Aircraft health state determination method and system based on digital twinning and transfer learning
Through digital twin and transfer learning technology, the aircraft's data transfer model and health assessment model are built, which solves the problem of aircraft health status assessment and realizes accurate assessment of aircraft health status and maintenance decision support.
Patent Information
- Application Number
- CN202410989143.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
AI Technical Summary
It is difficult for the aircraft to monitor its health status in real time during long-term storage, and the harsh environment leads to the deterioration of its health status. There is a lack of effective evaluation methods to provide decision-making support for maintenance and spare parts replacement.
Using a method based on digital twins and transfer learning, a data transfer model of the aircraft is constructed, and the source and target domain characteristics are aligned in the high-video Hilbert space through migration component analysis, and empirical reasoning is used to fusion of index features to evaluate the health status of the aircraft.
The accurate assessment of the health status of the aircraft is achieved, reducing the dependence on the data volume of new operating conditions training samples, and improving the generalization ability of the evaluation and the actual consistency of the results.
Smart Images

Figure CN118940507A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of aircraft health status assessment, and in particular to a method and system for determining the health status of an aircraft based on digital twins and transfer learning. Background Art
[0002] Aircraft have the characteristics of "long-term storage, small amount of testing, and one-time use". During the entire life cycle from leaving the factory to scrapping, they are in storage for a long time, and the mission execution time is relatively short. Due to the long detection time interval, the sensor cannot monitor the parameters of the aircraft's health status in real time, and there are more normal data and less degradation data. At the same time, under the influence of the harsh battlefield environment, the health status of the aircraft will inevitably degrade and it will be difficult to meet the mission requirements. Therefore, it is necessary to take effective measures to grasp the health status of the aircraft and provide decision support for the maintenance and replacement of spare parts of the aircraft. Digital twin (DT) was first proposed by Professor Michael Grieves, aiming to establish a virtual digital model that can reflect the performance status of physical products in real time. On the one hand, the product in the physical space provides real-time test data for the twin model in the virtual space. On the other hand, the virtual space uses the acquired test data for modeling and simulation, thereby providing full life cycle services for the physical space products. Since the concept of digital twin technology is very broad, the present invention aims to use digital twin technology to achieve health assessment of aircraft. Based on the idea of digital twin model and data fusion, the effect of aircraft health status monitoring is improved by establishing a digital twin model that maps virtuality with the physical aircraft. For different aircraft, due to differences in manufacturers, delivery time, workpiece materials, and different training tasks, the performance degradation trends of different aircraft are also different. Therefore, considering migrating the monitoring data of different aircraft to the aircraft to be evaluated based on the idea of transfer learning, and building its digital twin model, it is beneficial to improve the generalization ability of aircraft health assessment. Compared with traditional machine learning methods, transfer learning methods can use a small number of samples and realize the knowledge transfer of aircraft, significantly reducing the model's dependence on the amount of training sample data under new conditions. Summary of the invention
[0003] The purpose of the present invention is to provide a method and system for determining the health status of an aircraft based on digital twins and transfer learning. Considering the lack of test data and expert experience of the aircraft, a data migration model of the aircraft is established with the help of the concept of digital twins, and then evidential reasoning is used to fuse indicator features to evaluate the health status of the aircraft, providing decision support for the spare parts ordering, repair and maintenance of the aircraft.
[0004] To achieve the above object, the present invention provides the following solutions:
[0005] A method for determining the health status of an aircraft based on digital twins and transfer learning, comprising:
[0006] Build a data migration model for aircraft;
[0007] Solving the data migration model of the aircraft to obtain an optimal mapping kernel matrix;
[0008] Using the optimal mapping kernel matrix to project and migrate the original characteristic data of the aircraft;
[0009] The original feature data after projection migration are fused to obtain the health status of the aircraft.
[0010] Optionally, the data migration model of the aircraft is expressed as follows:
[0011]
[0012] Where trace(·) represents the trace of the matrix KL, W represents the optimal mapping kernel matrix, K S,S , K S,T are the kernel matrices of the source domain and the target domain respectively, N1 is the number of samples in the source domain, N2 is the number of samples in the target domain, H represents the center matrix, μ>0 is the balance parameter, I M ∈R M×M is the identity matrix and T represents the transpose.
[0013] Optionally, the expression of the central matrix is:
[0014]
[0015] in, is a column vector whose elements are all 1, is the identity matrix.
[0016] Optionally, the original feature data after the projection migration is fused to obtain the health status of the aircraft using the following formula:
[0017]
[0018] Among them, S(θ) represents the distributed health assessment result after all evidences are integrated, p θ It represents the confidence of the proposition θ after fusion, and Θ represents the health assessment identification framework.
