An electrically powered truck drive assembly and method of monitoring the same

Abstract

The application relates to the technical field of new energy, in particular to an electric pallet truck driving assembly and a monitoring method thereof. The driving assembly effectively isolates the interference of the external environment on the internal transmission system by placing a three-stage reduction gear set and transmission components in two independent closed chambers respectively, and ensures the sealing and stability of the chambers by the cooperation of sealing elements and bearings, thereby reducing the possibility of external pollutants such as dust and moisture entering the chambers and prolonging the service life of the transmission system. The monitoring method realizes accurate monitoring and risk early warning of the sealing state by converting multi-modal data into time-frequency domain feature vectors in real time and combining a graph neural network to construct a sealing failure prediction model. Meanwhile, the maintenance plan is dynamically adjusted based on the early warning signal and risk atlas, and an optimization instruction is fed back, so that accurate allocation and efficient utilization of maintenance resources are realized.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, and in particular to an electric transport vehicle drive assembly and its monitoring method. Background Technology

[0002] With the rapid development of industrial logistics, electric pallet trucks are increasingly widely used in warehousing, manufacturing, and other fields. The performance of their drive assemblies directly affects the operating efficiency and reliability of the pallet trucks. Traditional electric pallet truck drive assemblies typically employ a single-chamber design, with low integration of the three-stage reduction gear set and seals. This makes the sealing performance susceptible to external environmental factors (such as dust and humidity), and lacks real-time monitoring methods for the sealing status, making it difficult to provide early warnings of seal failures or batch defects. Furthermore, existing monitoring methods largely rely on single-machine data analysis, failing to identify systemic risks through multi-node collaborative learning, resulting in high maintenance costs and slow fault response. To address these issues, this invention proposes an electric pallet truck drive assembly and its monitoring method. Summary of the Invention

[0003] This invention overcomes the shortcomings of the prior art and provides an electric transport vehicle drive assembly and its monitoring method.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] The first aspect of the present invention discloses an electric transport vehicle drive assembly, which includes a housing, a housing cover, wheels, a motor, a motor shaft, an electromagnetic brake, a three-stage reduction gear set, a wheel hub, a chamber I, and a chamber II;

[0006] The wheels are equipped with support components and wheel pads; the box body is equipped with a keel.

[0007] The three-stage reduction gear set includes a first gear shaft, a first-stage gear, a second gear shaft, a third gear shaft, and a large gear; the large gear is fixedly connected to the hub, and the hub is fixedly connected to the wheel's support component;

[0008] Chamber I is formed by a housing, a cover, a first seal, and a first bearing; Chamber II is formed by a housing, a support on the wheel, a second seal, a third seal, and a first bearing.

[0009] The gear sections of the first gear shaft, the first-stage gear, the second gear shaft, and the second-stage gear are located in chamber I; the gear sections of the third gear shaft and the large gear are located in chamber II.

[0010] Furthermore, a protective plate is provided on the rear end cover of the motor, which serves to protect the electromagnetic brake from impact.

[0011] Furthermore, the rear end cover of the motor has a protrusion that protrudes into the motor, and the back of the protrusion forms a groove, on which the electromagnetic brake is mounted.

[0012] Furthermore, the large gear is an external gear, and the large gear is fixedly connected to the hub.

[0013] Furthermore, the hub, wheel, and second bearing inside the hub are fitted onto the keel of the housing.

[0014] The second aspect of this invention discloses a monitoring method for an electric pallet truck drive assembly, applicable to any of the electric pallet truck drive assemblies described in any one of the claims, comprising the following steps:

[0015] Multimodal data related to the sealing state of the drive assembly chamber in each electric transport vehicle are collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance.

[0016] A sealing failure prediction model is built and deployed on the local nodes of each electric transport vehicle based on a graph neural network.

[0017] The extracted time-frequency domain feature vectors are imported into the sealing failure prediction model in the corresponding local node for prediction, and the leakage risk value of the electric transport vehicle in each local node is obtained; the local node with the leakage risk value greater than the preset threshold is marked as a warning node.

[0018] By analyzing the spatiotemporal correlation of cross nodes, leakage risk characteristics common to the early warning nodes are identified, and batch defect early warning signals and risk distribution maps are generated.

[0019] Based on the batch defect warning signals and risk distribution map, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node.

[0020] Furthermore, multimodal data related to the sealing state of the drive assembly chamber is collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance, specifically:

[0021] Multiple types of sensor arrays are deployed in chambers I and II of the drive assembly, including embedded pressure sensors, triaxial vibration sensors and infrared temperature measurement units;

[0022] The data acquisition module acquires the raw signals output by each sensor in real time, including dynamic changes in chamber pressure, vibration spectrum data of the three-stage reduction gear set, and temperature gradient distribution data of the seal surface and surrounding area.

[0023] The vibration spectrum data is converted into time-frequency domain features using short-time Fourier transform; at the same time, the periodic features of chamber pressure fluctuations and the spatial distribution features of temperature gradients are extracted as statistical features through statistical analysis of the dynamic pressure change data and temperature gradient distribution data.

[0024] The extracted time-frequency domain features are fused with statistical features to generate a multi-dimensional feature vector characterizing the sealing performance, thus obtaining the time-frequency domain feature vector characterizing the sealing performance.

[0025] Furthermore, a seal failure prediction model is constructed and deployed on the local nodes of each electric transport vehicle based on a graph neural network, specifically as follows:

[0026] By introducing a graph neural network, the time-frequency domain feature vectors are mapped to graph structure data, where nodes represent specific features collected by the sensor and edges represent the physical relationships between specific features;

[0027] Calculate the Pearson correlation coefficient between each specific feature, construct an adjacency matrix based on the Pearson correlation coefficient between each specific feature, and assign corresponding connection strengths to each node through the adjacency matrix;

[0028] The weight matrix of the graph neural network is initialized by using the time-frequency domain feature vector as the initial feature input of the nodes, and the initial weights are generated by random initialization to capture the dynamic interaction between the seal and the three-stage reduction gear set.

[0029] The node features are iteratively updated using graph convolutional layers to obtain the updated node features, and the higher-order feature representations of the updated node features are extracted.

