A method and apparatus for assessing the health of a turbine engine system based on an RVE

Abstract

This application relates to the field of turbine engine system health assessment technology, and discloses a method and apparatus for assessing the health status of a turbine engine system based on Remaining Useful Life (RUL). The method includes: acquiring various sensor parameter data throughout the entire life cycle of the turbine engine system, performing basic preprocessing operations on the dataset, constructing each batch of data using the sliding window method and dividing it into training set, validation set and test set, then using the training set data as input to construct a model, constructing a loss function for the remaining useful life prediction task, training the turbine engine system health status assessment model using an improved loss function and Adam optimizer, and finally inputting historical data of the turbine engine system to be assessed into the model, outputting the remaining useful life (RUL) degradation process of the turbine engine system until complete failure as the health status assessment curve, thereby making the assessment of the system health status more accurate.

Description

Technical Field

[0001] This application relates to the field of turbine engine system health assessment technology, and in particular to a method and apparatus for assessing the health status of a turbine engine system based on RVE. Background Technology

[0002] Monitoring the health of mechanical systems is crucial in any field; however, traditional strategies, such as periodic preventative maintenance or corrective maintenance for failures, are increasingly failing to meet the growing demands of industry for efficiency and reliability. Therefore, system health indicators such as remaining service life (RUL) have been established as key elements for maintaining mechanical systems and preventing engineering safety issues.

[0003] In the context of aviation safety, turbine engine systems are susceptible to overall system failure due to hardware degradation. Data is typically collected from various built-in sensors to monitor turbine engine system operation and calculate the system's remaining useful life (RUL). This helps engineers develop preventative maintenance plans and avoid operating turbine engines under hardware deterioration conditions, effectively increasing turbine engine flight time and reducing maintenance costs.

[0004] However, over the years, with the continuous increase in the amount of information that can be collected and the increasing precision requirements for preventive maintenance, existing technologies have problems such as low utilization rate of multi-dimensional sensor parameter features and low accuracy of life prediction indicators. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide a method and apparatus for assessing the health status of a turbine engine system based on RVE.

[0006] Firstly, a method for assessing the health status of a turbine engine system based on Relative Energy Evaluation (RVE) is provided, including:

[0007] Acquire data X collected by sensors throughout the entire lifecycle of a turbine engine;

[0008] Preprocess the data X, construct dataset health indicators as data labels, and divide the inputs and corresponding outputs of each batch into training set, validation set and test set by using the sliding window method;

[0009] Construct a health status assessment model for turbine engine systems based on RVE;

[0010] An improved loss function and Adam optimizer were used to train a turbine engine system health assessment model;

[0011] Historical data of the turbine engine system to be evaluated is input into the trained turbine engine system health status assessment model, and the remaining service life degradation process of the turbine engine system until complete failure is output as the health status assessment curve.

[0012] Furthermore, the data X = [x1, x2, ..., x N ],in, n is the batch size, and d is the input dimension size, which depends on the number of sensors selected. Represents an n-row, d-column real matrix.

[0013] Furthermore, a turbine engine system health status assessment model based on RVE is constructed, including: using the training set as input, mapping the full life cycle data features of the turbine engine system to the latent space, learning the remaining service life corresponding to different values ​​of sensor parameter data, the turbine engine system health status assessment model includes a decoder composed of a variational autoencoder, a latent space, and a regression model, the variational autoencoder includes a bidirectional long short-term memory network layer and two fully connected layers, the latent space contains reparameter operations, and the decoder composed of the regression model includes a fully connected layer, a tanh activation function, and an output layer.

[0014] Furthermore, the calculation process of the turbine engine system health status assessment model is as follows:

[0015] The variational autoencoder includes a bidirectional long short-term memory network layer and two structurally identical fully connected layers. These two structurally identical fully connected layers are used to learn the mean μ = (μ1, ..., μ) of the training set data. l ) and variance σ=(σ1,…,σ l ), where l is the size of the latent space dimension. In the computation of the bidirectional long short-term memory network layer, the dimension of the input is (seq_len, batch_size, input_size), resulting in a hidden state sequence with the same sequence length as the sensor parameter data. as well as Output of bidirectional hidden layer In Perform a splicing operation and input two fully connected layers with the same structure to obtain the encoder outputs mu and var;

