Method and apparatus for estimating quantities of manufacturing resources for an aeroengine

Abstract

A method and apparatus for estimating aeroengine manufacturing resource quantities is disclosed. The method can include receiving a historical data set for aeroengines, the historical data set including a plurality of performance parameters associated with each of a plurality of aeroengines and a manufacturing resource quantity used for each aeroengine; generating virtual data based on the historical data set, including generating a set of random values corresponding to the plurality of performance parameters and an interpolated manufacturing resource quantity interpolated from the historical data set; and training an aeroengine manufacturing resource quantity estimation model based at least on the virtual data. The trained aeroengine manufacturing resource quantity estimation model can be used to generate an estimated manufacturing resource quantity for a prospective aeroengine.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aero-engines, and in particular to an aero-engine manufacturing resource quantity estimation method and device. BACKGROUND

[0002] An aero-engine is a highly complex device, and a single aero-engine can contain tens of thousands of components. Due to different requirements for material selection, surface roughness, dimensional tolerance, and processing technology, the manufacturing resource requirements of each type of component will be quite different. Insufficient or shortage of key manufacturing resources will directly cause the aero-engine development and trial production to be stalled. On the contrary, excessive storage of resources will also cause waste, and storage, maintenance, and inventory management will also consume a large amount of funds for the enterprise. If the manufacturing resource quantity of an aero-engine can be estimated before or during the design of the aero-engine, raw materials can be stored as needed, and the risk of delay in the development of the aero-engine can be reduced.

[0003] There are three main types of existing aero-product manufacturing resource quantity estimation methods: analogy method, engineering method, and parameter method. The analogy method is based on the experience of experts and the resource quantity information of similar models, and the estimation accuracy depends on the judgment of the experts, and the stability of the estimation results is also not high. The engineering method is based on the existing engine manufacturing, and needs to collect and accumulate from the bottom unit, which is time-consuming and requires a lot of manpower, financial and material resources, so it is rarely used in the development stage. The parameter method is usually based on multiple linear regression models and exponential regression models to fit the data, which has the advantages of simple data collection, fast calculation speed, and the ability to conduct sensitivity analysis on the technical parameters of the engine based on the regression model. However, most of the existing resource quantity estimation methods based on the parameter method are only applicable to aircraft or specific types of aero-engines, and there is currently no widely applicable manufacturing resource quantity estimation method for aero-engines.

[0004] The design and manufacturing of aero-engines have very high confidentiality requirements and technical complexity, and the types and quantities of manufacturing resources required in each link are not disclosed, and can only be accumulated by the manufacturing enterprises themselves. When the number of aero-engine models produced by the manufacturing enterprise is small, the sample size of the manufacturing resource quantity data that can be accumulated will be small. In the case of small sample size and low quality of the aero-engine manufacturing resource usage samples obtained, the estimation accuracy and stability of the existing methods are poor.

[0005] Therefore, there is a need for an improved aero-engine manufacturing resource quantity estimation method and device. SUMMARY

[0006] The present application provides an aero-engine manufacturing resource amount estimation method and device, which can expand the sample capacity by using a virtual data generation method on the basis of the use of existing aero-engine manufacturing resources, and establish an engine manufacturing resource amount estimation model based on a neural network. By using the engine manufacturing resource amount estimation model, the estimated manufacturing resource amount of the expected aero-engine can be generated based on the performance parameters of the expected aero-engine, so as to facilitate and simplify the design process of the aero-engine.

[0007] In one embodiment of the present application, an aero-engine manufacturing resource amount estimation method is provided, which includes: receiving a historical data set of aero-engines, the historical data set including a plurality of performance parameters associated with each of a plurality of aero-engines and a manufacturing resource amount used by each aero-engine; generating virtual data based on the historical data set, wherein generating the virtual data includes: generating a set of random values corresponding to the plurality of performance parameters, sorting each aero-engine in the historical data set and the virtual data for the plurality of performance parameters based on the degree of association between each performance parameter in the plurality of performance parameters and the manufacturing resource amount to determine the sorting position of the virtual data in the historical data set, and generating an interpolated manufacturing resource amount of the virtual data according to the manufacturing resource amounts associated with two or more aero-engines adjacent to the virtual data in the historical data set; and training an aero-engine manufacturing resource amount estimation model based at least in part on the virtual data containing the set of random values and the interpolated manufacturing resource amount.

[0008] In one aspect, the plurality of performance parameters of the aero-engine includes two or more of the following: take-off thrust, bypass ratio, total pressure ratio, turbine inlet temperature, specific fuel consumption, air flow, fan diameter, engine overall length, engine net weight.

