Electromechanical hydraulic equipment fault diagnosis method and system based on bonding graph structure

Abstract

The invention provides an electromechanical hydraulic equipment fault diagnosis method and system based on a bonding graph structure, and belongs to the field of hydraulic equipment fault diagnosis. The problems that in current electro-hydraulic system fault diagnosis, integration of physical information and a data driving method is insufficient, and the capacity of tracking a composite fault source is limited are solved. According to the strong coupling characteristic of the electro-hydraulic system, a power bonding graph is adopted as a physical constraint carrier, the core characteristics of a system flow transmission path and component interaction logic are converted into quantifiable mathematical constraint equations, and the equations clearly express the inherent physical relation of the strong coupling system; through the framework, accurate diagnosis and interpretable tracking of the composite fault are realized, so that the model makes a fundamental breakthrough in the aspects of physical interpretability, composite fault tracking and cross-subsystem correlation analysis.

Description

Technical Field

[0001] This invention relates to the field of hydraulic equipment fault diagnosis technology, and more specifically, to a method and system for diagnosing electromechanical hydraulic equipment based on a bond graph structure. Background Technology

[0002] As the core power and control unit of engineering machinery, electromechanical hydraulic systems undertake critical tasks including energy transmission and motion actuation, playing a decisive role in ensuring operational continuity and stability. These systems are highly complex, involving deep coupling of multiple subsystems, with dynamic interactions between energy and control flows among various components and branches. System failures can lead not only to significant economic losses due to downtime but also to equipment damage and even personal injury, severely hindering effective engineering operations and safety assurance. Therefore, conducting research on fault diagnosis for electromechanical hydraulic systems is of great importance for improving reliability, reducing maintenance costs, and mitigating safety risks.

[0003] In the field of electromechanical and hydraulic fault diagnosis, several persistent industry challenges remain unresolved. Regarding complex fault diagnosis, while identifying anomalies from a single sensor is relatively straightforward, diagnosing complex situations such as multi-pump coordination failures is far more challenging. Due to the deep coupling between subsystems, the causal relationships between different fault sources exhibit complex interdependencies, which traditional diagnostic methods often cannot effectively address. Existing research indicates that traditional methods suffer from high false alarm and false false alarm rates when handling such complex faults, and expert systems struggle to respond promptly to complex fault modes. Similarly, because single-model methods can only capture local features, they lack the ability to comprehensively analyze fault propagation paths across subsystems.

[0004] Traditional physics-based methods, such as power bond graph modeling and AMESim simulation, aim to diagnose faults by analyzing the fundamental physical principles of a system. Power bond graph methods, based on the principle of energy conservation, use graphical symbols to represent energy transfer and conversion within the system. This approach effectively reveals the physical interactions between system components, thus providing a theoretical foundation for physical-level fault diagnosis. AMESim, as a complex multi-domain system modeling and simulation platform, can realize comprehensive virtual prototypes of electro-hydraulic systems. Comparing simulation results with operational data helps to effectively identify faults. The effectiveness of these physics-based methods depends on modeling accuracy, posing challenges to complex electro-hydraulic systems under dynamic conditions. Inaccurate modeling often fails to capture the actual system state, severely impairing diagnostic performance. Alternative methods based on rules and expert systems rely on domain expertise encoded in a knowledge base and predefined fault location rules. As systems evolve, these methods face significant challenges in maintaining relevance; their static knowledge structures struggle to adapt to new fault modes and complex multi-fault scenarios. Manual knowledge updates struggle to match constantly emerging fault modes, and the predefined rule framework lacks robustness in analyzing cross-domain fault propagation. The constraints of traditional methods highlight the necessity of adaptive intelligent algorithms, which bridge the gap between physical principles and data-driven insights in modern complex systems. Summary of the Invention

[0005] The technical problem to be solved by this invention is:

[0006] To address the current shortcomings in the integration of physical information and data-driven methods in electro-hydraulic system fault diagnosis, as well as the limited ability to trace complex fault sources.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0008] This invention provides a fault diagnosis method for electromechanical and hydraulic equipment based on bond graph structure, comprising the following steps:

[0009] S100. Construct the power key graph of electromechanical and hydraulic equipment, including defining potential energy nodes, flow nodes and sensor nodes, defining edges and elements, and establishing power conservation constraints for the power system, pilot pump branch, front pump branch and pump branch.