[0030] The pooling layer aggregates the higher-order feature representations into a graph-level feature vector, which serves as the input for sealing failure prediction.

[0031] The graph-level feature vector is input into the fully connected layer, and the model is trained based on the historical fault data of the local transport electric vehicle until the model prediction accuracy meets the preset requirements, and then a sealing failure prediction model based on local data is generated.

[0032] The corresponding seal failure prediction model is deployed to the local node of the corresponding electric transport vehicle.

[0033] Furthermore, through cross-node spatiotemporal correlation analysis, leakage risk characteristics common to the early warning nodes are identified, and batch defect early warning signals and risk distribution maps are generated, specifically:

[0034] Collect the leakage risk values ​​of all early warning nodes and their corresponding time-frequency domain feature vectors, and extract key features related to sealing performance, including chamber pressure fluctuation frequency, vibration spectrum peak value and temperature gradient change rate;

[0035] Calculate the cosine similarity between key features in each early warning node, and then weight the cosine similarity between key features in each early warning node to obtain the correlation between each early warning node.

[0036] If the correlation between two sets of early warning nodes is greater than a preset correlation threshold, then these two sets of early warning nodes will be marked as early warning node groups with common leakage risk characteristics.

[0037] All early warning node groups with common leakage risk characteristics are divided into the same batch, and common characteristics of the same batch are extracted, including pressure fluctuations and vibration patterns at specific frequencies.

[0038] Based on the common characteristics of the same batch, a batch defect early warning signal is generated, and a risk distribution map is drawn by combining the geographical location and operating conditions of the early warning node.

[0039] Furthermore, based on the batch defect early warning signals and risk distribution maps, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node, specifically as follows:

[0040] Based on batch defect early warning signals, corresponding leakage risk values ​​and spatiotemporal hotspots in the risk distribution map are extracted;

[0041] Based on the extracted leakage risk values ​​and spatiotemporal hotspot areas, and combined with the available resources in the fleet maintenance resource library, a maintenance priority list is generated;

[0042] The early warning nodes are dynamically grouped according to the leakage risk value, and maintenance weights are assigned to each dynamic group to obtain the task urgency; and maintenance paths are formulated based on the geographical location of the spatiotemporal hotspot areas and the task urgency.

[0043] Based on the maintenance path and maintenance priority list, and combined with the repair efficiency and cost constraints in historical maintenance data, a dynamic maintenance strategy is formulated, which includes the batch of seal replacement, maintenance time window and resource allocation ratio.

[0044] The dynamic maintenance strategy is decomposed into node-level optimization instructions, and these instructions are sent to each node via an encrypted communication protocol.

[0045] This invention addresses the technical deficiencies in the prior art and offers the following advantages: This drive assembly, through its dual-chamber design, optimized gear-hub connection, integrated electromagnetic brake, and protective plate, effectively improves the overall performance of the electric pallet truck drive assembly. The dual-chamber structure effectively isolates the internal transmission system from external environmental interference, reduces contaminant intrusion and lubricant leakage, and extends the service life of the transmission system. The gear-hub connection design enhances transmission stability and efficiency, while the integrated electromagnetic brake improves the braking system's response speed and reliability. The protective plate further protects the electromagnetic brake from external impacts, improving equipment safety and durability. The compact overall design facilitates installation and disassembly, reduces maintenance costs, and enables the electric pallet truck to operate more stably and efficiently under complex working conditions. Simultaneously, this monitoring method improves the operational reliability and maintenance efficiency of the electric pallet truck drive assembly, reducing the risk of equipment failure and maintenance costs due to seal failure. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the first three-dimensional structure of the drive assembly;

[0048] Figure 2 This is a schematic diagram of the second three-dimensional structure of the drive assembly;

[0049] Figure 3 This is a schematic diagram of the overall structure of the three-stage reduction gear set in this drive assembly;

[0050] Figure 4 This is a schematic diagram of the first cross-sectional structure of the drive assembly;

[0051] Figure 5 This is a schematic diagram of the second cross-sectional structure of the drive assembly;

[0052] Figure 6 This is a partial structural diagram of the three-stage reduction gear set in this drive assembly;

[0053] Figure 7 This is an enlarged structural diagram of the rotating ring and protective plate in this drive assembly;

[0054] In the diagram: 1. Housing; 1.1. Keel; 2. Housing cover; 3. Wheel; 3.1. Support component; 3.2. Wheel skin; 4. Motor; 4.1. Motor shaft; 4.2. Rear end cover; 4.2.1. Protective plate; 4.2.2. Protrusion; 5. Electromagnetic brake; 6. Three-stage reduction gear set; 6.1. Gear shaft 1; 6.2. First gear; 6.3. Gear shaft 2; 6.4. Second gear; 6.5. Gear shaft 3; 6.6. Large gear; 7. Transition cover; 8. Connecting sleeve; 9. First seal; 10. Second seal; 11. Third seal; 12. Wheel hub; 13. First bearing. Detailed Implementation

[0055] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0057] The first aspect of the present invention discloses an electric transport vehicle drive assembly, which includes a housing, a housing cover, wheels, a motor, a motor shaft, an electromagnetic brake, a three-stage reduction gear set, a wheel hub, a chamber I, and a chamber II.

[0058] like Figure 1 , Figure 4 and Figure 5 The wheel 3 is provided with a support 3.1 and a wheel skin 3.2; the box body 1 is provided with a keel 1.1.

[0059] like Figure 1 , Figure 4 and Figure 5 As shown, chamber I is formed by housing 1, housing cover 2, first seal 9 and first bearing 13, and chamber II is formed by housing 1, support member 3.1 on wheel 3, second seal 10, third seal 11 and first bearing 13.

[0060] like Figure 3 and Figure 4 As shown, the gear sections of gear shaft 6.1, primary gear 6.2, gear shaft 6.3, and secondary gear 6.4 are located in chamber I; the gear sections of gear shaft 6.5 and large gear 6.6 are located in chamber II.

