Aero-engine oil quantity prediction method based on EMD-CNN

Abstract

The application discloses an aero-engine oil quantity prediction method based on EMD-CNN, calculates correlation coefficients of engine state parameters, oil system parameters and other unmodeled factors and oil quantity based on a Spearman coefficient, selects parameters strongly correlated with the oil quantity as input parameters, decomposes the oil quantity signal by using empirical mode decomposition (EMD), obtains a plurality of intrinsic mode functions (IMFs), selects specific IMFs to reconstruct the signal, normalizes input data and oil quantity label data, builds a convolutional neural network, trains the neural network model by using test data, and obtains an oil quantity prediction model to predict the oil quantity of the aero-engine. The method can effectively solve the problem of difficulty in establishing a traditional oil quantity analysis model, has the advantage that a neural network can approximate any nonlinear function, and has the characteristic that a convolutional neural network has strong feature extraction capability, so that the oil quantity is predicted.

Description

Technical Field

[0001] This invention relates to the field of engine lubrication systems, and in particular to a method for predicting the lubrication quantity of aero-engines based on EMD-CNN. Background Technology

[0002] The transmission lubrication system is a crucial subsystem of the engine, responsible for providing sufficient, appropriately temperature-controlled, and clean lubricating oil during engine operation. It forms an oil film on the surfaces of transmission components, carrying away heat, preventing overheating, reducing surface damage, and lowering frictional resistance, thereby improving engine efficiency. The engine's onboard system monitors oil level parameters in real time and issues an alarm when warning conditions are met, thus alerting the pilot.

[0003] Currently, most domestic military aircraft lubricating oil systems are monitored using conventional methods that combine lubricating oil parameters (such as pressure, temperature, and level), oil filter interception, and visual observation. Although this method is simple and easy to implement, it is not sensitive enough, it is difficult to guarantee its accuracy during measurement, and the measurement is easily affected by the oil and gas environment, resulting in large measurement noise. It is also difficult to establish an analytical model for oil quantity.

[0004] To address this issue, there is an urgent need to develop a method to reduce the impact of the oil and gas environment, measurement uncertainty, and noise on oil level measurement.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To address the aforementioned technical problems, the objective of this invention is to provide a method for predicting the lubricating oil quantity of aero-engines based on EMD-CNN. This method utilizes historical test data to establish a lubricating oil quantity prediction model, and through the trained model, predicts the real-time lubricating oil quantity based on engine state parameters and lubricating oil system parameters.

[0007] To achieve the above objectives, the present invention adopts the following approach.

[0008] A method for predicting lubricating oil quantity in aero-engines based on EMD-CNN, the method comprising the following steps:

[0009] Step S100: Calculate the correlation coefficients between engine state parameters, lubricating oil system parameters, and other unmodeled factors and lubricating oil quantity signals based on Spearman coefficients, and select parameters that are strongly correlated with the lubricating oil quantity signal as input parameters;

[0010] Step S200: Decompose the lubricating oil quantity signal using Empirical Mode Decomposition (EMD) technology to obtain several intrinsic mode functions (IMFs); select the IMF according to requirements to reconstruct the lubricating oil quantity signal to reduce the impact of oil and gas environment and sensor measurement noise on the measurement.

[0011] Step S300: Perform max-min normalization on the input parameters and lubricating oil volume label data to eliminate the dimensional differences between features;

[0012] Step S400: Construct a CNN convolutional neural network, which includes: a convolutional layer for extracting features from the input parameters; and a fully connected layer for regressing the lubricating oil volume signal.

[0013] Step S500: The input parameters are used as inputs, and the lubricating oil quantity label data is used as outputs to train a convolutional neural network to obtain a lubricating oil quantity prediction model; and the lubricating oil quantity prediction model is used to predict the lubricating oil quantity of the aero-engine.

[0014] Optionally, step S100 includes:

[0015] Step S101: Select data obtained from a test run, including data under stable engine conditions, intermediate conditions, low afterburner, partial afterburner, and maximum afterburner conditions; and extract engine state parameters, lubrication system parameters, and other unmodeled factors.

