Marine electromechanical equipment fault diagnosis method based on causal relationship network

Abstract

The invention discloses a ship electromechanical equipment fault diagnosis method based on a causal relationship network, and the method comprises the following steps: 1) collecting electromechanical equipment monitoring data, and carrying out the data preprocessing; 2) according to related design and mechanism analysis of the electromechanical equipment, judging a causal relationship existing in monitoring variables of the electromechanical equipment, and obtaining a determinable causal relationship existing among part of monitoring characteristic variables; 3) mining the causal relationship between the feature variables according to the preprocessed data, and constructing a corresponding causal relationship network; 4) taking the preprocessed monitoring data and the constructed causal relationship network as input, and establishing a fault diagnosis model; 5) training the fault diagnosis model, and optimizing and adjusting model parameters; and 6) performing fault diagnosis by using the trained model. Compared with an existing fault diagnosis method based on correlativity, the fault diagnosis method based on the correlativity has higher interpretability by mining the causal relationship between the monitoring characteristic variables.

Description

Technical Field

[0001] The invention relates to a ship electromechanical equipment fault diagnosis technology, and in particular to a ship electromechanical equipment fault diagnosis method. Background Art

[0002] At present, there are many electromechanical devices on ships. The complexity of the equipment itself makes it difficult for even professionals to judge the status of the equipment by appearance. Once a failure occurs, it may cause the equipment to be unable to be used normally, and even cause a chain reaction affecting other equipment or the entire ship, causing huge economic losses and catastrophic consequences.

[0003] By analyzing, processing and utilizing data mining technology, we can discover the fault rules and patterns hidden behind the data resources, find potential problems in advance, and take targeted measures, which is of great significance for the fault diagnosis, repair and maintenance of ship equipment and improving the normal operation of equipment. However, at the same time, artificial intelligence algorithms are mostly black boxes and lack interpretability, resulting in low credibility of the results. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for diagnosing faults of ship electromechanical equipment based on a causal relationship network in view of the defects in the prior art.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a method for diagnosing faults of ship electromechanical equipment based on a causal relationship network, comprising the following steps:

[0006] 1) Collect monitoring data of electromechanical equipment and perform data preprocessing;

[0007] The data preprocessing includes preliminary deletion and denoising of the monitoring data of the electromechanical equipment;

[0008] 2) According to the design and mechanism analysis of electromechanical equipment, determine the causal relationship between the monitoring variables of the electromechanical equipment, and obtain the causal relationship between some of the monitoring characteristic variables that can be determined;

[0009] 3) Mining the causal relationship between the preprocessed data and monitoring the characteristic variables, and constructing the corresponding causal relationship network;

[0010] 4) Establish a fault diagnosis model using the preprocessed monitoring data and the constructed causal network as input;

[0011] 5) Train the fault diagnosis model and optimize and adjust the model parameters;

[0012] 6) Use the trained model for fault diagnosis.

[0013] According to the above scheme, in step 3), the corresponding causal relationship network is constructed as follows:

[0014] The preprocessed monitoring data is taken as input, the causal relationships determined by analysis are used as constraints, and a data-driven causal discovery algorithm is used to construct a causal relationship network between monitoring feature variables.

[0015] According to the above scheme, in step 4), the graph attention network model is used as the fault diagnosis model.

[0016] According to the above scheme, in step 4), the graph attention network model includes an input layer, multiple graph attention layers and a fully connected layer. The multiple graph attention layers are used to realize information aggregation and extraction of input data, and the fully connected layer is used to realize fault diagnosis.

[0017] According to the above scheme, in step 4), for each graph attention layer, the graph attention network calculates the attention e between the i-th node and the j-th node through a shared weight matrix ij , as shown below:

[0018]

[0019] In the formula, Represents the F-dimensional features of the i-th node and the j-th node respectively; W∈R F′×F is a shared weight matrix that transforms the input node F-dimensional features into F′ dimensions; || represents the vector concatenation operation; is a weight vector that maps the concatenated 2F′-dimensional vector to a real value; LeakyReLU is the activation function;

[0020] For a node in the network graph, after calculating the attention coefficient of the node and its neighboring nodes, a weighted sum can be performed on all neighboring nodes to obtain the output features of each node in the network graph after aggregating the information of its respective neighboring nodes;

[0021] The output feature calculation formula of the i-th node is as follows:

[0022]

[0023] In the formula, σ represents the activation function, α ij is the attention coefficient between two nodes, N i is the set of neighbor nodes of node i.

