A Predictive Modeling Method for Health Management Oriented to Airborne Deployment

Abstract

An embodiment of the present invention discloses a predictive modeling method for health management for airborne deployment, including: Step 1, performing data processing and model training on airborne data and ground data on a ground platform to obtain multiple training models; Step 2, evaluating, selecting, and making decisions on the training models obtained in Step 1 in a principle prototype to obtain a training model that meets the task requirements; Step 3, deploying the training model that meets the task requirements obtained in Step 2 on the airborne platform; through the above Steps 1 to 3, an all-cycle PHM modeling strategy for the airborne platform is obtained; further including: constructing an airborne PHM operator library, and sorting out the PHM operator library according to each stage in Steps 1 to 3. The technical solution provided by the embodiment of the present invention solves the problem of inconsistent predictive maintenance specifications for complete fault prediction and health management (referred to as PHM) for airborne and ground in aviation.

Description

Technical Field

[0001] The present invention relates to, but is not limited to, the field of aviation system reliability, and involves ground / airborne health management, especially a predictive modeling method for health management for airborne deployment. Background Art

[0002] In an aviation health management and fault diagnosis system, the computing resources and constraints of airborne and ground platforms are different, and different stages of Predictive and Health Management (PHM) processing need to run in different hardware environments; specifically, the ground platform has unrestricted computing performance, storage space, and a rich data analysis toolkit, etc., while the airborne environment has limited computing resources, low power consumption requirements, and small storage space.

[0003] Currently, there is no complete set of PHM modeling methods and corresponding operator libraries suitable for airborne deployment in aviation. Therefore, for PHM of an aviation system, a complete set of modeling strategies needs to be established to achieve standardized data collection and analysis, give full play to the advantages of the ground platform for data processing and model training, ensure airborne running time and accuracy during model evaluation and model decision-making, and finally deploy the selected optimal model on the airborne side. Summary of the Invention

[0004] The object of the present invention: In view of the PHM deployment requirements for airborne, considering the differences between the ground and airborne environments, the present invention proposes a predictive modeling method for health management for airborne deployment, which is a full-cycle PHM modeling strategy that performs data processing and model training in the ground stage, conducts evaluation, selection, and decision-making in a prototype, and finally deploys it on the airborne side.

[0005] The technical solution of the present invention: An embodiment of the present invention proposes a predictive modeling method for health management for airborne deployment, including:

[0006] Step 1, perform data processing and model training on airborne data and ground data on a ground platform to obtain multiple training models;

[0007] Step 2, conduct evaluation, selection, and decision-making on the training models obtained in Step 1 in a prototype to obtain a training model that meets the task requirements;

[0008] Step 3, deploy the training model that meets the task requirements obtained in Step 2 on the airborne side;

[0009] Through the above Steps 1 to 3, a full-cycle PHM modeling strategy for the airborne side is obtained.

[0010] Optionally, in the above-mentioned predictive modeling method for health management for airborne deployment, Step 1 includes: a data import stage, a data processing stage, and a model training stage running on a ground platform; specifically including:

[0011] Step 1.1, data import phase, including: importing various types of structured and unstructured data in a high-performance ground platform;

[0012] Step 1.2, data processing phase, including: data exploration and analysis sub-phase, data preprocessing sub-phase, and feature engineering sub-phase;

[0013] Step 1.3, model training phase, including: training multiple algorithm models for specific PHM tasks through model training based on physical or empirical models and data-driven model training.

[0014] Optionally, in the health management predictive modeling method for airborne deployment as described above, in step 1.2,

[0015] The data exploration and analysis sub-phase includes: visually exploring and simply mining the imported data to obtain the number of samples, the number of features, data distribution characteristics, feature trends, and correlations;

[0016] The data preprocessing sub-phase is used to improve data quality, including: data cleaning, data denoising, and data standardization;

[0017] The feature engineering sub-phase includes: feature mining, feature selection, and feature extraction of data to extract the most useful features for specific tasks (such as fault diagnosis and remaining life prediction).