[0010] S200. Establish a causal flow graph structure model, including constructing a node feature matrix based on each sensor node of the electromechanical hydraulic equipment, constructing an edge index matrix and an edge attribute matrix, and constructing a spatiotemporal graph based on the constructed node feature matrix, edge index matrix and edge attribute matrix as input to the causal flow graph structure model;

[0011] S300. Construct a causal flow graph neural network model, including designing a mask graph attention mechanism, defining the physical influence strength of adjacent nodes, establishing a graph convolutional network with pooling and multi-layer aggregation, and defining the label classification loss.

[0012] S400 defines physical and causal constraints, including supplementing missing parameters based on the power conservation law, constructing a physical loss term, establishing a causal loss function based on the causal relationship between nodes, and finally integrating physical loss, causal loss, and classification loss, and using the Adam optimizer to optimize the total loss of network parameters.

[0013] 2. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 1, characterized in that: in step S100, it includes,

[0014] S110, Define node 0 as a potential energy related node, define node 1 as a flow node, and define node Df as a Df sensor node;

[0015] S120 defines edges and elements, including TF, which is the engine to the transmission system, speed adaptation; GY, which is the front pump to the boom, converting hydraulic energy into mechanical energy; R1 is the engine friction loss; C is the oil circuit energy storage.

[0016] S130. Establish power conservation constraints for the power system, pilot pump branch, front pump branch, and rear pump branch. :

[0017]

[0018] in, This represents the power conservation constraint, i.e., physical loss. Indicates engine power. Indicates the power of the front pump. Indicates the power of the rear pump. This indicates the predicted power of the pilot pump. Indicates power loss. This indicates the pilot pump's predicted pressure. and These represent the actual pressures of the front and rear pumps, respectively. These are the weighting coefficients.

[0019] 3. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 2, characterized in that: in step S200, it includes,

[0020] S210. Extract the six-dimensional features of each sensor node of the electromechanical and hydraulic equipment as the node feature matrix. This includes node type, potential variable, flow variable, physical properties, and timestamp:

[0021]

[0022] in, Indicates the type of node. Quantify the potential state of the node. Characterizing flow intensity, Represents the additional attributes of a node. Indicates the current moment;

[0023] S220, Define the edge index matrix This includes the source node and the target node:

[0024]

[0025] in, These represent the source node and the target node connected via a power key, respectively.

[0026] S230, Define the edge attribute matrix Including causal direction Power flow Edge type and physical parameters :

[0027]

[0028] S240. A spatiotemporal graph is constructed based on the node feature matrix, edge index matrix, and edge attribute matrix as the network input.

[0029] 4. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 3, characterized in that: in step S300, it includes,

[0030] S310. Design a masked graph attention mechanism to adjust the contributions of adjacent nodes and update the state of the central node as follows:

[0031]

[0032] in, and Representing nodes respectively In the lth layer and the +1 layer feature representation; Indicates the first Layer nodes Feature representation; This represents the activation function. Indicates the first The learnable parameter matrix of the layer; Represents from adjacent nodes Passed to node The message aggregation process, Represents a node The set of neighboring nodes; Represents the attention coefficient, with the following constraints: ; Indicates from node To the node Attribute transfer items;

[0033] S320, will Defined as the intensity of physical influence between adjacent nodes, to regulate the actual energy transfer patterns:

[0034]

[0035] MLP is used to perform nonlinear transformations on the input. It represents the physical constraints or features of the entire graph; all information is connected as input to the MLP; the MLP output uses... A conversion is performed to ensure a positive attention coefficient; Indicates the first The layer in the first Hidden layer state in the next iteration Indicates the first The layer in the first Hidden layer states in the next iteration;

[0036] S330. Establish a graph convolutional network with pooling and multi-layer aggregation to extract local features and global patterns at key nodes:

[0037]

[0038]

[0039] in, express Node at the The layer is located at the first Hidden layer state at each time index Indicates the time step. Indicates global average pooling. Indicates global max pooling. Indicates global max pooling. Indicates the hidden layer dimension. Represents the set of real numbers;

[0040] S340, Define label classification loss Used to determine the diagnostic performance of the model:

[0041]

[0042] in, Indicates the number of samples. Indicates the number of fault categories. and They represent the first Fault categories of each sample The truth labels and predicted failure probabilities.

