Method, device and equipment for predicting remaining service life of aero-engine and medium

Abstract

The invention relates to the technical field of aviation, and provides a residual service life prediction method, device and equipment of an aero-engine and a medium, the residual service life prediction method of the aero-engine comprises the following steps: acquiring sensor data of the aero-engine; inputting the sensor data into a preset residual service life prediction model for prediction to obtain the predicted residual service life of the aero-engine; inputting the sensor data into a preset residual service life prediction model for prediction, and obtaining the predicted residual service life of the aero-engine, wherein the method comprises the following steps: extracting spatial features based on a graph convolutional network; extracting time features based on the path signature layer; and based on the prediction layer, obtaining the predicted remaining service life through a fusion feature obtained by fusing the spatial feature and the time feature. According to the method, the spatial features and the time features are considered at the same time, and the relationship between the structural features of the aero-engine and the time sequence data is combined, so that the remaining service life is predicted more accurately.

Description

Technical Field

[0001] The present invention relates to the field of aviation technology, and particularly to a method, device, equipment and medium for predicting the remaining useful life of an aeroengine. Background Art

[0002] As a core component of an aircraft, the safety of an aeroengine plays a crucial role in flight. However, affected by factors such as a complex and changeable working environment, the aeroengine is prone to performance degradation or even failure. If the remaining useful life (RUL) of a turbofan engine can be predicted timely and accurately, and effective maintenance measures can be taken in time, irreversible damage to the equipment can be effectively prevented, thereby increasing its life and greatly improving the overall efficiency and safety of the equipment. Summary of the Invention

[0003] The present invention provides a method, device, equipment and medium for predicting the remaining useful life of an aeroengine to solve the defects in the prior art.

[0004] The present invention provides a method for predicting the remaining useful life of an aeroengine, including: Obtaining sensor data of the aeroengine; Inputting the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aeroengine; wherein, the remaining useful life prediction model includes a graph convolutional network, a path signature layer and a prediction layer; The step of inputting the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aeroengine includes: Extracting spatial features from the sensor data based on the graph convolutional network; Extracting temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtaining the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features.

[0005] According to the method for predicting the remaining useful life of an aeroengine provided by the present invention, the prediction layer includes an autoencoder, a long short-term memory network and a fully connected layer connected in sequence; the step of obtaining the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features based on the prediction layer includes: Processing the fused features based on the autoencoder to obtain a first output result; Process the first output result based on the long short - term memory network and the fully - connected layer to obtain the predicted remaining useful life.

[0006] According to a method for predicting the remaining useful life of an aero - engine provided by the present invention, the extracting of spatial features from the sensor data based on the graph convolutional network includes: Based on the sensor data, construct a graph structure from the sensor dimension; Calculate the distance between any two nodes in the graph structure to obtain a symmetric positive - semi - definite matrix, and perform parametric processing on the symmetric positive - semi - definite matrix through linear transformation; Process the symmetric positive - semi - definite matrix through a Gaussian kernel to obtain an adjacency matrix; Extract the spatial features from the adjacency matrix through the graph convolutional network.

[0007] According to a method for predicting the remaining useful life of an aero - engine provided by the present invention, the extracting of temporal features from the sensor data based on the path signature layer includes: Based on a sliding window and a sliding step, segment the sensor data into multiple sensor data segments from the time dimension; Based on the path signature layer, perform path signature transformation on each of the sensor data segments respectively to obtain multiple signature features; Perform multi - layer non - linear transformation on each of the signature features to obtain the temporal features.

[0008] According to a method for predicting the remaining useful life of an aero - engine provided by the present invention, the construction process of the remaining useful life prediction model includes: Obtain a training set and a validation set; wherein, the training set and the validation set respectively include multiple samples, and each sample includes sample data and a sample label; Input the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model; wherein, the initial remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; Validate the remaining useful life prediction model after the training is completed through the validation samples in the validation set.