[0061] It should be noted that by placing the three-stage reduction gear set and transmission components in two independent enclosed chambers, interference from the external environment to the internal transmission system is effectively isolated. Simultaneously, the cooperation of seals and bearings ensures the sealing and stability of the chambers, reducing the possibility of external contaminants such as dust and moisture entering the chambers, thereby extending the service life of the transmission system. Furthermore, the dual-chamber design effectively improves the reliability and durability of the electric pallet truck's drive assembly. The enclosed structure of Chamber I and Chamber II not only effectively protects the internal three-stage reduction gear set and transmission components from external environmental corrosion, but also, through the rational layout of seals and bearings, reduces internal lubricant leakage and the intrusion of external contaminants, thus lowering maintenance costs. In addition, this design optimizes the operating environment of the transmission system, reduces energy loss due to friction and contamination, improves the transmission efficiency and overall performance of the drive assembly, enabling the electric pallet truck to operate more stably and efficiently under complex working conditions.

[0062] like Figure 4 , Figure 5 and Figure 6 As shown, the large gear 6.6 is fixedly connected to the hub 12, and the hub 12 is fixedly connected to the support 3.1 of the wheel 3. The power is output from the large gear 6.6 to the wheel 3 through the hub 12.

[0063] like Figure 6 As shown, the large gear 6.6 is an external gear, and the large gear 6.6 is fixed to the hub 12.

[0064] like Figure 4 As shown, the hub 12, the wheel 3, and the second bearing inside the hub 12 are fitted onto the keel 1.1 of the housing 1.

[0065] like Figure 4 As shown, there is a protrusion 4.2.2 on the rear end cover 4.2 of the motor 4. The protrusion 4.2.2 faces into the motor, and the back of the protrusion forms a groove, in which the electromagnetic brake 5 is installed.

[0066] It should be noted that by fixing the large gear 6.6 to the hub 12 and the hub 12 to the support 3.1 of the wheel 3, a highly efficient transmission path is achieved, where power is directly transmitted from the large gear 6.6 to the wheel 3 via the hub 12. The large gear 6.6 is designed as an external gear, further optimizing the connection strength between the gear and the hub, ensuring the stability and reliability of the transmission. Simultaneously, the hub 12, the wheel 3, and the second bearing inside the hub are fitted onto the keel 1.1 of the housing 1. This structural design enhances the support and stability of the overall assembly, reducing vibration and misalignment during operation. Furthermore, the rear end cover 4.2 of the motor 4 has a protrusion 4.2.2, with a groove formed on its back for mounting the electromagnetic brake 5. This integrated design not only saves space but also improves the matching accuracy between the brake and the motor, enhancing the response speed and reliability of the braking system.

[0067] like Figure 7 As shown, a protective plate 4.2.1 is provided on the rear end cover 4.2 of the motor 4. Figure 7 In the rotating ring formed around the center of the cross axis, the top of the protective plate 4.2.1 is closer to the rotating ring than the top of the electromagnetic brake 5, so that the protective plate 4.2.1 can protect the electromagnetic brake 5 from impacts during use of the drive assembly.

[0068] It should be noted that by setting a protective plate 4.2.1 on the rear end cover 4.2 of the motor 4, its design position is within the rotating ring formed by the center of the cross shaft, and the top of the protective plate 4.2.1 is closer to the rotating ring than the top of the electromagnetic brake 5. This layout allows the protective plate 4.2.1 to effectively cover the outer area of ​​the electromagnetic brake 5 during the operation of the drive assembly, forming a physical barrier to prevent external objects or components from directly impacting or interfering with the electromagnetic brake 5 during rotation. The design of the protective plate not only optimizes the structural layout of the rear end of the motor, but also, through its ingenious design of position and shape, maximizes the protection of the safety and functionality of the electromagnetic brake, avoiding damage or performance degradation caused by external impacts.

[0069] like Figure 2 , Figure 3 , Figure 4 and Figure 5 As shown, during operation, the power of the drive assembly is transmitted from the motor 4 to the connecting sleeve 8 via the motor shaft 4.1, and then from the connecting sleeve to the gear shaft 6.1 of the three-stage reduction gear set 6. The gear shaft 6.1 transmits the power to the first-stage gear 6.2, and then through the gear shaft 6.3 fixed to the first-stage gear 6.2, to the second-stage gear 6.4. Finally, through the gear shaft 6.5 fixed to the second-stage gear 6.4, the power is transmitted to the large gear 6.6, and then from the large gear 6.6 to the wheel 3 via the hub.

[0070] It should be noted that both chamber I and chamber II are closed cavities.

[0071] In summary, this drive assembly effectively improves the overall performance of the electric pallet truck drive assembly through its dual-chamber design, optimized gear and hub connection, integrated electromagnetic brake, and protective pads. The dual-chamber structure effectively isolates the internal transmission system from external environmental interference, reduces contaminant intrusion and lubricant leakage, and extends the service life of the transmission system. The gear and hub connection improves transmission stability and efficiency, while the integrated electromagnetic brake enhances the braking system's response speed and reliability. The protective pads further protect the electromagnetic brake from external impacts, improving the equipment's safety and durability. The compact overall design facilitates installation and disassembly, reduces maintenance costs, and enables the electric pallet truck to operate more stably and efficiently under complex working conditions.

[0072] The second aspect of this invention discloses a monitoring method for an electric pallet truck drive assembly, applicable to any of the electric pallet truck drive assemblies described in any one of the claims, comprising the following steps:

[0073] Multimodal data related to the sealing state of the drive assembly chamber in each electric transport vehicle are collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance.

[0074] A sealing failure prediction model is built and deployed on the local nodes of each electric transport vehicle based on a graph neural network.

[0075] The extracted time-frequency domain feature vectors are imported into the sealing failure prediction model in the corresponding local node for prediction, and the leakage risk value of the electric transport vehicle in each local node is obtained; the local node with the leakage risk value greater than the preset threshold is marked as a warning node.

[0076] By analyzing the spatiotemporal correlation of cross nodes, leakage risk characteristics common to the early warning nodes are identified, and batch defect early warning signals and risk distribution maps are generated.

[0077] Based on the batch defect warning signals and risk distribution map, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node.

[0078] It should be noted that a local node refers to an independent monitoring and computing unit for each electric pallet truck, which is responsible for monitoring the sealing status of the drive assembly in real time and providing local data support for global risk warning. It can be understood that each electric pallet truck in the target monitoring area (such as the production workshop) is an independent local node.

