x t′ -x t′-1 ,t=2,…,N (1); D 2 x t′ =D 1 x t′ -D 1 x t′-1 ,t=3,…,N (2); Among them, x t′ represents the t′th data in the original signal, and N represents a positive integer; S3.2, divide the original signal x into M bins according to the value range. Each bin corresponds to a state, where m = 1, 2, ..., M. The bin division is completed by formula (3): Among them, S m Indicates the range of the mth quantile bin, Q m is the mth quantile, Indicates that x has The data is less than or equal to the quantile of this value; Each value x in the signal is binned according to the quantile t′ Mapped to the corresponding number s t : s t =m,x t′ ∈S m (4); S3.3, the state transition matrix P represents the transition from state S i Transfer to state S j The probability P ij : Among them, n ij Indicates state S i To state S j The number of transitions, J is the total number of states; The state transition matrix P is constructed as I, where the pixel value I ij Represents the state transition probability: I ij =P ij ,i,j=1,2,…,J (6); Where I is a single-channel Markov transfer field; S3.4, for the original signal x, first-order difference Δ 1 x t′ , second-order difference Δ 2 x t′ Execute the Markov method separately to obtain three single-channel transfer field images, and merge the three single-channel transfer field images into a three-channel image I 3C : I 3C =Concat(I (0) ,I (1) ,I (2) ) (7); Among them, I (0) is the transfer field of the original signal, I (1) is the first-order difference transfer field, I (2) is the transfer field of the second-order difference, Concat represents the channel merging operation; S3.5, scale the pixel value range of the image and normalize it according to the mean and standard deviation.

2. The method for diagnosing rotor water-spraying faults of a cross-domain vehicle according to claim 1, characterized in that: Step S4 specifically includes the following steps: S4.1, input three-channel two-dimensional image After ResNet50 extracts multi-scale features, we get the feature map F f , where f = 2, 3, 4, 5; S4.2, for each feature map F f , where f = 2, 3, 4, 5, through the expansion convolution operation with different expansion rates k, three feature maps of different scales are obtained, and then the feature maps are spliced ​​along the channel dimension to form a multi-scale fusion feature map; S4.3, perform global average pooling on the multi-scale fusion feature map, generate an attention weight vector through a fully connected layer, and then perform channel-wise weighted fusion to generate an attention-enhanced feature map; S4.4, the deepest attention-enhanced feature map is used as the top-level feature map, and the shallow features are upsampled and fused in turn to form the final multi-scale fused feature set, which is finally converted into a feature vector through a global average pooling operation.

3. The method for diagnosing rotor water-spraying faults of a cross-domain vehicle according to claim 1, characterized in that: Step S5 specifically includes the following steps: S5.1, divide the fault feature data into 80% training set and 20% test set, extract the support set S and query set Q from the training set, where the support set contains labeled samples and unlabeled samples; for each labeled sample, the label of the sample is encoded by one-hot encoding. Indicates that Splice and generate the node feature vector V of the labeled sample node : Where C and d represent the vector dimensions; For unlabeled samples, the label vector is initialized to a zero vector Node feature V′ of unlabeled samples node Expressed as: Among them, Concat represents the concatenation operation; S5.2, Node Feature Matrix in Instance Graph It is composed of the node features of the support set and the query set, where N ins is the total number of samples in the support set and query set; the node features of the labeled samples in the support set are given by The node features of the unlabeled samples in the support set are composed of The label vector is set to zero vector; the node features of the query set samples are composed of Composition; all node features are combined into the instance graph node feature matrix S5.3, use K-means clustering algorithm to cluster F final Divided into N dis distribution clusters and determine the center point C of the distribution cluster k , concatenate the center point with the initialization label vector as the node feature matrix in the distribution graph Represents the distribution information of support set and query set samples in the feature space; S5.4, Instance Graph Edge Weight Matrix Initialized by feature similarity between node features, specifically expressed as: in, and They represent the features of the i-th and j-th nodes in the instance graph, σ′ is the scale parameter that controls the similarity of features; exp is the natural exponential function; Distribution graph edge weight matrix Initialized by feature similarity between distribution centers: in, and Represent the features of the kth and lth nodes in the distribution graph respectively.