[0009] According to a method for predicting the remaining useful life of an aero - engine provided by the present invention, the inputting the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model includes: Input the training samples in the training set into the initial remaining useful life prediction model to obtain a first remaining useful life prediction value and the reconstruction error of the autoencoder in the prediction layer; Calculate the mean square error loss value based on the first remaining useful life prediction value and the sample label, and calculate the training loss based on the mean square error loss value and the reconstruction error; Iteratively train the initial remaining useful life prediction model based on the training loss until the training is completed to obtain the remaining useful life prediction model.

[0010] According to a method for predicting the remaining useful life of an aero-engine provided by the present invention, validating the remaining useful life prediction model that has completed training through the validation samples in the validation set includes: Input the validation samples in the validation set into the remaining useful life prediction model that has completed training to obtain a second remaining useful life prediction value; Calculate the root mean square error and the prediction score based on the second remaining useful life prediction value and the sample label; Validate the remaining useful life prediction model based on the root mean square error and the prediction score.

[0011] The present invention also provides a device for predicting the remaining useful life of an aero-engine, including: A first acquisition module configured to acquire sensor data of an aero-engine; A prediction module configured to input the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aero-engine; wherein, the remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The prediction module includes: A first extraction sub-module configured to extract spatial features from the sensor data based on the graph convolutional network; A second extraction sub-module configured to extract temporal features from the sensor data based on the path signature layer; A prediction sub-module configured to obtain the predicted remaining useful life based on the fusion features after fusing the spatial features and the temporal features through the prediction layer.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, it implements the method for predicting the remaining useful life of an aero-engine as described in any one of the above.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for predicting the remaining service life of an aero-engine as described in any one of the above.

[0014] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for predicting the remaining service life of an aero-engine as described in any one of the above.

[0015] The method, device, equipment and medium for predicting the remaining service life of an aero-engine provided by the present invention acquire sensor data of the aero-engine, and a remaining service life prediction model is pre-trained. The remaining service life prediction model includes a graph convolutional network, a path signature layer and a prediction layer. The relationship between sensors in the aero-engine is considered, and the graph convolutional network is used to extract spatial features from the sensor data. The path signature technology has uniqueness, translation invariance and reparameterization invariance, and the temporal features extracted by the path signature technology can fully retain the temporal information in the sensor data. The spatial features and temporal features are fused to obtain fused features. The fused features are input into the prediction layer for predicting the remaining service life, and the predicted remaining service life of the aero-engine is obtained. The present invention considers both spatial features and temporal features, combines the relationship between the structural features of the aero-engine and the time series data, so as to realize more accurate prediction of the remaining service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flowchart of the method for predicting the remaining service life of an aero-engine provided by the present invention.

[0018] Figure 2 It is a schematic diagram of the method for predicting the remaining service life of an aero-engine provided by the present invention.

[0019] Figure 3 It is a schematic diagram of the simulation structure of a turbofan engine involved in an embodiment of the present invention.

[0020] Figure 4 It is a schematic diagram of the sensor data involved in an embodiment of the present invention.

[0021] Figure 5 It is a schematic diagram of the root mean square error and prediction score of the test set involved in an embodiment of the present invention.

[0022] Figure 6 It is a schematic structural diagram of a remaining service life prediction device for an aeroengine provided by the present invention.

[0023] Figure 7 It is a schematic structural diagram of an electronic device provided by the present invention. Detailed implementation manners

[0024] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without making creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0025] Next, in conjunction with Figures 1-7 describe the remaining service life prediction method, device, equipment and medium of the aeroengine of the present invention.

[0026] Figure 1 It is a flowchart of a method for predicting the remaining service life of an aeroengine shown according to an exemplary embodiment. As Figure 1 shown, in an exemplary embodiment, the method for predicting the remaining service life of an aeroengine includes steps 110 to 120, which are introduced in detail as follows.

[0027] Step 110, obtain sensor data of the aeroengine.

[0028] In an embodiment of the present invention, sensor data of the entire life cycle during the operation of the aeroengine is collected, and the sensor data that is always a constant is deleted. The remaining sensor data is normalized using the maximum-minimum normalization method. For each sensor data , the normalized sensor data is: ; wherein, and are respectively the minimum value and the maximum value of the sensor data at all time points.