[0079] It should be noted that this method achieves accurate monitoring and risk warning of sealing status by real-time acquisition of multimodal data and conversion into time-frequency domain feature vectors, combined with graph neural networks to construct a sealing failure prediction model. Through cross-node spatiotemporal correlation analysis, common leakage risk characteristics are identified, and batch defect warning signals and risk distribution maps are generated, improving the ability to identify systemic risks. Simultaneously, maintenance plans are dynamically adjusted based on warning signals and risk maps, and optimization instructions are fed back, achieving precise allocation and efficient utilization of maintenance resources.

[0080] Furthermore, multimodal data related to the sealing state of the drive assembly chamber is collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance, specifically:

[0081] Multiple types of sensor arrays are deployed in chambers I and II of the drive assembly, including embedded pressure sensors, triaxial vibration sensors and infrared temperature measurement units;

[0082] Among them, the pressure sensor is placed near the contact surface of the seal to capture dynamic pressure fluctuations, the vibration sensor is installed on the support structure of the three-stage reduction gear set to collect meshing vibration signals, and the infrared temperature measurement unit is used to monitor the temperature gradient on the surface of the seal.

[0083] The data acquisition module acquires the raw signals output by each sensor in real time, including dynamic changes in chamber pressure, vibration spectrum data of the three-stage reduction gear set, and temperature gradient distribution data of the seal surface and surrounding area.

[0084] The vibration spectrum data is converted into time-frequency domain features using short-time Fourier transform; at the same time, the periodic features of chamber pressure fluctuations and the spatial distribution features of temperature gradients are extracted as statistical features through statistical analysis of the dynamic pressure change data and temperature gradient distribution data.

[0085] The extracted time-frequency domain features are fused with statistical features to generate a multi-dimensional feature vector characterizing the sealing performance, thus obtaining the time-frequency domain feature vector characterizing the sealing performance.

[0086] It should be noted that the collected vibration spectrum data is preprocessed to eliminate environmental interference through denoising and normalization. The preprocessed vibration spectrum data is then divided into multiple time windows using short-time Fourier transform, and the energy distribution of frequency components is calculated within each window to generate a time-frequency domain feature matrix. Simultaneously, periodic analysis is performed on the dynamic changes in chamber pressure, extracting the dominant frequency and harmonic components of pressure fluctuations using fast Fourier transform as periodic features. Next, spatial interpolation is performed on the temperature gradient distribution data to generate a temperature gradient field, and the spatial distribution characteristics of the temperature gradient are extracted by calculating the local extrema and rate of change of the gradient field. Finally, the time-frequency domain feature matrix is ​​integrated with the periodic and spatial distribution features to generate a multi-dimensional feature vector characterizing the sealing performance.

[0087] It should be noted that traditional electric pallet truck drive assemblies lack precise monitoring methods for the chamber sealing status, making it difficult to obtain multi-dimensional characteristics of sealing performance in real time, thus hindering early warning of sealing failure risks. This invention, supported by a multi-type sensor array, collects multi-modal data such as chamber pressure, vibration spectrum, and temperature gradient in real time. It then uses short-time Fourier transform and statistical analysis to convert this data into time-frequency domain feature vectors, generating multi-dimensional feature vectors characterizing sealing performance. This achieves a comprehensive and accurate characterization of the sealing status, providing a data foundation for subsequent sealing failure prediction.

[0088] Furthermore, a seal failure prediction model is constructed and deployed on the local nodes of each electric transport vehicle based on a graph neural network, specifically as follows:

[0089] By introducing a graph neural network, the time-frequency domain feature vectors are mapped to graph structure data, where nodes represent specific features collected by the sensor (such as chamber pressure, vibration spectrum, and temperature gradient), and edges represent the physical correlation between specific features (such as the coupling relationship between chamber pressure and gear vibration).

[0090] Calculate the Pearson correlation coefficient between each specific feature, construct an adjacency matrix based on the Pearson correlation coefficient between each specific feature, and assign corresponding connection strengths to each node through the adjacency matrix;

[0091] It should be noted that the Pearson correlation coefficient between specific features (such as chamber pressure, vibration spectrum, and temperature gradient) is calculated based on multimodal data collected by sensors, serving as a quantitative indicator of the connection strength between nodes. For example, the correlation coefficient between chamber pressure and gear vibration spectrum is 0.85, while the correlation coefficient between temperature gradient and chamber pressure is 0.72. The correlation coefficient matrix is ​​used as the initial value of the adjacency matrix, where the matrix element (A_{ij}) represents the connection strength between node (i) and node (j). For example, if the correlation coefficient between node 1 (chamber pressure) and node 2 (vibration spectrum) is 0.85, then in the adjacency matrix (A_{12}=0.85), the connection strength between node 1 and node 2 is assigned as 0.85.

[0092] The weight matrix of the graph neural network is initialized by using the time-frequency domain feature vector as the initial feature input of the nodes, and the initial weights are generated by random initialization to capture the dynamic interaction between the seal and the three-stage reduction gear set.

[0093] It should be noted that the size of the weight matrix is ​​defined according to the input layer dimension of the graph neural network (determined by the dimension of the time-frequency domain feature vectors). For example, if the node feature dimension is (n), then the size of the weight matrix is ​​n multiplied by m, where (m) is the hidden layer dimension. Then, a random initialization method (such as Xavier initialization or He initialization) is used to generate the initial values ​​of the weight matrix, ensuring that the weights are randomly distributed within a reasonable range to avoid gradient vanishing or exploding. The time-frequency domain feature vectors are used as the initial features of the nodes and input into the graph neural network, where the feature vector of each node corresponds to a node in the graph structure. The initial node features are mapped to the hidden layer through the weight matrix to capture the dynamic interaction between the seal and the three-stage reduction gear set.

[0094] The node features are iteratively updated using graph convolutional layers to obtain the updated node features, and the higher-order feature representations of the updated node features are extracted.

[0095] It should be noted that the initial node features and the adjacency matrix are input into the graph convolutional layer. Weighted aggregation of node features is calculated through matrix multiplication, where the adjacency matrix defines the connection strength between nodes. The aggregated node features are multiplied by the weight matrix and transformed using a non-linear activation function (such as ReLU) to generate updated node features. Then, the updated node features are used as input to the next layer of graph convolution, and the above steps are repeated iteratively to gradually extract higher-order feature representations. Finally, through multi-layer graph convolution operations, the complex interactions between nodes are captured, generating node feature vectors containing higher-order feature representations.

