Fault diagnosis method, device, equipment and medium for hydrogen fuel heavy truck motor

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a fault diagnosis method and device of a hydrogen fuel heavy truck motor, computer equipment and a medium, and the fault diagnosis method of the hydrogen fuel heavy truck motor comprises the steps: obtaining detection frequency spectrum data corresponding to a plurality of detection vibration signals of the hydrogen fuel heavy truck motor; constructing a local geometric adjacent structure according to the adjacency relation among the multiple pieces of detection spectrum data; and inputting the local geometric adjacent structure into a target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor. The beneficial effect is that the accuracy of fault diagnosis of the hydrogen fuel heavy truck motor is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent fault diagnosis technology, and in particular to a fault diagnosis method, device, equipment and medium for a hydrogen fuel heavy truck motor. Background Art

[0002] Hydrogen fuel heavy-duty trucks are a clean and efficient means of transportation. Powered by hydrogen fuel, they offer advantages such as zero emissions, high energy density, and long driving range. Hydrogen fuel heavy-duty trucks are considered a key development direction for heavy-duty trucks in the future, and their application in industries such as logistics and transportation is increasing.

[0003] Motor fault diagnosis for hydrogen fuel cell heavy-duty trucks is crucial for ensuring safe vehicle operation. Timely diagnosis of motor faults not only prevents power outages and improves transportation efficiency, but also mitigates safety hazards such as hydrogen leaks or explosions caused by the unique characteristics of hydrogen fuel systems. Relevant technologies use shallow machine learning methods for hydrogen fuel cell heavy-duty truck motor fault diagnosis, but their accuracy remains to be improved when processing large-scale and complex hydrogen fuel cell heavy-duty truck motor fault data.

[0004] Therefore, it is urgent to propose a new fault diagnosis method for hydrogen fuel heavy truck motors. Summary of the Invention

[0005] The present application provides a method, device, equipment and medium for fault diagnosis of a hydrogen fuel heavy truck motor, which solves the technical problem in the related art that the accuracy of fault diagnosis of a hydrogen fuel heavy truck motor needs to be improved, and achieves the technical effect of improving the accuracy of fault diagnosis of a hydrogen fuel heavy truck motor.

[0006] In order to achieve the above objectives, the main technical solutions adopted in this application include:

[0007] In a first aspect, an embodiment of the present application provides a method for diagnosing a fault in a hydrogen fuel heavy truck motor, the method comprising:

[0008] Acquire detection spectrum data corresponding to each of a plurality of detection vibration signals of the hydrogen fuel heavy truck motor;

[0009] Constructing a local geometric proximity structure based on the adjacency relationship between the plurality of detected spectrum data; wherein the local geometric proximity structure is used to describe the local geometric properties between the plurality of detected vibration signals that meet a preset proximity condition;

[0010] The local geometric neighborhood structure is input into a target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor; wherein, the target fault diagnosis model is obtained by training an initial fault diagnosis model using a vibration signal sample set, and the vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label, and the proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold, and the vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure.

[0011] Optionally, the pseudo label is obtained by propagating the true label along the edge of the preset type graph structure to samples in a similar field with the help of a smoothing assumption.

[0012] Optionally, a smoothing assumption is made on the true label in the following manner:

[0013] Determining neighboring nodes of the node having the true label;

[0014] Propagate to the neighboring nodes according to the real labels along the edges in the preset type graph structure to obtain the pseudo labels corresponding to the neighboring nodes.

[0015] Optionally, the target fault diagnosis model includes a first multi-receptive field map convolution layer and a first adaptive feature fusion layer; inputting the local geometric neighborhood structure into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor includes:

[0016] Performing multi-receptive field feature extraction on the local geometric neighborhood structure through the first multi-receptive field graph convolution layer to obtain a first multi-receptive field feature set;

[0017] Using a first adaptive feature fusion layer to perform feature fusion on the first multi-receptive field feature set to obtain a first fusion node feature;

[0018] A prediction is performed based on the first fusion node feature to obtain a fault type of the hydrogen fuel heavy truck motor.

[0019] Optionally, the target fault diagnosis model further includes a second multi-receptive field map convolution layer and a second adaptive feature fusion layer; the prediction based on the first fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor includes:

[0020] Performing multi-receptive field feature extraction on the first fusion node feature through the second multi-receptive field graph convolution layer to obtain a second multi-receptive field feature set;

[0021] Using the second adaptive feature fusion layer to perform feature fusion on the second multi-receptive field feature set to obtain a second fusion node feature;

[0022] A prediction is performed based on the second fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

[0023] Optionally, the local geometric neighbor structure adopts a KNN graph structure; and the adjacency relationship between the plurality of detected spectrum data is determined by the following method, including:

[0024] determining distance data between any two of the plurality of detected spectrum data;

[0025] An adjacency relationship between the plurality of detected spectrum data is determined based on the distance data.

[0026] Optionally, the preset type graph structure corresponding to the vibration signal sample set is obtained by:

[0027] Obtaining normal state vibration signals and fault state vibration signals;

[0028] Performing a fast Fourier transform on the normal-state vibration signal to obtain normal frequency spectrum data corresponding to the normal-state vibration signal;

[0029] Performing a fast Fourier transform on the fault state vibration signal to obtain fault spectrum data corresponding to the fault state vibration signal;

[0030] The preset type graph structure is obtained by performing a proximity relationship analysis based on the normal spectrum data corresponding to the normal state vibration signal and the fault spectrum data corresponding to the fault state vibration signal.

[0031] In a second aspect, an embodiment of the present application provides a fault diagnosis device for a hydrogen fuel heavy truck motor, the device comprising:

[0032] A data acquisition module, configured to acquire detection spectrum data corresponding to each of a plurality of detection vibration signals of the hydrogen fuel heavy truck motor;

[0033] A proximity structure construction module, configured to construct a local geometric proximity structure based on the adjacency relationship between the plurality of detected spectrum data; wherein the local geometric proximity structure is used to describe the local geometric properties between the plurality of detected vibration signals that satisfy a preset proximity condition;

[0034] A fault type prediction module is used to input the local geometric neighborhood structure into a target fault diagnosis model for prediction, so as to obtain the fault type of the hydrogen fuel heavy-duty truck motor; wherein, the target fault diagnosis model is obtained by training an initial fault diagnosis model using a vibration signal sample set, and the vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label, and the proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold, and the vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure.

