Attention mechanism driven belief rule base aviation component performance evaluation method

Abstract

The invention discloses an attention mechanism driven belief rule base aviation component performance evaluation method, which comprises the following steps: acquiring observation data of an aviation component inertial navigation system, and inputting the observation data into an aviation component performance evaluation model to obtain a performance evaluation result; the aviation component performance evaluation model is trained according to the following steps: acquiring various monitoring data of an aviation component inertial navigation system to construct a confidence rule base, and setting default weights of precondition attributes and rules; using a feature attention module and monitoring data to optimize default weights of the premise attributes; based on the optimization weight of the premise attribute, determining an activated rule in a confidence rule base by using a rule attention module; and determining a final evaluation result after the various monitoring data are input into the activated rule by using an evidence reasoning algorithm, and optimizing parameters of the confidence rule base by using the final evaluation result and a preset objective function until the final evaluation result reaches preset precision, thereby obtaining the aviation component performance evaluation model, and improving the evaluation precision.

Description

Technical Field

[0001] This invention belongs to the field of performance evaluation technology, specifically relating to a performance evaluation method for aerospace components based on a confidence rule base driven by an attention mechanism. Background Technology

[0002] Inertial navigation systems (INS) are core navigation devices for aircraft. Using sensors such as gyroscopes and accelerometers, they autonomously calculate the aircraft's attitude, position, and velocity, offering the unique advantage of not relying on external signals and providing stable navigation data under various adverse weather conditions. However, the accuracy and reliability of INS are susceptible to factors such as environment, temperature, mechanical vibration, long-term wear, and aging of gyroscope bearings, leading to the accumulation of bias errors, system malfunctions, and even aviation accidents. Therefore, performance evaluation of aircraft components has significant engineering value for aircraft performance evaluation.

[0003] Confidence rule bases are a cutting-edge performance evaluation method. Compared with data-driven or expert knowledge-driven performance evaluation methods, they can effectively integrate expert knowledge and monitoring data, do not rely on large amounts of monitoring data, and reduce the impact of expert experience judgment on performance evaluation results. However, existing confidence rule bases have the following two problems: (1) fixed feature weights, that is, the importance of sensor features varies significantly under different fault modes, and the weights need to be preset manually, which leads to an increased misjudgment rate; (2) redundancy of the rule base, that is, the combination of multi-source features leads to an exponential increase in the number of rules, and the efficiency of full rule traversal is low, making it difficult to meet real-time requirements. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides a method for evaluating the performance of aerospace components based on an attention mechanism-driven confidence rule base. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a confidence rule base-based method for evaluating the performance of aerospace components based on an attention mechanism, comprising: Collect observation data from the inertial navigation system of aircraft components; The observed data is input into a pre-trained aircraft component performance evaluation model to obtain performance evaluation results; wherein, the aircraft component performance evaluation model is trained according to the following steps: Acquire various monitoring data from the inertial navigation system of aircraft components; A confidence rule base is constructed based on the aforementioned multiple monitoring data, and default weights for prerequisite attributes and rules are set in the confidence rule base. Using the feature attention module and the various monitoring data, the default weights of the prerequisite attributes are optimized to obtain the optimized weights of the prerequisite attributes. Based on the optimized weights of the aforementioned prerequisite attributes, the rules activated in the confidence rule base are determined using the rule attention module. The final evaluation result after inputting various monitoring data into the activated rules is determined using an evidence reasoning algorithm. The parameters of the confidence rule base are then optimized using the final evaluation result and a preset objective function until the final evaluation result reaches a preset accuracy, thereby obtaining a trained aviation component performance evaluation model.

[0005] In one embodiment of the present invention, the confidence rule base contains the first... The rule is expressed as follows: ; In the formula, , The number of rules in the confidence rule base. They represent The first moment Such monitoring data, They represent the first The prerequisite attributes are Reference values ​​for monitoring data at any given time. Representation and operation They represent the first One evaluation result, They represent In the first The confidence level in the rule, They represent the first The default weight of the first prerequisite attribute, the first The default weight of this rule is .

[0006] In one embodiment of the present invention, the feature attention module includes a normalization unit and a weight calculation unit; The step of optimizing the default weights of the prerequisite attributes using a feature attention module and the various monitoring data to obtain optimized weights for the prerequisite attributes includes: The normalization unit is used to normalize various monitoring data:

[0007] In the formula, Indicates the first Such monitoring data, , Indicates the first The mean of various monitoring data, Indicates the first The standard deviation of the monitoring data Indicates the first Species-normalized monitoring data; Based on the Species Normalization Monitoring Data The optimization weights of each prerequisite attribute are determined using the weight calculation unit. ; ; In the formula, Indicates the first Optimization weights of the prerequisite attributes, express Feature attention weights, Indicates the first The default weight of each prerequisite attribute. , These represent the weights of the first and second convolutional layers in the weight calculation unit, respectively. , These are the bias terms for the first and second convolutional layers, respectively.