[0016] Engine status parameters include: compressor speed, fan speed, throttle lever angle, compressor inlet total temperature, compressor inlet total pressure, fan inlet total temperature, fan inlet total pressure, and fuel flow rate;

[0017] Lubricating oil system parameters include: supply oil temperature, return oil temperature, lubricating oil pressure, and lubricating oil volume;

[0018] Other unmodeled factors include: altitude, atmospheric pressure, atmospheric temperature, angle of attack, and sideslip angle;

[0019] Step S102: Considering the distribution of lubricating oil data in multiple test runs, the Spearman correlation coefficient is selected to measure the correlation between each variable and the lubricating oil quantity. The Spearman coefficient can be used not only for non-normally distributed data but is also insensitive to outliers. The Spearman coefficient calculation formula is as follows:

[0020] (1)

[0021] In the formula, Represents the number of samples. Representing the Variables In variables The ranking in Representing the Variables In variables The ranking in Representative variable Rank of the average grade of all values Representative variable Rank of all values ​​in the average grade;

[0022] Step S103: Based on the calculation results of step S102, select the high-pressure rotor speed, low-pressure rotor speed, return oil temperature, fuel flow rate, lubricating oil pressure, fan inlet total temperature, fan inlet total pressure, and throttle lever angle as input parameters for subsequent steps.

[0023] Optionally, step S200 includes:

[0024] Step S201: Use Empirical Mode Decomposition (EMD) technology to decompose the lubricating oil quantity signal and extract the intrinsic mode function and residual signal;

[0025] Step S202: Calculate the lubricating oil quantity signal All local maxima and local minima are analyzed using spline interpolation to construct the upper envelope and... Lower envelope ;

[0026] calculate and average :

[0027] (2)

[0028] Step S203: Use the lubricating oil quantity signal minus The first decomposed signal is obtained. ;

[0029] (3)

[0030] Step S204: Determine the decomposition signal Does the intrinsic mode function (IMF) condition satisfy the given condition? If so, decompose the signal. This is the first IMF component; if it does not meet the requirements, then... Repeat step S202 as the decomposed signal;

[0031] Step S205: Continue the decomposition process as described above until the IMF convergence condition is met. The IMF convergence condition is:

[0032] (4)

[0033] This is a constant, usually taken as 0.2 or 0.3. and Representing the first The eigenmode functions of the second decomposition and the th The eigenmode functions of the second decomposition; after Lubricating oil volume signal after secondary decomposition Represented as:

[0034] (5)

[0035] In the formula, Representing the indivual Quantity, Indicates residual signal;

[0036] Step S206: In the EMD decomposition result, ignore the first few high-frequency IMF components and synthesize the subsequent low-frequency IMF components. The synthesized signal for:

[0037] (6)

[0038] Optionally, step S300 includes:

[0039] Step S301: Normalize the maximum and minimum values ​​of the input parameters in step S100 to scale the data to the range of [0,1].

[0040] (7)

[0041] In the formula, For the first Measured values ​​of each input variable This represents the maximum value of the input variable. The minimum value of the input variable; For the first The normalized values ​​of the input variables;

[0042] Step S302: Perform maximum and minimum value normalization on the lubricating oil volume signal reconstructed in step S200, and scale the data to the range of [0,1].

[0043] (8)

[0044] In the formula, For the first Measured values ​​of each output variable To output the maximum value of the variable; To output the minimum value of the variable; For the first The normalized values ​​of the output variables.

[0045] Optionally, step 400 includes:

[0046] Step 401: Construct the dataset , This is a dataset from a certain engine ground test. for A two-dimensional matrix, The number of input features is 10. The data is normalized lubricating oil volume label data, of which 70% of the data is used as the training set and 30% as the test set;

[0047] Step 402: Construct a convolutional neural network; convolutional layers can automatically extract important features from input data. Convolutional layers use local receptive fields, focusing only on local areas of the input data, enabling the network to capture local features; the input of the convolutional neural network model is 8-dimensional, including: compressor speed, fan speed, lubricating oil temperature, lubricating oil pressure, fan inlet total temperature, fuel flow rate, fan inlet total pressure, and throttle lever angle; the output is the lubricating oil quantity;

[0048] Convolution operation is:

[0049] (9)

[0050] In the formula, and These are the positional variables of the input and output features, respectively. For the first Layered networks; Network layer number; operators This represents the input variables for each layer. With weight variables Performing a convolution operation yields the corresponding result. Output feature mapping , Indicates bias;

[0051] Step 403: Construct the activation function; after convolution calculation, introduce... The function is used as an activation function; the activation function as follows:

[0052] (10)

[0053] Step 404: Construct fully connected layers; flatten the output of the last convolutional layer into a one-dimensional array and use it as the input to the fully connected layer; construct a three-layer fully connected network as the final lubricant volume regression layer, with 32, 16, and 1 neurons in each layer, and select the activation function... The fully connected network layer operations are as follows:

[0054] (11)

[0055] In the formula, For the input value of the k-th layer, For weight, This is the output of the k-th layer;

[0056] Step 405: The output object is the amount of lubricating oil, which is a regression task. Mean Squared Error is used as part of the loss function. Additionally, to prevent overfitting, an L1 norm penalty term for the weights is added to the loss function. The total loss function is as follows:

[0057] (12)

[0058] In the formula, For the true value of the sample, For the sample predicted value, These are the network weight coefficients. is the coefficient of the regularization term.