[0035] In a third aspect, an embodiment of the present application provides a computer device, including:

[0036] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method described in any of the above embodiments by executing the computer instructions.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to execute the method described in any of the above embodiments.

[0038] In an embodiment of the present application, first, a vibration signal sample set is constructed using a first sample set with real labels and a second sample set with pseudo labels. The initial fault diagnosis model is trained using the vibration signal sample set to obtain a target fault diagnosis model. Second, the detection spectrum data corresponding to each of the multiple detection vibration signals of the hydrogen fuel heavy truck motor is obtained. A local geometric proximity structure is constructed based on the adjacency relationship between the multiple detection spectrum data. Finally, the local geometric proximity structure is input into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor. The proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold. By performing label propagation on the real label along the path represented by the preset type graph structure corresponding to the vibration signal sample set, pseudo labels are obtained. This not only reduces the dependence on the real label and improves the efficiency of the training process, but also helps to improve the accuracy of the fault diagnosis of the target fault diagnosis model. By using the local geometric proximity structure of the detection spectrum data to predict the fault type of the hydrogen fuel heavy truck motor, it helps to explore the topological structure information between the detection vibration signals, which can further improve the accuracy of the fault diagnosis of the target fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific implementation methods of the present application or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the specific implementation methods or the description of the prior art. Obviously, the drawings described below are some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0040] Figure 1a A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0041] Figure 1b Statistical graphs of test results of different algorithm models provided in the embodiments of this specification;

[0042] Figure 1c Statistical graphs of test results of different algorithm models provided in the embodiments of this specification;

[0043] Figure 2 A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0044] Figure 3 A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0045] Figure 4 A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0046] Figure 5 Statistical graphs of test results of different algorithm models provided in the embodiments of this specification;

[0047] Figure 6 A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0048] Figure 7a A flow chart of a method for diagnosing a fault in a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0049] Figure 7b A schematic diagram of the construction process of the local geometric proximity structure provided in the embodiments of this specification;

[0050] Figure 8 A schematic diagram of a fault diagnosis device for a hydrogen fuel heavy truck motor provided in an embodiment of this specification;

[0051] Figure 9 A schematic diagram of the structure of a computer device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0052] To make the purpose, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0053] Motor fault diagnosis for hydrogen fuel cell heavy-duty trucks is crucial to ensuring safe vehicle operation. Some related technologies rely on shallow machine learning for hydrogen fuel cell heavy-duty truck motor fault diagnosis, but these methods are primarily applicable to smaller data sets and are ineffective for processing large, complex data. Others analyze the temporal characteristics of hydrogen fuel cell heavy-duty truck motor vibration signals. These techniques fail to fully exploit the potential information in the signals and often overlook the topological structure between signals, which can also provide important auxiliary information for fault diagnosis and improve the accuracy of fault identification. Some models are based on graph structures, but their effective training relies on large amounts of labeled data, making it difficult to obtain sufficient labeled data for heavy-duty truck motor fault diagnosis.

[0054] Based on this, the present application proposes a fault diagnosis method for a hydrogen fuel heavy truck motor. First, the detection spectrum data corresponding to each of the multiple detection vibration signals of the hydrogen fuel heavy truck motor is obtained; second, a local geometric proximity structure is constructed based on the adjacency relationship between the multiple detection spectrum data; finally, the local geometric proximity structure is input into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor. Among them, the local geometric proximity structure is used to describe the local geometric properties between the multiple detection vibration signals that meet the preset proximity conditions; the target fault diagnosis model is obtained by training the initial fault diagnosis model using a vibration signal sample set. The vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label. The proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold. The vibration signal sample set corresponds to a preset type graph structure. The pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure. By employing a targeted fault diagnosis model for diagnosis, large-scale and complex detection vibration signals can be efficiently processed. When the proportion of the first sample set within the vibration signal sample set is less than or equal to a preset ratio threshold, pseudo labels are generated by propagating true labels along a path represented by a preset type graph structure corresponding to the vibration signal sample set. This not only reduces reliance on true labels and improves the efficiency of the training process, but also helps improve the accuracy of fault diagnosis using the targeted fault diagnosis model. By leveraging the local geometric neighborhood structure of detection spectrum data to predict the fault type of hydrogen fuel heavy-duty truck motors, this helps to explore the topological structure information between detection vibration signals, further improving the accuracy of fault diagnosis using the targeted fault diagnosis model.

[0055] According to an embodiment of the present application, an embodiment of a fault diagnosis method for a hydrogen fuel heavy-duty truck motor is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0056] See also Figure 1a In this embodiment, a method for diagnosing a fault of a hydrogen fuel heavy truck motor is provided, the method comprising:

[0057] S110 , obtaining detection spectrum data corresponding to each of a plurality of detection vibration signals of a hydrogen fuel heavy truck motor.

[0058] The detected vibration signal can be the time-domain vibration signal of a hydrogen fuel heavy-duty truck motor. This is the vibration signal generated during motor rotation due to mechanical movement, component friction, and other factors. The detected spectrum data is frequency distribution data obtained by processing the detected vibration signal, demonstrating the intensity of the detected vibration signal at different frequencies.

[0059] In some embodiments, a vibration sensor can be installed on a hydrogen fuel heavy truck motor to collect vibration signals during motor rotation and store them in a computer via serial communication (such as RS485, CAN, or USB). The collected vibration signals are then subjected to a Fast Fourier Transform (FFT) to obtain corresponding spectrum data.

[0060] S120: Construct a local geometric proximity structure according to the adjacency relationship between the plurality of detected spectrum data.