[0043] 5. A fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure according to claim 4, characterized in that: step S400 includes,

[0044] S410. Since the pilot pump lacks direct sensor parameters, its flow parameters are deduced based on the law of conservation of power.

[0045] S420, Construct the physical loss term to estimate the deviation between the minimum power conservation deviation and the pilot pump flow rate;

[0046] S430. Based on the causal relationships between nodes, establish a causal loss function. :

[0047]

[0048] in, Indicates the number of edges. Indicates the first The attention weight of the edge Indicates the first Causal direction label of the edge; Indicates element-wise multiplication;

[0049] S440 integrates physical loss, causal loss, and classification loss, and uses the Adam optimizer to optimize the total loss of the network parameters:

[0050]

[0051] in Indicates the total loss. Represents classification loss, Indicates physical loss. Indicates causal loss. and These represent the control weights for physical loss and causal loss, respectively.

[0052] A fault diagnosis system for electromechanical and hydraulic equipment based on a bond graph structure is provided. The system has a program module corresponding to the above steps and executes the steps in the above-described fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure when running.

[0053] A computer-readable storage medium storing a computer program configured to, when invoked by a processor, implement the steps of a method for diagnosing faults in electromechanical equipment based on a bond graph structure.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] This invention uses power bond graphs as the physical carrier to transform electro-hydraulic system components (such as pumps, valves, and motors) and energy flow transmission paths into a computable node / edge structure, establishing mathematical constraints for energy conservation and component interaction. Therefore, it has strong physical interpretability. By embedding the energy flow and causal logic of the electro-hydraulic system into a GNN through power bond graphs, the diagnostic results are completely consistent with the physical mechanism, solving the black box problem of traditional deep learning.

[0056] This invention utilizes a causal flow-aware graph neural network (CFAGNN) that employs power conservation as a hard constraint in the loss function and encodes the causal logic between components into adjacency weights of GNN nodes, ensuring consistency between diagnostic results and physical mechanisms. It extracts fault-sensitive features through "TopK pooling + global statistical aggregation," locates the root cause of the fault using a causal flow attention mechanism, and transforms the fault propagation path into a visualized topology graph. Furthermore, it employs a multi-channel parallel structure or parameter adaptation strategy to handle composite faults (multi-component faults) in electro-hydraulic systems, achieving simultaneous identification and differentiation of different fault sources.

[0057] This invention offers high accuracy in composite fault diagnosis. Its causal flow attention mechanism distinguishes between fault sources and chain reactions, and experimental verification shows that its diagnostic accuracy is significantly superior to other neural network methods. The physical constraint module completes missing sensor data, reducing reliance on redundant sensors. The attention mechanism focuses on key fault features, resulting in highly efficient model parameters that adapt to dynamic operating conditions of electromechanical systems. It also boasts strong scalability, applicable not only to excavator systems but also extending to other electro-hydraulic equipment such as loaders and tunnel boring machines by adjusting power key graph elements, without requiring large-scale model reconstruction. Attached Figure Description

[0058] Figure 1 This is a flowchart of a fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure, as described in an embodiment of the present invention.

[0059] Figure 2 This is a diagram showing the multi-pump branch connection and power key of the electro-hydraulic system of an excavator in an embodiment of the present invention;

[0060] Figure 3 This is a block diagram of the causal flow graph structured neural network model in an embodiment of the present invention;

[0061] Figure 4 This is a QQ graph showing the comparison of diagnostic accuracy in an embodiment of the present invention;

[0062] Figure 5 This is a source tracing analysis diagram for a specific fault case in an embodiment of the present invention. Detailed Implementation

[0063] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0064] Specific Implementation Plan 1: Combining Figures 1 to 3 As shown, this invention provides a fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure, comprising the following steps:

[0065] S100, Construct the power key diagram for electromechanical and hydraulic equipment, including,

[0066] S110. Define node 0 as a potential energy-related node, node 1 as a flow node, and node Df as a sensor node.

