Space-air equipment key component life cycle prediction-oriented digital-analog fusion method

Abstract

The application discloses a numerical model fusion method for spaceflight equipment key component life cycle prediction, belongs to the field of spaceflight equipment key component health management and intelligent fault prediction, and comprises the following steps: constructing an equipment multi-dimensional tree-like hierarchical retrieval mechanism and a quantitative mapping relationship of motion mode, operation temperature and failure mode, adopting a tree-like CNN (convolutional neural network) driven graph neural network algorithm to build a prediction model intelligent retrieval architecture; designing a data relationship generator to fuse small sample test data and basic working condition data, training a dynamic prediction model based on time series reinforcement learning, and realizing real-time life prediction of key components. Through deep fusion of numerical models and multi-technology cooperation, the application improves the life cycle prediction accuracy and real-time performance of key components, and provides core support for equipment design optimization, health management and operation and maintenance support.

Description

Technical Field

[0001] This invention belongs to the field of health management and intelligent fault prediction of key components of aerospace equipment, and specifically relates to a numerical-analog fusion method for life cycle prediction of key components of aerospace equipment. Background Technology

[0002] In the research and development of key components for aerospace equipment, extensive physical testing is required after design to verify lifespan and reliability under various scenarios. However, there are pain points such as difficulty in constructing test environments (requiring simulation of extreme conditions, complex platforms, and lengthy debugging), high costs (large consumption of test pieces, with special components costing 5-10 times more than conventional ones), and long cycles (averaging over 3 months for full-condition testing), which restrict research and development efficiency. For example, a neural network-based engine remaining life prediction method and model disclosed in Chinese patent application CN202411991492.5 only focuses on component-level life prediction and does not extend to the part level; a digital-analog linkage aerospace bearing remaining life prediction method disclosed in Chinese patent application CN202311423213.0 only focuses on a single bearing component and lacks the ability to dynamically predict similar components. Summary of the Invention

[0003] To address the challenges of constructing physical testing environments, incurring high testing costs, and experiencing long testing cycles in the development of key components for aerospace equipment, this invention provides a numerical-model fusion method for lifecycle prediction of key components. It utilizes a graph neural network (GNN) to capture the topological relationships between operating conditions and failures, enabling rapid matching and adaptive invocation of prediction models for different models and operating conditions. Based on temporal reinforcement learning, it achieves dynamic tracking of the entire lifecycle status and real-time prediction of failure times. This invention, through the fusion of digital testing and small-sample physical testing, reduces the frequency of physical testing while ensuring the accuracy of test results, thus improving the efficiency of aerospace equipment development and iteration.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A numerical-model fusion method for lifecycle prediction of key components of aerospace equipment includes the following steps:

[0006] S1. Construct a multi-dimensional tree-structured hierarchical retrieval mechanism and a quantitative mapping and association mechanism for motion mode, operating temperature, and failure mode of aerospace equipment components;

[0007] S2. Build a graph neural network algorithm driven by a tree-like convolutional neural network, and construct an intelligent retrieval architecture for a dynamic life cycle prediction model under different models and working conditions;

[0008] S3. Design a data relationship generator that uses real-time test data and mechanical test life detection data as small sample data sources. Through generative data enhancement and spatiotemporal alignment and fusion of multi-source data, it deeply integrates the actual working condition operation data of aerospace equipment with small sample test data at the spatiotemporal and semantic levels to generate an expanded dataset.

[0009] S4. Construct a neural network based on temporal reinforcement learning, train the life cycle dynamic prediction model of each component, and quickly call the matching life cycle dynamic prediction model by name through an intelligent retrieval architecture. Continuously ingest real-time test data or online operation data to realize dynamic tracking of the full life cycle status of components and real-time update of failure time.

[0010] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described numerical-analog fusion method for life cycle prediction of key components of aerospace equipment.

[0011] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment.

[0012] Beneficial effects:

[0013] This invention, based on a tree-structured CNN-GNN intelligent retrieval architecture, forms an organically organized predictive model library. New tasks can quickly obtain adapted models through a retrieval-transfer-fine-tuning process, greatly improving model deployment efficiency and adaptability to new models and operating conditions. This enables continuous accumulation and sharing of predictive knowledge, laying the foundation for building an evolvable and reusable intelligent model ecosystem. Through a data relationship generator embedded with physical laws, small-sample experimental data and multi-dimensional operational data are deeply integrated at the spatiotemporal and semantic levels, generating an expanded dataset that conforms to both data distribution and physical laws. This fundamentally solves the problem of model overfitting and inaccurate predictions caused by data scarcity, thereby achieving high accuracy and robustness of the predictive model under small-sample conditions. The predictive model based on temporal reinforcement learning can integrate real-time monitoring data online, continuously adjusting and optimizing its lifespan prediction. This not only provides the current health status but also enables rolling, real-time updates of the remaining useful life, providing dynamic decision-making basis for condition-based maintenance and health management of aerospace equipment, realizing a leap from "static assessment" to "dynamic tracking" in lifespan prediction. In summary, this invention addresses the specific bottleneck problem of lifespan prediction for aerospace equipment by creatively designing a new closed-loop collaborative paradigm of data-model fusion that integrates "intelligent model retrieval, multi-source data fusion, and dynamic sequence prediction." Attached Figure Description