[0009] In one aspect, the aero-engine manufacturing resource amount includes one or more of the following: manufacturing time consumption, number of parts, number of manufacturing processes, amount of grinding tool / cutter required, or a weighted value of two or more of the above.

[0010] In one aspect, the aero-engine manufacturing resources include at least a first manufacturing resources and a second manufacturing resources, and generating the virtual data further includes: determining a first sorting position of the virtual data in the historical dataset based on the correlation between each of the plurality of performance parameters and the first manufacturing resources, and generating a first interpolated manufacturing resources of the virtual data based on the first manufacturing resources associated with two or more aero-engines adjacent to the virtual data in the historical dataset; and determining a second sorting position of the virtual data in the historical dataset based on the correlation between each of the plurality of performance parameters and the second manufacturing resources, and generating a second interpolated manufacturing resources of the virtual data based on the second manufacturing resources associated with two or more aero-engines adjacent to the virtual data in the historical dataset.

[0011] On the one hand, the correlation between each performance parameter and the amount of manufacturing resources is predetermined or determined using grey relational analysis based on the historical dataset.

[0012] On one hand, the range of random values ​​in the virtual data corresponding to the plurality of performance parameters is selected from at least one of the following: within a predetermined range of the corresponding performance parameter; and between the upper and lower bounds of the value of the corresponding performance parameter in the original dataset.

[0013] In one aspect, the method for estimating the amount of aero-engine manufacturing resources also includes: testing the trained aero-engine manufacturing resource estimation model based on at least a portion of the historical dataset.

[0014] In one aspect, the method for estimating the manufacturing resources of an aero-engine further includes: inputting multiple performance parameters associated with the expected aero-engine into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the expected aero-engine.

[0015] On one hand, the aero-engine manufacturing resource estimation model includes feedforward neural networks, recurrent neural networks, or symmetric connected neural networks.

[0016] In one embodiment of the present invention, a computer-readable medium storing a computer program is provided, which, when executed by a processor, causes the processor to perform the method described above.

[0017] In one embodiment of the present invention, an aero-engine manufacturing resource estimation apparatus is provided, comprising: a data receiving component that receives a historical dataset of aero-engines, the historical dataset including multiple performance parameters associated with each of a plurality of aero-engines and the amount of manufacturing resources used by each aero-engine; a virtual data generation component that generates virtual data based on the historical dataset, wherein generating the virtual data includes: generating a set of random values ​​corresponding to the plurality of performance parameters; sorting each aero-engine in the historical dataset and the virtual data for multiple key values ​​for the plurality of performance parameters based on the correlation between each of the plurality of performance parameters and the amount of manufacturing resources to determine the sorting position of the virtual data in the historical dataset; and generating an interpolated amount of manufacturing resources for the virtual data based on the amount of manufacturing resources associated with two or more aero-engines adjacent to the virtual data in the historical dataset; and a model training component that trains an aero-engine manufacturing resource estimation model based at least in part on the virtual data containing the set of random values ​​and the interpolated amount of manufacturing resources.

[0018] On one hand, the various performance parameters of the aero-engine include two or more of the following: takeoff thrust, bypass ratio, overall pressure ratio, turbine inlet temperature, fuel consumption rate, airflow, fan diameter, overall engine length, and engine net weight.

[0019] On one hand, the amount of aero-engine manufacturing resources includes one or more of the following: manufacturing time, number of parts, number of manufacturing processes, demand for molds / tools, or a weighted average of two or more of the above.

[0020] In one aspect, the aero-engine manufacturing resources include at least a first manufacturing resource quantity and a second manufacturing resource quantity, and the virtual data generation component is configured to: determine a first sorting position of the virtual data in the historical dataset based on the correlation between each of the plurality of performance parameters and the first manufacturing resource quantity; and generate a first interpolated manufacturing resource quantity of the virtual data based on the first manufacturing resource quantity associated with two or more aero-engines adjacent to the virtual data in the historical dataset; and determine a second sorting position of the virtual data in the historical dataset based on the correlation between each of the plurality of performance parameters and the second manufacturing resource quantity; and generate a second interpolated manufacturing resource quantity of the virtual data based on the second manufacturing resource quantity associated with two or more aero-engines adjacent to the virtual data in the historical dataset.

[0021] On the one hand, the correlation between each performance parameter and the amount of manufacturing resources is predetermined or determined using grey relational analysis based on the historical dataset.

[0022] In one aspect, the aero-engine manufacturing resource estimation device further includes a prediction component that inputs multiple performance parameters associated with the expected aero-engine into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the expected aero-engine.