[0067] S120, Define edges and elements: TF (engine → transmission system, speed adaptation), GY (front pump → boom, hydraulic energy → mechanical energy), R1 = engine friction loss, C = oil circuit energy storage;

[0068] S130. Establish power conservation constraints for the power system, pilot pump branch, front pump branch, and rear pump branch:

[0069]

[0070] in, This represents the power conservation constraint, i.e., physical loss. Indicates engine power. Indicates the power of the front pump. Indicates the power of the rear pump. This indicates the predicted power of the pilot pump. This indicates the pilot pump's predicted pressure. Indicates power loss. and These represent the actual pressures of the front and rear pumps, respectively. These are the weighting coefficients;

[0071] S200. Establish a causal flow graph structure model, including:

[0072] S210, extracting each sensor node of the electro-hydraulic equipment as a node feature matrix. This includes node type, potential variables, flow variables, physical properties, and timestamps:

[0073]

[0074] in, Indicates the type of node. Quantify the potential state of the node. Characterizing flow intensity, Represents the additional attributes of a node. Indicates the current moment;

[0075] S220, Define the edge index matrix Includes source node and target node:

[0076]

[0077] in, These represent the source node and the target node connected via a power key, respectively.

[0078] S230, Edge Attribute Matrix It contains four attributes: causal direction Power flow Edge type and physical parameters :

[0079]

[0080] S240. A spatiotemporal graph is constructed based on the node feature matrix, edge index matrix, and edge attribute matrix as the network input.

[0081] S300, causal flow graph neural network model construction, including,

[0082] S310. Design a masked graph attention mechanism to adjust the contributions of adjacent nodes and update the state of the central node:

[0083]

[0084] in, and Representing nodes respectively In the lth layer and the +1 layer feature representation; Indicates the first Layer nodes Feature representation; This represents the activation function. Indicates the first The learnable parameter matrix of the layer; Represents a node The set of neighboring nodes, Represents from adjacent nodes Passed to node Message aggregation process; attention coefficient Used for adjacent nodes The transmitted messages are weighted so that the model can focus on more important neighbors, with the following constraints: . Indicates from node To the node Attribute transfer items;

[0085] S320, Defined as the intensity of physical influence between adjacent nodes, to regulate the actual energy transfer patterns:

[0086]

[0087] MLP is used to perform nonlinear transformations on the input. It represents the physical constraints or features of the entire graph; all information is connected as input to the MLP; the MLP output uses... A conversion is performed to ensure a positive attention coefficient; Indicates the first The layer in the first Hidden layer state in the next iteration Indicates the first The layer in the first Hidden layer states in the next iteration;

[0088] S330. Establish a graph convolutional network with pooling and multi-layer aggregation to extract local features and global patterns at key nodes:

[0089]

[0090]

[0091] in, express Node at the The layer is located at the first Hidden layer state at each time index Indicates the time step. Indicates global average pooling. Indicates global max pooling. Indicates global max pooling. Indicates the hidden layer dimension. Represents the set of real numbers;

[0092] S340, Define label classification loss Used to determine the diagnostic performance of the model:

[0093]

[0094] in Indicates the number of samples. Indicates the number of fault categories; and They represent the first Fault categories of each sample The truth labels and predicted failure probabilities;

[0095] S400. Define physical and causal constraints, including supplementing missing parameters based on the power conservation law, constructing physical loss terms, establishing causal loss functions based on causal relationships between nodes, and finally integrating physical loss, causal loss, and classification loss. The Adam optimizer is then used to optimize the total loss of network parameters.

[0096] include,

[0097] S410. For equipment components with missing sensor parameters, the missing physical variables are deduced based on the law of power conservation; a physical loss term is constructed to estimate the deviation by minimizing the power conservation deviation and the pilot pump flow rate.

[0098] S420. Based on the causal relationships between nodes, establish a causal loss function. The focus is on tracking the energy transfer process between nodes;

[0099]

[0100] in Indicates the number of edges. Indicates the first The attention weight of the edge Indicates the first Causal direction label of the strip; symbol This represents element-wise multiplication, ensuring that the direction of the attention mechanism's learning aligns with the causal relationships in the bond graph;

[0101] S430 integrates physical loss, causal loss, and classification loss, and uses the Adam optimizer to optimize the total loss of the network parameters.