[0014] Figure 1 This is a flowchart of the numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to the present invention.

[0015] Figure 2 This is a comparative simulation diagram of an embodiment of the numerical-analog fusion method for life cycle prediction of key components of aerospace equipment according to the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0017] like Figure 1 As shown, this invention proposes a numerical-model fusion method for lifecycle prediction of key components of aerospace equipment, comprising the following steps:

[0018] S1. Construct a multi-dimensional retrieval and association mechanism for aerospace equipment components, namely, construct a multi-dimensional tree-structured hierarchical retrieval mechanism for aerospace equipment components and a quantitative mapping and association mechanism for motion mode, operating temperature, and failure modes of key components.

[0019] S2. Build a tree-structured CNN (convolutional neural network) driven GNN (graph neural network) algorithm, construct an intelligent retrieval architecture for prediction models under different models and working conditions, and build a fast retrieval mechanism to enable the rapid acquisition of adapted models for retrieval, migration and fine-tuning of new tasks.

[0020] S3. Design a data relationship generator. Using real-time test data and mechanical test life detection data as small sample data sources, through generative data augmentation and spatiotemporal alignment fusion of multi-source data, and through a data relationship generator embedded with physical laws, the actual operating data of equipment and small sample test data are deeply integrated at the spatiotemporal and semantic levels to generate an expanded dataset that conforms to both data distribution and physical laws. This fundamentally solves the problem of model overfitting and inaccurate prediction caused by data scarcity, thereby achieving high accuracy and high robustness of the prediction model under small sample conditions.

[0021] S4. Construct a neural network based on Temporal Reinforcement Learning (TRL) to train a dynamic lifecycle prediction model for each component. The core innovation of this dynamic lifecycle prediction model lies in its "plug-and-play" dynamic prediction mechanism: supported by a unified relationship interface built by a data relationship generator, a matching dynamic lifecycle prediction model can be quickly invoked by name (model + operating condition) through an intelligent retrieval architecture. This dynamic lifecycle prediction model can continuously ingest real-time test or online operation data (such as vibration, temperature, stress, etc.). Real-time test or online operation data (such as vibration, temperature, stress, etc.) are used as dynamic input to achieve a leap from static prediction to closed-loop adaptive inference, ultimately completing the dynamic tracking of the entire lifecycle status of components and real-time updates of failure time.

[0022] Furthermore, S1 includes:

[0023] When constructing a multi-dimensional tree-structured hierarchical retrieval mechanism, the tree-structured hierarchy refers to system-subsystem-component. The level of detail in the "system-subsystem-component" hierarchy is determined based on the complexity of the aerospace equipment and the component management requirements. To ensure accurate location of any key component, the matching degree of the hierarchical retrieval is scored using the following formula:

[0024] ;

[0025] In the formula, S represents the search matching score, and n represents the number of attribute dimensions for hierarchical retrieval. The weight of the i-th attribute dimension (satisfying) ), To retrieve input attribute x i and database attribute y i The similarity is between [0,1].

[0026] When establishing a quantitative mapping relationship between motion mode, operating temperature, and failure modes of key components, the failure probability and failure mode characteristic parameters of key components under different combinations of motion mode and operating temperature are determined by analyzing a large amount of historical operating data and failure data, combined with small-sample physical experiments. The core idea is to quantify and encode the operating condition variable "motion mode," and use it together with the continuous variable "operating temperature" as input. Through a joint probability and regression model, the estimated values ​​of the failure probability and key failure characteristic parameters under specific operating condition combinations are output.

[0027] First, motion mode A is represented using one-hot encoding as follows:

[0028] ;

[0029] in, , The element representing A is where n is the preset number of motion types, and p represents the index value of the motion type. For example, for "uniform rotation," its encoding might be: .

