Fault analysis method for aviation fuel pump

Abstract

The invention discloses an aviation fuel pump fault analysis method, and belongs to the technical field of fault recognition. The identification method comprises the following steps: respectively converting evaluation results of a plurality of experts into fuzzy numbers; constructing a graph neural network by taking the fuzzy number as a node; performing feature aggregation on the graph neural network through graph convolution; and inputting the aggregation characteristics into a fuel pump fault tree to obtain a fault analysis result. According to the method, subjectivity and opinion divergence in expert evaluation can be reduced, and fault features of the fuel pump can be captured more accurately, so that stability, accuracy and reliability of an analysis result are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of fault analysis, and more particularly to a method for analyzing the fault of an aviation fuel pump. Background Art

[0002] As the core component of the aircraft fuel system, the stability and reliability of the aviation fuel pump are crucial to flight safety. However, due to the complexity of the fuel pump's working environment and wear and tear during long-term operation, the fuel pump may have various faults.

[0003] Traditional fault analysis methods usually rely on sensor data and historical maintenance records, but these methods often cannot fully reflect the actual working status of the fuel pump, especially when facing complex or unknown faults. In recent years, with the development of artificial intelligence and machine learning technologies, data-driven fault diagnosis methods have gradually gained attention. These methods identify potential failure modes by analyzing large amounts of operating data. However, for critical equipment such as aviation fuel pumps, it is often difficult to obtain sufficient high-quality data, and the data may be affected by noise and uncertainty.

[0004] In order to improve the accuracy of fault diagnosis, a multi-expert evaluation system was introduced into fault analysis. By bringing together the knowledge and experience of experts in multiple fields, the working condition of the fuel pump can be evaluated more comprehensively. However, expert evaluation results are usually subjective and there may be differences of opinion. Therefore, how to effectively integrate these evaluation results and extract valuable information has become a challenge. Summary of the invention

[0005] In view of this, the present invention provides an aviation fuel pump fault analysis method, which aims to effectively process multiple expert evaluation results, thereby improving the accuracy and reliability of aviation fuel pump fault diagnosis.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A method for analyzing aviation fuel pump failure, comprising the steps of:

[0008] The evaluation results of multiple experts are converted into fuzzy numbers to reduce the uncertainty of subjective judgment and make the evaluation results more objective;

[0009] Graph neural networks are constructed with fuzzy numbers as nodes to effectively integrate the knowledge and experience of experts in multiple fields and make fault diagnosis more comprehensive;

[0010] Feature aggregation of graph neural networks through graph convolution;

[0011] The aggregated features are input into the fuel pump fault tree to obtain the fault analysis results.

[0012] Furthermore, the evaluation results of multiple experts are converted into fuzzy numbers respectively by using triangular fuzzy numbers or trapezoidal fuzzy numbers. Due to the complex working environment of the aviation fuel pump, the data may be affected by noise and uncertainty. The present invention uses fuzzy numbers to process these uncertain data, thereby improving the robustness of fault diagnosis.

[0013] Furthermore, when constructing a graph neural network, the weights of the edges between nodes are determined according to the following formula;

[0014]

[0015] In the formula, w ij represents the weight of the edge between the i-th node and the j-th node, and dis(i,j) represents the distance measure between the i-th node and the j-th node.

[0016] Furthermore, the distance between nodes is measured based on the expert experience years weight; wherein, the expert experience years weight calculation formula is:

[0017]

[0018] Where W exp (i) is the experience weight of the ith expert, Y(i) is the working years of the ith expert, and Y min The minimum working experience among all experts, Y max This is the longest working experience among all experts.

[0019] Furthermore, the distance between nodes is measured based on the expert education level weight;

[0020] When the expert is a bachelor’s degree holder, the weight of the expert’s education level is 0.2;

[0021] When the expert is a master’s degree holder, the weight of the expert’s education level is 0.3;

[0022] When the expert is a doctor, the weight of the expert's education level is 0.5.

[0023] Furthermore, the distance between nodes is measured based on the total weight of the years of expert experience and the degree of expert education; the total weight of the years of expert experience and the degree of expert education is:

[0024]

[0025] Where W edu (i) represents the expert education level weight of the i-th node, and m represents W edu The coefficient of (i), W exp (i) represents the expert experience weight of the i-th node, and n represents W exp The coefficient of (i).

[0026] This application provides a more reasonable calculation method for the edge weights between nodes in graph neural networks by considering the experts' years of experience and education level, so that the network can better reflect the importance of expert opinions.