[0018] Optionally, in the health management predictive modeling method for airborne deployment as described above, the PHM tasks in step 1.3 include: condition monitoring, fault diagnosis, fault prediction, and remaining life prediction;

[0019] For a specific PHM task type, the way to analyze and deconstruct the problem is as follows:

[0020] Method 1: According to the understanding of the analysis object, when there is an easy-to-solve physical or empirical model for the object, a corresponding physical model is established for the specific task for model training and solution;

[0021] Method 2: When the object structure is complex and the failure or degradation mechanism is difficult to obtain, a data-driven method is adopted, and corresponding machine learning, statistical analysis, etc. methods are used to select algorithms and perform model training;

[0022] Among them, in the model training phase of step 1.3, multiple algorithms are adopted to train multiple models to obtain multiple trained algorithm models.

[0023] Optionally, in the health management predictive modeling method for airborne deployment as described above, step 2 includes: the model evaluation phase and the model decision phase running on a prototype with a principle similar to the airborne software and hardware environment; specifically including:

[0024] Step 2.1, the model evaluation phase, evaluate multiple trained algorithm models in the prototype. According to the algorithm model type and task requirements, select multiple appropriate model evaluation metrics, run the multiple algorithm models trained on the ground in the prototype, and obtain a model evaluation table;

[0025] Step 2.2, the model decision phase, formulate model decision rules according to the specified task requirements, and combine with the obtained model evaluation table to make a final decision on the model and select the "optimal model" for this specific task.

[0026] Optionally, in the health management predictive modeling method for airborne deployment as described above, step 3 includes:

[0027] In the model deployment phase, encapsulate the selected "optimal model" using a language supported by airborne hardware, and deploy the encapsulated "optimal model" to the airborne side.

[0028] Optionally, in the health management predictive modeling method for airborne deployment as described above, it further includes:

[0029] Construct an airborne PHM operator library, including: sorting out the PHM operator library according to each phase in steps 1 to 3;

[0030] Among them, the airborne PHM operator library is classified according to the data processing flow and includes: a process module and a supporting module;

[0031] The airborne PHM operator library is classified according to the operator function and includes: general data processing operators, PHM task-specific operators, and integrated modular aircraft component-level / system-level / aircraft-level PHM operators.

[0032] Optionally, in the health management predictive modeling method for airborne deployment as described above,

[0033] The general data processing operators include: a data import operator unit, a data basic operation operator unit, a data preprocessing operator unit, a data exploration and analysis operator unit, a feature engineering operator unit, a machine learning operator unit, and a hyperparameter optimization operator unit;

[0034] The PHM task-specific operators include: PHM task-specific operators such as an expert system operator unit, an anomaly monitoring operator unit, a fault diagnosis operator unit, a remaining useful life prediction operator unit, etc.; among them, the expert system operator unit contains expert knowledge, experience-based models, physics-of-failure mechanism-based models, etc.

[0035] The integrated and modular aircraft component-level / system-level / aircraft-level PHM operators are highly integrated state monitoring, fault diagnosis, and remaining useful life prediction operators for specific aircraft components, specific systems, or the entire aircraft; the aircraft component-level / system-level / aircraft-level PHM operators are modular operators of components or systems built by selecting specific data processing general operators and PHM-specific operators in the corresponding functional operators according to the full-process airborne PHM algorithm processing strategy.

[0036] Optionally, in the health management predictive modeling method for airborne deployment as described above,

[0037] The process module includes: a data import operator unit, a data exploration and analysis operator unit, a data preprocessing operator unit, a feature engineering operator unit, a model training operator unit, a model evaluation operator unit, a model decision-making operator unit, and a model deployment operator unit;

[0038] The supporting module includes: a data basic operation operator unit, a machine learning operator unit, an expert system operator unit, and a hyperparameter optimization operator unit;

[0039] Among them, the data basic operation operator unit is the basic operator that supports the entire process module; the machine learning operator unit, the expert system operator unit, and the hyperparameter optimization operator unit are the supporting operators that support the model training operator unit.