[0096] The pooling layer aggregates the higher-order feature representations into a graph-level feature vector, which serves as the input for sealing failure prediction.

[0097] It should be noted that the high-order feature representations of each node output from the graph convolutional layer are input into the pooling layer. Global average pooling or global max pooling is used to aggregate the features of all nodes, generating a single-dimensional graph-level feature vector. Next, the pooled graph-level feature vector is normalized to ensure its feature values ​​are within a reasonable range and to avoid numerical instability. Then, the normalized graph-level feature vector is used as input to the sealing failure prediction model for subsequent classification or regression tasks.

[0098] The graph-level feature vector is input into the fully connected layer, and the model is trained based on the historical fault data of the local transport electric vehicle until the model prediction accuracy meets the preset requirements, and then a sealing failure prediction model based on local data is generated.

[0099] It should be noted that the graph-level feature vector output from the pooling layer is input into the fully connected layer. A preliminary prediction result is generated through a linear transformation of the weight matrix and bias vector. Then, a training set is constructed based on historical fault data of the local transport vehicle (including records of seal leakage, wear data of the three-stage reduction gear set, abnormal chamber pressure records, abnormal vibration spectrum data, and abnormal temperature gradient data). The graph-level feature vector is matched with the corresponding fault labels. The error between the predicted result and the true label is calculated using a loss function (such as cross-entropy loss or mean squared error), and the weights and biases of the fully connected layer are updated using the backpropagation algorithm. Finally, through iterative training, the model's prediction accuracy reaches a preset requirement (e.g., accuracy exceeds 95% or the loss function value is below a threshold), generating a seal failure prediction model based on local data.

[0100] It's important to note that the primary purpose of building and deploying a seal failure prediction model based on local data is to accurately predict seal failure risks by utilizing real-time and historical operating data of the local electric pallet truck drive assembly, providing highly tailored support for maintenance decisions on local equipment. The advantage of local data lies in its ability to fully reflect the unique operating conditions of specific equipment in a specific environment (such as load, temperature, and humidity), as well as the wear and aging patterns of local seals, thereby ensuring the accuracy and applicability of the prediction model. Compared to remote data, local data more closely reflects the actual operating state of the equipment, capturing changes in seal performance caused by differences in location, usage frequency, and maintenance conditions, avoiding prediction biases due to insufficient data generalization. Therefore, by training the model based on local data, customized seal failure warnings and maintenance recommendations can be provided for local equipment, improving the accuracy of predictions and the timeliness of maintenance.

[0101] The corresponding seal failure prediction model is deployed to the local node of the corresponding electric transport vehicle.

[0102] It should be noted that in the field of electric pallet trucks, traditional seal failure prediction models often struggle to capture the complex correlations between multimodal features, resulting in insufficient prediction accuracy. Therefore, this method introduces a graph neural network to map time-frequency domain feature vectors into graph-structured data. It then utilizes the Pearson correlation coefficient to construct an adjacency matrix and combines graph convolutional and pooling layers to extract high-order feature representations, generating graph-level feature vectors as input for seal failure prediction. This enables accurate modeling of the dynamic interaction between the seal and the three-stage reduction gear set, improving the accuracy and reliability of seal failure prediction and providing intelligent support for the stable operation and maintenance decisions of electric pallet trucks.

[0103] Furthermore, through cross-node spatiotemporal correlation analysis, leakage risk characteristics common to the early warning nodes are identified, and batch defect early warning signals and risk distribution maps are generated, specifically:

[0104] Collect the leakage risk values ​​of all early warning nodes and their corresponding time-frequency domain feature vectors, and extract key features related to sealing performance, including chamber pressure fluctuation frequency, vibration spectrum peak value and temperature gradient change rate;

[0105] Calculate the cosine similarity between key features in each early warning node, and then weight the cosine similarity between key features in each early warning node to obtain the correlation between each early warning node.

[0106] It should be noted that key features related to sealing performance are extracted from all early warning nodes, including chamber pressure fluctuation frequency, vibration spectrum peak value, and temperature gradient change rate. Next, the cosine similarity of the key features between each early warning node is calculated. For example, the cosine similarity between node A and node B is 0.92 for chamber pressure fluctuation frequency, 0.85 for vibration spectrum peak value, and 0.78 for temperature gradient change rate. Based on the degree of influence of each key feature on sealing performance, corresponding weights are assigned in advance (e.g., chamber pressure fluctuation frequency weight is 0.5, vibration spectrum peak value weight is 0.3, and temperature gradient change rate weight is 0.2). Finally, the cosine similarity of each key feature is multiplied by its weight and summed to obtain the weighted correlation value between nodes. For example, the weighted correlation value between node A and node B is (0.9 x 0.5 + 0.85 x 0.3 + 0.78 x 0.2 = 0.876).

[0107] If the correlation between two sets of early warning nodes is greater than a preset correlation threshold, then these two sets of early warning nodes will be marked as early warning node groups with common leakage risk characteristics.

[0108] All early warning node groups with common leakage risk characteristics are divided into the same batch, and common characteristics of the same batch are extracted, including pressure fluctuations and vibration patterns at specific frequencies.

[0109] Based on the common characteristics of the same batch, a batch defect early warning signal is generated, and a risk distribution map is drawn by combining the geographical location and operating conditions of the early warning node.

[0110] It should be noted that, firstly, the geographical location information (such as latitude and longitude coordinates) and operational data (such as load, operating time, and ambient temperature) of all early warning nodes are acquired. Common leakage risk characteristics of the nodes (such as pressure fluctuations at specific frequencies and vibration patterns) are correlated with the geographical location information to generate spatial data points containing risk levels. Then, using a Geographic Information System (GIS) or visualization tools, the spatial data points are mapped onto a map and color-coded according to risk level (e.g., high risk in red, medium risk in yellow, and low risk in green). Finally, combined with the operational data, key operational information (such as high-load areas and high-temperature areas) is marked on the risk distribution map, generating an intuitive risk distribution map.