[0030] Let T represent the operating temperature. Based on the Logistic regression model, under the (A,T) operating condition, what is the probability P of component failure? f The calculation formula is as follows:

[0031] ;

[0032] The intermediate function z(A,T) uses a linear combination of terms to describe the effect of operating speed and temperature on the failure probability, as follows:

[0033] ;

[0034] in, For the intercept term; For the first The contribution coefficient of each motion mode to the failure probability; This is the contribution coefficient of standardized temperature to the failure probability; This represents the interaction coefficient between motion mode and temperature. These are weighting factors for different motion modes, used to adjust for differences in temperature sensitivity. Model parameters. , , , , It is determined by using maximum likelihood estimation on historical failure data (recording operating conditions and whether failure occurred).

[0035] Further, S2 includes:

[0036] In the intelligent retrieval architecture that integrates tree-structured CNN and GNN, parameters such as the number of network layers in the tree-structured CNN, the size of the convolutional kernels, and the node feature dimensions of the GNN are optimized and determined through methods such as cross-validation and model performance evaluation.

[0037] ;

[0038] in, .

[0039] In the above formula, The combination of hyperparameters to be optimized (such as the number of network layers, convolutional kernel size, and node feature dimension). This refers to the globally optimal combination of hyperparameters found through the optimization process. This is the preset search space for hyperparameters. These are the weight parameters of the model (such as the weights of convolutional kernels and fully connected layers in CNN and GNN networks). To be in a given hyperparameter The optimal model weights are obtained by optimizing the training set data. The loss function (such as cross-entropy loss) computed on the training set is used to train the model weights. . The composite loss function (i.e., the previously defined loss that includes retrieval accuracy and model quality), computed on the validation set, is used to evaluate and optimize hyperparameters. .

[0040] Specifically, the root mean square error (RMSE) is used as the indicator for model performance evaluation, and the calculation formula is as follows:

[0041] ;

[0042] In the formula, N is the number of validation samples. y represents the retrieval fit predicted by the model. j is the actual retrieval fit label, and j is the validation sample index value.

[0043] Further, S3 includes:

[0044] In the data relationship generator, generative data augmentation employs operating condition migration mapping rules and lifetime feature interpolation algorithms, which are formulated based on the distribution characteristics of small sample data and basic data, as well as the physical relationships between the data.

[0045] In two known "data-lifetime" samples and In between, intelligently generate a new, physically plausible sample. The core of this approach is to perform linear interpolation within the damage domain, specifically through the following two synchronously executed formulas. First, linear interpolation is performed in the physical feature space of the data (such as spectral features and stress features characterizing the damage state) to ensure the physical plausibility of the generated data:

[0046] ;

[0047] in, Let L represent the data, L represent the lifetime, and subscripts a and b represent different samples; Encode is the encoder that maps the raw data to the physical feature space, and Decode is the decoder that maps the physical features back to the data space; t is the interpolation weight, 0 <t<1。

[0048] Lifetime interpolation is based on Miner's linear cumulative damage theory. Interpolation is performed in the damage domain (the inverse of lifetime), and the result is then converted back to lifetime to obtain the converted lifetime. The formula is as follows:

[0049] ;

[0050] Further, S4 includes:

[0051] When using a time-series reinforcement learning-based training lifecycle dynamic prediction model, the reward function for the reinforcement learning agent is designed based on metrics such as improved prediction accuracy and the closeness of the prediction results to the actual failure time. The reward function is as follows:

[0052] ;

[0053] In the formula, R t Let be the reward value at time t. Let t be the model's predicted failure time. Acc is the failure time for testing. t Let be the prediction accuracy of the model at time t, α, β, and γ be the reward coefficients, and exp() denote the exponential function. This represents the prediction accuracy of the model at time t-1.

[0054] Specifically, this also includes steps for continuously updating and optimizing the lifecycle dynamic prediction model. Using newly acquired experimental and operational data, model parameters are adjusted periodically or in real-time to improve the model's predictive ability for new operating conditions and failure modes. The model parameter update uses the gradient descent method, and the parameter update rules are as follows:

[0055] ;

[0056] In the formula, θ t+1 For the updated model parameters, θ t The parameters are the model parameters before the update, and η is the learning rate, which ranges from [0,1]. The loss function is applied to the model parameters θ t The gradient.

[0057] The lifecycle dynamic prediction model is integrated into the health management system of aerospace equipment to achieve real-time display of lifecycle prediction results for key components, early warning push notifications, and interaction with the equipment operation and maintenance decision-making system. The early warning threshold is dynamically adjusted using a formula:

[0058] ;

[0059] in, Let T be the warning threshold at time t. base The basic early warning threshold (determined by operation and maintenance requirements) is set. The remaining lifetime predicted by the model. This is the threshold adjustment coefficient, with a value range of (0,1), and max indicates the maximum value.