[0061] The local geometric proximity structure describes the local geometric properties between multiple detected vibration signals that satisfy preset proximity conditions. It can be a geometric structure established by the spatial relationship between detected spectral data, reflecting the inherent connection between the detected vibration signals. An adjacency relationship can be a relationship between two detected spectral data points, based on preset adjacency conditions, to determine whether they are "adjacent" or "near neighbors." This relationship reflects the similarity of the detected spectral data in feature space.

[0062] In some embodiments, first, the Euclidean distance is used to calculate the distance between each pair of detected spectral data. Then, based on a preset adjacency condition (e.g., less than a distance threshold), spectral data that meets the condition is considered an adjacent node, and spectral data that does not meet the condition is considered a non-adjacent node. For example, spectral data whose distance to spectral data A is less than the distance threshold is considered to be its adjacent node, and spectral data whose distance is greater than or equal to the distance threshold is considered to be a non-adjacent node. Finally, a local geometric proximity structure is constructed based on the adjacency relationship between multiple nodes.

[0063] S130: Input the local geometric neighborhood structure into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor.

[0064] Among them, the target fault diagnosis model is obtained by training the initial fault diagnosis model using the vibration signal sample set. The vibration signal sample set includes a first sample set with real labels and a second sample set with pseudo labels. The proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold. The vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation (LP) of the real label along the path represented by the preset type graph structure.

[0065] The preset type graph structure is a node-edge relationship network constructed based on the preset adjacency relationship between sample data in the vibration signal sample set, where each node represents a sample data and the edge represents the degree of similarity between sample data. This preset type graph structure is used to reveal the similarity between samples in order to better capture the intrinsic relationship between sample data.

[0066] In some embodiments, the initial fault diagnosis model includes an input layer, a first multi-receptive field graph convolution layer, a first adaptive feature fusion layer, a second multi-receptive field graph convolution layer, a second adaptive feature fusion layer, a fully connected layer, and an output layer. The input layer is used to input a local geometric neighborhood structure; the first multi-receptive field graph convolution layer is used to perform multi-receptive field feature extraction on the local geometric neighborhood structure to obtain a first multi-receptive field feature set; the first adaptive feature fusion layer is used to perform feature fusion on the first multi-receptive field feature set to obtain a first fused node feature; the second multi-receptive field graph convolution layer is used to perform multi-receptive field feature extraction on the first fused node feature to obtain a second multi-receptive field feature set; the second adaptive feature fusion layer is used to perform feature fusion on the second multi-receptive field feature set to obtain a second fused node feature; the fully connected layer is used to linearly transform the features of the nodes in the second fused node feature, mapping them from the input feature space to the output fault type space; and the output layer is used to process the output of the fully connected layer and output the fault type corresponding to each node.

[0067] In some embodiments, a vibration signal sample set is constructed based on the normal state vibration signal and the fault state vibration signal of a hydrogen fuel heavy truck motor in a normal state by obtaining the normal state vibration signal and the fault state vibration signal of the hydrogen fuel heavy truck motor in a faulty state.

[0068] In some embodiments, a small number of vibration signal samples in the vibration signal sample set are manually labeled, and these vibration signals constitute a first sample set with true labels, and the remaining vibration signals belong to a second sample set. The first sample set covers vibration signals in normal state and each fault state, and its proportion in the vibration signal sample set is less than or equal to a preset proportion threshold (for example, 2%). For example, 0 represents a normal motor, 1 represents an unbalanced motor fault, 2 represents a warped motor fault, 3 represents an outer ring bearing motor fault, and 4 represents a rolling bar bearing motor fault. The true labels can be represented by 0, 1, 2, 3, and 4.

[0069] In some embodiments, a fast Fourier transform is performed on the vibration signals in the vibration signal sample set to obtain normal spectrum data and fault spectrum data (collectively referred to as spectrum data); further, a proximity analysis is performed on the spectrum data to obtain a preset type graph structure. Because the vibration signal sample set includes a first sample set with true labels, some nodes (i.e., spectrum data) in the preset type graph structure have true labels. To improve the training effect of the initial fault diagnosis model, label propagation can be performed on the true labels based on the preset type graph structure to generate pseudo labels for nodes without true labels (i.e., the second sample set).

[0070] In some embodiments, a preset type graph structure is input into the initial fault diagnosis model to train it. During the training process, the model calculates the prediction result through forward propagation and compares it with the real label or pseudo label. Then, the cross entropy loss function is used to calculate the difference between the prediction result and the actual label, and the cross entropy loss value is used as the optimization target. Then, the gradient descent algorithm or its variant (such as Adam optimizer) is used to backpropagate the loss function to update the parameters of the model to minimize the prediction error and improve the accuracy of the model. This process can be iterated until the model performs well enough on the validation set. After the training is completed, the target fault diagnosis model is obtained. Among them, the cross entropy loss function formula is:

[0071]

[0072] Among them, I is the number of labels (including real labels and pseudo labels), C is the number of fault types, is the value of the c-th dimension of the label (one-hot encoding), is the value of the c-th dimension of the predicted label (one-hot encoding).

[0073] In some implementations, a local geometric proximity structure is first constructed based on the adjacency relationship between multiple detection spectrum data points. This local geometric proximity structure is then input into a target fault diagnosis model for prediction, thereby obtaining the fault type of the hydrogen fuel heavy-duty truck motor corresponding to each node in the local geometric proximity structure. The local geometric proximity structure and the preset type graph structure are constructed based on the same preset adjacency conditions.