[0016] In the latent space, the reparameter technique is used on the encoder outputs mu and var to make the sampling operation differentiable instead of non-differentiable. First, sampling is performed from a Gaussian distribution with a mean of 0 and a standard deviation of 1. Then, the latent variable Z is obtained by scaling and shifting and is used for the forward propagation operation in the second half. The specific calculation process is as follows:

[0017]

[0018] Where, μ i σ is the i-th dimension value of the mean μ learned from the training set. i Z is the i-th dimension value of the mean σ learned from the training set, where ∈ is a tensor that follows a normal distribution. i It is the value of the i-th dimension of the latent variable Z;

[0019] The decoder, which consists of a regression model, includes a fully connected layer, a hyperbolic tangent activation function, and another fully connected layer as the output layer.

[0020] Furthermore, before training the label-based turbine engine system health status assessment model, the parameter weights of all network layers are initialized to 0, and a loss function and optimizer are constructed for the remaining service life prediction task, with the goal of minimizing the loss function to obtain the trained model.

[0021] Furthermore, the preprocessing of data X includes: denoising the data X, then using the Laida criterion to remove outliers, and finally performing Z-Score normalization.

[0022] Secondly, a turbine engine system health status assessment device based on RVE is provided, comprising:

[0023] The acquisition module is used to acquire data X collected by sensors throughout the entire life cycle of the turbine engine;

[0024] The preprocessing module is used to preprocess the data X, construct dataset health indicators as data labels, and divide the inputs and corresponding outputs of each batch into training set, validation set and test set by using the sliding window method.

[0025] The model building module is used to build a health status assessment model for turbine engine systems based on RVE;

[0026] The model training module is used to train a turbine engine system health assessment model using an improved loss function and the Adam optimizer.

[0027] The output module is used to input historical data of the turbine engine system to be evaluated into the trained turbine engine system health status assessment model, and output the remaining service life degradation process of the turbine engine system until complete failure as a health status assessment curve.

[0028] Furthermore, the turbine engine system health status assessment model includes a decoder composed of a variational autoencoder, a latent space, and a regression model. The variational autoencoder includes a bidirectional long short-term memory network layer and two fully connected layers. The latent space contains reparameter-based operations. The decoder composed of the regression model includes a fully connected layer, a tanh activation function, and an output layer.

[0029] Thirdly, a computer-readable storage medium is provided that stores program code for execution by a device, the program code including steps for performing a method as described in any implementation of the first aspect.

[0030] Fourthly, an electronic device is provided, the electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as in any of the implementations of the first aspect.

[0031] This application has the following beneficial effects: This application innovatively proposes to use the RVE model to construct the remaining service life degradation curve to assess the health status of a turbine engine system, making full use of the parameter characteristics of multi-dimensional sensors, and making the assessment of the health status of the turbine engine system more accurate. This method can be applied to the health status management and maintenance of turbine engine systems and has strong practicality. Attached Figure Description

[0032] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0033] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of the RVE-based turbo engine system health status assessment method according to Embodiment 1 of this application;

[0035] Figure 2 This is a structural block diagram of the turbine engine system health status assessment model in the RVE-based turbine engine system health status assessment method of Embodiment 1 of this application;

[0036] Figure 3This is a comparative diagram of the degradation functions used to construct the remaining service life of the health assessment index label in the RVE-based turbine engine system health status assessment method of Embodiment 1 of this application;

[0037] Figure 4 This is a schematic diagram of the training iteration of the turbine engine system health status assessment model in the RVE-based turbine engine system health status assessment method of Embodiment 1 of this application;

[0038] Figure 5 This is a schematic diagram of the prediction results of the turbine engine system health status assessment model in the RVE-based turbine engine system health status assessment method of Embodiment 1 of this application;

[0039] Figure 6 This is a simplified schematic diagram of the engine simulation results using the IEEE PHM08 turbofan engine degradation simulation data set in Embodiment 1 of this application. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example 1

[0042] This application discloses a method for assessing the health status of a turbine engine system based on Residual Vessel Analysis (RVE). The method includes: acquiring data X collected by sensors throughout the turbine engine's lifecycle; preprocessing the data X to construct dataset health indicators as labels, and dividing the dataset into training, validation, and test sets by using a sliding window method to partition the inputs and corresponding outputs of each batch; constructing an RVE-based turbine engine system health status assessment model; training the model using an improved loss function and the Adam optimizer; inputting historical data of the turbine engine system to be assessed into the trained model, and using the remaining service life degradation process of the turbine engine system until complete failure as the health status assessment curve. This method assesses the health status of a turbine engine system by constructing a remaining service life degradation curve using the RVE model, fully utilizing the parameter characteristics of multi-dimensional sensors, and significantly improving the accuracy of predicted service life indicators. This method can be applied to the health status management and maintenance of turbine engine systems and has strong practicality.