[0024] According to the above scheme, in step 4), the attention coefficient between node i and its neighboring nodes is calculated as shown in the following formula:

[0025]

[0026] Where N i is the set of neighbor nodes of node i;

[0027] According to the above scheme, in step 4), the graph attention network introduces a multi-head attention mechanism to improve the representation ability of the model itself; the output feature is the concatenation or averaging of the results obtained by the multi-head attention.

[0028] The beneficial effects produced by the present invention are:

[0029] 1. The present invention explores the causal relationship between monitoring feature variables and reflects the real influence relationship between variables based on the causal relationship. Compared with the existing correlation, the fault diagnosis method based on the causal relationship network of the present invention has the advantages of being directed and sparse.

[0030] 2. Based on the combination of causal network and graph neural network model, fault diagnosis can be better performed and the accuracy of diagnosis can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0032] Figure 1 is a method flow chart of an embodiment of the present invention;

[0033] Figure 2 It is a schematic diagram of the principle of an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0035] like Figure 1 and Figure 2 As shown, a method for diagnosing faults of ship electromechanical equipment based on a causal relationship network comprises the following steps:

[0036] 1) Collect monitoring data of electromechanical equipment and perform data preprocessing;

[0037] The electromechanical equipment monitoring data are several types of data that are usually concerned in the fault diagnosis and monitoring of electromechanical equipment. In this embodiment, the electromechanical equipment monitoring data includes vibration data, sound signal data, temperature data, and electrical data;

[0038] Data preprocessing includes:

[0039] Deleting monitoring variable data whose status or value remains unchanged;

[0040] For monitoring data containing noise, common filtering methods are used to remove noise, such as wavelet analysis, data smoothing, empirical mode decomposition, etc.

[0041] 2) According to the design and mechanism analysis of electromechanical equipment, determine the causal relationship between the monitoring variables of the electromechanical equipment, and obtain the causal relationship between some of the monitoring characteristic variables that can be determined;

[0042] 3) Mining the causal relationship between the preprocessed data and monitoring the characteristic variables, and constructing the corresponding causal relationship network;

[0043] The preprocessed monitoring data is used as input, the causal relationships determined by analysis are used as constraints, and a data-driven causal discovery algorithm is used to construct a causal relationship network between monitoring feature variables;

[0044] 4) Establish a fault diagnosis model using the preprocessed monitoring data and the constructed causal network as input;

[0045] Since the causal relationship network in the present invention is a directed graph, and the graph attention network (GAT) introduces an attention mechanism so that the model can learn different weights for different neighbor nodes, the graph attention network is used as a fault diagnosis model;

[0046] In this embodiment, a multi-layer graph attention network is constructed, such as Figure 2 Taking layer 3 as an example, the graph attention layer realizes information aggregation and extraction of input data, and realizes diagnosis or prediction through the final fully connected layer.

[0047] The graph attention network calculates the attention e between the i-th node and the j-th node through a shared weight matrix ij , as shown below:

[0048]

[0049] In the formula, Represents the F-dimensional features of the i-th node and the j-th node respectively; W∈R F′×F is a shared weight matrix that transforms the input node F-dimensional features into F′ dimensions; || represents the vector concatenation operation; is a weight vector that maps the concatenated 2F′-dimensional vector to a real value; LeakyReLU is the activation function;

[0050] When calculating the attention coefficient between node i and its neighboring nodes, the attention coefficient calculation result needs to be normalized by the softmax function, as shown in the following formula:

[0051]

[0052] In the above formula: N i is the set of neighbor nodes of node i;

[0053] The above process of calculating the attention coefficient between two nodes is as follows Figure 2 shown.

[0054] For a node in the network graph, after calculating the attention coefficient of the node and its neighboring nodes, a weighted sum can be performed on all neighboring nodes to obtain the output features of each node in the network graph after aggregating the information of its respective neighboring nodes;

[0055] The calculation formula for the output feature of the i-th node is shown as follows.

[0056]

[0057] In the above formula: σ represents the activation function.

[0058] In order to make the self-attention mechanism able to stably represent nodes, the graph attention network introduces a multi-head attention mechanism to improve the representation ability of the model itself. For the output features of the intermediate layer, multiple weight matrices W are used to calculate the self-attention coefficients, and then the results obtained by the multi-head attention are spliced ​​or averaged.