4. The method for diagnosing rotor water-spraying faults of a cross-domain vehicle according to claim 3, characterized in that: Step S5 further includes: S5.5, Input instance graph node features and instance graph edge weights The instance graph node features are enhanced through the dual-graph collaborative Transformer method. The process of enhancing the instance graph node features is the same as that of enhancing the distribution graph node features. The specific steps of enhancing the instance graph node features are as follows: S5.5.1, the instance graph node features Mapped to instance graph query vector Q respectively ins , key vector K ins Sum value vector V ins , the calculation formula is: in, is the mapping weight matrix, h is the number of heads of multi-head attention, d a The dimension of a single head; S5.5.2, use the instance graph mask matrix and the instance graph edge weight matrix to calculate the attention weights between nodes; S5.5.3, the instance graph attention weight α ins and the instance graph value vector V ins Multiply them together to get the instance graph attention output X att,ins , and mapped back to the original feature dimension through linear transformation: in, is the weight matrix of the output mapping; S5.5.4, output the instance graph attention X att,ins Instance graph node features Perform residual connection and layer normalization to obtain the first step enhanced instance graph node feature X res1,ins : Among them, LayerNorm is the layer normalization operation; S5.5.5, input X res1,ins To the two-layer feedforward network FFN, and perform residual connection and layer normalization again, and finally obtain the instance graph enhanced feature :

5. The method for diagnosing rotor water-spraying faults of a cross-domain vehicle according to claim 4, characterized in that: Step S5.5.2 specifically includes the following steps: First, the query vector Q of the instance graph node i is ins,i and the key vector K of instance graph node j ins,j The dot product between them calculates the raw attention score: Among them, α raw,ij is the attention score between node i and node j, is the scaling factor, It's K ins,j The transpose of Then, combined with the edge weight matrix between node i and node j in the instance graph and instance image mask matrix M ins,ij , shield and adjust the attention score to get the final attention weight α ins,ij : Among them, Softmax is the activation function, and the instance image mask matrix M ins,ij Defined as:

6. The method for diagnosing rotor water-spraying faults of a cross-domain vehicle according to claim 4, characterized in that: Step S5 further includes: S5.6, using a collaborative attention mechanism to perform cross-graph interaction between the instance graph and the distribution graph; the cross-graph interaction process between the instance graph and the distribution graph is consistent; Cross-graph interaction of instance graphs specifically includes the following steps: S5.6.1, Enhance the features of the instance graph Mapping to cross-graph interaction attention query vector Q ins2dis , the distribution map is enhanced Mapped to the cross-graph interaction attention key vector K dis2ins and the cross-graph interaction attention value vector V dis2ins , the specific calculation is: in, is the linear mapping matrix of cross-graph interactive attention, h is the number of heads of multi-head attention, d a is the characteristic dimension of a single head; S5.6.2, by Q ins2dis and K dis2ins Calculate cross-graph attention weight α ins←dis : Among them, α ins←dis,ij represents the cross-graph attention weight between the i-th node in the instance graph and the j-th node in the distribution graph, is the scaling factor, Q ins2dis,i Enhance features for instance graphs The query vector of the i-th node, K dis2ins,j Enhance features for distribution maps The key vector of the j-th node, K dis2ins,j The transpose of S5.6.3, combine the cross-graph attention weights with the value vector V of the distribution graph dis2ins Multiply them together to get the cross-graph fusion feature V of the instance graph node ins←dis : in, Represents the representation of the i-th node in the instance graph after fusion of distribution graph features; S5.6.4, cross-graph fusion features V of instance graph nodes ins←dis Enhanced features with instance graphs Perform residual connection and layer normalization to obtain instance graph cross-graph interaction features :