[0019] Optionally, the expression of the confidence level pointing to the proposition θ after fusion is as follows:
[0020]
[0021] Among them, mθ,1 =w1p θ,1 , m θ,2 =w1p θ,2 , p θ,e(2) represents the confidence of the fusion of two pieces of evidence, w1 represents the weight of evidence e1, w2 represents the weight of evidence e2, r1 represents the reliability of evidence e1, r2 represents the reliability of evidence e2, m B,1 represents the probability mass function of evidence e1 pointing to any proposition or subset, m C,2 It represents the probability mass function of evidence e2 pointing to any proposition or subset, and D is a variable representing the traversal of the propositions in Θ. It means to traverse the propositions in Θ after fusion, p θ,1 represents the confidence level of the evidence e1 pointing to the proposition θ, p θ,2 represents the confidence of the proposition θ in the evidence e2, B and C are any different propositions in the identification framework Θ, and θ represents the proposition in the identification framework Θ. It represents the probability mass of the proposition θ after the two pieces of evidence are combined.
[0022] Based on the above method in the present invention, the present invention further provides an aircraft health status determination system based on digital twin and transfer learning, comprising:
[0023] A data migration model building module is used to build a data migration model for an aircraft;
[0024] A solution module, used for solving the data migration model of the aircraft to obtain an optimal mapping kernel matrix;
[0025] A projection migration module, used for projecting and migrating the original feature data of the aircraft using the optimal mapping kernel matrix;
[0026] The feature fusion module is used to fuse the original feature data after projection migration to obtain the health status of the aircraft.
[0027] Optionally, the expression of the data migration model construction module is as follows:
[0028]
[0029] Where trace(·) represents the trace of the matrix KL, W represents the optimal mapping kernel matrix, K S,S , K S,T are the kernel matrices of the source domain and the target domain respectively, N1 is the number of samples in the source domain, N2 is the number of samples in the target domain, H represents the center matrix, μ>0 is the balance parameter, I M ∈R M×M is the identity matrix and T represents the transpose.
[0030] Optionally, the expression of the central matrix is:
[0031]
[0032] in, is a column vector whose elements are all 1, is the identity matrix.
[0033] Optionally, the feature fusion module specifically adopts the following formula:
[0034]
[0035] Among them, S(θ) represents the distributed health assessment result after all evidences are integrated, p θ It represents the confidence of the proposition θ after fusion, and Θ represents the health assessment identification framework.
[0036] Optionally, the expression of the confidence level pointing to the proposition θ after fusion is as follows:
[0037]
[0038] Among them, m θ,1 =w1p θ,1 , m θ,2 =w1p θ,2 , p θ,e(2) represents the confidence of the fusion of two pieces of evidence, w1 represents the weight of evidence e1, w2 represents the weight of evidence e2, r1 represents the reliability of evidence e1, r2 represents the reliability of evidence e2, m B,1 represents the probability mass function of evidence e1 pointing to any proposition or subset, m C,2 It represents the probability mass function of evidence e2 pointing to any proposition or subset, and D is a variable representing the traversal of the propositions in Θ. It means to traverse the propositions in Θ after fusion, p θ,1 represents the confidence level of the evidence e1 pointing to the proposition θ, p θ,2 represents the confidence of the proposition θ in the evidence e2, B and C are any different propositions in the identification framework Θ, and θ represents the proposition in the identification framework Θ. It represents the probability mass of the proposition θ after the two pieces of evidence are combined.
[0039] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0040] The present invention provides a method and system for determining the health status of an aircraft based on digital twins and transfer learning. By constructing a digital twin model of an aircraft, the health status of the aircraft is evaluated. First, the migration component analysis method is used to migrate data to achieve feature alignment between the source domain and the target domain in the high-dimensional Hilbert space; secondly, the indicator features are fused based on evidence reasoning to achieve knowledge fusion at the feature layer; finally, according to the implementation steps of the proposed method, an aircraft health assessment model is established. The present invention can comprehensively utilize digital twins and transfer learning technologies to achieve the evaluation of the health status of the aircraft and ensure that the evaluation results are consistent with the actual situation. 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0042] Figure 1 A flow chart of a method for determining the health status of an aircraft based on digital twins and transfer learning provided by the present invention;
[0043] Figure 2 A schematic diagram of the aircraft health assessment results provided by the present invention. DETAILED DESCRIPTION
[0044] 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.