[0029] Step 120, input the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine; wherein, the remaining service life prediction model includes a graph convolutional network, a path signature layer and a prediction layer; The inputting the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine includes: Extract spatial features from the sensor data based on the graph convolutional network; Extract temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtain the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features.

[0030] In the embodiment of the present invention, a remaining useful life prediction model is pre-trained, and the sensor data is input into the preset remaining useful life prediction model for prediction, and the predicted remaining useful life of the aero-engine can be obtained.

[0031] As Figure 2 shown, considering the relationship between sensors, the graph convolutional network (Graph Convolutional Network, GCN) is used to extract spatial features. A graph is a language for describing and analyzing related entities. It does not regard entities as a series of isolated points, but as interrelated. Constructing a graph structure for sensors can provide a more effective way to simulate the spatial relationship between them.

[0032] The path signature layer has uniqueness, translation invariance, and reparameterization invariance. Compared with other neural network-based feature extractions, different time series data may extract similar features and lose discriminative power. The temporal features extracted by the path signature technology can fully retain the temporal information.

[0033] Fuse the extracted spatial features and temporal features to obtain fused features.

[0034] Input the fused features into the prediction layer for remaining useful life prediction to obtain the predicted remaining useful life of the aero-engine. The embodiment of the present invention considers both spatial features and temporal features, combines the relationship between the structural features of the aero-engine and the time series data, and can model the complex patterns and the relationship evolving over time between engine components, so as to achieve more accurate prediction of the remaining useful life.

[0035] In an exemplary embodiment of the present invention, the prediction layer includes an autoencoder, a long short-term memory network, and a fully connected layer connected in sequence; the obtaining the predicted remaining useful life based on the prediction layer through the fused features after fusing the spatial features and the temporal features includes: Process the fused features based on the autoencoder to obtain a first output result; Based on the long short-term memory network and the fully connected layer, process the first output result to obtain the predicted remaining useful life.

[0036] In the embodiments of the present invention, as Figure 2 shown, an auto - encoder (AE) learns an effective encoding (or representation) of a set of data in an unsupervised manner. The auto - encoder maps the input data to the feature space to obtain the corresponding encoding. Through the auto - encoder, an effective data representation can be obtained. Therefore, in the remaining service life prediction model after training, only the output of the auto - encoder is retained as the input of the subsequent long short - term memory network. During training, it is desired to reconstruct the original data based on the obtained encoding, and then obtain the corresponding reconstruction error.

[0037] The structure of the auto - encoder is divided into two parts: an encoder and a decoder. Among them, the encoder ; the decoder .

[0038] If the dimension of the feature space is less than the dimension of the original space, the auto - encoder is equivalent to a dimensionality reduction or feature extraction method.

[0039] The long short - term memory network (LSTM) introduces a gating mechanism to control the path of information transmission. The forget gate controls how much information from the internal state at the previous moment needs to be forgotten; the input gate controls how much information from the candidate state at the current moment needs to be saved; the output gate controls how much information from the internal state at the current moment needs to be output to the external state . The calculation methods of the three gates are as follows: ; ; ; ; ; ; Among them, is the element - wise product of vectors, is the Logistic function, t is the current moment, is the input at the current moment, is the external state at the previous moment, is the output at the current moment, 、 、 respectively represent the weight matrices corresponding to the input gate, forget gate, and output gate, , , respectively represent the weight matrices corresponding to the input gate, forget gate, and output gate for processing the external state at the previous moment, , and respectively represent the bias terms corresponding to the input gate, forget gate, and output gate, represents the candidate cell state, represents the weight matrix corresponding to the candidate state, represents the weight matrix corresponding to the candidate state for processing the external state at the previous moment, represents the bias term of the candidate state.

[0040] Use the output of the long short-term memory network as the input of the fully connected layer to obtain the predicted remaining useful life.

[0041] In an exemplary embodiment of the present invention, extracting spatial features from the sensor data based on the graph convolutional network includes: Based on the sensor data, construct a graph structure from the sensor dimension; Calculate the distance between any two nodes in the graph structure to obtain a symmetric positive semi-definite matrix, and perform parametric processing on the symmetric positive semi-definite matrix through a linear transformation; Process the symmetric positive semi-definite matrix through a Gaussian kernel to obtain an adjacency matrix; Extract the spatial features from the adjacency matrix through the graph convolutional network.