[0102]

[0103] in Indicates the total loss. Represents classification loss, Indicates physical loss. Indicates causal loss. and These represent the control weights for physical loss and causal loss, respectively.

[0104] Specific Implementation Scheme 2: The present invention provides a fault diagnosis system for electromechanical and hydraulic equipment based on a bond graph structure. The system has a program module corresponding to the above steps, and executes the steps in the above-mentioned fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure when running.

[0105] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.

[0106] Specific Implementation Scheme 3: The present invention provides a computer-readable storage medium, characterized in that: the computer-readable storage medium stores a computer program, the computer program being configured to implement, when called by a processor, the steps of a fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure.

[0107] The other combinations and connections in this implementation scheme are the same as in Specific Implementation Scheme 1.

[0108] Simulation Experiment

[0109] Composite Fault Location Test

[0110] Fault classification tests were conducted to verify the fault diagnosis performance. To comprehensively validate the performance of the proposed model, additional neural networks were introduced for comparison (MCDBN multi-scale cascaded belief network, ConDenseNet convolutional dense network, Res2Net residual network, MCICN multi-scale contextual information convolutional network, Diversify network, Fs-learning few-shot learning network, SSSRL-M semi-supervised representation learning, S3M sparse semi-supervised model network, LASSC label-assisted self-supervised neural network, CFAGNN attribution graph neural network), with particular attention paid to the utilization of computational resources, such as... Figure 4 As shown, CFAGNN performs exceptionally well in both F1 score and parameter efficiency rankings. Despite its relatively simplified architecture, the model achieves higher fault diagnosis accuracy. This validates that physically constrained embedding effectively balances model complexity and predictive performance, while achieving accurate fault feature extraction without the accumulation of redundant parameters. Although traditional pure data-driven models such as ConDenseNet and Res2Net have impressive performance in certain dimensions, their overall rankings are relatively low. This reflects their need for a large number of parameters and high computational complexity to achieve accuracy, and their lower engineering practicality compared to methods that incorporate physical constraints.

[0111] Combination Figure 5 As shown, the attention weights of the output nodes and edges are mapped onto the graph structure of a specific case, and the fault propagation path is visualized through a heatmap. According to the fault description, the corresponding nodes of the graph structure are mapped to the actual physical topology. Based on the analysis of the physical topology, the source and propagation path of the specific fault of the excavator (a kind of electromechanical hydraulic equipment) are consistent with the model diagnosis results, proving that the model proposed in this invention has the ability to diagnose complex faults and visualize fault tracing.

[0112] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A fault diagnosis method for electromechanical and hydraulic equipment based on bond graph structure, characterized in that, Includes the following steps: S100. Construct the power key graph of electromechanical and hydraulic equipment, including defining potential energy nodes, flow nodes and sensor nodes, defining edges and elements, and establishing power conservation constraints for the power system, pilot pump branch, front pump branch and pump branch. S200. Establish a causal flow graph structure model, including constructing a node feature matrix based on each sensor node of the electromechanical hydraulic equipment, constructing an edge index matrix and an edge attribute matrix, and constructing a spatiotemporal graph based on the constructed node feature matrix, edge index matrix and edge attribute matrix as input to the causal flow graph structure model; S300. Construct a causal flow graph neural network model, including designing a mask graph attention mechanism, defining the physical influence strength of adjacent nodes, establishing a graph convolutional network with pooling and multi-layer aggregation, and defining the label classification loss. S400 defines physical and causal constraints, including supplementing missing parameters based on the power conservation law, constructing a physical loss term, establishing a causal loss function based on the causal relationship between nodes, and finally integrating physical loss, causal loss, and classification loss, and using the Adam optimizer to optimize the total loss of network parameters.

2. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 1, characterized in that: In step S100, the following are included: S110, Define node 0 as a potential energy related node, define node 1 as a flow node, and define node Df as a Df sensor node; S120 defines edges and elements, including TF, which is the engine to the transmission system, speed adaptation; GY, which is the front pump to the boom, converting hydraulic energy into mechanical energy; R1 is the engine friction loss; C is the oil circuit energy storage; S130. Establish power conservation constraints for the power system, pilot pump branch, front pump branch, and rear pump branch. : in, This represents the power conservation constraint, i.e., physical loss. Indicates engine power. Indicates the power of the front pump. Indicates the power of the rear pump. This indicates the predicted power of the pilot pump. Indicates power loss. This indicates the pilot pump's predicted pressure. and These represent the actual pressures of the front and rear pumps, respectively. These are the weighting coefficients.

3. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 2, characterized in that: In step S200, the following are included: S210. Extract the six-dimensional features of each sensor node of the electromechanical and hydraulic equipment as the node feature matrix. This includes node type, potential variable, flow variable, physical properties, and timestamp: in, Indicates the type of node. Quantify the potential state of the node. Characterizing flow intensity, Represents the additional attributes of a node. Indicates the current time; S220, Define the edge index matrix This includes the source node and the target node: in, These represent the source node and the target node connected via a power key, respectively. S230, Define the edge attribute matrix Including causal direction Power flow Edge type and physical parameters : S240. A spatiotemporal graph is constructed based on the node feature matrix, edge index matrix, and edge attribute matrix as the network input.

4. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 3, characterized in that: In step S300, the following are included: S310. Design a masked graph attention mechanism to adjust the contributions of adjacent nodes and update the state of the central node as follows: in, and Representing nodes respectively In the lth layer and the +1 layer feature representation; Indicates the first Layer nodes Feature representation; This represents the activation function. Indicates the first The learnable parameter matrix of the layer; Represents from adjacent nodes Passed to node The message aggregation process, Represents a node The set of neighboring nodes; Represents the attention coefficient, with the following constraints: ; Indicates from node To the node Attribute transfer items; S320, will Defined as the intensity of physical influence between adjacent nodes, to regulate the actual energy transfer patterns: MLP is used to perform nonlinear transformations on the input. It represents the physical constraints or features of the entire graph; all information is connected as input to the MLP; the MLP output uses... A conversion is performed to ensure a positive attention coefficient; Indicates the first The layer in the first Hidden layer state in the next iteration Indicates the first The layer in the first Hidden layer states in the next iteration; S330. Establish a graph convolutional network with pooling and multi-layer aggregation to extract local features and global patterns at key nodes: in, express Node at the The layer is located at the first Hidden layer state at each time index Indicates the time step. Indicates global average pooling. Indicates global max pooling. Indicates global max pooling. Indicates the hidden layer dimension. Represents the set of real numbers; S340, Define label classification loss Used to determine the diagnostic performance of the model: in, Indicates the number of samples. Indicates the number of fault categories. and They represent the first Fault categories of each sample The truth labels and predicted failure probabilities.

5. The method for fault diagnosis of electromechanical and hydraulic equipment based on bond graph structure according to claim 4, characterized in that: In step S400, the following are included: S410. Since the pilot pump lacks direct sensor parameters, its flow parameters are deduced based on the law of conservation of power. S420, Construct the physical loss term to estimate the deviation between the minimum power conservation deviation and the pilot pump flow rate; S430. Based on the causal relationships between nodes, establish a causal loss function. : in, Indicates the number of edges. Indicates the first The attention weight of the edge Indicates the first Causal direction label of the edge; Indicates element-wise multiplication; S440 integrates physical loss, causal loss, and classification loss, and uses the Adam optimizer to optimize the total loss of the network parameters. in Indicates the total loss. Represents classification loss, Indicates physical loss. Indicates causal loss. and These represent the control weights for physical loss and causal loss, respectively.

6. A fault diagnosis system for electromechanical and hydraulic equipment based on bond graph structure, characterized in that: The system has a program module corresponding to the steps of any one of the claims 1-5 above, and executes the steps in the above-described method for diagnosing faults in electromechanical and hydraulic equipment based on a bond graph structure when it is run.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program configured to, when invoked by a processor, implement the steps of the fault diagnosis method for electromechanical and hydraulic equipment based on a bond graph structure as described in any one of claims 1-5.

Images