[0027] Furthermore, the graph neural network is feature aggregated through graph convolution, including aggregating the neighbor nodes of the i-th node layer by layer using the following formula;

[0028]

[0029] In the formula, represents the feature vector of node i in the k+1 layer, σ is the activation function, and w ij is the edge weight between node i and node j, is the set of neighbor nodes of node i, represents the feature vector of node i in layer k, q i is the bias term of node i.

[0030] Furthermore, the NOA algorithm is used to optimize the parameters of graph convolution to further improve the performance of the fault diagnosis model.

[0031] Furthermore, the centroid method is used to defuzzify the aggregated features to make the final fault analysis results clearer and more certain.

[0032] Furthermore, fault analysis includes probability importance analysis and critical importance analysis to help maintenance personnel better understand the possibility of fault occurrence and key influencing factors, so as to formulate more effective maintenance strategies;

[0033] The calculation formula for probability importance analysis is:

[0034]

[0035] In the formula, IP(X i ) represents the failure probability of the bottom event, P(T) represents the probability of the top event obtained by the fault tree based on the aggregation characteristics, Represents aggregate features;

[0036] The calculation formula for critical importance analysis is:

[0037]

[0038] Where IC(i) is the critical importance of the i-th bottom event, P(T) represents the probability of occurrence of the top event obtained by the fault tree based on the aggregated features, Represents an aggregate feature.

[0039] It can be seen from the above technical solutions that the present invention discloses a method for analyzing aviation fuel pump faults. By converting the evaluation results of multiple experts into fuzzy numbers and using graph neural networks for feature aggregation, the present invention can more accurately capture the fault characteristics of the fuel pump, thereby improving the accuracy of fault diagnosis;

[0040] Compared with the existing technology, this application significantly improves the accuracy and efficiency of fault diagnosis through data-driven objective analysis. It not only reduces the subjectivity and possible differences of opinion in expert evaluation, ensuring the stability and reliability of the analysis results; it also improves the maintenance efficiency of aviation fuel pumps and reduces maintenance costs. This application is of great significance for ensuring flight safety and improving the maintenance efficiency of aviation fuel systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0042] Figure 1 The present invention is a flow chart of the aviation fuel pump fault analysis method. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Reference Figure 1 , Figure 1 This is a flow chart of the aviation fuel pump fault analysis method for this application;

[0045] The aviation fuel pump fault analysis method in this application includes the following steps:

[0046] The evaluation results of multiple experts are converted into fuzzy numbers respectively;

[0047] Construct graph neural network with fuzzy numbers as nodes;

[0048] Feature aggregation of graph neural networks through graph convolution;

[0049] The aggregated features are input into the fuel pump fault tree to obtain the fault analysis results.

[0050] In one embodiment, the mechanism of the aviation fuel pump is first analyzed to construct a fault tree; then multiple experts are invited to obtain natural language evaluations of the multiple experts on the events at the bottom of the fault tree.

[0051] further,

[0052] 1. Use triangular fuzzy numbers or trapezoidal fuzzy numbers to convert the evaluation results of multiple experts into fuzzy numbers. In one embodiment, the membership function of the triangular fuzzy number is:

[0053]

[0054] The trapezoidal membership function is:

[0055]

[0056] In the formula, a i is the lowest estimate of the failure probability of the bottom event by expert i, b i is the most likely estimate of the failure probability of the bottom event by expert i, c i is the highest estimate of the failure probability of the bottom event by expert i, and x represents the expert’s evaluation result. represents the membership degree of the i-th expert evaluation, d i is the maximum possible limit value of the failure probability of the bottom event for expert i.

[0057] 2. Building a graph neural network

[0058] In this embodiment, each expert corresponds to a node, and the initial feature of the node is the fuzzy number evaluation of the expert; in the graph neural network, expert nodes are connected by edges, and the edges are ij represents the relationship between expert i and expert j, and its weight is obtained by the following formula;

[0059]

[0060] In the formula, w ij represents the weight of the edge between the i-th node and the j-th node, and dis(i,j) represents the distance measure between the i-th node and the j-th node.

[0061] In one embodiment, in order to more reasonably obtain the fault data of the bottom event, different weights are given to experts with different education levels and years of work experience, and the distance between nodes is measured based on the weights;

[0062] Among them, the expert experience years weight is converted into a value between [0.1, 0.5] through linear normalization, and the calculation formula is:

[0063]

[0064] Where Wexp (i) is the experience weight of the ith expert, Y(i) is the working years of the ith expert, and Y min The minimum working experience among all experts, Y max This is the longest working experience among all experts.