[0045] The purpose of the present invention is to provide a method and system for determining the health status of an aircraft based on digital twins and transfer learning. Considering the lack of test data and expert experience of the aircraft, a data migration model of the aircraft is established with the help of the concept of digital twins, and then evidential reasoning is used to fuse indicator features to evaluate the health status of the aircraft, providing decision support for the spare parts ordering, repair and maintenance of the aircraft.
[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0047] Embodiment 1
[0048] Figure 1The flowchart of the method for determining the health status of an aircraft based on digital twin and transfer learning provided by the present invention is as follows: Figure 1 As shown, the method of the present invention comprises:
[0049] Step 1: Build a data migration model for the aircraft.
[0050] The data migration model of the aircraft is constructed to achieve the alignment of heterogeneous data, with the goal of constructing an optimization objective function to achieve consistency in the distribution of source domain data and target domain data in high-dimensional space.
[0051] For aircraft A and aircraft B with the same specifications and models, the production and delivery time is the same, but the mission profiles they have experienced are different. Contains s samples. For sample The health status label of the source domain is , and Υ={1,2,...,k} is the health status space, which contains k aircraft health statuses. Belongs to the sample space X s , obeys the marginal probability distribution P(X s ). The health monitoring data set of aircraft B is Due to the high cost of dataset annotation, there is only a small amount of annotation information in the target domain dataset. Belongs to the sample space X t , obeys the marginal probability distribution P(X t ), due to the different service environments and operating conditions, the source domain P(X s ) and the target domain P(X t ) have different marginal probability distributions.
[0052] In order to realize the migration of data samples from aircraft A (source domain) to aircraft B (target domain), we first need to consider the actual data distribution characteristics of the source domain and the target domain. The data distribution of multiple aircraft of the same model delivered in the same batch is not only similar in overall, but also similar in each health state. The former is called the marginal distribution of the data, and the latter is called the conditional distribution of the data. In order to solve the problem of data migration, a transfer component analysis (TCA) method is proposed. This method can adjust the importance of the marginal distribution and conditional distribution of the data according to the actual data distribution, and realize the migration and adaptation of data from the source domain to the target domain. The specific implementation steps are as follows:
[0053] When applying the migration component analysis method, the sample data of the source domain and the target domain are first mapped to the same feature subspace, and the distribution metric criterion is used to reduce the difference in feature distribution: that is, assuming that the source domain is Ds ={X s , Y s}, where X s is the source domain sample set, Y s is a label sample; the target domain is D T ={X T}, where X T is a target domain sample set with unknown labels. Through the feature mapping function Φ, the marginal probability distribution after mapping is made as similar as possible, that is, P(Φ(X s ))≈Q(Φ(X T )).
[0054] Among them, Φ(X s ) and Φ(X T ) is the source domain feature sample set and the target domain feature sample set after kernel Hilbert space mapping. The distance between the source domain and the target domain is expressed as:
[0055]
[0056] In the formula, ||·|| H is the RKHS norm; N1 is the number of source domain samples; N2 is the number of target domain samples; The above mapping function Φ is transformed into a kernel learning problem, and the distance between the source domain and the target domain after mapping is expressed as:
[0057] dis(Φ(X s ),Φ(X T ))=trace(KL)
[0058] Where trace(·) represents the trace of the matrix KL, where K S,S , K S,T are the kernel matrices of the source domain and the target domain respectively,
[0059]
[0060] Define the matrix, M, N1+N2 maps the kernel function to a high-dimensional space, and its kernel matrix is transformed into:
[0061]
[0062] In the formula, Substituting the kernel matrix, we can get the distance between the source domain and the target domain after mapping:
[0063] dis(Φ(X s ),Φ(X T ))=trace((KWW T K)L)=trace(WT KKW)
[0064] From the above formula, the sample data mapped to the high-dimensional space is W T K, the covariance matrix is W T KHKW. Among them, It is the central array. is a column vector whose elements are all 1, is the identity matrix.
[0065] In order to minimize the distance between different domain distributions, in addition to minimizing the distance between the source domain and the target domain, it is also necessary to add a regularization term μ·trace(W T W) to control the complexity of the matrix W, and the optimization objective is transformed into:
[0066]
[0067] wxya T KHKW=I M
[0068] Where: μ>0 is the equilibrium parameter, I M ∈R M×M is the identity matrix.
[0069] According to the constraint condition W T KHKW=I M , the constrained optimization function can be transformed into the objective function solution:
[0070]
[0071] The above formula can be used to solve (I+μKLK) -1 KHK obtains the best mapping kernel matrix W, sample matrix X * = KW. The optimal mapping kernel matrix W is to project and migrate the original data, thereby achieving distribution adaptation between the source domain and the target domain. In the next step, the projected features are fused using evidence reasoning.