[0008] In one embodiment of the present invention, the rule attention module includes a similarity calculation unit and an activation probability calculation unit; The step of determining the activated rules in the confidence rule base using the rule attention module based on the optimized weights of the aforementioned prerequisite attributes includes: According to the normalized first Various monitoring data The similarity calculation unit is used to calculate The first moment Species Normalization Monitoring Data With the k Rule-based cosine similarity : ; In the formula, ; Based on the cosine similarity The activation probability calculation unit is used to calculate the first... Rule in Activation probability at time: ; In the formula, , , , The number of rules in the confidence rule base. , Indicates the monitoring duration. Indicates the temperature coefficient. ; The activation probability With preset threshold Compare; when At that time, the first Rule in It is activated at any moment.

[0009] In one embodiment of the present invention, the step of determining the final evaluation result after inputting multiple monitoring data into the activated rule using an evidence reasoning algorithm includes: calculate The first moment Various monitoring data With the first Matching degree of the rule ; Based on the matching degree Determine the input of the various monitoring data. The final evaluation result obtained after the rules are activated at a certain time.

[0010] In one embodiment of the present invention, calculation The first moment Various monitoring data With the first Matching degree of the rule The steps include: Sure The first moment Various monitoring data The monitoring data reference range that fell into ,in, They represent the first The prerequisite attributes are The first moment Reference values ​​for monitoring data; according to and ,calculate With the first The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ;in, when hour, ; when hour, ; when , Indicates the first The prerequisite attributes are The number of reference values ​​for monitoring data at any given time; according to With the first The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ,calculate With the first Matching degree of the rule: ; In the formula, .

[0011] In one embodiment of the present invention, based on the matching degree Determine the input of the various monitoring data. Before the step of obtaining the final evaluation result after the rule is activated at a certain time, it also includes: calculate The weights of each rule that is activated at any given time: ; In the formula, express The moment that was activated The activation probability of the rule. Indicates in The number of rules that are activated at any given time. Indicates in The moment that was activated The default weight of the rule, express With the The moment that was activated The matching degree of the rules, where, , .

[0012] In one embodiment of the present invention, based on the matching degree Determine the input of the various monitoring data. The steps for obtaining the final evaluation result after the rule is activated at a given time include: According to the above The weights of the rules activated at each moment and the confidence level of each evaluation result are used to determine the various monitoring data inputs. After the rule is activated at a given time, the overall confidence level of each evaluation result is: ; in, ; In the formula, As an intermediate variable, Indicates the first The overall confidence level of the evaluation results Indicates the first The evaluation results are in The moment that was activated The confidence level in the rule, ; according to Each evaluation result and the total confidence level of each evaluation result are used to determine... Final evaluation results at the moment: ; In the formula, Indicates the first One evaluation result.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a confidence rule base-based method for evaluating the performance of aerospace components based on an attention mechanism. During the training of the aerospace component performance evaluation model, the default weights of prerequisite attributes are optimized using a feature attention module and various monitoring data. Then, a rule attention module determines the activated rules. Finally, an evidence reasoning algorithm is used to determine the final evaluation result after inputting various monitoring data into the activated rules. This invention enables performance evaluation of aerospace components using multi-source monitoring data, improves the evaluation accuracy of the aerospace component performance evaluation model, and simplifies the model structure. Furthermore, this invention can be used for intelligent decision-making in inertial navigation systems, demonstrating significant engineering application value.

[0014] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] Figure 1 This is a flowchart of an aviation component performance evaluation method based on an attention mechanism-driven confidence rule base provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the training process of the aviation component performance evaluation model provided in an embodiment of the present invention; Figure 3 This is an example diagram of multi-source monitoring data from the gyroscope and accelerometer of an inertial navigation system; Figure 4 This is a comparison chart of the performance evaluation results output by the aviation component performance evaluation model and the actual values ​​of the inertial navigation system performance status level. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0017] Figure 1 This is a flowchart of an aviation component performance evaluation method based on an attention mechanism-driven confidence rule base, provided in an embodiment of the present invention. Figure 1 As shown, this embodiment of the invention provides a method for evaluating the performance of aerospace components based on an attention mechanism-driven confidence rule base, comprising: S1. Collect observation data from the inertial navigation system of aircraft components; S2. Input the observed data into the pre-trained aircraft component performance evaluation model to obtain the performance evaluation results; the aircraft component performance evaluation model is trained according to the following steps: S21. Acquire various monitoring data from the inertial navigation system of aircraft components.