[0060] Example:

[0061] This invention takes aero-engine turbine blades as an example. Through its intelligent retrieval architecture, it accurately matches historical models within 1 second, and uses a data relationship generator to fuse the only 15 sets of small sample measured data with basic data within 2 hours, generating 350 sets of physically reliable enhanced data. Figure 2 Comparative simulation diagrams of embodiments of a numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment, such as... Figure 2 As shown, the accuracy of the blade life prediction model was ultimately improved to 91.7%, which is a significant improvement of 15.2% compared with direct training on basic data. This effectively achieved high-precision and high-efficiency dynamic life prediction under small sample conditions.

[0062] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the above-described numerical-analog fusion method for life cycle prediction of key components of aerospace equipment.

[0063] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment.

[0064] Those skilled in the art will understand that embodiments of the present invention can be provided as methods or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0065] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0066] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0067] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0068] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0069] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0070] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention for those skilled in the art.

Claims

1. A numerical-model fusion method for lifecycle prediction of key components of aerospace equipment, characterized in that, Includes the following steps: S1. Construct a multi-dimensional tree-structured hierarchical retrieval mechanism and a quantitative mapping and association mechanism for motion mode, operating temperature, and failure mode of aerospace equipment components; First, motion mode A is represented using one-hot encoding as follows: ； in, , The element representing A is n, which is the preset number of movement types, and p represents the index value of the movement type. Let T represent the operating temperature. Based on a logistic regression model, what is the probability P of component failure under operating condition (A,T)? f The calculation formula is as follows: ； The intermediate function z(A,T) uses a linear combination of terms to describe the effect of operating speed and temperature on the failure probability, as follows: ； in, For the intercept term; For the first The contribution coefficient of each motion mode to the failure probability; This is the contribution coefficient of standardized temperature to the failure probability; This represents the interaction coefficient between motion mode and temperature. These are weighting factors for different motion modes, used to adjust for differences in temperature sensitivity among different motion modes; model parameters. , , , , Determined by maximum likelihood estimation of historical fault data; S2. Build a graph neural network algorithm driven by a tree-like convolutional neural network, and construct an intelligent retrieval architecture for a dynamic life cycle prediction model under different models and working conditions; S3. Design a data relationship generator that uses real-time test data and mechanical test life detection data as small sample data sources. Through generative data enhancement and spatiotemporal alignment and fusion of multi-source data, it deeply integrates the actual working condition operation data of aerospace equipment with small sample test data at the spatiotemporal and semantic levels to generate an expanded dataset. S4. Construct a neural network based on temporal reinforcement learning, train the life cycle dynamic prediction model of each component, and quickly call the matching life cycle dynamic prediction model by name through an intelligent retrieval architecture. Continuously ingest real-time test data or online operation data to realize dynamic tracking of the full life cycle status of components and real-time update of failure time.

2. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 1, characterized in that, In S1, the multi-dimensional tree-structured hierarchical retrieval mechanism is divided into three levels: system, subsystem, and component. The retrieval matching degree is calculated through a hierarchical retrieval matching degree scoring formula. The quantitative mapping association mechanism is verified by combining historical operation data and fault data with small-sample physical experiments to determine the failure probability and failure mode characteristic parameters of components under different motion modes and operating temperature combinations.

3. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 1, characterized in that, In S2, the number of network layers, kernel size, and node feature dimension of the tree-structured convolutional neural network are determined through cross-validation and model performance evaluation optimization, and the root mean square error is used as the model performance evaluation index.

4. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 1, characterized in that, In S3, generative data augmentation employs a condition migration mapping rule and a life feature interpolation algorithm to generate a new augmented dataset by performing linear interpolation in the damage domain between two known samples.

5. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 1, characterized in that, In S4, the reward function of the reinforcement learning agent is designed based on the improvement of prediction accuracy and the closeness of the prediction result to the actual failure time, including a prediction accuracy improvement term and a failure time closeness term.

6. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 1, characterized in that, S4 also includes continuously updating and optimizing the life cycle dynamic prediction model, using newly acquired experimental data and operational data to periodically or in real time adjust the model parameters of the life cycle dynamic prediction model, and the model parameter update adopts the gradient descent method.

7. The numerical-analog fusion method for lifecycle prediction of key components of aerospace equipment according to claim 4, characterized in that, The data relationship generator maps the original data to the physical feature space through an encoder for linear interpolation, and then maps the physical features back to the data space through a decoder. The lifetime interpolation is performed in the damage domain based on the linear cumulative damage theory.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the numerical-analog fusion method for life cycle prediction of key components of aerospace equipment as described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the numerical-analog fusion method for life cycle prediction of key components of aerospace equipment as described in any one of claims 1-7.

Images