[0042] In the embodiment of the present invention, define the graph structure of the sensor data as , represents the sensor data, where represents the length of the time series, represents the set of edges in the graph, represents the set of nodes in the graph, that is, sensors. Calculate the distance between any two nodes. The distance can be Euclidean distance, cosine distance, Pearson correlation coefficient, Mahalanobis distance, etc., to obtain a symmetric positive semi-definite matrix , and perform parameterization on it using a linear transformation to obtain trainable weights. The distance metric can be adaptively learned according to the characteristics of the feature space. Then, by using a Gaussian kernel, obtain an adaptive adjacency matrix , , where represents the distance between node i and node j, σ represents the scale parameter. The adjacency matrix reflects the connection between sensors, where, Represents the sensor and The connection weight between them.

[0043] Next, a graph convolutional network is used to extract spatial features from the adjacency matrix. The node representation obtained through the graph convolutional network is defined as: , where is the normalized adjacency matrix with self-connections added, is defined as , represents the model parameters, is the activation function, and I is the identity matrix. The obtained by the graph convolutional network encodes the direct neighborhood of each node in the graph.

[0044] In an exemplary embodiment of the present invention, extracting temporal features from the sensor data based on the path signature layer includes: Based on a sliding window and a sliding step size, from the time dimension, the sensor data is segmented into multiple sensor data segments; Based on the path signature layer, path signature transformation is performed on each of the sensor data segments to obtain multiple signature features; Performing multi-layer non-linear transformation on each of the signature features to obtain the temporal features.

[0045] In the embodiment of the present invention, the sliding window size time_window_size and the sliding step size time_step_size are set, and the sensor data of each engine is segmented and extracted to obtain multiple sensor data segments.

[0046] The present invention uses the path signature transformation method to extract the temporal features of sensor data. The path signature method is derived from rough path theory and interprets multivariate time series as the discretization of a potential continuous path, and it can be applied to generate a known real-valued feature vector. Specifically, the path signature is a set of infinitely long integrals along dimension combinations, and each integral term is integrated in the order of the corresponding dimension combination.

[0047] Let represent the time series data of sensors with a time step of and a time span of , also known as a path. The path signature of the time series is a set of iterated integrals of each order. The path signature of the sequence consists of 1 and the signature terms of each order, where 1 represents the zero-order signature term, and the signature terms of each order are composed of iterated integrals of the path in the specified dimension order, expressed as: ; wherein, represents the th order path signature of the sequence represents the zero - order path signature. Since the path signature is an infinitely long sequence, generally, in practice, a truncation number is defined, and then the path signature method is applied within the range of orders. The path signature sequence of the th order usually has terms, which are composed of iteratively integrating the path in the specified dimension order, and are formalized as shown in the following formula: ; wherein, represents a single signature term for integrating in the dimension order of , and is formalized as shown in the following formula: ; In the embodiments of the present invention, path signature transformation is performed on each sensor data segment to extract signature features, and multi - layer non - linear transformation is performed on the obtained signature features to further extract features, thereby obtaining time features.

[0048] In an exemplary embodiment of the present invention, the process of constructing the remaining useful life prediction model includes: Obtaining a training set and a validation set; wherein, the training set and the validation set each include a plurality of samples, and the samples include sample data and sample labels; Inputting the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed, thereby obtaining the remaining useful life prediction model; wherein, the initial remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; Validating the remaining useful life prediction model after the training is completed through the validation samples in the validation set.

[0049] In the embodiments of the present invention, a training set and a validation set are obtained. The samples in the training set and the validation set are both the same sensor data as described above. The samples in the training set are training samples, and the samples in the validation set are validation samples. The sample labels are the true remaining useful life of the corresponding aero - engine.

[0050] Input the training samples in the training set into the initial remaining useful life prediction model for iterative training until the training is completed, obtaining the remaining useful life prediction model. The initial remaining useful life prediction model and the remaining useful life prediction model have the same structure, and the only difference lies in the parameters of their graph convolutional network, path signature layer, autoencoder, long short-term memory network, and fully connected layer.