[0018] S22. Construct a confidence rule base based on multiple monitoring data, and set the default weights of the premise attributes and the default weights of the rules in the confidence rule base.

[0019] A Confidence Rule Base (CRB) is an intelligent system modeling algorithm based on evidence-based reasoning. In a CRB, rules and their premise attributes have a hierarchical dependency relationship. Firstly, in each rule, multiple premise attributes are logically combined to form the premise part of the rule. Each premise attribute is assigned a weight, representing its relative importance in the rule. The evaluation result and confidence distribution constitute the conclusion part of the rule, with the confidence level characterizing the reliability of the corresponding evaluation result. Secondly, a rule is activated only when all its premise attributes are simultaneously satisfied. For example, the rule form is: IF [Premise] THEN [Conclusion] WITH [Confidence Distribution]. More specifically, the first rule in the CRB... The rule is expressed as follows: ; In the formula, , The number of rules in the confidence rule base. They represent The first moment Such monitoring data, They represent the first The prerequisite attributes are Reference values ​​for monitoring data at any given time. Representation and operation They represent the first One evaluation result, They represent In the The confidence level in the rule, They represent the first The default weight of the first prerequisite attribute, the first The default weight of this rule is .

[0020] S23. Using the feature attention module and various monitoring data, optimize the default weights of the premise attributes to obtain the optimized weights of the premise attributes.

[0021] In this embodiment, the feature attention module includes a normalization unit and a weight calculation unit; Specifically, the steps for optimizing the default weights of the premise attributes using a feature attention module and various monitoring data to obtain optimized weights for the premise attributes include: S2301. Normalize various monitoring data using a normalization unit:

[0022] In the formula, Indicates the first Such monitoring data, Indicates the first The mean of various monitoring data, Indicates the first The standard deviation of the monitoring data Indicates the first Species-normalized monitoring data, That is, the number of premise attributes in the aforementioned confidence rule base is equal to the number of types of monitoring data. Of course, in some other embodiments of this application, correlation analysis, mutual information and other algorithms can also be used to select several types of monitoring data that have a significant impact on the output as premise attributes.

[0023] S2302, based on the first Species Normalization Monitoring Data The optimization weights of each prerequisite attribute are determined using the weight calculation unit: ; ; In the formula, Indicates the first Optimization weights of the prerequisite attributes, express Feature attention weights, Indicates the first The default weight of each prerequisite attribute. , These represent the weights of the first and second convolutional layers in the weight calculation unit, respectively. , These are the bias terms for the first and second convolutional layers, respectively.

[0024] S24. Based on the optimized weights of the premise attributes, the rule attention module is used to determine the activated rules in the confidence rule base.

[0025] Optionally, the rule-based attention module includes a similarity calculation unit and an activation probability calculation unit; The steps for determining the activated rules in the confidence rule base based on the optimized weights of the premise attributes and using the rule attention module include: S2401, according to the normalized first Various monitoring data Calculate using similarity calculation unit The first moment Species Normalization Monitoring Data With the k Rule-based cosine similarity : ; In the formula, ; S2402, Based on cosine similarity The activation probability calculation unit is used to calculate the first Rule in Activation probability at time: ; In the formula, , , , The number of rules in the confidence rule base. , Indicates the monitoring duration. Indicates the temperature coefficient. ; S2403, Activation Probability With preset threshold Compare; when At that time, the first Rule in It is activated at any moment.

[0026] S25. Use the evidence reasoning algorithm to determine the final evaluation result after the rules are activated by various monitoring data inputs, and use the final evaluation result and the preset objective function to optimize the parameters of the confidence rule base until the final evaluation result reaches the preset accuracy, and obtain the trained aviation component performance evaluation model.

[0027] Optionally, the step of using evidence reasoning algorithms to determine the final evaluation result after activating rules from multiple monitoring data inputs includes: S2501, Calculation The first moment Various monitoring data With the Matching degree of the rule .

[0028] Specifically, first determine The first moment Various monitoring data The monitoring data reference range that fell into ,in, They represent the first The prerequisite attributes are The first moment Reference values ​​for monitoring data; Next, according to and ,calculate With the The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ;in, when hour, ; when hour, ; when , Indicates the first The prerequisite attributes are The number of reference values ​​for monitoring data at any given time; Finally, according to With the The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ,calculate With the Matching degree of the rule: ; In the formula, .