[0040] The beneficial effects of the present invention: The health management predictive modeling method for airborne deployment provided by the embodiments of the present invention, on the one hand, realizes the analysis of massive heterogeneous data, problem modeling, and model decision-making by leveraging the high-performance computer on the ground platform and the principle prototype with the same principle as the airborne environment, and finally selects the "optimal model" for specific tasks for model deployment to achieve the full-cycle PHM modeling strategy for airborne deployment; on the other hand, by concretizing the full-cycle PHM modeling strategy for airborne deployment into specific operators, a PHM operator library is proposed to fully support the PHM modeling process for airborne. Description of the Drawings

[0041] The drawings are used to provide a further understanding of the technical solutions of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present application to explain the technical solutions of the present invention, and do not constitute a limitation to the technical solutions of the present invention.

[0042] Figure 1Schematic diagram of the principle of a health management predictive modeling method for airborne deployment provided by an embodiment of the present invention;

[0043] Figure 2 Schematic diagram of the principle of the model evaluation stage and the model decision-making stage in an embodiment of the present invention;

[0044] Figure 3 Schematic diagram of the airborne PHM operator library in an embodiment of the present invention. Detailed implementation manners

[0045] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that, without conflict, the embodiments and features in the embodiments of this application can be combined with each other arbitrarily.

[0046] The present invention provides several specific embodiments that can be combined with each other. For the same or similar concepts or processes, they may not be repeated in some embodiments.

[0047] The objective of the embodiment of the present invention is to establish a PHM modeling method for airborne deployment and build a PHM operator library, which can fully support the analysis and modeling process of aviation ground / on-board PHM data, quickly and conveniently realize fault diagnosis and life analysis of aircraft component-level / system-level / whole-aircraft-level members, and evaluate the health status of the aircraft.

[0048] Figure 1 Schematic diagram of the principle of a health management predictive modeling method for airborne deployment provided by an embodiment of the present invention. The health management predictive modeling method for airborne deployment provided by the embodiment of the present invention adopts the following technical solutions to implement; specifically, it includes two parts, namely, forming an airborne PHM algorithm processing strategy and a PHM operator library.

[0049] First part: Forming an airborne PHM algorithm processing strategy:

[0050] The implementation manners of this part include:

[0051] Step 1: Process the airborne data and ground data on the ground platform and perform model training to obtain multiple training models;

[0052] Step 2: Evaluate, select, and make decisions on the training models obtained in Step 1 in the principle prototype to obtain the training models that meet the task requirements.

[0053] Step 3: Deploy the training models that meet the task requirements obtained in Step 2 on the airborne platform.

[0054] The above-mentioned first part specifically includes the following steps: a data import stage, a data processing stage, and a model training stage running on a ground platform; a model evaluation stage and a model decision-making stage running on a prototype, and a model deployment stage running on an airborne device; as Figure 1 shown in the operation of each of the above stages.

[0055] For each of the above stages in Step 1, namely the data import stage, the data processing stage, and the model training stage, they all run on a ground high-performance computing platform (which can be abbreviated as: ground platform in this article). In this Step 1, for massive heterogeneous data such as airborne operation historical data, test data generated by a test bench, simulation data generated by a simulation model platform, maintenance and repair data, etc., by using the rich data analysis toolkits on the ground platform and relying on the high-performance computers on the ground platform, in-depth mining and analysis of the data are realized; model training for specific PHM tasks (such as fault diagnosis, life prediction, etc.) is carried out to provide a trained model for the specific PHM tasks on the aircraft and lay a foundation for realizing the PHM tasks on the aircraft. The specific implementation methods of each stage in Step 1 are described below respectively.

[0056] Step 1.1: Data import stage.

[0057] This data import stage runs on a ground high-performance platform and can include the import of various types of structured data (such as data files in txt, csv, etc. formats) and unstructured data (such as tables, images, etc. files).

[0058] Step 1.2: Data processing stage.