[0051] Use the validation samples in the validation set to validate the remaining useful life prediction model after the training is completed, and verify the accuracy and effectiveness of the model prediction results.

[0052] In an exemplary embodiment of the present invention, the step of inputting the training samples in the training set into the initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model includes: Input the training samples in the training set into the initial remaining useful life prediction model to obtain the first remaining useful life prediction value and the reconstruction error of the autoencoder in the prediction layer; Calculate the mean squared error loss value based on the first remaining useful life prediction value and the sample label, and calculate the training loss based on the mean squared error loss value and the reconstruction error; Based on the training loss, perform iterative training on the initial remaining useful life prediction model until the training is completed to obtain the remaining useful life prediction model.

[0053] In the embodiment of the present invention, under the PyTorch deep learning framework, input the training samples in the training set into the initial remaining useful life prediction model for training. The Adam optimization algorithm is used. During each batch of training, calculate the mean squared error loss value based on the first remaining useful life prediction value and the sample label, and the loss function is the mean squared error (MSE).

[0054] The learning objective of the autoencoder is to minimize the reconstruction error: ; where N represents the number of samples, represents the latent space representation, represents the decoder, represents the encoder.

[0055] After the autoencoder obtains the corresponding encoding based on the input data, it performs reconstruction based on this encoding to obtain the reconstructed data, and calculates the reconstruction error between the reconstructed data and the input data.

[0056] Add the calculated mean squared error loss value and the reconstruction error as the training loss to perform backpropagation and update the trainable parameters in the initial remaining useful life prediction model until the training is completed.

[0057] In an exemplary embodiment of the present invention, validating the remaining useful life prediction model that has been trained through the validation samples in the validation set includes: Input the validation samples in the validation set into the trained remaining useful life prediction model to obtain a second remaining useful life prediction value; Based on the second remaining useful life prediction value and the sample label, calculate the root mean square error and the prediction score; Based on the root mean square error and the prediction score, validate the remaining useful life prediction model.

[0058] In the embodiments of the present invention, in order to verify the accuracy and effectiveness of the model prediction results, the present invention selects two indicators, the root mean square error (RMSE) and the prediction score (Score), for verification.

[0059] RMSE is a commonly used evaluation index in regression problems. It calculates the average difference between the predicted values and the true values, and the unit of measurement is the same as the predicted values. The range of RMSE is [0, +∞). The smaller the value, the more accurate the model prediction result. The specific calculation formula is as follows: ; where N represents the number of samples, represents the predicted remaining useful life value of sample i, represents the true remaining useful life value of sample i, that is, the sample label.

[0060] Score is the official evaluation standard provided by NASA for this research problem. It is an asymmetric index that uses different measurement methods for equal absolute errors when they are on the high side and the low side. Considering the actual situation, when the predicted remaining useful life value is less than the true remaining useful life value, it means that the predicted engine failure time is earlier than the true failure time. The maintenance decision made according to this prediction result is relatively conservative and safe. Therefore, the obtained Score score is smaller. And when the predicted remaining useful life value is greater than the true remaining useful life value, it means that the predicted failure time is later than the true failure time. According to this prediction result, it is impossible to formulate an inspection plan in time, which may lead to accidents and more serious consequences. Therefore, the obtained Score score is larger. The range of Score is [0, +∞). The smaller the value, the more accurate the model prediction result. The specific calculation formula is as follows: ; Among them, represents the predicted remaining useful life value of sample i, represents the true remaining useful life value of sample i.

[0061] After calculating the root mean square error and the prediction score, the smaller both of them are, the better the accuracy and effectiveness of the model prediction result are.