[0043] Specifically, Figure 1A flowchart of the RVE-based turbine engine system health assessment method in Embodiment 1 of the application is shown, including:

[0044] S101. Acquire data X collected by sensors throughout the entire life cycle of the turbine engine;

[0045] Specifically, data X includes HPC outlet total temperature (T30), LPT outlet total temperature (T50), HPC outlet total pressure (P30), engine pressure ratio P50 / P2 (epr), HPC outlet static pressure (Ps30), and the ratio of fuel flow rate to Ps30 (phi). Data X = [x1, x2, ..., x N ],in, n is the batch size, and d is the input dimension size, which depends on the number of sensors selected. Represents an n-row, d-column real matrix.

[0046] S102. Preprocess the data X, construct dataset health indicators as data labels, and divide the inputs and corresponding outputs of each batch into training set, validation set, and test set using the sliding window method. This includes the following steps:

[0047] S201. The collected data is denoised, then outlier removal is performed using the Laida criterion, and finally Z-score normalization is performed. The specific steps are as follows:

[0048] S2010. Denoise the normalized dataset to obtain the preprocessed dataset.

[0049] S2011. Outlier removal is performed on the collected data. Here, the Laida criterion is used, retaining only the x values ​​of each dimension that are within the range (μ-3σ, μ+3σ). i , where μ j σ is the mean of the j-th dimension of X. j It is the standard deviation of the j-th dimension of X.

[0050] S2012. Calculate the normalized dataset after Z-Score processing. The normalization process for the dataset is as follows:

[0051]

[0052] Where, x i,j x represents the i-th time step. i The value of the j-th feature dimension, μ j σ is the mean of the j-th dimension of X. j It is the standard deviation of the j-th dimension of X;

[0053] S202. Construct a health metric "Remaining Usable Life (RUL)" as the dataset label for the preprocessed dataset in step S201, such as... Figure 3 As shown, since the system always tends to deteriorate, it is usually necessary to pre-assume the degradation trend and use labels based on these assumptions to construct the target RUL in a supervised manner to guide model training and enhance its prediction accuracy. Here, a "piecewise linear degradation function" is adopted. The first 125 cycles of the full life cycle data of the turbine engine system are taken as the maximum RUL label. Typically, the maximum RUL of the system under single-condition is 130, and the maximum RUL of the system under multiple conditions is 150. The RUL label of the single-condition system after the 125th cycle is adopted, and the label range is [130,0]. The RUL label of the multi-condition system after the 125th cycle is adopted, and the label range is [150,0]. This results in a preprocessed dataset with the label "Remaining Service Life (RUL)".

[0054] S203. Using the sliding window method, select sensor data [x] within each time window of size 30. i ,x i+1 ,…,x i+29 ] Form a high-dimensional feature vector to be used as input sample, and x i+30 As the corresponding output of the samples, the model is guided to learn a better health status assessment effect. Finally, all the samples obtained by the sliding window method are first grouped according to the engine number, and then divided according to the ratio of 7:2:1 to obtain the training set, validation set and test set.

[0055] S103. Construct a turbine engine system health status assessment model based on RVE. Please refer to [link / reference]. Figure 2 Specifically, it includes the following steps:

[0056] S301. Using the training set data divided in step S102 as input, the remaining service life (RUL) corresponding to different values ​​of sensor parameter data is learned by mapping the full life cycle data features of the turbine engine system to the potential space.