[0059] This data processing stage runs on a ground high-performance platform and includes: a data exploration and analysis sub-stage, a data preprocessing sub-stage, and a feature engineering sub-stage.

[0060] The data exploration and analysis sub-stage can be implemented through many mature algorithms. For example, the correlation analysis of data can be realized by calculating the Pearson correlation coefficient, Kendall correlation coefficient, etc. between pairwise features of the data. At the same time, by drawing scatter pair plots, correlation heat zone plots, etc., the linear correlation between pairwise features of the data can also be intuitively displayed; the trend analysis can be intuitively displayed by drawing a line chart of each feature. The preliminary exploration of the data distribution can be realized by calculating various statistical features (mean, variance, mode, kurtosis, skewness, root mean square, etc.) of the data, or by drawing a box plot and frequency histogram of the data to visually display the data distribution.

[0061] The data preprocessing sub-phase can improve data quality, including: data cleaning, data denoising, data normalization, etc. Among them, data cleaning is mainly carried out from the following three aspects: consistency check, handling of invalid values and missing values, and handling of duplicate values. For invalid values and missing values, three handling methods are adopted according to different situations: when the number of missing values is small and the importance of the attribute is low, if the attribute is numerical data, simple filling is performed using the mean, median, etc. according to the data distribution; if the missing rate is high and the importance of the attribute is low, the attribute can be directly deleted; if the missing rate is high and the importance of the attribute is high, interpolation methods and modeling methods are used. For the judgment of duplicate values, the records in the dataset are first sorted according to certain rules, and then whether the adjacent records are similar is compared to detect whether the records are duplicate. The common methods of data normalization are min-max normalization (normalizing the data of different features into the interval [0, 1]) and z-score normalization (normalizing the data of different features into data that follows the standard normal distribution).

[0062] The feature engineering sub-phase can adopt many algorithms, which are specifically divided into methods based on experience and physical models and data-driven methods. Among them, the methods based on experience and physical models are for specific objects, using existing empirical knowledge or establishing corresponding physical models for them, so as to select or construct the most useful features for the target task; the data-driven methods only start from the perspective of data, and select or construct the most meaningful features for the target by analyzing the trend of different feature data, the correlation with the target label, etc. Specific feature selection methods include: variance-based methods, relief, fisher-score, sparse learning-based methods, etc., and feature extraction methods include principal component analysis PCA, linear discriminant analysis LDA, independent component analysis ICA, locally linear embedding LLE, isometric mapping Isomap, etc.

[0063] Step 1.3: Model training phase.

[0064] This model training phase runs on a high-performance ground platform and includes: training multiple algorithm models for specific PHM tasks through model training based on physical or empirical models and data-driven model training.

[0065] The PHM tasks in this step include, for example: condition monitoring, fault diagnosis (identifying whether there is a fault, identifying the fault type, identifying the fault location), fault prediction (predicting when a fault will occur), remaining useful life prediction (predicting the degradation trend of components, predicting when components will fail), etc.

[0066] For a specific PHM task type, the way to analyze and decompose its problems can be as follows: First, according to the understanding of the analysis object, if there is an easy-to-solve physical or empirical model for the object, then for the specific task, establish a corresponding physical model for the object to perform model training and solution; if the object structure is complex and the failure or degradation mechanism is difficult to obtain, then adopt a data-driven method, and use corresponding machine learning, statistical analysis and other methods to select algorithms and perform model training.

[0067] It should be noted that in the model training stage of step 1.3 of the embodiment of the present invention, multiple algorithms can be adopted to train multiple models to obtain multiple trained models.

[0068] For each of the above stages in step 2, that is, the model evaluation stage and the model decision-making stage, they run in a prototype similar to the airborne software and hardware environment. After completing the model training stage of step 1, several trained models are obtained, and these models need to be decided to obtain the "optimal model" finally deployed on the aircraft. Since the concept of the "optimal model" is closely related to the actual application and aims at airborne deployment, when evaluating the model, it is necessary to specify the hardware environment for the model to run and construct a prototype similar to the airborne software / hardware environment to simulate the airborne operating environment. The following will separately describe the specific implementation manners of each stage in step 2 above.