[0074] In some embodiments, the fault diagnosis effects of the target fault diagnosis model (denoted as LP-MRF-GCN, label propagation and multi-receptive field graph convolutional neural network) are experimentally compared with five other advanced algorithm models in the field of fault diagnosis. These five algorithm models are: graph attention network (GAT), graph convolutional neural network (GCN), label propagation graph attention convolutional neural network (LP-GAT), label propagation and graph convolutional neural network (LP-GCN) and multi-receptive field graph convolutional neural network (MRF-GCN). The main hyperparameters of each algorithm model are set as follows: the number of iterations is set to 400 times, the learning rate is set to 0.01, and it decays by 0.1 times when it iterates to 200 and 300 times, respectively. In order to ensure the reliability of the fault diagnosis results, 5 experiments are conducted on each algorithm, and the average and standard deviation of the experimental results of each experiment are calculated. Please see Table 1, and the corresponding bar chart is shown in Table 1. Figure 1b . As can be seen from the table, the average value of the LP-MRF-GCN algorithm model in the five fault diagnosis experiments is higher than that of other algorithm models, indicating that it has the best fault diagnosis effect. At the same time, the standard deviation is much lower than that of other algorithm models, indicating that it has the best stability. In addition, it can be seen that the average value of the prediction results of LP-GAT is higher than that of GAT, the average value of the prediction results of LP-GCN is higher than that of GCN, and the average value of the prediction results of LP-MRF-GCN is higher than that of MRF-GCN, indicating that the label propagation strategy can improve the accuracy of the fault diagnosis results of different algorithm models. In summary, the LP-MRF-GCN model not only outperforms other algorithm models in fault diagnosis effect, but also has significantly improved stability. The label propagation strategy it adopts is also very effective in improving the accuracy of fault diagnosis.

[0075] Table 1 Fault diagnosis effect data

[0076]

[0077] In some implementations, the single fault recognition accuracy of LP-MRF-GCN is compared with that of GAT, GCN, LP-GAT, LP-GCN, and MRF-GCN algorithm models through confusion matrix. Figure 1cIn the figure, label 0 represents a normal motor, 1 represents an unbalanced motor fault, 2 represents a warped motor fault, 3 represents an outer race bearing motor fault, and 4 represents a rolling bar bearing motor fault. As can be seen from the figure, the LP-MRF-GCN model achieves 100% recognition accuracy for unbalanced motor faults, warped motor faults, outer race bearing motor faults, and rolling bar bearing motor faults. The recognition rate under normal conditions is also relatively high at 97.89%. Although the MRF-GCN model also achieves 100% recognition accuracy for normal motors, unbalanced motor faults, warped motor faults, and outer race bearing motor faults, its recognition accuracy for rolling bar bearing motor faults is only 91.58%. The recognition accuracy of the other algorithm models is even lower. This shows that the LP-MRF-GCN model also has a significant advantage in the recognition accuracy of single faults.

[0078] In the above embodiment, the proportion of the first sample set in the vibration signal sample set is less than or equal to a preset ratio threshold. Propagating labels along the path represented by the preset type graph structure corresponding to the vibration signal sample set to obtain pseudo labels not only reduces reliance on true labels and improves the efficiency of the training process, but also helps improve the accuracy of fault diagnosis in the target fault diagnosis model. By utilizing the local geometric neighborhood structure of the detected spectral data to predict the fault type of the hydrogen fuel heavy truck motor, it helps to explore the topological structure information between the detected vibration signals, further improving the accuracy of the target fault diagnosis model.

[0079] In some embodiments, pseudo labels are obtained by propagating true labels to samples in similar domains along the edges of a preset type of graph structure using a smoothing assumption.

[0080] The pseudo labels can be labels automatically generated by algorithms. Similar domain samples can be sample data represented by nodes connected to each other by edges in a preset type of graph structure, and these sample data have high similarity in the feature space.

[0081] The smoothing hypothesis can be an assumption method that assumes that the labels of adjacent nodes in a preset type of graph structure should be the same. Therefore, label information can be propagated to adjacent nodes (i.e., similar neighborhood samples) along the edges in the preset graph structure, while label propagation is not performed for non-adjacent nodes. It can be expressed as:

[0082]

[0083] Among them, ym is the true label of node hm, and yn is the pseudo label of node hn obtained by smoothing assumption.

[0084] In some embodiments, a preset type graph structure is constructed based on the similarity between nodes. For example, the sample distance between each pair of sample data is first calculated using the Euclidean distance formula. Then, based on a preset distance threshold, sample data with a sample distance less than the distance threshold is considered to be adjacent nodes, and data with a distance greater than or equal to the distance threshold is considered to be non-adjacent points. Finally, the preset type graph structure is constructed based on this adjacency relationship, with an edge constructed between adjacent points and no edge constructed between non-adjacent points. Therefore, based on this preset type graph structure, the true label can be propagated along the edges of the preset type graph structure to adjacent nodes, i.e., samples in similar domains, with the help of a smoothing assumption.

[0085] See also Figure 2 In some embodiments, a smoothing assumption is made on the true label by:

[0086] S310: Determine neighboring nodes of the node with the true label.

[0087] S320 , propagating to neighboring nodes according to the real labels along the edges in the preset type graph structure to obtain pseudo labels corresponding to the neighboring nodes.

[0088] Among them, the neighboring nodes can be nodes without real labels in a preset type of graph structure, and are connected to nodes with real labels by edges.

[0089] In some embodiments, the edges in the preset type graph structure are provided with weight values, which can be obtained based on the similarity between the nodes. First, the nodes with real labels are used as seed nodes, and the label information of these nodes is initialized. For unlabeled nodes, their pseudo labels are initialized to a default value, such as a uniformly distributed probability vector. Then, the real label of the seed node is propagated to its neighboring nodes along the edge of the graph in an iterative manner. In each iteration, each neighboring node updates its own pseudo label based on the label information of its neighboring node and the weight value of the edge, ensuring that the label information of the neighboring node can be integrated into its pseudo label according to the weight contribution. When the termination condition is reached, such as reaching a preset number of iterations or the change in the pseudo label is less than a certain threshold, the label propagation process is stopped. After label propagation, the pseudo label of the neighboring node will be updated to the value of the real label of its closer neighboring node. These pseudo labels can be used for subsequent model training tasks to enhance the model's learning ability for unlabeled data.

[0090] See also Figure 3 In some embodiments, the target fault diagnosis model includes a first multi-receptive field map convolution layer and a first adaptive feature fusion layer; the local geometric neighborhood structure is input into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor, including:

[0091] S410 , performing multi-receptive field feature extraction on the local geometric neighborhood structure through a first multi-receptive field graph convolution layer to obtain a first multi-receptive field feature set.