[0062] Based on the method for predicting the remaining useful life of an aeroengine provided by the present invention, experiments were carried out on a Windows 10 64-bit operating system, with the CPU model being i7-9750 and the running memory being 24GB. Programming was carried out using Python 3.9 on Jupyter Notebook, and the CPU version of Pytorch 2.3.1 was used. The CMAPSS dataset publicly available from NASA Ames Research Center was used. This dataset was obtained on the CMAPSS commercial aviation system propulsion simulation platform developed by it. Figure 3 shows the structure of the simulated turbofan engine. Among them, Figure 3 Fan, Combustor, N1, LPT, LPC, HPC, N2, HPT, Nozzle in

[0063] respectively represent the fan, combustor, fan speed, low-pressure turbine, low-pressure compressor, high-pressure compressor, high-pressure rotor speed, high-pressure turbine, and nozzle.

[0064] Table 1

[0065] Table 2

[0066] In this embodiment, the FD001 sub-dataset is used as the training set and the test set respectively. The specific process of applying the method provided by the present invention for predicting the remaining useful life is as follows: Plot the images of the data of 21 sensors of the first engine in the training set changing with time, such asFigure 4 As shown, the values of sensors No. 1, 5, 6, 10, 16, 18, and 19 are always constants and are irrelevant values. Delete them and only retain the data of the remaining 14 sensors.

[0067] Add the remaining useful life corresponding to each sample in the training set, that is, the maximum flight cycles of each engine minus the current flight cycles. Further, use the piecewise linear degradation method to reset the remaining useful life, set the maximum remaining useful life value to 125, and if the remaining useful life of the sample exceeds this value, set it to the maximum value.

[0068] Group according to the value of working condition 1, and use the maximum-minimum normalization method to normalize the data of the 14 sensors respectively.

[0069] Set the sliding window size time_window_size to 50 and the sliding step size time_step_size to 1 to segment the sensor data.

[0070] Build the network structure. The output dimension of the graph convolutional network is set to 100; the order of the path signature in the path signature layer is set to 3, the window size sig_window_size is set to 10, the sliding step size sig_step_size is set to 3, and it contains 3 non-linear layers with output dimensions of 2000, 1000, and 100 respectively, and the activation function is Leaky relu; in the autoencoder, the decoder contains 4 non-linear layers with output dimensions of 1000, 500, 250, and 100 respectively, the encoder also contains 4 non-linear layers with output dimensions of 500, 250, 1000, and 100 respectively, and the activation functions are all Relu; the hidden layer dimension of the long short-term memory network is 100; the input dimension of the fully connected layer is 2800 and the output dimension is 1.

[0071] Input the training samples of the training set into the network model. The loss function is the mean squared error, the optimization method is Adam, the learning rate is set to 0.001, and the weight decay is set to 0.0001. Finally, obtain the remaining useful life prediction model.

[0072] Input the test set samples into the trained remaining useful life prediction model to obtain the prediction results, as Figure 5 shown, where the RMSE is 11.8225 and the Score is 234.1745.

[0073] The remaining service life prediction device of the aero-engine provided by the present invention will be described below. The remaining service life prediction device of the aero-engine described below can be correspondingly referred to the remaining service life prediction method of the aero-engine described above. It should be noted that the device provided in the following embodiments and the method provided in the above embodiments belong to the same concept. The specific ways in which each module and unit perform operations have been described in detail in the method embodiments, and will not be repeated here.

[0074] In an exemplary embodiment of the present invention, please refer to Figure 6 , Figure 6 which is a remaining service life prediction device of an aero-engine shown according to an exemplary embodiment, and includes the following modules.

[0075] The first acquisition module 610 is configured to acquire sensor data of the aero-engine; The prediction module 620 is configured to input the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aero-engine; wherein, the remaining service life prediction model includes a graph convolutional network, a path signature layer and a prediction layer; The prediction module includes: The first extraction sub-module is configured to extract spatial features from the sensor data based on the graph convolutional network; The second extraction sub-module is configured to extract temporal features from the sensor data based on the path signature layer; The prediction sub-module is configured to obtain the predicted remaining service life based on the fusion features after fusing the spatial features and the temporal features based on the prediction layer.

[0076] In an exemplary embodiment of the present invention, the prediction layer includes an autoencoder, a long short-term memory network and a fully connected layer connected in sequence; the prediction sub-module includes: The first processing unit is configured to process the fusion features based on the autoencoder to obtain a first output result; The second processing unit is configured to process the first output result based on the long short-term memory network and the fully connected layer to obtain the predicted remaining service life.