[0059] Optionally, step S500 includes:

[0060] Step 501: Select an L1 regularization coefficient of 0.01, randomly discard 40% of the neurons in each fully connected layer; set the batch size to 32, the epoch to 50, the learning rate to 0.0001, and select the Adam optimizer to optimize the weights and bias parameters through backpropagation.

[0061] Step 502: To evaluate the model's performance, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination are used. As an evaluation indicator:

[0062] (13)

[0063] (14)

[0064] (15)

[0065] in, Represents the actual value. Represents the predicted value. The average of the actual values.

[0066] A computer-readable storage medium for storing a computer program configured to implement the method when invoked by a processor.

[0067] An electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program. Compared with the prior art, the present invention has the following beneficial technical effects:

[0068] Based on engine state parameters and lubricating oil system parameters, a predictive model for lubricating oil quantity was established using a neural network method. This model has good generalization ability and can realize the evolution from monitoring the lubricating oil quantity status to predicting the lubricating oil quantity.

[0069] This invention predicts the oil quantity of aero-engine lubricating oil, providing a fundamental, feasible, and effective oil quantity prediction model for numerous research works on engine lubricating oil system parameter state estimation and practical applications. Attached Figure Description

[0070] The accompanying drawings illustrate exemplary embodiments of the invention and, together with the description thereof, serve to explain the principles of the invention. These drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification.

[0071] Figure 1 This is a flowchart of an aero-engine lubricating oil quantity prediction based on an EMD-CNN deep neural network according to an embodiment of this disclosure;

[0072] Figure 2 This is a correlation heatmap of test data provided in one embodiment of this disclosure;

[0073] Figure 3 This is an empirical mode decomposition diagram of the lubricating oil quantity signal provided in one embodiment of this disclosure;

[0074] Figure 4 This is a graph showing the correlation coefficients between various IMF components and the lubricating oil volume signal, provided in one embodiment of this disclosure.

[0075] Figure 5 This is a noise-reduced lubricating oil volume signal diagram provided in one embodiment of this disclosure;

[0076] Figure 6 This is a diagram of the lubricating oil volume prediction model architecture provided in one embodiment of this disclosure;

[0077] Figure 7 This is a graph showing the relationship between the loss value and the number of iterations provided in one embodiment of this disclosure;

[0078] Figure 8 These are the predicted and actual values ​​of the EMD-CNN model provided in one embodiment of this disclosure on the new test data dataset1;

[0079] Figure 9 These are the predicted and actual values ​​of the EMD-CNN model provided in one embodiment of this disclosure on the new test data dataset2;

[0080] Figure 10 This is a density scatter plot of the SVR (Support Vector Machine Regression) method provided in one embodiment of this disclosure on test data dataset1 and dataset2;

[0081] Figure 11 This is a density scatter plot of the MLP (Multilayer Perceptron) method provided in one embodiment of this disclosure on test data dataset1 and dataset2;

[0082] Figure 12 This is a density scatter plot of the EMD-CNN method provided in one embodiment of this disclosure on test data dataset1 and dataset2. Detailed Implementation

[0083] The following is in conjunction with the appendix Figures 1 to 12 The present invention will be further described in detail below with reference to the embodiments. It is to be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the present invention are shown in the accompanying drawings.

[0084] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The technical solution of this invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0085] Unless otherwise stated, the exemplary embodiments / exemplifications shown are to be understood as providing exemplary features of various details that provide ways in which the technical concept of the invention can be implemented in practice. Therefore, unless otherwise stated, the features of the various embodiments / exemplifications may be additionally combined, separated, interchanged and / or rearranged without departing from the technical concept of the invention.