[0092] S420 , using a first adaptive feature fusion layer to perform feature fusion on the first multi-receptive field feature set to obtain a first fusion node feature.

[0093] S430: Predict based on the first fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

[0094] Among them, the first multi-receptive field graph convolution layer can be a convolution layer structure that introduces multiple receptive fields (i.e., multiple neighborhood ranges) in graph neural networks (GNNs). The core idea is to capture the information of nodes and edges in the graph from multiple scales by combining receptive fields (neighborhood ranges) of different sizes, thereby enhancing the expressive power of the model. The first multi-receptive field feature set can be a set of multiple features obtained by feature extraction through the first multi-receptive field graph convolution layer. Since each receptive field outputs a feature, the features output by multiple receptive fields can constitute a feature set. The first adaptive feature fusion layer is used to perform feature fusion on the first multi-receptive field feature set, which can highlight the receptive field features that are more useful for the current task, while suppressing unimportant receptive field features.

[0095] In some implementations, the target fault diagnosis model includes an input layer, a first multi-receptive field graph convolution layer, a first adaptive feature fusion layer, a fully connected layer, and an output layer.

[0096] In some embodiments, the local geometric neighborhood structure is first input through the input layer, and then the first multi-receptive field graph convolution layer performs multi-receptive field feature extraction on the local geometric neighborhood structure. The feature output by each receptive field can be expressed as:

[0097]

[0098] Among them, X represents the node features in the local geometric neighborhood structure, H j represents the features extracted by the jth receptive field. Each receptive field corresponds to a specific neighborhood range, so different receptive fields can capture the dependencies between nodes in different ranges. σ represents the activation function, which can be a nonlinear function such as ReLU, which is used to introduce nonlinearity so that the model can learn more complex feature representations. β k are the coefficients of the polynomial; j represents the order of the polynomial; Indicates the graph convolution operation on the node feature X, is the normalized graph Laplacian matrix, which is used to describe the structural information of the graph; is a polynomial function used to transform the graph Laplacian matrix to achieve graph convolutions of different orders; v represents the number of receptive fields, that is, the number of parallel graph convolution layers. The first multi-receptive field feature set can be expressed as: {H1,H2,...,Hv}, H v ∈R N×F , where N represents the number of nodes and F represents the node feature dimension.

[0099] In some embodiments, the first adaptive feature fusion layer receives the first multi-receptive field feature set from the first multi-receptive field graph convolution layer, and first obtains the importance of each receptive field feature to the node through adaptive weight calculation (AWC), that is, the attention coefficient {w1, w2, ..., w v For example, a single-layer feedforward neural network can be used to calculate the attention coefficient, and the formula is:

[0100]

[0101] Where tanh is the activation function and α is the parameter of the feedforward neural network. The first adaptive feature fusion layer is used to fuse the first multi-receptive field feature set to obtain the first fusion node feature, which can be expressed as:

[0102]

[0103] In some embodiments, the first fused node features are input to a fully connected layer, which performs a linear transformation on the features of each node in the first fused node features, mapping them from the input feature space to the output fault type space. The output layer then processes the output of the fully connected layer through a softmax function, which converts the output value of each node into a probability distribution. This probability distribution represents the probability that the node belongs to each different fault type, i.e., the higher the probability value, the greater the possibility that the node is predicted to be of that fault type. Finally, for each node, the fault type with the highest probability is selected as the fault type corresponding to the node. In other embodiments, the classification result of the output layer is expressed by the formula: Z = softmax(FC(H 2 )), where Z represents the classification result, FC(H 2 ) represents the output of the fully connected layer.

[0104] In the above embodiment, the multi-receptive field feature extraction of the local geometric neighborhood structure is performed through the first multi-receptive field graph convolution layer, which can capture the complex relationship between nodes from different scales and ranges. The obtained first multi-receptive field feature set contains rich local geometric information. Then, the first multi-receptive field feature set is fused using the first adaptive feature fusion layer, and weights are dynamically assigned to features of different receptive fields through the attention mechanism, so that the model can automatically emphasize important features and suppress irrelevant features according to task requirements, thereby obtaining more accurate first fusion node features. Finally, prediction based on the fused features can more accurately diagnose the fault type of the hydrogen fuel heavy truck motor corresponding to each node, effectively improving the accuracy and reliability of fault diagnosis.

[0105] See also Figure 4 In some embodiments, the target fault diagnosis model further includes a second multi-receptive field map convolution layer and a second adaptive feature fusion layer; prediction based on the first fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor includes:

[0106] S510 , performing multi-receptive field feature extraction on the first fusion node feature through a second multi-receptive field graph convolution layer to obtain a second multi-receptive field feature set.

[0107] S520 , using a second adaptive feature fusion layer to perform feature fusion on the second multi-receptive field feature set to obtain a second fusion node feature.

[0108] S530: Predict based on the second fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

[0109] The second multi-receptive field map convolution layer may have the same structure as the first multi-receptive field map convolution layer.

[0110] In some embodiments, the target fault diagnosis model includes an input layer, a first multi-receptive field map convolution layer, a first adaptive feature fusion layer, a second multi-receptive field map convolution layer, a second adaptive feature fusion layer, a fully connected layer, and an output layer.

[0111] In some embodiments, the first multi-receptive field graph convolution layer has three receptive fields, and the first multi-receptive field feature set output by the first multi-receptive field convolution layer is processed by the first adaptive feature fusion layer to obtain the first fusion node feature, which can be expressed as:

[0112]

[0113] Among them, H 1 Represents the first fusion node feature.

[0114] Furthermore, the second multi-receptive field graph convolution layer also has three receptive fields. The first multi-receptive field feature set output by it is processed by the first adaptive feature fusion layer, and the second fusion node feature obtained can be expressed as:

[0115]

[0116] Among them, H 2 Represents the second fusion node feature.