[0057] S302. The health status assessment model for this turbine engine system consists of three parts: a variational autoencoder, a latent space, and a decoder composed of a regression model. The variational autoencoder includes a bidirectional long short-term memory (Bi-LSTM) network layer with a hidden layer size of 300 and two fully connected layers with an input size of 600 and an output size of 2. The latent space only contains "re-parameter" operations. The decoder includes a fully connected layer with an input size of 2 and an output size of 200, a tanh activation function, and an output layer with an input size of 200 and an output size of 1. The specific calculation process for each part is as follows:

[0058] S3021. The variational autoencoder includes a bidirectional long short-term memory network layer (Bi-LSTM) and two fully connected layers with identical structures, used to learn the mean μ = (μ1,…,μ) of the training set data. l ) and variance σ=(σ1,…,σ l ), where l is the size of the potential space dimension;

[0059] In the computation of the Bidirectional Long Short-Term Memory (Bi-LSTM) network layer, the input dimension is (seq_len, batch_size, input_size), and we obtain a hidden state sequence with the same length as the sensor parameter data sequence. as well as Output of bidirectional hidden layer In Perform a splicing operation and input two fully connected layers with the same structure to obtain the encoder outputs mu and var.

[0060] S3022. In the latent space part, the encoder outputs mu and var from step S3021 are used with the "reparameter trick" to make the sampling operation differentiable instead of non-differentiable. First, samples are taken from a Gaussian distribution with a mean of 0 and a standard deviation of 1, and then the latent variable Z is obtained by scaling and shifting for the forward propagation operation in the second half. The specific calculation process is as follows:

[0061]

[0062] Where, μ i σ is the i-th dimension value of the mean μ learned from the training set. i Z is the i-th dimension value of the mean σ learned from the training set, where ∈ is a tensor that follows a normal distribution. i It is the value of the i-th dimension of the latent variable Z.

[0063] S3023. The decoder composed of the regression model contains a fully connected layer (the input dimension is the size of the latent space dimension), a hyperbolic tangent activation function, and a fully connected layer (the output dimension is 1) as the output layer.

[0064] S104. The improved loss function and Adam optimizer are used to train the turbine engine system health status assessment model;

[0065] For example, the training set data and remaining lifespan labels partitioned in step S102 are input into the health status assessment model for training. Before training the model, the parameter weights of all network layers are initialized to 0, and a loss function and optimizer are constructed for the remaining lifespan prediction task. The trained model is obtained by minimizing the loss function. Figure 4 As shown, the specific steps are as follows:

[0066] S401. Initialize the parameter weights of all network layers in step S302 to 0 to obtain the initial model before training.

[0067] S402. In the loss function for the RUL prediction task, to improve the model's prediction accuracy, the root mean square error (RMSE) is added as part of the loss function, and Kullback-Leibler divergence is used to learn the reconstruction of the training set to achieve better feature extraction results. The specific formula is as follows:

[0068]

[0069] Where X represents the training set data, Z represents the hidden variable vector, and θ and φ represent the parameters of the encoder and decoder, respectively. It is the predicted remaining useful life (RUL), y i It is a Remaining Useful Life (RUL) label;

[0070] S403. Use the Adam Optimizer to train the model with the minimum loss function of the health status assessment model as the optimization objective, and obtain the trained model.

[0071] S105. Input the historical data of the turbine engine system to be evaluated into the trained turbine engine system health status assessment model, and use the remaining service life degradation process of the turbine engine system until complete failure as the health status assessment curve. Figure 5 As shown.

[0072] like Figure 6 As shown, a specific implementation example involves data acquisition on the 2008 IEEE PHM08 turbofan engine degradation simulation dataset C-MAPSS. First, six sensor parameters T30, T50, P30, epr, Ps30, and phi were selected as input data. The data underwent denoising, outlier removal, and normalization. A piecewise linear degradation function was used to construct the hypothetical health indicator "Remaining Usable Life (RUL)" as the dataset label. Then, using a sliding window method, the inputs and corresponding outputs of each batch were divided into two parts with a sliding window of size 30. Since the simulation dataset already provides the test set of the turbofan engine to be evaluated and its labels, the dataset was divided into training and validation sets in an 8:2 ratio.

[0073] Then, a health assessment model for a turbine engine system based on Regression-Encoder (RVE) is established. Sensor parameter features are learned through variational autoencoders, and "reparameter re-operation" is performed in the latent space for the forward propagation of the latter half. Finally, the health assessment of the turbine engine system is completed in the decoder composed of regression models. The model prediction effect is optimized by using a loss function for the RUL prediction task. Finally, the test set of the turbine engine to be evaluated is input into the trained prediction model, and the remaining service life (RUL) degradation process of the turbine engine system until complete failure is output as the health assessment curve.