[0111] It should be noted that "the same batch" refers to a group of early warning nodes with common leakage risk characteristics. That is, these early warning nodes show high similarity in key characteristics (such as chamber pressure fluctuation frequency, vibration spectrum peak value, and temperature gradient change rate), indicating that they may have the same sealing defects or failure modes. Suppose there are 10 electric pallet trucks in a fleet, and 5 of them show high similarity in chamber pressure fluctuation frequency and vibration spectrum peak value. Moreover, these vehicles use the same batch of seals. Then these 5 vehicles will be classified as the same batch, and a batch defect early warning signal will be generated, indicating that the seals in this batch may have systemic defects.

[0112] It should be noted that traditional monitoring methods struggle to identify common leakage risk characteristics across multiple nodes, resulting in a lack of systematic early warning for batch defects. Therefore, this invention employs cross-node spatiotemporal correlation analysis to calculate the cosine similarity of key features among early warning nodes, identifying node groups with common leakage risks and generating batch defect early warning signals and risk distribution maps. This enables systematic identification and visualization of batch defects, enhancing the early warning capability for common leakage risks, providing a scientific basis for the dynamic adjustment of fleet maintenance plans, and effectively reducing equipment failure risks and maintenance costs caused by batch defects.

[0113] Furthermore, based on the batch defect early warning signals and risk distribution maps, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node, specifically as follows:

[0114] Based on batch defect early warning signals, corresponding leakage risk values ​​and spatiotemporal hotspots in the risk distribution map are extracted;

[0115] It should be noted that spatiotemporal hotspots refer to areas with a high concentration of leakage risk in both time and space. These are determined by analyzing leakage risk values ​​and their geographical location and temporal information in a risk distribution map. Specifically, spatiotemporal hotspots include geographical areas (such as high-frequency operation areas in a warehouse or specific equipment concentration areas in a manufacturing workshop) where multiple high-leakage-risk nodes appear within a specific time period (e.g., high-load operation periods or high-temperature environment periods). These areas are typically closely related to abnormal wear of seals, increased vibration of three-stage reduction gear sets, or fluctuations in chamber pressure, and are high-incidence areas for batch defects.

[0116] Based on the extracted leakage risk value and spatiotemporal hotspot areas, and combined with the available resources in the fleet maintenance resource library (such as spare parts inventory and manpower dispatch), a maintenance priority list is generated.

[0117] For example, firstly, based on the batch defect warning signal, the leakage risk value (e.g., high risk, medium risk, low risk) and spatiotemporal hotspot areas (e.g., high load areas, high temperature areas) in the risk distribution map are extracted for each warning node. Next, available resources in the fleet maintenance resource library are queried, including spare parts inventory (e.g., quantity of seals, inventory of three-stage reduction gear sets) and manpower scheduling (e.g., number of technicians, availability of maintenance vehicles). Then, the warning nodes are prioritized according to the leakage risk value; for example, high-risk nodes are prioritized over medium-risk nodes, and medium-risk nodes are prioritized over low-risk nodes. Simultaneously, the geographical location of the spatiotemporal hotspot areas is considered to optimize maintenance routes to reduce time costs. Finally, the priority ranking results are matched with maintenance resources to generate a maintenance priority list; for example, high-risk node A prioritizes the use of inventory seals and technician A, while medium-risk node B prioritizes the use of maintenance vehicle B.

[0118] The early warning nodes are dynamically grouped according to the leakage risk value, and maintenance weights are assigned to each dynamic group to obtain the task urgency; and maintenance paths are formulated based on the geographical location of the spatiotemporal hotspot areas and the task urgency.

[0119] For example, the nodes are first dynamically grouped based on their leakage risk values ​​(e.g., high risk, medium risk, low risk). For instance, high-risk nodes are assigned to group A, medium-risk nodes to group B, and low-risk nodes to group C. Next, a maintenance weight is assigned to each group; for example, group A has a maintenance weight of 0.6, group B 0.3, and group C 0.1. These weights reflect the urgency of the task. Then, based on the geographical location of spatiotemporal hotspots (e.g., high-frequency operation areas in a warehouse or specific equipment concentration areas in a manufacturing workshop) and the urgency of the task, an optimal maintenance path is determined. For example, technicians are prioritized to start maintenance from the high-weight group A, and the path is optimized according to geographical location to reduce travel time, such as from node A1 to node A2 and then to node A3, finally addressing the nodes in the medium and low-weight groups.

[0120] Based on the maintenance path and maintenance priority list, and combined with the repair efficiency and cost constraints in historical maintenance data, a dynamic maintenance strategy is formulated, which includes the batch of seal replacement, maintenance time window and resource allocation ratio.

[0121] It should be noted that, based on the maintenance path and priority list, the maintenance sequence and schedule for each early warning node are determined. Historical maintenance data is queried to assess repair efficiency (e.g., average time for a single seal replacement) and cost constraints (e.g., spare parts cost, labor cost), calculating the maintenance time window and resource requirements for each node. Then, according to the maintenance sequence and time window, the seal replacement task is divided into multiple batches to ensure that each batch of maintenance tasks is reasonably allocated in terms of time and resources. Finally, resource allocation ratios are optimized based on cost constraints; for example, allocating more technical personnel and spare parts resources to high-priority nodes and fewer resources to low-priority nodes, generating a dynamic maintenance strategy.

[0122] The dynamic maintenance strategy is decomposed into node-level optimization instructions (such as adjusting the lubrication cycle of a specific node and updating sensor calibration parameters), and these instructions are sent to each node through an encrypted communication protocol.

[0123] It should be noted that traditional fleet maintenance plans lack the ability to dynamically respond to batch defect risks, resulting in low maintenance efficiency and unreasonable resource allocation. Therefore, this invention, based on batch defect early warning signals and risk distribution maps, combined with a maintenance resource library and historical data, dynamically generates a maintenance priority list, maintenance paths, and optimization strategies, and feeds back optimization instructions to each node. This enables intelligent adjustment of the fleet maintenance plan, improves the utilization efficiency of maintenance resources, reduces the risk of equipment failure due to batch defects, and ensures the efficient execution of maintenance strategies through the precise issuance of node-level optimization instructions, providing a reliable guarantee for the stable operation of the fleet.