[0117] It should be noted that in this embodiment, the two multi-receptive field graph convolution layers (MRF-GCN) and the two adaptive feature fusion layers each have their own focus and function, and together improve the accuracy of the target fault diagnosis model in predicting the fault type of hydrogen fuel heavy truck motors. The first MRF-GCN layer focuses on the extraction of multi-receptive field features of local geometric neighborhood structures, and can capture the local geometric relationship between nodes from different scales and ranges to obtain the first multi-receptive field feature set. Through the setting of multiple receptive fields, this layer can simultaneously consider the node's immediate and distant neighborhood information, providing rich basic features for subsequent feature fusion. Then, the first adaptive feature fusion layer fuses these features and dynamically assigns weights to features of different receptive fields through the attention mechanism, so that the model can automatically emphasize important features and suppress irrelevant features according to task requirements, thereby obtaining a more accurate first fusion node feature. The second MRF-GCN layer further performs multi-receptive field feature extraction based on the first fusion node feature output by the first adaptive feature fusion layer. It can abstract and integrate features at a higher level, capture more complex and advanced relationships between nodes, and obtain a second multi-receptive field feature set. Then, the second adaptive feature fusion layer fuses the features of the second multi-receptive field feature set to further optimize the feature expression, so that the model can more comprehensively characterize the characteristics of the node and improve the expression ability and discrimination of the node features.

[0118] After processing by two MRF-GCN layers and two adaptive feature fusion layers, the target fault diagnosis model can fully mine useful information in the local geometric neighborhood structure from different levels and angles, significantly improving the accuracy and reliability of hydrogen fuel heavy truck motor fault diagnosis.

[0119] In some embodiments, t-Distributed Stochastic Neighbor Embedding (t-SNE) is used to visualize the feature extraction effects of LP-MRF-GCN, GAT, GCN, LP-GAT, LP-GCN, and MRF-GCN algorithm models. Figure 5. As can be seen from the figure, the LP-MRF-GCN model can effectively cluster the features of different fault types and form a clear boundary. In contrast, although the MRF-GCN model can better cluster fault features, in some cases, the distance between the features of different fault types after aggregation is small, thereby increasing the risk of misclassification; the other four algorithm models have the problem of overlapping of different types of fault features, which will reduce the accuracy of their fault diagnosis. In summary, the LP-MRF-GCN algorithm model has good fault feature extraction capabilities, which can improve the accuracy and reliability of hydrogen fuel heavy truck motor fault diagnosis.

[0120] See also Figure 6 In some embodiments, the local geometric neighbor structure adopts a KNN graph structure; the adjacency relationship between the plurality of detected spectrum data is determined by the following method, including:

[0121] S610: Determine distance data between any two of a plurality of detected spectrum data.

[0122] S620: Determine an adjacency relationship between the plurality of detected spectrum data based on the distance data.

[0123] The KNN graph (k-Nearest Neighbor Graph) is a graph structure built based on the proximity relationships between data points. It effectively reflects the local similarities and overall distribution structure between data points, helping to discover patterns and features in the data. In a KNN graph, each node is connected to the k nearest other nodes, where k is a preset positive integer.

[0124] In some embodiments, the distance between any two detected spectrum data is determined using the Euclidean distance formula, which is:

[0125]

[0126] Where Lmn represents the distance between the detection spectrum data hm and the detection spectrum data hn, Represents the i-th dimension feature of the detection spectrum data hm, and d is the feature dimension of the input detection spectrum data.

[0127] In some embodiments, after determining the distance data between any two detected spectrum data points, the distances between each detected spectrum data point and all other detected spectrum data points are sorted in ascending order based on the calculated distance data. Next, for each detected spectrum data point, the first k detected spectrum data points with the smallest distances after sorting are selected as its nearest neighbors. Finally, a KNN graph is constructed, in which each node represents a detected spectrum data point. If a node is one of the k nearest neighbors of another node, an edge is established between the two nodes. This results in a KNN graph of the detected spectrum data.

[0128] In some embodiments, the formula for determining the edges in the KNN graph of the detected spectrum samples is:

[0129] A mn =KNN(k,L mn ,Ω m ),A m,n ∈A

[0130] Among them, Ω i ={L m1 ,L m2, ...,L mn}, represents the distance data set of the detected spectrum data hm and all other detected spectrum data, and k represents the hyperparameter of the spectrum sample KNN graph. If Lmn belongs to Ω i If there are k smallest values in the set, KNN(·)=1, otherwise KNN(·)=0.

[0131] In the above embodiment, a KNN graph structure is constructed by determining the distance data between any two pieces of multiple detection spectrum data and, based on this distance data, determining the adjacency relationship between the detection spectrum data. This process not only effectively extracts the features of the detection spectrum data but also, through the KNN graph structure, clearly demonstrates the similarities between the detection spectrum data. This provides high-quality input data for training the initial fault diagnosis model, helping to improve the model's fault recognition ability and diagnostic accuracy.

[0132] See also Figure 7a and Figure 7b In some embodiments, the preset type graph structure corresponding to the vibration signal sample set is obtained by:

[0133] S710: Acquire a normal state vibration signal and a fault state vibration signal.

[0134] The normal state vibration signal can be the vibration signal of a hydrogen fuel heavy truck motor operating normally, and can be relatively stable, with uniform energy distribution, low amplitude, and no significant sudden changes. The fault state vibration signal can be the vibration signal of a hydrogen fuel heavy truck motor operating normally. Since hydrogen fuel heavy truck motors have multiple fault types, the fault state vibration signal can include multiple fault types, and can be unstable, with the possibility of non-periodic shocks, etc.

[0135] In some embodiments, a vibration sensor is installed on a hydrogen fuel heavy truck motor to collect vibration signals during the rotation of the motor, and the vibration signals are stored in a computer device through serial communication (such as RS485, CAN or USB). For example, a normal hydrogen fuel heavy truck motor is selected, and a vibration signal is collected every 5 seconds by a vibration sensor to obtain a plurality of normal state vibration signals. For example, hydrogen fuel heavy truck motors with different fault types (such as unbalanced motor fault, warping motor fault, outer ring bearing motor fault and rolling bar bearing motor fault) are selected respectively, and a vibration signal is collected every 5 seconds by a vibration sensor to obtain a plurality of fault state vibration signals of different fault types.