[0065] For the weight of experts’ education level,

[0066] When the expert is a bachelor’s degree holder, the weight of the expert’s education level is 0.2;

[0067] When the expert is a master’s degree holder, the weight of the expert’s education level is 0.3;

[0068] When the expert is a doctor, the weight of the expert's education level is 0.5.

[0069] In one embodiment, the distance between nodes is measured based on the total weight of the expert's years of experience and the expert's education level; the calculation formula is:

[0070]

[0071] Where W edu (i) represents the expert education level weight of the i-th node, and m represents W edu The coefficient of (i) is preferably 0.6, W exp (i) represents the expert experience weight of the i-th node, and n represents W exp The coefficient of (i) is preferably 0.4.

[0072] 3. Perform feature aggregation on the graph neural network through graph convolution, including aggregating the neighbor nodes of the i-th node layer by layer until the K-th layer using the following formula, where K is the total number of layers of the graph neural network; specifically, the feature representation of node i at the k+1th layer is It is the weighted sum of its neighbor node information, and the formula is expressed as:

[0073]

[0074] In the formula, represents the feature vector of node i in the k+1 layer, σ is the activation function, and w ij is the edge weight between node i and node j, is the set of neighbor nodes of node i, that is, all other experts related to the expert, represents the feature vector of node i in layer k, q i is the bias term of node i (to improve the expressiveness of the model).

[0075] In order to further aggregate multi-layer information, multiple rounds of graph convolution are usually performed. Each round of graph convolution will update the feature representation of the node. Finally, each node i in the network will get an aggregated feature representation , where K represents the number of graph convolution layers.

[0076] To further optimize the above technical solution, this application uses the NOA algorithm to optimize the parameters of graph convolution, mainly including the bias term, to reduce the uncertainty of expert knowledge and obtain better fuzzy probability.

[0077] The optimized objective function is:

[0078]

[0079] n represents the total number of nodes in the network, that is, the number of experts or samples. During the optimization process of NOA, there are two main stages: the global search stage and the local search stage.

[0080] The global search phase aims to explore the entire search space and find potential high-quality solutions. The update formula for each particle is:

[0081]

[0082] where y i (t) represents the position of the ith particle in the tth generation, y m (t) is the center position y of the current particle swarm A (t) and y B (t) are two positions randomly selected from the particle swarm, γ and is the control parameter, r2 is a random number in the range [0,1], and U and L are the upper and lower bounds respectively.

[0083] The local development phase aims to fine-tune the found high-quality solutions. The update formula for each particle is:

[0084]

[0085] where y C (t) is the current global optimal position, and δ is a small probability value used to control random jumps. Then, the steps of calculating fitness, updating individual optimal and global optimal, and updating speed and position are repeated until the predetermined number of iterations or convergence conditions are reached.

[0086] As a preferred embodiment, the centroid method is used to defuzzify the aggregated features.

[0087] After multiple layers of aggregation through the graph neural network, each node will have a new feature representation ,It contains the aggregate information of all experts. This information is usually a fuzzy number, which represents the aggregate value of the final failure probability of the event. This application uses the centroid method to defuzzify the fuzzy number to obtain the final failure probability.

[0088] In this embodiment, the centroid (i.e., weighted average) of the fuzzy number is calculated by the following formula:

[0089]

[0090] 4. The aggregated features are input into the fuel pump fault tree for fault analysis. In order to understand the impact of each change in the bottom event failure probability on the occurrence of the top event failure, a sensitivity analysis is performed; the sensitivity analysis includes probability importance analysis and critical importance analysis;

[0091] Probabilistic importance analysis, used to measure the absolute impact of the probability of a single bottom event on the probability of a top event;

[0092] The calculation formula is:

[0093]

[0094] In the formula, IP(X i ) represents the failure probability of the bottom event or the bottom event after defuzzification, P(T) represents the probability of the top event, which is calculated by the logical relationship between the fault tree model and the bottom event probability. Represents an aggregate feature.

[0095] Critical importance analysis is used to measure the relative impact of changes in the probability of occurrence of bottom events on the probability of occurrence of top events, taking into account the relative importance of bottom events in the entire system.

[0096] The calculation formula is:

[0097]

[0098] Where IC(i) is the critical importance of the i-th bottom event, P(T) represents the probability of occurrence of the top event obtained by the fault tree based on the aggregated features, Represents an aggregate feature.