[0086] The use of crosshairs and / or shading in the accompanying drawings is generally used to clarify the boundaries between adjacent components. Thus, unless otherwise stated, the presence or absence of crosshairs or shading does not convey or indicate any preference or requirement for the specific material, material properties, dimensions, proportions, commonalities between the illustrated components, or any other characteristics, properties, etc., of the components. Furthermore, in the accompanying drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0087] When a component is referred to as being "on" or "above" another component, "connected to," or "joined to" another component, the component may be directly on, directly connected to, or directly joined to the other component, or there may be intermediate components. However, when a component is referred to as being "directly on" another component, "directly connected to," or "directly joined to" another component, there are no intermediate components. Therefore, the term "connection" can refer to a physical connection, an electrical connection, etc., and may or may not have intermediate components.

[0088] For descriptive purposes, the present invention may use spatial relative terms such as “below,” “under,” “below,” “down,” “above,” “above,” “higher,” and “side (e.g., in a “sidewall”)” to describe the relationship between one component and another component as shown in the accompanying drawings. In addition to the orientations depicted in the drawings, the spatial relative terms are also intended to encompass different orientations of the device during use, operation, and / or manufacture. For example, if the device in the drawings is flipped, a component described as “below” or “under” another component or feature would subsequently be positioned “above” said other component or feature. Thus, the exemplary term “below” can encompass both “above” and “below” orientations. Furthermore, the device may be otherwise positioned (e.g., rotated 90 degrees or in other orientations), thus interpreting the spatial relative descriptive terms used herein accordingly.

[0089] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0090] In one embodiment, such as Figure 1 As shown, this invention provides a method for predicting the lubricating oil quantity of aero-engines based on EMD-CNN, comprising the following steps:

[0091] Step S100: Calculate the correlation coefficients between engine state parameters, lubricating oil system parameters, and other unmodeled factors and lubricating oil quantity signals based on Spearman coefficients, and select parameters that are strongly correlated with the lubricating oil quantity signal as input parameters;

[0092] Step S200 uses Empirical Mode Decomposition (EMD) technology to decompose the lubricating oil quantity signal to obtain several intrinsic mode functions (IMFs); high-frequency noise and low-frequency noise components are separated, and a specific IMF is selected according to the requirements to reconstruct the lubricating oil quantity signal, so as to reduce the influence of oil and gas environment and sensor measurement noise on the measurement.

[0093] Step S300: Perform max-min normalization on the input parameters and lubricating oil volume label data to eliminate the dimensional differences between features;

[0094] Step S400: Construct a convolutional neural network, which includes: a convolutional layer for extracting features from the input parameters; and a fully connected layer for regressing the lubricating oil volume signal.

[0095] Step S500: Train the convolutional neural network using historical test data to obtain a lubricating oil quantity prediction model; and use the lubricating oil quantity prediction model to predict the lubricating oil quantity of the aero-engine on new test data. The new test data refers to data that the model has never seen before.

[0096] Step S100 includes:

[0097] Step S101: Select data obtained from a test run. The test run data includes data under the following conditions: engine stable state, intermediate state, small afterburner, partial afterburner, and maximum afterburner. Extract engine state parameters, lubricating system parameters, and other unmodeled factors.

[0098] Engine status parameters include: compressor speed, fan speed, throttle lever angle, compressor inlet total temperature, compressor inlet total pressure, fan inlet total temperature, fan inlet total pressure, and fuel flow rate.

[0099] The parameters of the lubricating oil system include: oil supply temperature, oil return temperature, lubricating oil pressure, and lubricating oil volume.

[0100] Other unmodeled factors include: altitude, atmospheric pressure, atmospheric temperature, angle of attack, and sideslip angle.

[0101] Step S102: Considering the distribution of lubricating oil data in multiple test runs, the Spearman correlation coefficient is selected to measure the correlation between each variable and the lubricating oil quantity. The Spearman coefficient can be used not only for non-normally distributed data but is also insensitive to outliers. The Spearman coefficient calculation formula is as follows:

[0102] (1)

[0103] In the formula Represents the number of samples. Representing the Variables In variables The ranking in Representing the Variables In variables The ranking in Representative variable Rank of the average grade of all values Representative variable Ranked by the average grade of all values.

[0104] Step S103: Based on the calculation results of step S102, select the high-pressure rotor speed, low-pressure rotor speed, return oil temperature, fuel flow rate, lubricating oil pressure, fan inlet total temperature, fan inlet total pressure, and throttle lever angle as input parameters for subsequent steps.

