Aero-engine life prediction method based on multi-source time sequence feature fusion

Abstract

The invention discloses an aero-engine life prediction method based on multi-source time sequence feature fusion. The method comprises the steps that an initial training data set is acquired, each piece of initial training data is full-life-cycle data of each airspace engine, and the full-life-cycle data of each airspace engine is 14 sensor parameters related to the remaining service life of the airspace engine at multiple moments; and constructing an F2Net network model, the F2Net network model comprising a GRU, a multi-head self-attention mechanism, a feature fusion module, a first full connection layer, a linear projection layer, an ReLU activation function, a first Dropout layer, a second full connection layer, a feature fusion connection layer, a second Dropout layer, an RUL linear regression layer and a prediction result. The technical problem that in the prior art, when the residual life of the aero-engine is predicted, enough valuable information cannot be extracted from mass data, and the prediction accuracy of the residual life of the aero-engine is low is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of aerospace big data intelligent analysis, and in particular, to a method for predicting the life of an aeroengine based on the fusion of multi-source time series features. Background Art

[0002] The operating state of an aeroengine is crucial to the flight safety of an aircraft. The remaining useful life prediction technology is an important task in the fault prediction and health management of an aeroengine. It can predict potential faults under certain conditions, so as to dynamically adjust the maintenance plan and extend the service life of the engine. How to extract sufficient valuable information from massive data and make full use of it is an important task for the final prediction of the remaining useful life of the system.

[0003] Generally speaking, although most existing methods perform well in some cases, the generality of the model is very poor and the modeling process is extremely complex. For example, Li et al. in "Li C J, Lee H. Gear Fatigue Crack Prognosis using Embedded Model, Gear Dynamic Model and Fracture Mechanics[J]. Mechanical Systems and Signal Processing ; 2005, 19(4): 836-846." proposed a method for predicting the remaining useful life of a gear with fatigue cracks by combining three models of rigid body, dynamics and fracture mechanics; and Ghodrati et al. in "Ghodrati B, Kumar U, Ahmadzadeh F. Remaining Useful Life Estimation of Mining Equipment: A Case Study[C]. New Delhi: International Symposium on Mine Planning and Equipment Selection , 2012." designed a scale-adaptive crowd counting network to estimate the remaining useful life of mining equipment through reliability analysis and provide maintenance suggestions accordingly. Although these models adopt a method combining statistics and data-driven, their generality is poor, and it is difficult to apply them in real time in large-scale industrial equipment. Therefore, they are not suitable for the task of predicting the remaining useful life of an aeroengine. The generality of a model means that the model needs to maintain good performance in a variety of application environments without significant adjustment of the model structure, parameters or methods. However, the limitations of the above methods (over-reliance on the physical characteristics of specific equipment and high-quality historical data, resulting in insufficient generality and complex calculations) hinder the application of this process in the task of predicting the remaining useful life of an aeroengine.

[0004] Recently, researchers have focused on using means such as deep learning to identify the performance degradation trend of the system from a large amount of sensor monitoring data, and use data-driven methods to predict the remaining useful life. For example, Li Hao et al. in "Li Hao, Wang Zhuo-jian, Li Zhe, et al. Prediction of Remaining Useful Life of Aero-Engine Based on Stacked Autoencoder and Deep-AR[J]. Journal of Propulsion Technology , 2022, 43(4):210645.)" proposed a method using a stacked autoencoder to extract features from the multivariate time series of the engine, and used a bidirectional LSTM to construct a DeepAR model for predicting the remaining useful life. Nie Lei et al. in "Nie Lei, Xu Shi-yi, Zhang Lyu-fan, et al. Remaining Useful Life Prediction of Aeroengine Based on Multi-Head Attention[J]. Journal of Propulsion Technology , 2023, 44(8): 2204040.)" established a one-dimensional convolutional neural network model for the multi-dimensional features of engine data and used a multi-head attention mechanism for weighted processing. However, although the above methods have improved the prediction accuracy of the remaining useful life of aero-engines, there are still problems with insufficient data feature extraction, unable to extract sufficient valuable information from these massive data, and there has been no breakthrough improvement in the overall performance and prediction accuracy of the model. Therefore, they are still not suitable for the prediction task of the remaining useful life of aero-engines. Summary of the Invention The embodiments of the present invention provide a method for predicting the life of an aero-engine based on multi-source time series feature fusion, so as to at least solve the technical problem that when predicting the remaining life of an aero-engine in the prior art, insufficient valuable information can be extracted from massive data, resulting in low prediction accuracy of the remaining life of the aero-engine.