[0124] In this embodiment, the monitoring method may further include the following steps:

[0125] Real-time acquisition of operating data of electric pallet trucks (such as ambient humidity, load intensity, and operating speed), and extraction of operating characteristic parameters related to sealing performance (such as the influence coefficient of humidity on the aging of seals and the frequency of load fluctuations).

[0126] A dynamic weight matrix is ​​constructed based on the operating condition characteristic parameters, where the weight values ​​are dynamically adjusted by the matching degree between the operating condition parameters and historical data (such as increasing the weight of temperature gradient characteristics under high humidity conditions).

[0127] By combining the dynamic weight matrix with the attention mechanism, the attention weights of abnormally sensitive frequency bands (such as specific vibration frequency ranges or pressure fluctuation cycles) in the time-frequency domain feature vector are calculated to generate the attention weight matrix.

[0128] By using an attention weight matrix to weight and fuse time-frequency domain feature vectors, the feature representation of abnormally sensitive frequency bands is enhanced, and anti-interference enhanced feature vectors are generated.

[0129] Specifically, the time-frequency domain feature vector is multiplied element-wise with the attention weight matrix, where the weights for abnormally sensitive frequency bands (such as specific vibration frequency ranges or pressure fluctuation cycles) are higher than those for other frequency bands. The weighted feature vector is then normalized to ensure that the numerical ranges of the features in each frequency band are consistent, avoiding numerical deviations caused by weight allocation. Next, a feature fusion layer superimposes the weighted feature vector with the original time-frequency domain feature vector to generate a fused feature vector containing enhanced frequency band information. Finally, a non-linear activation function is used to transform the fused feature vector to generate an anti-interference enhanced feature vector.

[0130] The anti-interference enhanced feature vector is input into the sealing failure prediction model for training and prediction. Based on the deviation between the prediction results and the actual leakage risk, the adjustment strategy of the dynamic weight matrix is ​​optimized in reverse to form a closed-loop feedback mechanism.

[0131] In summary, this method combines dynamic weight allocation driven by operating condition data with an attention mechanism to achieve precise enhancement of key characteristic frequency bands, effectively suppressing the interference of environmental noise and operating condition fluctuations on the prediction model, and further improving the robustness and accuracy of the sealing failure prediction model.

[0132] In this embodiment, the monitoring method may further include the following steps:

[0133] Extract the key trajectory nodes of the electric transport vehicle in the work area (such as loading and unloading points, charging stations, and high-frequency work areas) corresponding to the early warning nodes, and construct the running trajectory topology network of the electric transport vehicle including its position coordinates and motion status;

[0134] The leakage risk values ​​in the risk distribution map are mapped to the corresponding nodes in the topology network to generate a topology network structure with risk weights. The spatiotemporal correlation strength between nodes in the topology network structure is analyzed by calculating trajectory similarity to generate a correlation strength matrix.

[0135] The correlation strength matrix is ​​used as the basis for the initial distribution of pheromones in the ant colony algorithm. The risk gradient change rate is introduced as a heuristic factor to iteratively optimize the overall cost of maintaining the path. Real-time operating data of the path nodes are collected synchronously in each iteration to dynamically adjust the pheromone volatilization coefficient and the path energy efficiency evaluation threshold.

[0136] Once the overall cost of the maintenance path is less than the preset cost threshold, the optimal maintenance path sequence that satisfies both risk coverage and response time constraints is output. The working parameters and energy consumption prediction values ​​of the path nodes are encapsulated into a scheduling instruction set, and multi-dimensional data real-time mapping is achieved with the fleet scheduling system through a digital twin interface.

[0137] Simultaneously, based on the changes in the path node status reported by the scheduling execution, the correlation strength parameters of the topology network model are corrected in reverse, forming a closed-loop optimization mechanism.

[0138] It should be noted that traditional maintenance path planning lacks the ability to dynamically adapt to the spatiotemporal correlation between the electric pallet truck's trajectory and leakage risk, resulting in low maintenance efficiency and significant resource waste. Therefore, this method constructs a trajectory topology network with risk weights, combines ant colony optimization with real-time operating data to dynamically optimize paths, and generates an optimal maintenance path sequence that meets both risk coverage and response time constraints. This sequence is then simultaneously updated to the fleet dispatch system. This enables precise planning and dynamic adjustment of maintenance paths, improving maintenance efficiency and resource utilization, reducing maintenance delays and cost waste caused by unreasonable paths, and continuously optimizing path planning strategies through a feedback mechanism, providing reliable support for intelligent operation and maintenance of the fleet.