[0069] Step 2.1: Model evaluation stage.

[0070] This model evaluation runs on the prototype. For different task types, multiple model evaluation indicators reflecting the model prediction accuracy can be used for model evaluation. The specific implementation manner is: for the regression problem model, the commonly used model evaluation indicators include root mean square error RMSE, mean absolute error MAE, goodness of fit R2, etc.; for the classification problem model, the commonly used model evaluation indicators include accuracy acc, precision, recall, F1 score, ROC curve, etc. In addition to the indicators reflecting the model prediction accuracy, there are also indicators reflecting the model operation efficiency: model prediction time, which is very necessary for strongly real-time tasks. Evaluate multiple trained models in the prototype and obtain a model evaluation table. As Figure 2 shown, it is a schematic diagram of the principle of the model evaluation stage and the model decision-making stage in the embodiment of the present invention.

[0071] Step 2.1: Model decision-making stage.

[0072] This model decision-making stage runs on the prototype. First, formulate model decision rules according to specific task requirements, and combine the obtained model evaluation table to make a final decision on the model and select the "optimal model" for this specific task, that is, the training model that meets the task requirements.

[0073] The processes of the model evaluation phase and the model decision-making phase are as Figure 2 shown.

[0074] Step 3 of the embodiment of the present invention is specifically the model deployment phase.

[0075] In this model deployment phase, the "optimal model" selected in step 2 is encapsulated in a language supported by airborne hardware, and the encapsulated "optimal model" is deployed on the airborne side.

[0076] Through the above steps 1 to 3, the full-cycle PHM modeling strategy for airborne in the first part is obtained.

[0077] Second part: Construct an airborne PHM operator library;

[0078] According to each stage and sub-stage of the airborne PHM algorithm processing strategy in the above first part, the corresponding content is sorted out to obtain a PHM operator library, as Figure 3 shown, which is a schematic diagram of the airborne PHM operator library in the embodiment of the present invention.

[0079] The airborne PHM operator library in the embodiment of the present invention can be classified according to the data processing flow and can include: a process module and a support module.

[0080] The airborne PHM operator library in the embodiment of the present invention can be classified according to the operator function and can include: general data processing operators, PHM task-specific operators, and integrated modular aircraft component-level / system-level / aircraft-level PHM operators.

[0081] As Figure 3 shown, the airborne PHM operator library contains four support modules: a basic data operation operator unit, a machine learning operator unit, an expert system operator unit, and a hyperparameter optimization operator unit.

[0082] The basic data operation operator unit is the basic operator that supports the entire process. It includes logical operations, element-wise operations, selection / replacement of rows / columns, etc.

[0083] The machine learning operator unit, the expert system operator unit, and the hyperparameter optimization operator unit are support operators for the support model training operator unit. The machine learning operator unit provides diverse and flexible classification, regression, and artificial neural network operators for subsequent fault diagnosis and life prediction. The expert system operator unit contains empirical solutions for specific components. Hyperparameter optimization is a parameter tuning method for hyperparameters in model training. Common model hyperparameters include the number of network layers and the number of nodes in a neural network; the maximum tree depth and regularization coefficient in a decision tree model, etc.; hyperparameter optimization methods include particle swarm optimization, genetic algorithms, simulated annealing algorithms, etc.

[0084] According to the processing method and strategy of the airborne PHM algorithm, specific general data - processing operators and PHM - specific operators are selected from the operator library in the order of the process modules in the operator library, and a full - process state monitoring, fault diagnosis, and life prediction process for a specific component or system is built as the integrated and modular operator for this component or system.