[0136] S720 : Perform fast Fourier transform on the normal-state vibration signal to obtain normal frequency spectrum data corresponding to the normal-state vibration signal.

[0137] S730 : Perform fast Fourier transform on the fault state vibration signal to obtain fault spectrum data corresponding to the fault state vibration signal.

[0138] The normal spectrum data can be the frequency distribution information of the normal state vibration signal, which represents the intensity distribution of the normal state vibration signal in the frequency domain and can reveal the periodic components and specific frequency characteristics in the normal state vibration signal. The fault spectrum data can be the frequency distribution information of the fault state vibration signal, which represents the intensity distribution of the fault state vibration signal in the frequency domain and can reveal the periodic components and specific frequency characteristics in the fault state vibration signal.

[0139] Specifically, both the normal state vibration signal and the fault state vibration signal are time domain signals. The normal state vibration signal is subjected to a fast Fourier transform to obtain its corresponding frequency domain signal, i.e., normal spectrum data; the fault state vibration signal is subjected to a fast Fourier transform to obtain its corresponding frequency domain signal, i.e., fault spectrum data.

[0140] S740 : Perform proximity analysis based on the normal spectrum data corresponding to the normal state vibration signal and the fault spectrum data corresponding to the fault state vibration signal to obtain a preset type graph structure.

[0141] The normal spectrum data and the fault spectrum data may be collectively referred to as sample spectrum data. The proximity relationship analysis may be to analyze the distance between each pair of sample spectrum data.

[0142] In some embodiments, the Euclidean distance is used to analyze the similarity between different sample spectrum data. The preset type graph structure can be a KNN graph structure. For example, the Euclidean distance formula is first used to calculate the distance between each pair of sample spectrum data; then, based on the calculated distance value, the distance between each sample spectrum data and all other sample spectrum data is sorted in ascending order; then, for each sample spectrum data, the first k sample spectrum data with the smallest distance after sorting are selected as its nearest neighbors. Finally, a KNN graph is constructed, in which each node represents a sample spectrum data. If a node is one of the k nearest neighbors of another node, an edge is established between the two nodes, thereby obtaining a KNN graph of the sample spectrum data.

[0143] In the above embodiment, vibration signals under normal and fault conditions are acquired and subjected to fast Fourier transforms to generate corresponding spectral data. A proximity analysis is performed based on this spectral data to construct a predefined graph structure. This process not only effectively extracts the characteristics of the vibration signals but also clearly represents the similarities and differences between the spectral data under normal and fault conditions through the graph structure. This provides high-quality input data for subsequent initial fault diagnosis model training, helping to improve the model's fault recognition capabilities and diagnostic accuracy.

[0144] See also Figure 8 In this embodiment, a fault diagnosis device 800 for a hydrogen fuel heavy truck motor is provided. The fault diagnosis device 800 for a hydrogen fuel heavy truck motor includes:

[0145] The data acquisition module 810 is used to obtain detection spectrum data corresponding to multiple detection vibration signals of the hydrogen fuel heavy truck motor;

[0146] A proximity structure construction module 820 is configured to construct a local geometric proximity structure based on the adjacency relationship between the plurality of detected spectrum data; wherein the local geometric proximity structure is configured to describe the local geometric properties between the plurality of detected vibration signals that satisfy a preset proximity condition;

[0147] The fault type prediction module 830 is used to input the local geometric neighborhood structure into the target fault diagnosis model for prediction, so as to obtain the fault type of the hydrogen fuel heavy truck motor; wherein, the target fault diagnosis model is obtained by training the initial fault diagnosis model using the vibration signal sample set, and the vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label. The proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold. The vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure.

[0148] In some embodiments, the pseudo labels in the fault type prediction module 830 are obtained by propagating the true labels along the edges of the preset type graph structure to samples in similar fields with the help of a smoothing assumption.

[0149] In some implementations, the fault type prediction module 830 further includes:

[0150] A neighboring node determination unit, configured to determine neighboring nodes of a node having a true label;

[0151] The pseudo-label acquisition unit is used to propagate to adjacent nodes according to the real labels along the edges in the preset type graph structure to obtain the pseudo-labels corresponding to the adjacent nodes.

[0152] In some embodiments, the target fault diagnosis model includes a first multi-receptive field map convolution layer and a first adaptive feature fusion layer; the fault type prediction module 830 also includes:

[0153] A first feature set acquisition unit is configured to perform multi-receptive field feature extraction on the local geometric neighborhood structure through a first multi-receptive field graph convolution layer to obtain a first multi-receptive field feature set;

[0154] A first fusion feature acquisition unit is used to perform feature fusion on the first multi-receptive field feature set using a first adaptive feature fusion layer to obtain a first fusion node feature;

[0155] The first fault type prediction unit is used to predict based on the first fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

[0156] In some embodiments, the target fault diagnosis model further includes a second multi-receptive field map convolution layer and a second adaptive feature fusion layer; the fault type prediction module 830 further includes:

[0157] A second feature set acquisition unit is configured to perform multi-receptive field feature extraction on the first fusion node feature through a second multi-receptive field graph convolution layer to obtain a second multi-receptive field feature set;

[0158] A second fusion feature acquisition unit is used to perform feature fusion on the second multi-receptive field feature set using a second adaptive feature fusion layer to obtain a second fusion node feature;

[0159] The second fault type prediction unit is used to predict based on the second fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

[0160] In some embodiments, the adjacent structure construction module 820 includes:

[0161] a distance data determining unit, configured to determine distance data between any two of the plurality of detected spectrum data;

[0162] The adjacency relationship determining unit is configured to determine the adjacency relationship between the plurality of detected spectrum data based on the distance data.