7. A cross-domain vehicle rotor water-spraying fault diagnosis method according to claim 6, characterized in that: Step S5 further includes: S5.7, fusing features of the instance graph and the distribution graph nodes; the process of fusing features of the instance graph and the distribution graph nodes is the same; The instance graph node feature fusion specifically includes the following steps: S5.7.1, Enhance the features of the instance graph Cross-graph interaction features with instance graphs Splice and get the intermediate feature representation Z of the instance graph ins : Concat(·) means to and Splice along the feature dimension to get Z ins ; S5.7.2, gate coefficient g by example graph ins Perform weighted fusion on the instance graph enhancement features and the instance graph cross-graph fusion features to obtain the final node features of the instance graph : in, is the instance graph weight matrix of the fusion gate; is the bias vector; σ(·) represents the Sigmoid activation function; where ⊙ represents element-wise multiplication.

8. The cross-domain vehicle rotor water-spraying fault diagnosis method according to claim 7, characterized in that: Step S5 further includes: S5.8, updating the instance graph edge weights and the distribution graph edge weights; the instance graph edge weight update process is consistent with the distribution graph weight update process; The instance graph edge weight update specifically includes the following steps: S5.8.1, calculate the square of the characteristic difference of the node: in, Represents the square of the feature difference between the i-th node and the j-th node in the instance graph; S5.8.2, As input, the instance graph edge weight score h is obtained through two layers of MLP mapping ins,ij , the calculation formula is: Among them, W e1 and W e2 is the weight matrix of MLP; b e1 and b e2 is the bias vector; ReLU(·) represents the activation function to ensure nonlinear mapping; S5.8.3, the instance graph edge weight score h ins,ij Through exponential mapping, the edge weights of the instance graph are ensured to be positive, which can be expressed as follows: in, Represents the edge weight of the instance graph after the lth round of update.

9. A cross-domain vehicle rotor water-spraying fault diagnosis method according to claim 8, characterized in that: Step S5 further includes: S5.9, training the few-shot dual-graph collaborative propagation network based on dual-graph collaborative iteration with improved dynamic early stopping; The specific steps include: S5.9.1, for node i that supports centralized labeling, the predicted distribution is calculated using the Gaussian kernel function between nodes and normalized by Softmax, specifically expressed as follows: Where t represents the number of current iterations, is the predicted label distribution of node i; is the edge weight between nodes i and j in the instance graph; φ(y j ) is the one-hot encoding representation of the true label of node j; s (·) is the Softmax function; S5.9.2, Instance Graph Loss Obtained by the Euclidean distance between the predicted distribution and the true label: Among them, y i is the true label of node i, S label represents the set of all marked nodes in the support set; S5.9.3, integrate the losses of instance graph and distribution graph, add regularization terms to constrain model parameters W and edge weights E, and the total loss function L (t+1) Defined as: Among them, λ ins and ξ dis are the weight parameters for instance graph and distribution graph loss, λ reg is the weight coefficient of the regularization term, is the loss of the distribution map; is the instance graph edge weight matrix at iteration t+1; is the edge weight matrix of the distribution graph at the t+1 iteration; S5.9.4, rate of change of loss function ΔL (t+1) Defined as: The cooperativity distance measures the difference between node features in the instance graph and the distribution graph. Defined as: in, is the instance graph key vector at iteration t+1, Represents the feature representation of the distribution center of the same cluster as instance node i after iteration; S5.9.5: Within the set observation window Ω, if any of the following conditions is met, early stopping is triggered: Among them, ∈ loss is the loss change rate threshold, ∈ coh is the cooperativity distance threshold; After the dynamic early stop is triggered, the final features V of nodes i and j are used final,i and V final,i Perform label propagation and fault classification; the label of unlabeled node i is weighted by the Gaussian similarity with the labeled nodes in the support set: Among them, I(y j =c) represents the true label y of node j j Whether it belongs to category c; The Gaussian kernel function measures the feature similarity between nodes; is the predicted label of node i, and arg max is a function that finds the parameter that maximizes the function value.

Images