[0005] According to one aspect of the embodiments of the present invention, a method for predicting the life of an aeroengine based on multi-source time series feature fusion is provided. The method may include: obtaining an initial training data set, where each initial training data is the full life cycle data of each space engine, and the full life cycle data of each space engine is 14 sensor parameters related to the remaining service life of the space engine at multiple moments; constructing an F2Net network model, where the F2Net network model includes: GRU, a multi-head self-attention mechanism, a feature fusion module, a first fully connected layer, a linear projection layer, a ReLU activation function, a first Dropout layer, a second fully connected layer, a feature fusion connection layer, a second Dropout layer, a RUL linear regression layer, and a prediction result; preprocessing the initial training data set to obtain a target training data set; inputting the target training data set into the F2Net network model to obtain a successfully trained F2Net network; obtaining a test data set, where each test data is the half life cycle data of each space engine; inputting the test data set into the successfully trained F2Net network to obtain the prediction result of each test data, where the prediction result is the remaining service life of each space engine.

[0006] Optionally, the preprocessing the initial training data set to obtain a target training data set includes: normalizing each sensor parameter related to the remaining service life of the space engine at each moment in each initial training data to obtain a target training data set.

[0007] Optionally, the inputting the target training data set into the F2Net network model to obtain a successfully trained F2Net network includes: processing the target training data set using a sliding window to obtain a plurality of target training samples, each target training sample being 14 sensor parameters related to the remaining service life of the space engine at a target number of moments, where the size of the sliding window is 30; using each target training sample to train the F2Net network model in turn to obtain a successfully trained F2Net network.

[0008] Optionally, training the F2Net network model with each target training sample in turn to obtain a successfully trained F2Net network includes: inputting each target training sample into the GRU to obtain the sequence feature of each target training sample; inputting the sequence feature of each target training sample into the multi-head self-attention mechanism to obtain the target attention score; inputting the sequence feature of each target training sample and the target attention score into the feature fusion module to obtain the first fusion feature; inputting the first fusion feature into the first fully connected layer to obtain the first target feature; inputting each target training sample into the linear projection layer, ReLU activation function, first Dropout layer and second fully connected layer in turn to obtain the second target feature; inputting the first target feature and the second target feature into the feature fusion connection layer to obtain the second fusion feature; inputting the second fusion feature into the second Dropout layer and the RUL linear regression layer in turn to obtain the prediction result of each target training sample, and performing iterative loop to obtain a successfully trained F2Net network.

[0009] Optionally, the expression of the process of inputting each target training sample into the GRU to obtain the sequence feature of each target training sample is:

[0010]

[0011]

[0012]

[0013] Among them, are 14 sensor parameters related to the remaining useful life of the aero-engine at time t in a target training sample, is the sequence feature of a target training sample at time t, is the candidate hidden state of a target training sample at time t, is the Sigmoid activation function, is the exclusive NOR operation, 、 、 are weight matrices, is the update gate at the th moment, is the reset gate at the th moment, is the sequence feature of a target training sample at time t - 1, is the dot product operation, is the activation function; concatenating the sequence features of each moment in each target training sample to obtain the sequence feature of each target training sample.

[0014] Optionally, after inputting the test data set into the successfully trained F2Net network to obtain the prediction results of each test data, the method further includes: comparing the prediction results with the true values, and calculating the root mean square error (RMSE) and performance metric Score of the F2Net network model.

[0015] Optionally, the optimizer of the F2Net network model is the Adam optimizer, and the learning rate of the optimizer is 0.001.