[0139] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A monitoring method for the drive assembly of an electric pallet truck, characterized in that: The electric pallet truck drive assembly includes a housing, a housing cover, wheels, a motor, a motor shaft, an electromagnetic brake, a three-stage reduction gear set, a wheel hub, chamber I, and chamber II; The wheels are equipped with support components and wheel pads; the box body is equipped with a keel. The three-stage reduction gear set includes a first gear shaft, a first-stage gear, a second gear shaft, a third gear shaft, and a large gear; the large gear is fixedly connected to the hub, and the hub is fixedly connected to the wheel's support component; Chamber I is formed by a housing, a cover, a first seal, and a first bearing; Chamber II is formed by a housing, a support on the wheel, a second seal, a third seal, and a first bearing. The gear sections of the first gear shaft, the first-stage gear, the second gear shaft, and the second-stage gear are located in chamber I; the gear sections of the third gear shaft and the large gear are located in chamber II. The monitoring method includes: Multimodal data related to the sealing state of the drive assembly chamber in each electric transport vehicle are collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance. A sealing failure prediction model is built and deployed on the local nodes of each electric transport vehicle based on a graph neural network. The extracted time-frequency domain feature vectors are imported into the sealing failure prediction model in the corresponding local node for prediction, and the leakage risk value of the electric transport vehicle in each local node is obtained; the local node with the leakage risk value greater than the preset threshold is marked as a warning node. By analyzing the spatiotemporal correlation of cross nodes, leakage risk characteristics common to the early warning nodes are identified, and batch defect early warning signals and risk distribution maps are generated. Based on the batch defect early warning signals and risk distribution map, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node; Specifically, cross-node spatiotemporal correlation analysis is used to identify leakage risk characteristics common to the early warning nodes, generating batch defect early warning signals and risk distribution maps, as follows: Collect the leakage risk values ​​of all early warning nodes and their corresponding time-frequency domain feature vectors, and extract key features related to sealing performance, including chamber pressure fluctuation frequency, vibration spectrum peak value and temperature gradient change rate; Calculate the cosine similarity between key features in each early warning node, and then weight the cosine similarity between key features in each early warning node to obtain the correlation between each early warning node. If the correlation between two sets of early warning nodes is greater than a preset correlation threshold, then these two sets of early warning nodes will be marked as early warning node groups with common leakage risk characteristics. All early warning node groups with common leakage risk characteristics are divided into the same batch, and common characteristics of the same batch are extracted, including pressure fluctuations and vibration patterns at specific frequencies. Based on the common characteristics of the same batch, a batch defect early warning signal is generated, and a risk distribution map is drawn by combining the geographical location and operating conditions of the early warning node; Also includes: Extract the key trajectory nodes of the electric transport vehicle in the work area corresponding to the early warning nodes, and construct the running trajectory topology network of the electric transport vehicle including its position coordinates and motion status; The leakage risk values ​​in the risk distribution map are mapped to the corresponding nodes in the topology network to generate a topology network structure with risk weights. The spatiotemporal correlation strength between nodes in the topology network structure is analyzed by calculating trajectory similarity to generate a correlation strength matrix. The correlation strength matrix is ​​used as the basis for the initial distribution of pheromones in the ant colony algorithm. The risk gradient change rate is introduced as a heuristic factor to iteratively optimize the overall cost of maintaining the path. Real-time operating data of the path nodes are collected synchronously in each iteration to dynamically adjust the pheromone volatilization coefficient and the path energy efficiency evaluation threshold. Once the overall cost of the maintenance path is less than the preset cost threshold, the optimal maintenance path sequence that satisfies both risk coverage and response time constraints is output. The working parameters and energy consumption prediction values ​​of the path nodes are encapsulated into a scheduling instruction set, and multi-dimensional data real-time mapping is achieved with the fleet scheduling system through a digital twin interface. Simultaneously, based on the changes in the path node status reported by the scheduling execution, the correlation strength parameters of the topology network model are corrected in reverse, forming a closed-loop optimization mechanism.

2. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that: The motor has a protective plate on its rear end cover, which protects the electromagnetic brake from impact.

3. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that: The rear end cover of the motor has a protrusion that protrudes into the motor, and the back of the protrusion forms a groove, on which the electromagnetic brake is mounted.

4. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that: The large gear is an external gear, and it is fixed to the hub.

5. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that: The hub, wheel, and second bearing inside the hub are fitted onto the keel of the housing.

6. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that, Multimodal data related to the sealing state of the drive assembly chamber in each electric transport vehicle is collected in real time at each local node, and the multimodal data is converted into a time-frequency domain feature vector characterizing the sealing performance, specifically: Multiple types of sensor arrays are deployed in chambers I and II of the drive assembly, including embedded pressure sensors, triaxial vibration sensors and infrared temperature measurement units; The data acquisition module acquires the raw signals output by each sensor in real time, including dynamic changes in chamber pressure, vibration spectrum data of the three-stage reduction gear set, and temperature gradient distribution data of the seal surface and surrounding area. The vibration spectrum data is converted into time-frequency domain features using the short-time Fourier transform. Simultaneously, by statistically analyzing the dynamic pressure change data and temperature gradient distribution data, the periodic characteristics of chamber pressure fluctuations and the spatial distribution characteristics of temperature gradients are extracted as statistical features. The extracted time-frequency domain features are fused with statistical features to generate a multi-dimensional feature vector characterizing the sealing performance, thus obtaining the time-frequency domain feature vector characterizing the sealing performance.

7. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that, A sealing failure prediction model is built and deployed on the local nodes of each electric transport vehicle based on a graph neural network, specifically as follows: By introducing a graph neural network, the time-frequency domain feature vectors are mapped to graph structure data, where nodes represent specific features collected by the sensor and edges represent the physical relationships between specific features; Calculate the Pearson correlation coefficient between each specific feature, construct an adjacency matrix based on the Pearson correlation coefficient between each specific feature, and assign corresponding connection strengths to each node through the adjacency matrix; The weight matrix of the graph neural network is initialized by using the time-frequency domain feature vector as the initial feature input of the nodes, and the initial weights are generated by random initialization to capture the dynamic interaction between the seal and the three-stage reduction gear set. The node features are iteratively updated using graph convolutional layers to obtain the updated node features, and the higher-order feature representations of the updated node features are extracted. The pooling layer aggregates the higher-order feature representations into a graph-level feature vector, which serves as the input for sealing failure prediction. The graph-level feature vector is input into the fully connected layer, and the model is trained based on the historical fault data of the local transport electric vehicle until the model prediction accuracy meets the preset requirements, and then a sealing failure prediction model based on local data is generated. The corresponding seal failure prediction model is deployed to the local node of the corresponding electric transport vehicle.

8. The monitoring method for the drive assembly of an electric pallet truck according to claim 1, characterized in that, Based on the batch defect early warning signals and risk distribution map, the fleet maintenance plan is dynamically adjusted and optimization instructions are fed back to each node, specifically: Based on batch defect early warning signals, corresponding leakage risk values ​​and spatiotemporal hotspots in the risk distribution map are extracted; Based on the extracted leakage risk values ​​and spatiotemporal hotspot areas, and combined with the available resources in the fleet maintenance resource library, a maintenance priority list is generated. The early warning nodes are dynamically grouped according to the leakage risk value, and maintenance weights are assigned to each dynamic group to determine the urgency of the task. Maintenance paths were developed based on the geographical location and task urgency of the spatiotemporal hotspots. Based on the maintenance path and maintenance priority list, and combined with the repair efficiency and cost constraints in historical maintenance data, a dynamic maintenance strategy is formulated, which includes the batch of seal replacement, maintenance time window and resource allocation ratio. The dynamic maintenance strategy is decomposed into node-level optimization instructions, and these instructions are sent to each node via an encrypted communication protocol.

Images