[0085] The technical solution provided by the embodiment of the present invention aims at aviation predictive maintenance, solves the problem of inconsistent predictive maintenance specifications for the complete fault prediction and health management (abbreviated as PHM) of airborne and ground in aviation, and includes a full - cycle PHM algorithm strategy for the ground platform, principle prototype, and airborne deployment. A PHM operator library that supports data analysis and modeling is formed, including a data import stage, a data processing stage, a model training stage, a model evaluation stage, a model decision - making stage, and a model deployment stage, and is implemented as a specific PHM operator library. This PHM operator library fully supports the on - board PHM modeling and data - analysis process, while taking into account flexibility and integration modularity. Through theoretical analysis and experiments, the operator - library framework developed under this invention patent can meet the aviation PHM requirements, and based on the operator - library framework, PHM processing for aircraft component - level / system - level / aircraft - level members facing airborne deployment can be realized.

[0086] The predictive modeling method for health management facing airborne deployment provided by the embodiment of the present invention, on the one hand, realizes the analysis of massive heterogeneous data, problem modeling, and model decision - making by leveraging the high - performance computer on the ground platform and the principle prototype with the same airborne environment, and finally selects the "optimal model" for specific tasks for model deployment to achieve the full - cycle PHM modeling strategy facing airborne deployment; on the other hand, by concretizing the full - cycle PHM modeling strategy facing airborne deployment into specific operators, a PHM operator library is proposed to fully support the PHM modeling process facing airborne.

[0087] The following uses a specific embodiment to schematically illustrate the implementation manner of the predictive modeling method for health management facing airborne deployment provided by the embodiment of the present invention.

[0088] For specific components of an aircraft, such as engines, lubricating oil modules, rotating components, etc., by using the integrated and modular aircraft component - level operators, the health status of aircraft components can be quickly and conveniently evaluated.

[0089] For certain types of data and faults that do not often occur on the aircraft, according to the airborne PHM algorithm processing strategy proposed by the embodiment of the present invention, it is supported to independently use the rich general data - processing operators and PHM - task - specific operators in the operator library, and according to the order of the process modules in the operator library, explore and analyze the data, and flexibly build a full process for the diagnosis or prediction of this data. The specific steps are as follows:

[0090] First, data import, data processing, and model training are carried out on the ground high-performance platform. Taking the Windows platform using the Python language as an example. In the ground platform, the data analysis and model training both call the operators in the PHM operator library. The numerical data collected from the test bench is exported in the form of files such as txt, excel, csv, etc., and the collected file data is imported into the Python environment. Using the data import operator, it is converted into data in the dataframe format.

[0091] Data cleaning is performed on the original data. Using the data cleaning operators in the data preprocessing part of the PHM operator library, check for missing data, duplicate data, and check for data inconsistency problems, and process these data problems according to the cleaning algorithm to improve data quality. Using the operators for data exploratory analysis in the PHM operator library, conduct a preliminary exploration of the data after quality improvement to obtain data characteristics such as data scale, distribution characteristics, and correlation, providing ideas and guidance for the subsequent steps. The operators in the data exploratory analysis part include: correlation heat map, line chart, box plot, data statistical characteristics, etc.

[0092] After the data exploratory analysis step, the subsequent analysis and modeling ideas are initially clarified. First, task-oriented, determine what kind of model to establish. According to the actual situation of the task, it can be divided into regression problems, classification problems, clustering problems, prediction problems, etc. Each type of problem requires a different model to be established. Regression and classification belong to the category of supervised learning, which means that the features to be sought (sample labels) in the training set data are known. Among them, the features to be sought in the regression problem are continuous values, while the features to be sought in the classification problem are discrete values. Clustering belongs to the category of unsupervised learning, that is, the features to be sought (sample labels) in the training set data are unknown. Clustering models are often used for problems such as anomaly detection. The prediction problem refers to given the trend of feature values for a period of time, predicting the feature value at a future moment, or predicting the failure moment of the device based on the feature values of certain features within a period of time.