[0023] This invention can solve one or more of the following technical problems:

[0024] (1) When the sample size of collected aero-engine manufacturing resource usage is usually small, it is not conducive to carrying out manufacturing resource quantity estimation. The virtual data generation method proposed in this invention can be used to expand the sample size and improve the stability of the estimation results.

[0025] (2) The excellent nonlinear mapping and generalization capabilities of neural networks are applied to the estimation of manufacturing resources, which improves the accuracy of the estimation of manufacturing resources.

[0026] (3) Estimate the amount of manufacturing resources required for the aero-engine before or during the design phase, and reserve raw materials as needed, which helps to reduce the risk of delays in the development of aero-engines. Attached Figure Description

[0027] Figure 1 This is a flowchart of a method for estimating the amount of manufacturing resources for an aero-engine according to an embodiment of the present invention.

[0028] Figure 2 This is an example of an implementation process for estimating the amount of manufacturing resources for an aero-engine according to an embodiment of the present invention.

[0029] Figure 3 This is a schematic diagram of the basic architecture of a BP neural network according to an embodiment of the present invention.

[0030] Figure 4 This is a schematic diagram illustrating the prediction results of an aero-engine manufacturing resource estimation method according to an embodiment of the present invention.

[0031] Figure 5 This is a comparative experiment on the predictive stability of an aero-engine manufacturing resource estimation method according to an embodiment of the present invention.

[0032] Figure 6 This is a functional block diagram of an aero-engine manufacturing resource estimation device according to an embodiment of the present invention. Detailed Implementation

[0033] The present invention will be further described below with reference to specific embodiments and accompanying drawings, but this should not be construed as limiting the scope of protection of the present invention.

[0034] This invention provides a method and apparatus for estimating the manufacturing resources required for aero-engines. Based on existing aero-engine manufacturing resource usage, it expands the sample size using virtual data generation methods, and then uses this virtual data to train an aero-engine manufacturing resource estimation model, improving the accuracy and stability of the estimation results. The input to the aero-engine manufacturing resource estimation model can include one or more performance parameters, and the output can include the estimated aero-engine manufacturing resources based on the input performance parameters. Using this model, the estimated manufacturing resources for a projected aero-engine can be generated based on its performance parameters, thereby facilitating and simplifying the aero-engine design process and reducing the risk of development delays.

[0035] This method and apparatus can be implemented using machine learning (ML). Machine learning is a crucial component of artificial intelligence and can encompass various machine learning techniques, such as CNN, RNN, LSTM, and GBDT. The operation of the machine learning model includes a training phase and an application phase. According to this disclosure, in the training phase, a virtual dataset can be used to train the engine manufacturing resource estimation model, and the model can optionally be validated or tested using a known dataset. In the application phase of the machine learning model, new performance parameter data can be input into the trained and tested engine manufacturing resource estimation model to obtain the engine manufacturing resource estimation result. Using the estimated engine manufacturing resources can effectively facilitate and simplify the design process of aero-engines.

[0036] Figure 1 This is a flowchart of a method for estimating the amount of manufacturing resources for an aero-engine according to an embodiment of the present invention.

[0037] Step 102: Receive the historical dataset of aero-engines. This dataset may include multiple performance parameters and manufacturing resource quantities associated with each of the multiple aero-engines. As an example, and not a limitation, the multiple performance parameters of an aero-engine may include two or more of the following: takeoff thrust, bypass ratio, overall pressure ratio, turbine inlet temperature, fuel consumption rate, airflow rate, fan diameter, overall engine length, engine net weight, etc. The manufacturing resource quantity for each aero-engine may be the collected resources used to manufacture that aero-engine. The manufacturing resource quantity for an aero-engine may include one or more of the following: manufacturing time, number of parts, number of manufacturing processes, mold / tool ​​requirements, or a weighted average of two or more of the above. This weighted average can reflect the overall manufacturing resource requirement level of the aero-engine. For example, for the critical component of an aero-engine—the high-pressure compressor—the weights of manufacturing time and number of parts can represent the direct costs required to produce a high-pressure compressor. The level of direct costs reflects the amount of manufacturing resources required to produce a high-pressure compressor.

[0038] In specific implementations, appropriate historical data can be selected as needed. For example, in one embodiment, the performance indicators of each data point in the historical dataset (corresponding to an aero-engine or a type of aero-engine) may include takeoff thrust, bypass ratio, overall pressure ratio, and turbine inlet temperature, while the manufacturing resource requirements may include manufacturing time. In another embodiment, the performance indicators of each data point in the historical dataset may include takeoff thrust and fan diameter, while the manufacturing resource requirements may include a weighted value of the number of parts and the number of manufacturing processes.