[0105] Figure 2 This is a correlation heatmap of the test data. Through the correlation heatmap, we can see the correlation between the lubricating oil quantity signal and physical quantities. The compressor speed, fan speed, lubricating oil pressure, fuel flow rate and throttle lever angle are all strongly negatively correlated with the lubricating oil quantity, while the fan intake total temperature and intake total pressure are both strongly positively correlated.

[0106] Step S200 includes:

[0107] Step S201: During actual engine operation, the surrounding environment will introduce noise of varying degrees, and the sensor itself will also have measurement noise. Moreover, the measurement of lubricating oil quantity is easily affected by the oil and gas environment, resulting in measurement errors. Therefore, the empirical mode decomposition technology is used to decompose the lubricating oil quantity signal and extract the intrinsic mode function and residual signal.

[0108] Step S202: Calculate the lubricating oil quantity signal All local maxima and local minima are analyzed using spline interpolation to construct the upper envelope and... Lower envelope ;

[0109] calculate and average :

[0110] (2)

[0111] Step S203: Use the lubricating oil quantity signal minus The first decomposed signal is obtained. ;

[0112] (3)

[0113] Step S204: Determine the first decomposition signal Does it satisfy the conditions for the Intrinsic Mode Function (IMF)? If it does, then it is the first IMF component; otherwise, it will be... Repeat step S202 as the decomposed signal;

[0114] Step S205: Continue the decomposition process as described above until the IMF convergence condition is met. The IMF convergence condition is:

[0115] (4)

[0116] This is a constant, usually taken as 0.2 or 0.3. and Representing the first The eigenmode functions of the second decomposition and the th The eigenmode functions of the second decomposition. The original signal after decomposition Represented as:

[0117] (5)

[0118] In the formula, Representing the indivual Quantity, This indicates a residual signal.

[0119] Step S206: In the EMD decomposition results, calculate the correlation coefficient between each IMF component and the original lubricating oil volume signal. The IMF component with the first minimum value is defined as the transition IMF component. IMF components before the transition IMF component are defined as high-frequency noise components, and those after are defined as low-frequency components. The correlation coefficient is calculated as follows:

[0120] (6)

[0121] In the formula, Representing the indivual Quantity, Represents the original lubricating oil signal. represent and Covariance between represent variance represent The variance.

[0122] Step S207: Ignoring the first few high-frequency IMF components and synthesizing the subsequent low-frequency IMF components effectively removes high-frequency noise interference, thereby reconstructing a smoother and more trend-oriented signal. This method can preserve the main information of the signal while suppressing noise, achieving signal denoising. The reconstructed signal... for:

[0123] (7)

[0124] In the formula, the first five IMF components are ignored.

[0125] Figure 3 This is the empirical mode decomposition diagram of the lubricating oil quantity signal. The lubricating oil quantity signal is decomposed into 12 IMF components and a residual signal. The IMF components are sorted in descending order of frequency. It can be seen that the first 5 IMF components contain high-frequency noise. Figure 4 This is a graph showing the correlation coefficients between each IMF component and the lubricating oil volume signal. A minimum value appears in the 5th IMF component, which is the transition IMF component. When reconstructing the lubricating oil volume signal, all IMF components following the transition IMF component are selected for reconstruction. Figure 5 This is a diagram of the lubricating oil volume signal after noise reduction.

[0126] Step S300 includes:

[0127] Step S301: Normalize the maximum and minimum values ​​of the input parameters in step S100 to scale the data to the range of [0,1].

[0128] (8)

[0129] In the formula, For the first Measured values ​​of each input variable This represents the maximum value of the input variable. The minimum value of the input variable; For the first The normalized values ​​of the input variables.

[0130] Step S302: Perform maximum and minimum value normalization on the lubricating oil quantity signal obtained after reconstruction in step S200, and scale the data to the range of [0,1].

[0131] (9)

[0132] In the formula, For the first Measured values ​​of each output variable To output the maximum value of the variable; To output the minimum value of the variable; For the first The normalized values ​​of the output variables.

[0133] Step 400 includes:

[0134] Step 401: Construct a dataset based on the input parameters obtained in step S100. , This is a dataset from a certain engine ground test. for A two-dimensional matrix, The input feature number is 8, including the normalized state of each engine and lubricating system parameters in step S301. The data consists of normalized lubricating oil volume labels, with 70% of the data used as the training set and 30% as the test set.

[0135] Step 402: Construct a convolutional neural network; convolutional layers can automatically extract important features from the input data. Using a local receptive field, the convolutional layer focuses only on local areas of the input data, enabling the network to capture local features. The convolutional neural network model has an 8-dimensional input, including: compressor speed, fan speed, lubricating oil temperature, lubricating oil pressure, fan inlet total temperature, fuel flow rate, fan inlet total pressure, and throttle lever angle; the output is the lubricating oil quantity.