[0016] Advantages of the present invention: The present invention proposes an aero-engine life prediction method based on multi-source time series feature fusion, and proposes a feature fusion network, which can fully extract and utilize the effective features of data from extremely high-dimensional and ultra-long-time massive sensor data to achieve the goal of accurately predicting the remaining useful life of aero-engines. Due to the adoption of a brand-new network architecture and sequence feature learning for the network, it can effectively capture short-term and long-term dependencies in the sequence, combine the features learned by the deep network with the original features obtained by linear mapping, and finally overcome the limitations of single feature extraction methods, providing a more comprehensive and integrated feature representation for the model; the method of the present invention can significantly improve the prediction accuracy in the prediction of the remaining useful life of aero-engines. Compared with the existing prediction methods, this method can effectively enhance the performance of predicting the remaining useful life. Brief Description of the Drawings

[0017] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of this application. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 is a flowchart of the aero-engine life prediction method based on multi-source time series feature fusion according to an embodiment of the present invention; Figure 2 is a framework diagram of the F2Net network model according to an embodiment of the present invention. Detailed Embodiments

[0018] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects and are used to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0020] Embodiment 1 According to an embodiment of the present invention, a method for predicting the life of an aero-engine based on multi-source time series feature fusion is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system including at least one set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0021] Figure 1 is a flowchart of a method for predicting the life of an aero-engine based on multi-source time series feature fusion according to an embodiment of the present invention, as Figure 1 shown, the method may include the following steps: Step S101, obtain an initial training data set, where each initial training data is the full life cycle data of each space engine, and the full life cycle data of each space engine is 14 sensor parameters related to the remaining service life of the space engine at multiple moments.

[0022] In the technical solution provided in step S101 of the present invention above, among the 21 sensor monitoring parameters of the FD001 data set, the sensor parameters irrelevant to the performance degradation parameters are removed, and 14 sensor parameters related to the remaining service life are accurately selected to improve the training efficiency of the network and used as the input of the network.

[0023] Step S102, construct an F2Net network model, where the F2Net network model includes: GRU, multi-head self-attention mechanism, feature fusion module, first fully connected layer, linear projection layer, ReLU activation function, first Dropout layer, second fully connected layer, feature fusion connection layer, second Dropout layer, RUL linear regression layer and prediction result.

[0024] In the technical solution provided in step S102 of the present invention above, Figure 2It is a framework diagram of the F2Net network model according to an embodiment of the present invention. As can be seen from Figure 2 that the F2Net network model includes: Gated Recurrent Units (abbreviated as GRU), multi-head self-attention mechanism, feature fusion module, first fully connected layer, linear projection layer, ReLU activation function, first Dropout layer, second fully connected layer, feature fusion connection layer, second Dropout layer, RUL linear regression layer and prediction results.

[0025] Step S103, preprocess the initial training dataset to obtain a target training dataset.

[0026] In the technical solution provided in step S103 of the present invention above, each initial training data in the initial training dataset is normalized to obtain a target training dataset.

[0027] Step S104, input the target training dataset into the F2Net network model to obtain a successfully trained F2Net network.

[0028] In the technical solution provided in step S104 of the present invention above, the F2Net network model trains the target training dataset to obtain a successfully trained F2Net network.

[0029] Step S105, obtain a test dataset, where each test data is the semi-life cycle data of each space engine.

[0030] In the technical solution provided in step S105 of the present invention above, obtain test set data, and each test data is the remaining service life to be predicted for each space engine.

[0031] Step S106, input the test dataset into the successfully trained F2Net network to obtain the prediction result of each test data, where the prediction result is the remaining service life of each space engine.

[0032] In the technical solution provided in step S106 of the present invention above, the successfully trained F2Net network processes the test dataset to obtain the remaining service life of each space engine in the test dataset.

[0033] The above method of this embodiment will be further introduced below.

[0034] As an optional embodiment, in step S103, the preprocessing of the initial training dataset to obtain a target training dataset includes: normalizing each sensor parameter related to the remaining service life of the space engine at each moment in each initial training data to obtain a target training dataset.

[0035] In this embodiment, the 14 selected sensor parameters are smoothed to reduce the impact of noise on model learning. At the same time, the normalization expression is as follows:

[0036] where is the maximum value of each sensor parameter before normalization, is the minimum value of each sensor parameter before normalization, is each sensor parameter before normalization, is each sensor parameter after normalization. Each sensor parameter is normalized to the range [0, 1] to eliminate the differences in the magnitudes of different sensor parameters and improve the stability and convergence efficiency of model training.

[0037] As an alternative embodiment, in step S104, inputting the target training data set into the F2Net network model to obtain a successfully trained F2Net network includes: processing the target training data set using a sliding window to obtain multiple target training samples, where each target training sample is 14 sensor parameters related to the remaining service life of the aeroengine at a target moment. The size of the sliding window is 30; using each target training sample to train the F2Net network model in turn to obtain a successfully trained F2Net network.