[0077] In an exemplary embodiment of the present invention, the first extraction sub-module includes: The construction unit is configured to construct a graph structure from the sensor dimension based on the sensor data; The first calculation unit is configured to calculate the distance between any two nodes in the graph structure to obtain a symmetric positive semi-definite matrix, and perform parametric processing on the symmetric positive semi-definite matrix through linear transformation; A third processing unit, configured to process the symmetric positive semi - definite matrix through a Gaussian kernel to obtain an adjacency matrix; An extraction unit, configured to extract the spatial features from the adjacency matrix through the graph convolutional network.

[0078] In an exemplary embodiment of the present invention, the first extraction sub - module includes: A segmentation unit, configured to segment the sensor data into a plurality of sensor data segments from the time dimension based on a sliding window and a sliding step; A signature transformation unit, configured to perform path signature transformation on each of the sensor data segments respectively based on the path signature layer to obtain a plurality of signature features; A linear transformation unit, configured to perform multi - layer non - linear transformation on each of the signature features to obtain the time features.

[0079] In an exemplary embodiment of the present invention, the remaining useful life prediction device for an aero - engine further includes: A second acquisition module, configured to acquire a training set and a validation set; wherein, the training set and the validation set respectively include a plurality of samples, and each sample includes sample data and a sample label; A training module, configured to input the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model; wherein, the initial remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; A validation module, configured to validate the remaining useful life prediction model after the training is completed through the validation samples in the validation set.

[0080] In an exemplary embodiment of the present invention, the training module includes: A first input sub - module, configured to input the training samples in the training set into the initial remaining useful life prediction model to obtain a first remaining useful life prediction value and a reconstruction error of the auto - encoder in the prediction layer; A first calculation sub - module, configured to calculate a mean square error loss value based on the first remaining useful life prediction value and the sample label, and calculate a training loss based on the mean square error loss value and the reconstruction error; A training sub - module, configured to perform iterative training on the initial remaining useful life prediction model based on the training loss until the training is completed to obtain the remaining useful life prediction model.

[0081] In an exemplary embodiment of the present invention, the validation module includes: A second input sub-module, configured to input the validation samples in the validation set into the trained remaining useful life prediction model to obtain a second predicted remaining useful life value; A second calculation sub-module, configured to calculate the root mean square error and the prediction score based on the second predicted remaining useful life value and the sample label; A validation sub-module, configured to validate the remaining useful life prediction model based on the root mean square error and the prediction score.

[0082] Figure 7 An example of a schematic physical structure diagram of an electronic device is shown as Figure 7 shown. The electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740. Among them, the processor 710, the communication interface 720, and the memory 730 complete mutual communication through the communication bus 740. The processor 710 can call the logical instructions in the memory 730 to execute the method for predicting the remaining useful life of an aeroengine. The method includes: Obtain the sensor data of the aeroengine; Input the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aeroengine; wherein, the remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The step of inputting the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aeroengine includes: Extract spatial features from the sensor data based on the graph convolutional network; Extract temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtain the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features.

[0083] In addition, when the logical instructions in the above-mentioned memory 730 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0084] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the remaining service life prediction method of an aeroengine provided by the above-mentioned various methods. The method includes: Obtain sensor data of the aeroengine; Input the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine. Among them, the remaining service life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The step of inputting the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine includes: Extract spatial features from the sensor data based on the graph convolutional network; Extract temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtain the predicted remaining service life through the fused features after fusing the spatial features and the temporal features.

[0085] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the remaining service life prediction method of an aeroengine provided by the above-mentioned various methods. The method includes: Obtain sensor data of the aeroengine; Input the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine; wherein, the remaining service life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The step of inputting the sensor data into a preset remaining service life prediction model for prediction to obtain the predicted remaining service life of the aeroengine includes: Extract spatial features from the sensor data based on the graph convolutional network; Extract temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtain the predicted remaining service life through the fused features after fusing the spatial features and the temporal features.