[0136] This example constructs a three-layer one-dimensional convolutional neural network. All convolutional kernels are (3, 1), with a stride of 1. After the convolution operation, zeros are used to pad the input and output dimensions to maintain consistency. The first convolutional layer has 16 kernels, the second has 32 kernels, and the third has 32 kernels. The convolution operation is as follows:

[0137] (10)

[0138] In the formula, and These are the positional variables of the input and output features, respectively. For the first Layered networks; Network layer number; operators This represents the input variables for each layer. With weight variables Performing a convolution operation yields the corresponding result. Output feature mapping , Indicates bias.

[0139] Step 403: Construct the activation function; after convolution calculation, to enhance the nonlinearity of the neural network, an activation function is introduced. The function is used as the activation function. Activation functions can effectively alleviate the vanishing gradient problem, have sparse activation properties, and can accelerate convergence. as follows:

[0140] (11)

[0141] Step 404: Construct a fully connected layer; flatten the output of the last convolutional layer into a one-dimensional array, which serves as the input to the fully connected layer. The output of the last convolutional layer is 10×32, which is flattened into a vector of length 320 and connected to the neurons in the first layer of the fully connected layer. The output of the first layer of the fully connected layer is then connected to the neurons in the second layer, and so on. Construct a three-layer fully connected network as the final lubricant volume regression layer, with 32, 16, and 1 neurons in each layer, and select the appropriate activation function. The fully connected network layer operations are as follows:

[0142] (12)

[0143] In the formula For the input value of the k-th layer, For weight, This is the output of layer k.

[0144] Step 405: The output object is the amount of lubricating oil, which is a regression task. Mean Squared Error (MSE) is used as part of the loss function. Additionally, to prevent overfitting, an L1 norm penalty term for the weights is added to the loss function to limit model complexity and improve generalization ability. L1 regularization has sparsity, causing some weights to become zero, meaning unimportant features are automatically ignored. The total loss function is as follows:

[0145] (13)

[0146] In the formula, For the true value of the sample, For the sample predicted value, These are the network weight coefficients. is the coefficient of the regularization term.

[0147] Figure 6 This is the architecture diagram of the lubricating oil volume prediction model, which includes: a data preprocessing module, a feature extractor, and a lubricating oil volume regressor. The data preprocessing module is used to perform feature selection on the test data and to reduce noise and normalize the lubricating oil volume data; the feature extractor is used to extract useful features from the data; and the lubricating oil volume regressor is used to predict the lubricating oil volume.

[0148] Step S500 includes:

[0149] Step 501: The L1 regularization coefficient is set to 0.01, and 40% of the neurons in each fully connected layer are randomly dropped to reduce the risk of overfitting and enhance the model's generalization ability. The batch size is set to 32, the epoch is set to 50, the learning rate is 0.0001, and the Adam optimizer is selected to optimize the weights and biases through backpropagation. The weights and biases obtained at this stage are a set of hyperparameters obtained after multiple training iterations. Figure 7 This is a graph showing the relationship between the loss value and the number of iterations.

[0150] Step 502: To evaluate the model's performance, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination are used. As an evaluation indicator:

[0151] (14)

[0152] (15)

[0153] (16)

[0154] in, Represents the actual value. Represents the predicted value. The average of the actual values.

[0155] In the lubricating oil prediction task, R 2 A score below 0.85 is considered unacceptable. If an unacceptable score is found, the learning rate, L1 regularization coefficient, batch size, number of convolutional kernels, and number of neurons in the fully connected neural network are adjusted continuously according to the specific task.

[0156] The prediction results are as follows. Tables 1 to 3 below show the generalization performance of the model on the test set, the new test data dataset1, and the new test data dataset2. Comparing the evaluation metrics of the three methods SVR (Support Vector Machine Regression), MLP (Multilayer Perceptron), and the EMD-CNN method of this invention on the test set and two different test datasets (dataset1 and dataset2), the results show that the EMD-CNN method performs best in all cases. On dataset1, the MSE, MAE, and R² of the EMD-CNN method are 0.0010, 0.0188, and 0.9612, respectively, all superior to SVR and MLP. On dataset2, EMD-CNN also outperforms the other two methods. Overall, EMD-CNN demonstrates superior prediction accuracy across all metrics, especially within a smaller error range, proving its effectiveness and reliability on the test dataset. The results show that the EMD-CNN method can effectively predict the amount of lubricating oil. This invention uses Spearman correlation analysis for feature selection, which can solve the problem of large input data dimensionality in traditional neural network prediction of lubricating oil volume. By removing high-frequency noise from the data through empirical mode decomposition, a one-dimensional convolutional neural network can extract deep features from the input data, thereby achieving the prediction of lubricating oil volume.