[0038] In this embodiment, the sliding window technique is used to slice the target training data set (i.e., time series data). Each window contains 30 consecutive time steps to capture the temporal patterns of the input data. The stride of the sliding window can be set according to specific experiments to generate multiple training samples while ensuring the temporal correlation between features. The F2Net network model processes the multi-target training samples to obtain a successfully trained F2Net network.

[0039] As an alternative embodiment, the process of training the F2Net network model with each target training sample in turn to obtain a successfully trained F2Net network includes: inputting each target training sample into a GRU to obtain the sequence feature of each target training sample; inputting the sequence feature of each target training sample into a multi-head self-attention mechanism to obtain a target attention score; inputting the sequence feature of each target training sample and the target attention score into a feature fusion module to obtain a first fusion feature; inputting the first fusion feature into a first fully connected layer to obtain a first target feature; inputting each target training sample into a linear projection layer, a ReLU activation function, a first Dropout layer, and a second fully connected layer in turn to obtain a second target feature; inputting the first target feature and the second target feature into a feature fusion connection layer to obtain a second fusion feature; inputting the second fusion feature into a second Dropout layer and a RUL linear regression layer in turn to obtain the prediction result of each target training sample, and performing iterative loops to obtain a successfully trained F2Net network.

[0040] In this embodiment, as Figure 2 shown, input each target training sample into a GRU to obtain the sequence feature of each target training sample; input the sequence feature of each target training sample into a multi-head self-attention mechanism to obtain a target attention score; input the sequence feature of each target training sample and the target attention score into a feature fusion module to obtain a first fusion feature; input the first fusion feature into a first fully connected layer to obtain a first target feature; input each target training sample into a linear projection layer, a ReLU activation function, and a first Dropout layer in turn to obtain the original feature of each target training sample, and input the original feature of each training sample into a second fully connected layer to obtain a second target feature; input the first target feature and the second target feature into a feature fusion connection layer to obtain a second fusion feature; input the second fusion feature into a second Dropout layer and a RUL linear regression layer in turn to obtain the prediction result of each target training sample, and perform iterative loops to obtain a successfully trained F2Net network.

[0041] Input the sequence feature of each target training sample into a multi-head self-attention mechanism (abbreviated as MHSAM). MHSAM calculates the importance weights of each time step in the input sequence to identify the key time points that have a significant impact on RUL prediction. Each head independently calculates the attention weights and extracts features; after concatenating the features of all heads, a linear transformation is performed to obtain the target attention score.

[0042] As an alternative embodiment, the expression of the process of inputting each target training sample into a GRU to obtain the sequence feature of each target training sample is:

[0043]

[0044]

[0045]

[0046] Among them, are 14 sensor parameters related to the remaining useful life of the aerospace engine at time t in a target training sample, is the sequence feature at time t of a target training sample, is the candidate hidden state at time t of a target training sample, is the Sigmoid activation function, is the exclusive NOR operation, , , are weight matrices, is at the time update gate, is at the time reset gate, is the sequence feature at time t-1 of a target training sample, is the dot product operation, is the activation function; the sequence features at each moment in each target training sample are concatenated to obtain the sequence feature of each target training sample.

[0047] In this embodiment, to effectively retain the temporal nature of the data, a gated recurrent unit (GRU) improved from the long short-term memory network (LSTM) structure is used to capture sequence dependencies with a very long time distance. First, each processed target training sample is input into the GRU network for sequence feature learning, which can effectively capture the dependencies in the sequence.

[0048] The preprocessed sliding window data is input into the GRU. The GRU effectively captures the short-term and long-term dependencies of the time series through its gating mechanism, while reducing the number of parameters of the traditional RNN and LSTM networks. The process of the GRU to extract temporal features is as follows: First, the update gate is used to control the retention degree of the hidden state at the previous time step; The reset gate controls the degree to which the hidden state at the previous time step is forgotten under the influence of the current input , and then the candidate hidden state is based on the current input and the hidden state after reset, and finally through the candidate hidden state and the update gate The weighted fusion of obtains the current hidden state ; The GRU dynamically adjusts the information flow between time steps to learn the key patterns of time series data.

[0049] As an alternative embodiment, after inputting the test data set into the successfully trained F2Net network to obtain the prediction result of each test data, the method further includes: comparing the prediction result with the true value, and calculating the mean square error RMSE and the performance metric Score of the F2Net network model.