[0074] Experimental results show that the health status assessment method proposed in this invention has high reference and guidance value. Compared with the test set label, the RUL assessment performance index RMSE of this model on the test set is 11.05, and the RUL degradation curve, which serves as the health status assessment curve, has high prediction accuracy.

[0075] Example 2

[0076] The second embodiment of this application relates to a turbine engine system health status assessment device based on RVE, comprising:

[0077] The acquisition module is used to acquire data X collected by sensors throughout the entire life cycle of the turbine engine;

[0078] The preprocessing module is used to preprocess the data X, construct dataset health indicators as data labels, and divide the inputs and corresponding outputs of each batch into training set, validation set and test set by using the sliding window method.

[0079] The model building module is used to build a health status assessment model for turbine engine systems based on RVE;

[0080] The model training module is used to train a turbine engine system health assessment model using an improved loss function and the Adam optimizer.

[0081] The output module is used to input historical data of the turbine engine system to be evaluated into the trained turbine engine system health status assessment model, and output the remaining service life degradation process of the turbine engine system until complete failure as a health status assessment curve.

[0082] In a further embodiment, the turbine engine system health status assessment model includes a decoder consisting of a variational autoencoder, a latent space, and a regression model. The variational autoencoder includes a bidirectional long short-term memory network layer and two fully connected layers. The latent space contains reparameter-restricted operations. The decoder consisting of the regression model includes a fully connected layer, a tanh activation function, and an output layer.

[0083] Example 3

[0084] The present application discloses a computer-readable storage medium that stores program code for execution by a device, the program code including steps for performing the method in any implementation of the present application, as described in the first embodiment of the present application.

[0085] The computer-readable storage medium may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM); the computer-readable storage medium may store program code, and when the program stored in the computer-readable storage medium is executed by a processor, the processor is used to perform the steps of the method in any of the implementations of Embodiment 1 of this application.

[0086] Example 4

[0087] An electronic device according to Embodiment 4 of this application includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the method in any of the implementations in Embodiment 1 of this application.

[0088] The processor can be a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), graphics processing unit (GPU), or one or more integrated circuits, used to execute relevant programs to implement the method in any of the implementations of Embodiment 1 of this application.

[0089] The processor can also be an integrated circuit electronic device with signal processing capabilities. In implementation, each step of the method in any of the implementations of Embodiment 1 of this application can be completed by the integrated logic circuitry in the processor's hardware or by software instructions.

[0090] The aforementioned processor can also be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the functions required by the units included in the data processing apparatus of the embodiments of this application, or executes the methods in any implementation of Embodiment 1 of this application.

[0091] The above are merely preferred embodiments of this application; however, the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and its improved concept, should be covered within the scope of protection of this application.

Claims

1. A RVE-based turbine engine system health assessment method, characterized by, include: Acquiring data collected by sensors throughout the life cycle of a turbine engine ; Data Preprocessing is performed to construct dataset health indicators as labels for the data, and the inputs and corresponding outputs contained in each batch are divided into training set, validation set and test set by using the sliding window method; Construct a health status assessment model for turbine engine systems based on RVE; An improved loss function and Adam optimizer were used to train a turbine engine system health assessment model; Historical data of the turbine engine system to be evaluated are input into the trained turbine engine system health status assessment model, and the remaining service life degradation process of the turbine engine system until complete failure is output as the health status assessment curve. The construction of a turbine engine system health status assessment model based on RVE includes: using the training set as input to map the full life cycle data features of the turbine engine system to the latent space, learning the remaining service life corresponding to different values ​​of sensor parameter data; the turbine engine system health status assessment model includes a decoder composed of a variational autoencoder, a latent space, and a regression model; the variational autoencoder includes a bidirectional long short-term memory network layer and two fully connected layers; the latent space contains reparameter operations; and the decoder composed of the regression model includes a fully connected layer, a tanh activation function, and an output layer. The calculation process of the turbine engine system health status assessment model is as follows: The variational autoencoder includes a bidirectional long short-term memory network layer and two structurally identical fully connected layers. The two structurally identical fully connected layers are used to learn the mean of the training set data. and variance , Given the potential spatial dimension, the dimension of the input is [value]. In the computation of bidirectional long short-term memory network layers, the input dimension is [value]. The hidden state sequence with the same length as the sensor parameter data is obtained. as well as Output of bidirectional hidden layer , In , Perform a splicing operation and input two structurally identical fully connected layers to obtain the encoder output. and ; In the latent space, the encoder output and The reparameter technique is used to make the sampling operation differentiable from non-differentiable. First, samples are taken from a Gaussian distribution with a mean of 0 and a standard deviation of 1, and then the latent variables are obtained by scaling and shifting. The calculation process for the forward propagation operation in the latter half is as follows: ； in, The mean obtained by learning from the training set The value of the i-th dimension, The mean obtained by learning from the training set The value of the i-th dimension, It is a tensor that follows a normal distribution. It is a latent variable The value of the i-th dimension; The decoder, which consists of a regression model, includes a fully connected layer, a hyperbolic tangent activation function, and another fully connected layer as the output layer.