[0163] In some embodiments, the fault diagnosis device 800 for a hydrogen fuel heavy truck motor further includes:

[0164] A signal acquisition module is used to acquire a normal state vibration signal and a fault state vibration signal;

[0165] A normal spectrum data acquisition module is used to perform fast Fourier transform on the normal state vibration signal to obtain normal spectrum data corresponding to the normal state vibration signal;

[0166] A fault spectrum data acquisition module is used to perform fast Fourier transform on the fault state vibration signal to obtain fault spectrum data corresponding to the fault state vibration signal;

[0167] The graph structure acquisition module is used to perform proximity relationship analysis based on the normal spectrum data corresponding to the normal state vibration signal and the fault spectrum data corresponding to the fault state vibration signal to obtain a preset type graph structure.

[0168] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0169] The fault diagnosis device for the hydrogen fuel heavy truck motor in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0170] See also Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 9As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 9 A processor 10 is taken as an example.

[0171] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0172] The memory 20 stores instructions that can be executed by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0173] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0174] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0175] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 20 can be connected via a bus or other means. Figure 9 The bus connection is taken as an example.

[0176] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touch pad, an indicator stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0177] The embodiments of the present application also provide a computer-readable storage medium. The above-mentioned method according to the embodiment of the present application can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0178] An embodiment of the present application provides a computer program product, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a method according to any embodiment of the present application.

[0179] Although the embodiments of the present application have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations shall fall within the scope defined by the appended claims.

[0180] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0181] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0182] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0183] The present application is described with reference to the flow chart and / or block diagram of the method, device (system), and computer program product according to the embodiment of the present application. It should be understood that each flow process and / or box in the flow chart and / or block diagram and the combination of the flow process and / or box in the flow chart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processing machine or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for realizing the function specified in one flow chart flow or multiple flows and / or one box or multiple boxes of the block diagram.

[0184] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0185] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0186] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0187] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0188] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

[0189] Although the embodiments of the present application have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present application, and such modifications and variations shall fall within the scope defined by the appended claims.

Claims

1. A method for fault diagnosis of a hydrogen fuel heavy truck motor, characterized in that: The method comprises: Acquire detection spectrum data corresponding to each of a plurality of detection vibration signals of the hydrogen fuel heavy truck motor; Constructing a local geometric proximity structure based on the adjacency relationship between the plurality of detected spectrum data; wherein the local geometric proximity structure is used to describe the local geometric properties between the plurality of detected vibration signals that meet a preset proximity condition; The local geometric neighborhood structure is input into a target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor; wherein, the target fault diagnosis model is obtained by training an initial fault diagnosis model using a vibration signal sample set, and the vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label, and the proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold, and the vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure.

2. The method according to claim 1, characterized in that The pseudo labels are obtained by propagating the true labels along the edges of the preset type graph structure to samples in similar fields with the help of a smoothing assumption.

3. The method according to claim 2, characterized in that A smoothing assumption is made on the true labels in the following way: Determining neighboring nodes of the node having the true label; Propagate to the neighboring nodes according to the real labels along the edges in the preset type graph structure to obtain the pseudo labels corresponding to the neighboring nodes.

4. The method according to claim 1, wherein The target fault diagnosis model includes a first multi-receptive field graph convolution layer and a first adaptive feature fusion layer; The inputting of the local geometric neighborhood structure into the target fault diagnosis model for prediction to obtain the fault type of the hydrogen fuel heavy truck motor includes: Performing multi-receptive field feature extraction on the local geometric neighborhood structure through the first multi-receptive field graph convolution layer to obtain a first multi-receptive field feature set; Using a first adaptive feature fusion layer to perform feature fusion on the first multi-receptive field feature set to obtain a first fusion node feature; A prediction is performed based on the first fusion node feature to obtain a fault type of the hydrogen fuel heavy truck motor.

5. The method according to claim 4, characterized in that The target fault diagnosis model further includes a second multi-receptive field map convolution layer and a second adaptive feature fusion layer; the prediction based on the first fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor includes: Performing multi-receptive field feature extraction on the first fusion node feature through the second multi-receptive field graph convolution layer to obtain a second multi-receptive field feature set; Using the second adaptive feature fusion layer to perform feature fusion on the second multi-receptive field feature set to obtain a second fusion node feature; A prediction is performed based on the second fusion node feature to obtain the fault type of the hydrogen fuel heavy truck motor.

6. The method according to claim 1, characterized in that The local geometric proximity structure adopts a KNN graph structure; The adjacency relationship between the plurality of detected spectrum data is determined by: determining distance data between any two of the plurality of detected spectrum data; An adjacency relationship between the plurality of detected spectrum data is determined based on the distance data.

7. The method according to claim 1, characterized in that The preset type graph structure corresponding to the vibration signal sample set is obtained by: Obtaining normal state vibration signals and fault state vibration signals; Performing a fast Fourier transform on the normal-state vibration signal to obtain normal frequency spectrum data corresponding to the normal-state vibration signal; Performing a fast Fourier transform on the fault state vibration signal to obtain fault spectrum data corresponding to the fault state vibration signal; The preset type graph structure is obtained by performing a proximity relationship analysis based on the normal spectrum data corresponding to the normal state vibration signal and the fault spectrum data corresponding to the fault state vibration signal.

8. A fault diagnosis device for a hydrogen fuel heavy truck motor, characterized in that: The device comprises: A data acquisition module, configured to acquire detection spectrum data corresponding to each of a plurality of detection vibration signals of the hydrogen fuel heavy truck motor; A proximity structure construction module, configured to construct a local geometric proximity structure based on the adjacency relationship between the plurality of detected spectrum data; wherein the local geometric proximity structure is used to describe the local geometric properties between the plurality of detected vibration signals that satisfy a preset proximity condition; A fault type prediction module is used to input the local geometric neighborhood structure into a target fault diagnosis model for prediction, so as to obtain the fault type of the hydrogen fuel heavy-duty truck motor; wherein, the target fault diagnosis model is obtained by training an initial fault diagnosis model using a vibration signal sample set, and the vibration signal sample set includes a first sample set with a real label and a second sample set with a pseudo label, and the proportion of the first sample set in the vibration signal sample set is less than or equal to a preset proportion threshold, and the vibration signal sample set corresponds to a preset type graph structure, and the pseudo label is obtained by label propagation of the real label along the path represented by the preset type graph structure.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images