[0099] Indicates the bottom event X i The degree to which the change in the probability of occurrence of the top event affects the probability of occurrence of the top event is a way to measure the sensitivity of the bottom event to system failure. It is the change in the probability of the top event, reflecting the impact of the change in the bottom event on the top event. Through the logical relationship in the fault tree model, the change in the probability of the top event can be calculated. is the probability of failure of the bottom event The change of is usually calculated by updating the node features in the graph neural network. Finally, The absolute impact of the bottom event on the top event can be obtained.

[0100] This embodiment constructs a fuzzy fault tree, obtains the natural language evaluation of the bottom event of the fault tree by the expert, and uses the graph neural network algorithm to aggregate the expert knowledge; uses the centroid method to defuzzify the fuzzy failure probability of the bottom event to obtain the failure probability of the bottom event; then performs sensitivity analysis through the fuzzy fault tree to obtain the probability importance and critical importance, and outputs the fault tree analysis results. At the same time, a graph neural network expert knowledge uncertainty optimization model based on the starfinch optimization algorithm is constructed, and the uncertainty of expert knowledge is reduced through the graph neural network of the starfinch optimization algorithm.

[0101] The method embodiment provided in this embodiment can be executed in a server, a computer, a mobile phone, or a similar computing device.

[0102] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0103] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for analyzing aviation fuel pump failure, characterized in that: The evaluation results of multiple experts are converted into fuzzy numbers respectively; Construct graph neural network with fuzzy numbers as nodes; Feature aggregation of graph neural networks through graph convolution; The aggregated features are input into the fuel pump fault tree to obtain the fault analysis results.

2. The fault analysis method according to claim 1, characterized in that: The evaluation results of multiple experts are converted into fuzzy numbers respectively using triangular fuzzy numbers or trapezoidal fuzzy numbers.

3. The fault analysis method according to claim 1, characterized in that: When building a graph neural network, the weights of the edges between nodes are determined according to the following formula; ； In the formula, w ij represents the weight of the edge between the i-th node and the j-th node, and dis(i,j) represents the distance measure between the i-th node and the j-th node.

4. The fault analysis method according to claim 3, characterized in that: The distance between nodes is measured based on the weight of years of expert experience; the calculation formula for the weight of years of expert experience is: ； Where W exp (i) is the experience weight of the ith expert, Y(i) is the working years of the ith expert, and Y min The minimum working experience among all experts, Y max This is the longest working experience among all experts.

5. The fault analysis method according to claim 3, characterized in that: The distance between nodes is measured based on the expert education level weight; When the expert is a bachelor’s degree holder, the weight of the expert’s education level is 0.2; When the expert is a master’s degree holder, the weight of the expert’s education level is 0.3; When the expert is a doctor, the weight of the expert's education level is 0.

5.

6. The fault analysis method according to claim 3, characterized in that: The distance between nodes is measured based on the total weight of the expert's years of experience and the expert's education level; the total weight of the expert's years of experience and the expert's education level is: ； Where W edu (i) represents the expert education level weight of the i-th node, and m represents W edu The coefficient of (i), W exp (i) represents the expert experience weight of the i-th node, and n represents W exp The coefficient of (i).

7. The fault analysis method according to claim 1, characterized in that: The graph neural network is featured by graph convolution, including layer-by-layer aggregation of neighbor nodes of the i-th node using the following formula; ； In the formula, represents the feature vector of node i in the k+1 layer, σ is the activation function, and w ij is the edge weight between node i and node j, is the set of neighbor nodes of node i, represents the feature vector of node i in layer k, q i is the bias term of node i.

8. The fault analysis method according to claim 7, characterized in that: The NOA algorithm is used to optimize the parameters of graph convolution.

9. The fault analysis method according to claim 1, characterized in that: The centroid method is used to defuzzify the aggregated features.

10. The fault analysis method according to claim 1, characterized in that: Failure analysis includes probability importance analysis and critical importance analysis; The calculation formula for probability importance analysis is: ； In the formula, IP(X i ) represents the failure probability of the bottom event, P(T) represents the probability of the top event obtained by the fault tree based on the aggregation characteristics, Represents aggregate features; critical The calculation formula for importance analysis is: ； Where IC(i) is the critical importance of the i-th bottom event, P(T) represents the probability of occurrence of the top event obtained by the fault tree based on the aggregated features, Represents an aggregate feature.

Images