2. The RVE-based turbine engine system health assessment method of claim 1, wherein, The data ,in, , For batch size, The input dimension size depends on the number of sensors selected. express OK A column of real numbers.

3. The method for assessing the health status of a turbine engine system based on RVE according to claim 1, characterized in that, Before training the label-based model for assessing the health status of a turbine engine system, the parameters and weights of all network layers are initialized to 0. A loss function and optimizer are constructed for the remaining service life prediction task, and the trained model is obtained by minimizing the loss function.

4. The method for assessing the health status of a turbine engine system based on RVE according to claim 1, characterized in that, Data The preprocessing includes: denoising the data X, removing outliers using the Laida criterion, and finally performing Z-Score normalization.

5. A turbine engine system health status assessment device based on RVE, characterized in that, include: The acquisition module is used to acquire data collected by sensors throughout the entire lifecycle of the turbine engine. ; The preprocessing module is used to process the data. Preprocessing is performed to construct dataset health indicators as labels for the data, and the inputs and corresponding outputs contained in each batch are divided into training set, validation set and test set by using the sliding window method; The model building module is used to build a health status assessment model for turbine engine systems based on RVE. The model training module is used to train a turbine engine system health assessment model using an improved loss function and the Adam optimizer. The output module is used to input the historical data of the turbine engine system to be evaluated into the trained turbine engine system health status assessment model, and output the remaining service life degradation process of the turbine engine system until complete failure as the health status assessment curve. The construction of a turbine engine system health status assessment model based on RVE includes: using the training set as input to map the full life cycle data features of the turbine engine system to the latent space, learning the remaining service life corresponding to different values ​​of sensor parameter data; the turbine engine system health status assessment model includes a decoder composed of a variational autoencoder, a latent space, and a regression model; the variational autoencoder includes a bidirectional long short-term memory network layer and two fully connected layers; the latent space contains reparameter operations; and the decoder composed of the regression model includes a fully connected layer, a tanh activation function, and an output layer. The calculation process of the turbine engine system health status assessment model is as follows: The variational autoencoder includes a bidirectional long short-term memory network layer and two structurally identical fully connected layers. The two structurally identical fully connected layers are used to learn the mean of the training set data. and variance , Given the potential spatial dimension, the dimension of the input is [value]. In the computation of bidirectional long short-term memory network layers, the input dimension is [value]. The hidden state sequence with the same length as the sensor parameter data is obtained. as well as Output of bidirectional hidden layer , In , Perform a splicing operation and input two structurally identical fully connected layers to obtain the encoder output. and ; In the latent space, the encoder output and The reparameter technique is used to make the sampling operation differentiable from non-differentiable. First, samples are taken from a Gaussian distribution with a mean of 0 and a standard deviation of 1, and then the latent variables are obtained by scaling and shifting. The calculation process for the forward propagation operation in the latter half is as follows: ； in, The mean obtained by learning from the training set The value of the i-th dimension, The mean obtained by learning from the training set The value of the i-th dimension, It is a tensor that follows a normal distribution. It is a latent variable The value of the i-th dimension; The decoder, which consists of a regression model, includes a fully connected layer, a hyperbolic tangent activation function, and another fully connected layer as the output layer.

6. A computer-readable storage medium, characterized in that, The computer-readable medium stores program code for execution by the device, the program code including steps for performing the method as described in any one of claims 1-4.

7. An electronic device, comprising: The electronic device includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in any one of claims 1-4.

Images