[0072] Step 2: Solve the data migration model of the aircraft to obtain the optimal mapping kernel matrix.
[0073] Step 3: Use the optimal mapping kernel matrix to project and migrate the original feature data of the aircraft.
[0074] Step 4: Fuse the original feature data after projection migration to obtain the health status of the aircraft.
[0075] The first two steps migrate the indicators of the aircraft to similar fields, thereby achieving the distribution alignment of indicator features in the Hilbert space. In order to achieve indicator fusion at the feature level, the evidence reasoning algorithm is used to fuse the knowledge of the aircraft, and the fusion result is the aircraft health status assessment result. The ER method has unique advantages in realizing multi-information fusion and handling uncertainty. Based on the ER method, the evaluation is carried out and the evaluation model is established according to the constructed comprehensive evaluation indicator system. The core algorithm of evidence reasoning is as follows:
[0076] Assume that after the indicator migration, an indicator feature can be used as evidence e i It is expressed as the following reliability distribution form:
[0077]
[0078] Among them, p θ,i Indicates evidence i The reliability of the supported proposition θ, Θ represents the health assessment identification framework. i The weight and reliability are w i and r i , 0≤w i ,r i ≤1. For two independent pieces of evidence e1 and e2,
[0079]
[0080] Among them, m θ,1 =w1p θ,1 , m θ,2 =w1p θ,2 , p θ,e(2) represents the confidence of the fusion of two pieces of evidence, and D is any proposition in the identification framework Θ. Then, the combination of L independent pieces of evidence can be expressed as:
[0081]
[0082] Among them, m θ,e(1) =m θ,1 , m P(Θ),e(1) =m P(Θ),1 The overall reliability of the evidence combination is calculated as follows:
[0083] r e(L) (w e(L) )=(1-m P(Θ),e(L) (1+w e(L) )) / (1-m P(Θ),e(L) )
[0084] By integrating the characteristics of the aircraft through the ER algorithm in the above steps, the evaluation result of the aircraft health status can be obtained, which can be expressed as the following confidence distribution form:
[0085]
[0086] Among them, S(θ) represents the distributed health assessment result after all evidences are integrated, p θ Represents the confidence of the proposition θ after fusion.
[0087] In order to verify the proposed method, the aircraft test data is used as experimental support to carry out the aircraft health status assessment. According to historical experience and the actual needs of health assessment, the aircraft health status is divided into 4 levels, namely healthy, normal, sub-healthy, and failure.
[0088] Table 1 Classification of health status
[0089]
[0090] Firstly, based on steps 1 and 2, the source domain and target domain aircraft data are mapped into the same Hilbert space through the TCA method, the maximum mean difference distance is used to measure the transferable data samples, and the features of the source domain data are extracted.
[0091] Secondly, after obtaining the features of the sample after migration, knowledge fusion is realized based on the evidence reasoning in steps 3 and 4 to obtain the distributed health assessment results, such as Figure 2 In engineering applications, after an aircraft leaves the factory and is put into service, it has a low probability of being in a healthy state after long-term use and duty missions, and is in a normal state after performance degradation; with further service, it will be in a sub-healthy state.
[0092] Embodiment 2
[0093] Based on the method for determining the health status of an aircraft based on digital twins and transfer learning in the first embodiment, this embodiment further provides a system for determining the health status of an aircraft based on digital twins and transfer learning, including:
[0094] A data migration model building module is used to build a data migration model for an aircraft;
[0095] A solution module, used for solving the data migration model of the aircraft to obtain an optimal mapping kernel matrix;
[0096] A projection migration module, used for projecting and migrating the original feature data of the aircraft using the optimal mapping kernel matrix;
[0097] The feature fusion module is used to fuse the original feature data after projection migration to obtain the health status of the aircraft.
[0098] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0099] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for determining the health status of an aircraft based on digital twins and transfer learning, characterized in that: include: Build a data migration model for aircraft; Solving the data migration model of the aircraft to obtain an optimal mapping kernel matrix; Using the optimal mapping kernel matrix to project and migrate the original characteristic data of the aircraft; The original feature data after projection migration are fused to obtain the health status of the aircraft.
2. The method for determining the health status of an aircraft based on digital twins and transfer learning according to claim 1, characterized in that: The expression of the data migration model of the aircraft is as follows: Where trace(·) represents the trace of the matrix KL, W represents the optimal mapping kernel matrix, K S,S , K S,T are the kernel matrices of the source domain and the target domain respectively, N1 is the number of samples in the source domain, N2 is the number of samples in the target domain, H represents the center matrix, μ>0 is the balance parameter, I M ∈R M×M is the identity matrix and T represents the transpose.