[0093] Combined with the conclusions drawn from the data exploratory analysis, preprocess the data and perform feature engineering operations. If the model working conditions are unknown in the actual situation, use the data merging operator in the operator library in the preprocessing step to merge all input working condition data files, and the working conditions are not used as model inputs. If the selected algorithm model is sensitive to the range differences of different input features, use the data standardization operator in the preprocessing operator library to standardize the data features into data that follows the standard normal distribution. In the feature engineering step, it is necessary to select or construct the input features of the model from the original data. If it is found in the exploratory analysis step that the variance of a certain feature is very small and there is almost no change trend, then delete this feature.

[0094] In the model training stage, a suitable algorithm is selected for model training. The operator library in this part includes algorithms such as linear regression, decision tree, support vector machine, gradient boosting tree, random forest, artificial neural network, etc., which support the construction of various regression, classification, and prediction models. According to the ideas provided in the data exploration and analysis, several suitable algorithms are selected to build models; all samples are divided into a training set and a test set in a suitable proportion, and the training set samples are sent into the several built models for model training to obtain several specific trained models respectively.

[0095] So far, the data analysis and model training stage using a high-performance server on the ground platform is completed. Due to limited airborne computing resources, it is considered to only perform the prediction part of the model on the aircraft: the models trained on the ground are lightly encapsulated using the C language and deployed in the embedded platform on the aircraft for real-time diagnosis and prediction on the aircraft. To simulate the airborne embedded environment, a suitable embedded development board or a suitable development board is built by oneself as a prototype. First, the several trained models are lightly encapsulated using the C language, and the several obtained trained models are burned into the prototype respectively. The test set data is input, and the model prediction is carried out in the board. This step is used as the environmental simulation of the on-aircraft prediction to implement the verification link of the algorithm.

[0096] The results predicted by the several models in the board are evaluated using various evaluation metrics. Commonly used regression model accuracy evaluation metrics include: RMSE, MAE, R2; commonly used classification model accuracy evaluation metrics include: precision, recall, accuracy, F1 value, etc. In addition to the model accuracy evaluation metrics, metrics such as model training time and model testing time are also required to evaluate the prediction real-time performance of the model. The test set data is respectively input into the trained models in the prototype, and the prediction is carried out in the board, and the values of several model evaluation metrics are calculated to obtain Figure 2 the model evaluation table shown. This evaluation table is used as a reference for the selection of the best model and airborne deployment. According to the requirements of the actual situation, the optimal model that best meets the actual requirements is selected, and then this model can be deployed on the aircraft.

[0097] Although the disclosed embodiments of the present invention are as above, the described content is only an embodiment adopted for the convenience of understanding the present invention and is not used to limit the present invention. Any person skilled in the art within the scope of the present invention can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed by the present invention. However, the patent protection scope of the present invention shall still be subject to the scope defined by the appended claims.

Claims

1. A predictive modeling method for health management for airborne deployment, characterized in that, Including: Step 1: Perform data processing and model training on airborne data and ground data on the ground platform to obtain multiple training models; Step 2: Evaluate, select, and make decisions on the training models obtained in Step 1 in the prototype to obtain the training models that meet the mission requirements; Step 3: Deploy the training models that meet the mission requirements obtained in Step 2 on the airborne platform; Through the above Steps 1 to 3, an airborne full-cycle PHM modeling strategy is obtained; Among them, Step 2 includes: a model evaluation stage and a model decision-making stage running on a prototype similar to the airborne software and hardware environment; specifically including: Step 2.1, Model evaluation stage: Evaluate multiple trained algorithm models in the prototype. According to the algorithm model type and mission requirements, select a variety of appropriate model evaluation indicators, and run multiple algorithm models trained on the ground in the prototype to obtain a model evaluation table; Step 2.2, Model decision-making stage: Formulate model decision rules according to the specified mission requirements, and combine the obtained model evaluation table to make a final decision on the model and select the optimal model for the specific mission.