[0157] Table 1 Evaluation metrics of our method and other methods on the test dataset.

[0158]

[0159] Table 2 Evaluation metrics of this method and other methods on new test data dataset 1

[0160]

[0161] Table 3 Evaluation metrics of this method and other methods on new test data dataset2

[0162]

[0163] Figure 8These are the predicted and actual values ​​of the publicly disclosed EMD-CNN method model on the new test data dataset1; it can be seen that the model still has good generalization performance on previously unseen test data. It also shows good prediction performance during engine acceleration and deceleration. Figure 9 These are the predicted and actual values ​​of the EMD-CNN model on the new test data dataset2; it can be seen that the model still has good generalization performance on previously unseen test data. It also shows good prediction performance during engine acceleration and deceleration.

[0164] Figure 10 This is a density scatter plot of the SVR (Support Vector Machine Regression) method on the test data dataset1 and dataset2. It can be seen that during the test, the lubricating oil volume data is mainly concentrated in 0.2-0.5 (after normalization). The predicted values ​​and the actual values ​​are relatively dispersed and do not cluster near the diagonal. Moreover, most of the predicted values ​​are higher than the actual values. The points in the figure are biased towards the upper part of the diagonal. Figure 11 This is a density scatter plot of the MLP (Multilayer Perceptron) method on test data dataset1 and dataset2. Figure 12 This is a density scatter plot of the publicly disclosed EMD-CNN method on test data sets 1 and 2. It can be seen that most data points are concentrated near the diagonal, the model's predictions are very close to the true values, the error is small, and the model performs well.

[0165] In one embodiment, the present invention also provides a computer-readable storage medium for storing a computer program configured to implement the method when invoked by a processor.

[0166] In one embodiment, the present invention also provides an electronic device, the electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein the processor implements the method when executing the program.

[0167] In the description of this specification, the references to terms such as "one embodiment / mode," "some embodiments / modes," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment / mode or example is included in at least one embodiment / mode or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment / mode or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments / modes or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments / modes or examples described in this specification, as well as the features of different embodiments / modes or examples.

[0168] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0169] Those skilled in the art should understand that the above embodiments are merely for illustrating the present invention and are not intended to limit the scope of the invention. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of the present invention.

Claims

1. A method for predicting the lubricating oil quantity of an aero-engine based on EMD-CNN, characterized in that, The method includes the following steps: Step S100: Calculate the correlation coefficients between engine state parameters, lubricating oil system parameters, and other unmodeled factors and lubricating oil quantity signals based on Spearman coefficients, and select parameters that are strongly correlated with the lubricating oil quantity signal as input parameters; Step S200: Decompose the lubricating oil quantity signal using Empirical Mode Decomposition (EMD) technology to obtain several intrinsic mode functions; The intrinsic mode function (IMF) is selected according to the requirements to reconstruct the lubricating oil quantity signal, so as to reduce the impact of oil and gas environment and sensor measurement noise on the measurement. Step S300: Perform max-min normalization on the input parameters and lubricating oil volume label data to eliminate the dimensional differences between features; Step S400: Construct a CNN convolutional neural network, which includes: a convolutional layer for extracting features from the input parameters; and a fully connected layer for regressing the lubricating oil volume signal. Step S500: The input parameters are used as inputs, and the lubricating oil quantity label data is used as outputs to train a convolutional neural network to obtain a lubricating oil quantity prediction model; and the lubricating oil quantity prediction model is used to predict the lubricating oil quantity of the aero-engine.