[0050] In this embodiment, the expressions for comparing the prediction result with the true value and calculating the mean square error RMSE and the performance metric Score of the F2Net network model are:

[0051]

[0052]

[0053]

[0054] Among them, is the sample size, is the true value of the th test sample, represents the difference between the th test sample's predicted value and the true value, represents the final score, represents the score of the

[0055] As an alternative embodiment, the optimizer of the F2Net network model is the Adam optimizer, and the learning rate of the optimizer is 0.001.

[0056] Experimental part: The present invention was run on an operation using NVIDIA GeForce GTX3090, Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz.

[0057] The dataset used in the experiment is C-MAPSS, which was proposed by Frederick et al. in the literature "D. K. Frederick, J. A. DeCastro, and J. S. Litt, ‘User’s Guide for the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS),’ NASA Technical Report , NASA / TM-2007-215026, 2007.”; C-MAPSS contains a total of four subsets, namely FD001, FD002, FD003, and FD004. Among them, the FD001 subset is used in this experiment and contains 200 sample data. The training set includes the full life cycle data of 100 engines, and the test set includes the partial operating cycle data of 100 engines.

[0058] 2. Experimental content First, select the FD001 subset in the C-MAPSS dataset as the experimental data, and use its training set to train the feature fusion network model; adjust the number of hidden nodes in the network according to the given training data; specifically, the input dimension of GRU is 17 and the number of hidden nodes is set to 50; the number of heads of MHSAM is set to 5; the output hidden nodes of both FC layers are set to 10; the learning rate of the Adam optimization algorithm is set to 0.001.

[0059] Then, use the trained model to infer the test set of the FD001 subset, predict the remaining useful life of each engine, and calculate the performance indicators of the model; in the experiment, the performance of each algorithm was measured in terms of RMSE and Score indicators to evaluate the prediction performance of the model; to avoid the contingency of the results, 5 experiments were conducted on the FD001 dataset and the average value was taken.

[0060] To prove the effectiveness of the algorithm, the performance of several mainstream remaining useful life prediction models, such as DLSTM, LSTM, and DCNN, was compared in the experiment. Among them, DLSTM is described in the literature "MA Qi-you, LIU Ke-wei, DU Jian, et al. Prediction of Residual Life of Engine Blades Based on Deep Short Term Memory Network [J]. Journal of Propulsion Technology, 2021, 42(8): 1888-1897.)” is described in detail; the D convolutional neural network was proposed by Li X et al. in the literature "Li X, Ding Q, Sun J Q. Remaining Useful Life Estimation in Prognostics Using Deep Convolution Neural Networks [J]. Reliability Engineering&System Safety , 2018, 172: 1-11." The present invention uses RMSE and Score to measure the prediction performance of the model. The experimental results show that the feature fusion network is superior to other comparison models in both metrics, with significant performance advantages. The comparison results are shown in Table 1: Table 1 Comparison of results of different algorithms

[0061] As can be seen from Table 1, compared with the optimal comparison method, the AG convolutional neural network, on the FD001 dataset, the feature fusion network framework has significant improvements in both RMSE and Score performance metrics. Among them, RMSE is reduced by 10.54% and Score is reduced by 23.45%, which proves the effectiveness of the proposed method in improving the prediction accuracy. At the same time, although the AG convolutional neural network shows strong feature extraction ability under specific conditions, the feature fusion network realizes more comprehensive feature fusion by combining GRU and MHSAM, significantly improving the modeling ability for complex time series data, proving that the present invention has significant advantages in effectiveness and robustness.

[0062] In the embodiments of the present invention, by obtaining an initial training data set, wherein each initial training data is the full life cycle data of each aerospace engine, and the full life cycle data of each aerospace engine is 14 sensor parameters related to the remaining service life of the aerospace engine at multiple moments; constructing an F2Net network model, wherein the F2Net network model includes: GRU, a multi-head self-attention mechanism, a feature fusion module, a first fully connected layer, a linear projection layer, a ReLU activation function, a first Dropout layer, a second fully connected layer, a feature fusion connection layer, a second Dropout layer, an RUL linear regression layer, and a prediction result; preprocessing the initial training data set to obtain a target training data set; inputting the target training data set into the F2Net network model to obtain a successfully trained F2Net network; obtaining a test data set, wherein each test data is the half life cycle data of each aerospace engine; inputting the test data set into the successfully trained F2Net network to obtain the prediction result of each test data, wherein the prediction result is the remaining service life of each aerospace engine. When the prior art predicts the remaining life of an aeroengine, it is unable to extract sufficient valuable information from the massive data, resulting in a low accuracy in predicting the remaining life of the aeroengine. The technical effect of improving the accuracy of predicting the remaining life of the aeroengine is achieved by predicting the remaining life of the aeroengine through the F2Net feature fusion network.