[0059] The results of the multi-head attention mechanism are spliced, that is, the features calculated under different weight matrices are spliced, as shown in the following formula:

[0060]

[0061] In the above formula: L is the number of heads of the attention mechanism, that is, the number of different weight matrices W.

[0062] The results of the multi-head attention mechanism are averaged, that is, the features calculated under different weight matrices are averaged as the final calculation result, as shown in the following formula:

[0063]

[0064] 5) Train the fault diagnosis model and optimize and adjust the model parameters;

[0065] The preprocessed monitoring data and the constructed causal network are used as the input of the graph learning diagnosis model. The constructed graph learning diagnosis model is continuously trained, and the model parameters are optimized and adjusted to finally achieve satisfactory fault diagnosis or prediction effects.

[0066] Based on the trained graph learning diagnosis model, the GNNExplainer and other graph neural network interpretability methods can be used to calculate the monitoring features that have the greatest impact on diagnosis. The preprocessed monitoring data and the trained diagnosis model are used as the input of the graph neural network interpretability method to calculate and mine the monitoring features that have the greatest impact on diagnosis during the diagnosis process.

[0067] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for diagnosing faults of ship electromechanical equipment based on a causal relationship network, characterized in that: The following steps are involved: 1) Collect monitoring data of electromechanical equipment and perform data preprocessing; The data preprocessing includes preliminary deletion and denoising of the monitoring data of the electromechanical equipment; 2) According to the design and mechanism analysis of electromechanical equipment, determine the causal relationship between the monitoring variables of the electromechanical equipment, and obtain the causal relationship between some of the monitoring characteristic variables that can be determined; 3) Mining the causal relationship between the preprocessed data and monitoring the characteristic variables, and constructing the corresponding causal relationship network; 4) Establish a fault diagnosis model using the preprocessed monitoring data and the constructed causal network as input; 5) Train the fault diagnosis model and optimize and adjust the model parameters; 6) Use the trained model for fault diagnosis.

2. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 1 is characterized in that: In the step 1), data preprocessing includes: Deleting monitoring variable data whose status or value remains unchanged; For monitoring data containing noise, filtering method is used to denoise it.

3. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 1 is characterized in that: In step 4), the graph attention network model is used as the fault diagnosis model.

4. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 3 is characterized in that: In step 4), the graph attention network model includes an input layer, multiple graph attention layers and a fully connected layer. The multiple graph attention layers are used to realize information aggregation and extraction of input data, and the fully connected layer is used to realize fault diagnosis.

5. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 4 is characterized in that: In step 4), for each graph attention layer, the graph attention network calculates the attention e between the i-th node and the j-th node through a shared weight matrix ij , as shown below: In the formula, Represents the F-dimensional features of the i-th node and the j-th node respectively; W∈R F′×F is a shared weight matrix that transforms the input node F-dimensional features into F′ dimensions; || represents the vector concatenation operation; is a weight vector that maps the concatenated 2F′-dimensional vector to a real value; LeakyReLU is the activation function; For a node in the network graph, after calculating the attention coefficient of the node and its neighboring nodes, a weighted sum can be performed on all neighboring nodes to obtain the output features of each node in the network graph after aggregating the information of its respective neighboring nodes; The output feature calculation formula of the i-th node is as follows: In the formula, σ represents the activation function, α ij is the attention coefficient between two nodes, N i is the set of neighbor nodes of node i.

6. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 5 is characterized in that In step 4), the attention coefficient between node i and its neighboring nodes is calculated as shown in the following formula: Where N i is the set of neighbor nodes of node i.

7. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 3 is characterized in that: In the step 4), the graph attention network introduces a multi-head attention mechanism to improve the representation ability of the model itself; the output feature is the concatenation or averaging of the results obtained by the multi-head attention.

8. The method for diagnosing faults of ship electromechanical equipment based on a causal relationship network according to claim 1, characterized in that: In step 3), the corresponding causal relationship network is constructed as follows: The preprocessed monitoring data is taken as input, the causal relationships determined by analysis are used as constraints, and a data-driven causal discovery algorithm is used to construct a causal relationship network between monitoring feature variables.

9. An electronic device, characterized in that: include: one or more processors; as well as a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images