2. The method according to claim 1, characterized in that, Preferably, step S100 includes: Step S101: Select data obtained from a test run, including data under stable engine conditions, intermediate conditions, low afterburner, partial afterburner, and maximum afterburner conditions; and extract engine state parameters, lubrication system parameters, and other unmodeled factors. Engine status parameters include: compressor speed, fan speed, throttle lever angle, compressor inlet total temperature, compressor inlet total pressure, fan inlet total temperature, fan inlet total pressure, and fuel flow rate; Lubricating oil system parameters include: supply oil temperature, return oil temperature, lubricating oil pressure, and lubricating oil volume; Other unmodeled factors include: altitude, atmospheric pressure, atmospheric temperature, angle of attack, and sideslip angle; Step S102: Considering the distribution of lubricating oil data in multiple test runs, the Spearman correlation coefficient is selected to measure the correlation between each variable and the lubricating oil quantity. The Spearman coefficient can be used not only for non-normally distributed data but is also insensitive to outliers. The Spearman coefficient calculation formula is as follows: （1）， In the formula, Represents the number of samples. Representing the Variables In variables The ranking in Representing the Variables In variables The ranking in Representative variable Rank of the average grade of all values Representative variable Rank of all values ​​in the average grade; Step S103: Based on the calculation results of step S102, select the high-pressure rotor speed, low-pressure rotor speed, return oil temperature, fuel flow rate, lubricating oil pressure, fan inlet total temperature, fan inlet total pressure, and throttle lever angle as input parameters for subsequent steps.

3. The method according to claim 1, characterized in that, Step S200 includes: Step S201: Use Empirical Mode Decomposition (EMD) technology to decompose the lubricating oil quantity signal and extract the intrinsic mode function and residual signal; Step S202: Calculate the lubricating oil quantity signal All local maxima and local minima are analyzed using spline interpolation to construct the upper envelope and... Lower envelope ; calculate and average : （2）， Step S203: Use the lubricating oil quantity signal minus The first decomposed signal is obtained. ; （3）， Step S204: Determine the decomposition signal Does the intrinsic mode function (IMF) condition satisfy the given condition? If so, decompose the signal. This is the first IMF component; if it does not meet the requirements, then... Repeat step S202 as the decomposed signal; Step S205: Continue the decomposition process as described above until the IMF convergence condition is met. The IMF convergence condition is: （4）， This is a constant, usually taken as 0.2 or 0.

3. and Representing the first The eigenmode functions of the second decomposition and the th The eigenmode functions of the second decomposition; after Lubricating oil volume signal after secondary decomposition Represented as: （5）， In the formula, Representing the indivual Quantity, Indicates residual signal; Step S206: In the EMD decomposition result, ignore the first few high-frequency IMF components and synthesize the subsequent low-frequency IMF components. The synthesized signal for: （6）。 4. The method according to claim 3, characterized in that, Step S300 includes: Step S301: Normalize the maximum and minimum values ​​of the input parameters in step S100 to scale the data to the range of [0,1]. （7）， In the formula, For the first Measured values ​​of each input variable This represents the maximum value of the input variable. The minimum value of the input variable; For the first The normalized values ​​of the input variables; Step S302: Perform maximum and minimum value normalization on the lubricating oil volume signal reconstructed in step S200, and scale the data to the range of [0,1]. （8）， In the formula, For the first Measured values ​​of each output variable To output the maximum value of the variable; To output the minimum value of the variable; For the first The normalized values ​​of the output variables.

5. The method according to claim 4, characterized in that, Step 400 includes: Step 401: Construct the dataset , This is a dataset from a certain engine ground test. for A two-dimensional matrix, This is the normalized lubricating oil volume label data; Step 402: Construct a convolutional neural network; the convolution operation is as follows: （9）， In the formula, and These are the positional variables of the input and output features, respectively. For the first Layered networks; Network layer number; operators This represents the input variables for each layer. With weight variables Performing a convolution operation yields the corresponding result. Output feature mapping , Indicates bias; Step 403: Construct the activation function; after convolution calculation, introduce... The function is used as the activation function; Step 404: Construct fully connected layers; flatten the output of the last convolutional layer into a one-dimensional array and use it as the input to the fully connected layer; construct a three-layer fully connected network as the final lubricating oil volume regression layer, and select the activation function. ; Step 405: The output object is the lubricating oil volume, using the mean squared error as part of the loss function.

6. The method according to claim 5, characterized in that, Step S500 includes: Step 501: The L1 regularization coefficient is set to 0.01, and some neurons in each fully connected layer are randomly dropped; the weights and bias parameters are optimized through backpropagation. Step 502: To evaluate the model's performance, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination are used. As an evaluation indicator: （10）， （11）， （12）， in, Represents the actual value. Represents the predicted value. The average of the actual values.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program configured to implement the method of any one of claims 1-6 when invoked by a processor.

8. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor; wherein, when the processor executes the program, it implements the method of any one of claims 1-6.

Images