[0086] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0087] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, also by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the remaining service life of an aeroengine, characterized in that Including: Obtaining sensor data of an aero-engine; Inputting the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aero-engine; wherein, the remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The step of inputting the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aero-engine includes: Extracting spatial features from the sensor data based on the graph convolutional network; Extracting temporal features from the sensor data based on the path signature layer; Based on the prediction layer, obtaining the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features.

2. The method for predicting the remaining service life of an aero-engine according to claim 1, characterized in that, The prediction layer includes an autoencoder, a long short-term memory network, and a fully connected layer connected in sequence; the step of obtaining the predicted remaining useful life through the fused features after fusing the spatial features and the temporal features based on the prediction layer includes: Processing the fused features based on the autoencoder to obtain a first output result; Processing the first output result based on the long short-term memory network and the fully connected layer to obtain the predicted remaining useful life.

3. The remaining service life prediction method of an aero-engine according to claim 1, characterized in that, The step of extracting spatial features from the sensor data based on the graph convolutional network includes: Based on the sensor data, constructing a graph structure from the sensor dimension; Calculating the distance between any two nodes in the graph structure to obtain a symmetric positive semi-definite matrix, and performing parametric processing on the symmetric positive semi-definite matrix through linear transformation; Processing the symmetric positive semi-definite matrix through a Gaussian kernel to obtain an adjacency matrix; Extracting the spatial features from the adjacency matrix through the graph convolutional network.

4. The method for predicting the remaining service life of an aero-engine according to claim 1, characterized in that The step of extracting temporal features from the sensor data based on the path signature layer includes: Based on a sliding window and a sliding step, dividing the sensor data into multiple sensor data segments from the time dimension; Performing path signature transformation on each of the sensor data segments based on the path signature layer to obtain multiple signature features; Performing multi-layer non-linear transformation on each of the signature features to obtain the temporal features.

5. The method for predicting the remaining service life of an aeroengine according to any one of claims 1 to 4, characterized in that, The construction process of the remaining useful life prediction model includes: Obtaining a training set and a validation set; wherein, the training set and the validation set each include multiple samples, and each sample includes sample data and a sample label; Inputting the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model; wherein, the initial remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; Validating the remaining useful life prediction model after the training is completed through the validation samples in the validation set.

6. The method for predicting the remaining service life of an aeroengine according to claim 5, characterized in that, The step of inputting the training samples in the training set into an initial remaining useful life prediction model for iterative training until the training is completed to obtain the remaining useful life prediction model includes: Input the training samples in the training set into the initial remaining useful life prediction model to obtain a first remaining useful life prediction value and the reconstruction error of the autoencoder in the prediction layer; Calculate the mean square error loss value based on the first remaining useful life prediction value and the sample label, and calculate the training loss based on the mean square error loss value and the reconstruction error; Perform iterative training on the initial remaining useful life prediction model based on the training loss until the training is completed to obtain the remaining useful life prediction model.

7. The method for predicting the remaining service life of an aero-engine according to claim 5, characterized in that The verification of the remaining useful life prediction model after training is completed through the verification samples in the verification set includes: Input the verification samples in the verification set into the remaining useful life prediction model after training to obtain a second remaining useful life prediction value; Calculate the root mean square error and the prediction score based on the second remaining useful life prediction value and the sample label; Verify the remaining useful life prediction model based on the root mean square error and the prediction score.

8. A remaining service life prediction device for an aeroengine, characterized in that, It includes: A first acquisition module configured to acquire sensor data of an aero-engine; A prediction module configured to input the sensor data into a preset remaining useful life prediction model for prediction to obtain the predicted remaining useful life of the aero-engine; wherein, the remaining useful life prediction model includes a graph convolutional network, a path signature layer, and a prediction layer; The prediction module includes: A first extraction sub-module configured to extract spatial features from the sensor data based on the graph convolutional network; A second extraction sub-module configured to extract temporal features from the sensor data based on the path signature layer; A prediction sub-module configured to obtain the predicted remaining useful life based on the prediction layer through the fused features after fusing the spatial features and the temporal features.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting the remaining useful life of an aero-engine according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for predicting the remaining useful life of an aero-engine according to any one of claims 1 to 7.

Images