[0029] Optionally, based on matching degree Determine multiple monitoring data inputs Before the step of obtaining the final evaluation result after the rule is activated at a certain time, it also includes: calculate The weights of each rule that is activated at any given time: ; In the formula, express The moment that was activated The activation probability of the rule. Indicates in The number of rules that are activated at any given time. Indicates in The moment that was activated The default weight of the rule, express and The moment that was activated The matching degree of the rules, where, , .

[0030] S2502, Based on matching degree Determine multiple monitoring data inputs The final evaluation result obtained after the rules are activated at a certain time.

[0031] In this step, firstly according to The weights of the rules activated at each time point and the confidence level of each evaluation result are used to determine the input of various monitoring data. After the rule is activated at a given time, the overall confidence level of each evaluation result is: ; in, ; In the formula, As an intermediate variable, Indicates the first The overall confidence level of the evaluation results Indicates the first The evaluation results are in The moment that was activated The confidence level in the rule, ; Furthermore, according to Each evaluation result and the overall confidence level of each evaluation result are used to determine... Final evaluation results at the moment: ; In the formula, Indicates the first One evaluation result.

[0032] This embodiment uses the mean squared error (MSE) objective function shown below to optimize the parameters of the confidence rule base: ; In the formula, It is a column vector composed of the parameters. This indicates the actual performance status of aircraft components.

[0033] To achieve the best possible fit, the following objective function is constructed. ; The constraints of the above objective function are as follows: .

[0034] The following simulation experiment demonstrates the aviation component diagnostic method based on attention mechanism driven by a confidence rule base provided by this invention.

[0035] like Figure 3 As shown in the figure, this embodiment takes the multi-source monitoring data of the gyroscope and accelerometer of the inertial navigation system as an example, and acquires a total of 600 sets of test data. The reference levels are set as "low", "lower", "higher", and "high", represented by "L", "S", "V", and "H" respectively. The correspondence between the reference values ​​and the reference levels is shown in Table 1 below. The output performance evaluation results include four types, such as "very low", "low", "medium", and "good". The values ​​corresponding to each level are shown in Table 2. Table 1

[0036] Table 2

[0037] Furthermore, a confidence rule base was constructed based on the aforementioned monitoring data. This confidence rule base includes 16 rules. The second column of Table 3 shows the optimized weights of the premise attributes calculated based on the feature attention module, and the last column of Table 3 shows the confidence level of the evaluation result for each rule. Table 3

[0038] like Figure 4 As shown, the line segment with the asterisk (*) represents the actual value of the inertial navigation system's performance status level, while the line segment with the dot (*) represents the performance evaluation result output from the aviation component performance evaluation model. Figure 4 It can be seen that the performance evaluation model of the aviation components provided in this application can effectively and accurately evaluate the performance status level of the inertial navigation system.

[0039] As can be seen from the above embodiments, the beneficial effects of the present invention are as follows: This invention provides a confidence rule base-based method for evaluating the performance of aerospace components based on an attention mechanism. During the training of the aerospace component performance evaluation model, the default weights of prerequisite attributes are optimized using a feature attention module and various monitoring data. Then, a rule attention module determines the activated rules. Finally, an evidence reasoning algorithm is used to determine the final evaluation result after inputting various monitoring data into the activated rules. This invention enables performance evaluation of aerospace components using multi-source monitoring data, improves the evaluation accuracy of the aerospace component performance evaluation model, and simplifies the model structure. Furthermore, this invention can be used for intelligent decision-making in inertial navigation systems, demonstrating significant engineering application value.

[0040] In the description of this invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0041] Although this application has been described herein in conjunction with various embodiments, other variations of the disclosed embodiments can be understood and implemented by those skilled in the art in carrying out the claimed application by reviewing the accompanying drawings, the disclosure, and the appended claims.

[0042] The foregoing description provides a further detailed explanation of the present invention in conjunction with specific preferred embodiments, but it should not be construed as limiting the specific implementation of the invention to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A performance evaluation method for aerospace components based on an attention-driven confidence rule base, characterized in that, include: Collect observation data from the inertial navigation system of aircraft components; The observed data is input into a pre-trained aircraft component performance evaluation model to obtain performance evaluation results; wherein, the aircraft component performance evaluation model is trained according to the following steps: Acquire various monitoring data from the inertial navigation system of aircraft components; A confidence rule base is constructed based on the aforementioned multiple monitoring data, and default weights for prerequisite attributes and rules are set in the confidence rule base. Using the feature attention module and the various monitoring data, the default weights of the prerequisite attributes are optimized to obtain the optimized weights of the prerequisite attributes. Based on the optimized weights of the aforementioned prerequisite attributes, the rules activated in the confidence rule base are determined using the rule attention module. The final evaluation result after inputting various monitoring data into the activated rules is determined using an evidence reasoning algorithm. The parameters of the confidence rule base are then optimized using the final evaluation result and a preset objective function until the final evaluation result reaches a preset accuracy, thereby obtaining a trained aviation component performance evaluation model.

2. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 1, characterized in that, The confidence rule base The rule is expressed as follows: ； In the formula, , The number of rules in the confidence rule base. They represent The first moment Various monitoring data, They represent the first The prerequisite attributes are Reference values ​​for monitoring data at any given time. Representation and operation They represent the first One evaluation result, They represent In the first The confidence level in the rule, They represent the first The default weight of the first prerequisite attribute, the first The default weight of this rule is .

3. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 2, characterized in that, The feature attention module includes a normalization unit and a weight calculation unit; The step of optimizing the default weights of the prerequisite attributes using a feature attention module and the various monitoring data to obtain optimized weights for the prerequisite attributes includes: The normalization unit is used to normalize various monitoring data: In the formula, Indicates the first Various monitoring data, , Indicates the first The mean of various monitoring data, Indicates the first The standard deviation of the monitoring data Indicates the first Species-normalized monitoring data; Based on the Species Normalization Monitoring Data The optimization weights of each prerequisite attribute are determined using the weight calculation unit. ； ； In the formula, Indicates the first Optimization weights of the prerequisite attributes, express Feature attention weights, Indicates the first The default weight of each prerequisite attribute. , These represent the weights of the first and second convolutional layers in the weight calculation unit, respectively. , These are the bias terms for the first and second convolutional layers, respectively.

4. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 3, characterized in that, The rule-based attention module includes a similarity calculation unit and an activation probability calculation unit; The step of determining the activated rules in the confidence rule base using the rule attention module based on the optimized weights of the aforementioned prerequisite attributes includes: According to the normalized first Various monitoring data The similarity calculation unit is used to calculate The first moment Species Normalization Monitoring Data With the k Rule-based cosine similarity : ； In the formula, ; Based on the cosine similarity The activation probability calculation unit is used to calculate the first... The rule is Activation probability at time: ； In the formula, , , , The number of rules in the confidence rule base. , Indicates the monitoring duration. Indicates the temperature coefficient. ; The activation probability With preset threshold Compare; when At that time, the first The rule is It is activated at any moment.

5. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 2, characterized in that, The steps for determining the final evaluation result after inputting multiple monitoring data into the activated rule using an evidence-based reasoning algorithm include: calculate The first moment Various monitoring data With the first Matching degree of the rule ; Based on the matching degree Determine the input of the various monitoring data. The final evaluation result obtained after the rules are activated at a given time.

6. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 5, characterized in that, calculate The first moment Various monitoring data With the first Matching degree of the rule The steps include: Sure The first moment Various monitoring data The monitoring data reference range that fell into ,in, They represent the first The prerequisite attributes are The first moment Reference values ​​for monitoring data; according to and ,calculate With the first The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ;in, when hour, ; when hour, ; when , Indicates the first The prerequisite attributes are The number of reference values ​​for monitoring data at any given time; according to With the first The prerequisite attributes are The first moment Matching degree of individual monitoring data reference values ,calculate With the first Matching degree of the rule: ； In the formula, .

7. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base as described in claim 6, characterized in that, Based on the matching degree Determine the input of the various monitoring data. Before the step of obtaining the final evaluation result after the rule is activated at a certain time, it also includes: calculate The weights of each rule that is activated at any given time: ； In the formula, express The moment that was activated The activation probability of the rule. Indicates in The number of rules that are activated at any given time. Indicates in The moment that was activated The default weight of the rule, express With the The moment that was activated The matching degree of the rules, where, , .

8. The performance evaluation method for aerospace components based on an attention mechanism-driven confidence rule base according to claim 7, characterized in that, Based on the matching degree Determine the input of the various monitoring data. The steps for obtaining the final evaluation result after the rule is activated at a given time include: According to the above The weights of the rules activated at each moment and the confidence level of each evaluation result are used to determine the various monitoring data inputs. After the rule is activated at a given time, the overall confidence level of each evaluation result is: ； in, ； In the formula, As an intermediate variable, Indicates the first The overall confidence level of the evaluation results Indicates the first The evaluation results are in The moment that was activated The confidence level in the rule, ; according to Each evaluation result and the total confidence level of each evaluation result are used to determine... Final evaluation results at the moment: ； In the formula, Indicates the first One evaluation result.

Images