2. The predictive modeling method for health management for airborne deployment according to claim 1, characterized in that, Step 1 includes: a data import stage, a data processing stage, and a model training stage running on the ground platform; specifically including: Step 1.1, Data import stage: Include importing various types of structured data and unstructured data in the ground high-performance platform; Step 1.2, Data processing stage: Include a data exploration and analysis sub-stage, a data preprocessing sub-stage, and a feature engineering sub-stage; Step 1.3, Model training stage: Include training multiple algorithm models for specific PHM tasks through model training based on physical or empirical models and data-driven model training.

3. The predictive modeling method for health management for airborne deployment according to claim 2, characterized in that, In Step 1.2, The data exploration and analysis sub-stage includes: visually exploring and simply mining the imported data to obtain the sample quantity, feature quantity, data distribution characteristics, feature trends, and correlations; The data preprocessing sub-stage is used to improve the data quality, including: data cleaning, data denoising, and data standardization; The feature engineering sub-stage includes: feature mining, feature selection, and feature extraction of the data to extract the features most useful for the specific task.

4. The predictive modeling method for health management for airborne deployment according to claim 3, characterized in that, The PHM tasks in Step 1.3 include: condition monitoring, fault diagnosis, fault prediction, and remaining useful life prediction; For a specific PHM task type, the method of problem analysis and decomposition is: Method 1: According to the understanding of the analysis object, when there is an easy-to-solve physical or empirical model for the object, then establish a corresponding physical model for the specific task for model training and solution; Method 2: When the object structure is complex and the failure or degradation mechanism is difficult to obtain, then use a data-driven method, and use corresponding machine learning and statistical analysis methods to select algorithms and perform model training; Among them, in the model training stage of Step 1.3, multiple algorithms are adopted to train multiple models to obtain multiple trained algorithm models.

5. The predictive modeling method for health management for airborne deployment according to claim 4, characterized in that, Step 3 includes: In the model deployment stage, the selected optimal model is encapsulated in a language supported by airborne hardware, and the encapsulated optimal model is deployed on the airborne side.

6. The predictive modeling method for health management for airborne deployment according to any one of claims 1 to 5, characterized in that, It also includes: Construct an airborne PHM operator library, including: sorting out the PHM operator library according to each stage in Steps 1 to 3; Among them, the airborne PHM operator library is classified according to the data processing flow and includes: a process module and a supporting module; The airborne PHM operator library is classified according to the operator function and includes: general data processing operators, PHM task-specific operators, and integrated modular aircraft component-level / system-level / aircraft-level PHM operators.

7. The predictive modeling method for health management for airborne deployment according to claim 6, characterized in that, The general data processing operators include: a data import operator unit, a data basic operation operator unit, a data preprocessing operator unit, a data exploration and analysis operator unit, a feature engineering operator unit, a machine learning operator unit, and a hyperparameter optimization operator unit; The PHM task-specific operators include: an expert system operator unit, an anomaly monitoring operator unit, a fault diagnosis operator unit, and a remaining useful life prediction operator unit; among them, the expert system operator unit contains expert knowledge, experience-based models, and physics-of-failure mechanism-based models; The integrated modular aircraft component-level / system-level / aircraft-level PHM operators are high-integration state monitoring, fault diagnosis, and remaining useful life prediction operators for specific aircraft components, specific systems, or the entire aircraft; the aircraft component-level / system-level / aircraft-level PHM operators are modular operators of components or systems built by selecting specific general data processing operators and PHM-specific operators in the corresponding functional operators according to the full-process airborne PHM algorithm processing strategy.

8. The predictive modeling method for health management for airborne deployment according to claim 6, wherein, The process module includes: a data import operator unit, a data exploration and analysis operator unit, a data preprocessing operator unit, a feature engineering operator unit, a model training operator unit, a model evaluation operator unit, a model decision-making operator unit, and a model deployment operator unit; The supporting module includes: a data basic operation operator unit, a machine learning operator unit, an expert system operator unit, and a hyperparameter optimization operator unit; Among them, the data basic operation operator unit is the basic operator that supports the entire process module; the machine learning operator unit, the expert system operator unit, and the hyperparameter optimization operator unit are the supporting operators that support the model training operator unit.

Images