[0063] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0064] In the above embodiments of the present invention, the descriptions of the various embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0065] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of units can be a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0066] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0067] In addition, in each embodiment of the present invention, each functional unit can be integrated into a first processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0068] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. The aircraft engine life prediction method based on multi-source time series feature fusion is characterized by: include: Acquire an initial training data set, wherein each initial training data is the full life cycle data of each aerospace engine, and the full life cycle data of each aerospace engine is 14 sensor parameters related to the remaining service life of the aerospace engine at multiple moments; Construct the F2Net network model, which includes: GRU, multi-head self-attention mechanism, feature fusion module, first fully connected layer, linear projection layer, ReLU activation function, first Dropout layer, second fully connected layer, feature fusion connection layer, second Dropout layer, RUL linear regression layer and prediction results; Preprocess the initial training data set to obtain the target training data set; Input the target training data set into the F2Net network model to obtain a successfully trained F2Net network; Acquire a test data set, wherein each test data is half life cycle data of each aerospace engine; The test data set is input into the successfully trained F2Net network to obtain the prediction results of each test data, where the prediction results are the remaining service life of each aerospace engine.

2. The method according to claim 1, characterized in that The preprocessing of the initial training data set to obtain the target training data set includes: Each sensor parameter related to the remaining service life of the aerospace engine at each moment in each initial training data is normalized to obtain a target training data set.

3. The method according to claim 2, characterized in that The step of inputting the target training data set into the F2Net network model to obtain a successfully trained F2Net network includes: The target training data set is processed using a sliding window to obtain multiple target training samples. Each target training sample is 14 sensor parameters related to the remaining service life of the aerospace engine at a target moment, where the size of the sliding window is 30; The F2Net network model is trained in turn using each target training sample to obtain a successfully trained F2Net network.

4. The method according to claim 3, characterized in that The method of training the F2Net network model in sequence using each target training sample to obtain a successfully trained F2Net network includes: Input each target training sample into GRU to obtain the sequence features of each target training sample; Input the sequence features of each target training sample into the multi-head self-attention mechanism to obtain the target attention score; Input the sequence features and target attention scores of each target training sample into the feature fusion module to obtain the first fusion feature; Input the first fusion feature into the first fully connected layer to obtain the first target feature; Each target training sample is sequentially input into the linear projection layer, the ReLU activation function, the first Dropout layer, and the second fully connected layer to obtain the second target feature; Inputting the first target feature and the second target feature into the feature fusion connection layer to obtain a second fusion feature; The second fusion feature is input into the second Dropout layer and the RUL linear regression layer in sequence to obtain the prediction result of each target training sample. The iterative cycle is repeated to obtain a successfully trained F2Net network.

5. The method according to claim 4, characterized in that Each target training sample is input into the GRU, and the expression of the process of obtaining the sequence features of each target training sample is: in, are the 14 sensor parameters related to the remaining service life of aerospace engines at time t in a target training sample, is the sequence feature of a target training sample at time t, is a candidate hidden state of a target training sample at time t, is the Sigmoid activation function, is the same or operation, , , is the weight matrix, For the The door of renewal at all times, For the The reset door of time, is the sequence feature of a target training sample at time t-1, is the dot product operation, is the activation function; The sequence features of each target training sample at each moment are concatenated to obtain the sequence features of each target training sample.

6. The method according to claim 1, characterized in that After inputting the test data set into the successfully trained F2Net network to obtain the prediction result of each test data, the method further includes: Compare the predicted results with the true values ​​and calculate the mean square error RMSE and performance index Score of the F2Net network model.

7. The method according to claim 6, characterized in that The optimizer of the F2Net network model is the Adam optimizer, and the learning rate of the optimizer is 0.

001.

8. A computer system, characterized in that include: One or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method of claim 1.

9. A computer-readable storage medium, characterized in that Computer executable instructions are stored, and when the instructions are executed, they are used to implement the method of claim 1.

Images