3. The method for determining the health status of an aircraft based on digital twins and transfer learning according to claim 2, characterized in that: The expression of the central matrix is: in, is a column vector whose elements are all 1, is the identity matrix.
4. The method for determining aircraft health status based on digital twin and transfer learning according to claim 1, characterized in that: The original feature data after the projection migration is fused to obtain the health status of the aircraft using the following formula: Among them, S(θ) represents the distributed health assessment result after all evidences are integrated, p θ It represents the confidence of the proposition θ after fusion, and Θ represents the health assessment identification framework.
5. The method for determining aircraft health status based on digital twin and transfer learning according to claim 4, characterized in that: The expression of the confidence level pointing to proposition θ after fusion is as follows: Among them, m θ,1 =w1p θ,1 , m θ,2 =w1p θ,2 , p θ,e(2) represents the confidence of the fusion of two pieces of evidence, w1 represents the weight of evidence e1, w2 represents the weight of evidence e2, r1 represents the reliability of evidence e1, r2 represents the reliability of evidence e2, m B,1 represents the probability mass function of evidence e1 pointing to any proposition or subset, m C,2 It represents the probability mass function of evidence e2 pointing to any proposition or subset, and D is a variable representing the traversal of the propositions in Θ. It means to traverse the propositions in Θ after fusion, p θ,1 represents the confidence level of the evidence e1 pointing to the proposition θ, p θ,2 represents the confidence of the proposition θ in the evidence e2, B and C are any different propositions in the identification framework Θ, and θ represents the proposition in the identification framework Θ. It represents the probability mass of the proposition θ after the two pieces of evidence are combined.
6. An aircraft health status determination system based on digital twins and transfer learning, characterized in that: include: A data migration model building module is used to build a data migration model for an aircraft; A solution module, used for solving the data migration model of the aircraft to obtain an optimal mapping kernel matrix; A projection migration module, used for projecting and migrating the original feature data of the aircraft using the optimal mapping kernel matrix; The feature fusion module is used to fuse the original feature data after projection migration to obtain the health status of the aircraft.
7. The aircraft health status determination system based on digital twin and transfer learning according to claim 6, characterized in that: The expression of the data migration model building module is as follows: Where trace(·) represents the trace of the matrix KL, W represents the optimal mapping kernel matrix, K S,S , K S,T are the kernel matrices of the source domain and the target domain respectively, N1 is the number of samples in the source domain, N2 is the number of samples in the target domain, H represents the center matrix, μ>0 is the balance parameter, I M ∈R M×M is the identity matrix and T represents the transpose.
8. The aircraft health status determination system based on digital twin and transfer learning according to claim 7, characterized in that: The expression of the central matrix is: in, is a column vector whose elements are all 1, is the identity matrix.
9. The aircraft health status determination system based on digital twin and transfer learning according to claim 6, characterized in that: The feature fusion module specifically adopts the following formula: Among them, S(θ) represents the distributed health assessment result after all evidences are integrated, p θ It represents the confidence of the proposition θ after fusion, and Θ represents the health assessment identification framework.
10. The aircraft health status determination system based on digital twin and transfer learning according to claim 9, characterized in that: The expression of the confidence level pointing to proposition θ after fusion is as follows: Among them, m θ,1 =w1p θ,1 , m θ,2 =w1p θ,2 , p θ,e(2) represents the confidence of the fusion of two pieces of evidence, w1 represents the weight of evidence e1, w2 represents the weight of evidence e2, r1 represents the reliability of evidence e1, r2 represents the reliability of evidence e2, m B,1 represents the probability mass function of evidence e1 pointing to any proposition or subset, m C,2 It represents the probability mass function of evidence e2 pointing to any proposition or subset, and D is a variable representing the traversal of the propositions in Θ. It means to traverse the propositions in Θ after fusion, p θ,1 represents the confidence level of the evidence e1 pointing to the proposition θ, p θ,2 represents the confidence of the proposition θ in the evidence e2, B and C are any different propositions in the identification framework Θ, and θ represents the proposition in the identification framework Θ. It represents the probability mass of the proposition θ after the two pieces of evidence are combined.
Citation Information
Cited By
Trusted evidence reasoning method, device and equipment for multi-source data fusion
CN120012944A
Unmanned aerial vehicle flight state simulation method based on digital twinning
CN121278577A