[0039] Step 104: Generate virtual data based on historical datasets. In one embodiment, generating virtual data may include: generating a set of random values ​​corresponding to multiple performance parameters; sorting each aero-engine in the historical dataset and the virtual data for multiple performance parameters based on the correlation between each performance parameter and the manufacturing resource quantity to determine the sorting position of the virtual data in the historical dataset; and generating interpolated manufacturing resource quantities for the virtual data based on the manufacturing resource quantities associated with two or more aero-engines adjacent to the virtual data in the historical dataset.

[0040] On the one hand, the range of random values ​​in the generated virtual data can be selected from at least one of the following: within a predetermined range of the corresponding performance parameter, and between the upper and lower bounds of the corresponding performance parameter in the original dataset.

[0041] On one hand, the correlation between each performance parameter and the manufacturing resource quantity can be predetermined or determined using grey relational analysis based on historical datasets. If the manufacturing resource quantity includes multiple parameters representing different manufacturing resources, the correlation between each performance parameter and different manufacturing resource quantities can be different. For example, if the performance parameters include the total boost ratio and fuel consumption rate of an aero-engine, and the manufacturing resource quantity includes manufacturing time and the number of parts, then the correlation between the total boost ratio and manufacturing time, the correlation between the total boost ratio and the number of parts, the correlation between the fuel consumption rate and manufacturing time, and the correlation between the fuel consumption rate and the number of parts can be received or determined. In one embodiment, grey relational analysis can be used to determine the absolute and relative correlation between each performance parameter and each manufacturing resource quantity, and the correlation between each performance parameter and each manufacturing resource quantity can be a weighted value of the absolute and relative correlation.

[0042] Therefore, for each manufacturing resource quantity to be interpolated, multiple aero-engines and virtual data in the historical dataset can be sorted by multi-key value based on the correlation between each performance parameter and that manufacturing resource quantity. Multi-key value sorting can begin by sorting according to the performance parameter with the highest correlation; if two or more data points have the same performance parameter, then sorting is done according to the second highest performance parameter, and so on. This yields the ranking position of the virtual data within the historical dataset. Generating the interpolated manufacturing resource quantity for the virtual data can include interpolation (e.g., weighted averaging) using the corresponding manufacturing resource quantities of one or more aero-engines preceding and following the virtual data. In one implementation, if the ranking position of the virtual data is higher (or lower) than all aero-engines in the original dataset, the interpolated manufacturing resource quantity for that virtual data can also be generated based on interpolation or data fitting.

[0043] If a piece of virtual data has multiple manufacturing resource quantities, the first ranking position of the virtual data in the historical dataset can be determined based on the first manufacturing resource quantity and its corresponding correlation, and the interpolated data for the first manufacturing resource quantity can be obtained accordingly. Then, the second ranking position of the virtual data in the historical dataset can be determined based on the second manufacturing resource quantity and its corresponding correlation, and the interpolated data for the second manufacturing resource quantity can be obtained accordingly. This process can be repeated to generate multiple interpolated manufacturing resource quantities for this single piece of virtual data. In one example, multiple pieces of virtual data can be generated similarly based on the historical dataset.

[0044] Historical datasets and the generated dummy datasets can be divided into training and test sets. In one example, the dummy datasets generated from the historical dataset can serve as the training set, while the historical dataset itself can be used as the test set. In another example, the historical dataset and dummy data can be divided into training and test sets in other ways, such as random partitioning, average partitioning, etc. For example, the test set may include a subset of the historical dataset and a subset of the dummy data, or it may include another subset of the historical dataset and another subset of the dummy data. In one example, the training set contains at least a portion of the dummy data, and the test set contains at least a portion of the historical dataset.

[0045] Step 106: Train the aero-engine manufacturing resource estimation model based on the training set. In one example, the aero-engine manufacturing resource estimation model can be trained at least partially based on dummy data containing random values ​​and interpolated manufacturing resource quantities. Model training is a cyclical iterative process of parameter learning and tuning, enabling the final aero-engine manufacturing resource estimation model to fit the training set data well. In one example, the aero-engine manufacturing resource estimation model includes feedforward neural networks, recurrent neural networks, or symmetric connected neural networks, etc.

[0046] Step 108: Validate the aero-engine manufacturing resource estimation model based on a test set. In one example, the trained aero-engine manufacturing resource estimation model is validated based on at least a portion of a historical dataset. If the validation passes (e.g., the accuracy of the model output reaches a threshold), the trained aero-engine manufacturing resource estimation model can be saved. Optionally, if the model fails validation, the process can return to step 104 to regenerate dummy data and train / validate the aero-engine manufacturing resource estimation model.

[0047] Optionally, during the model application phase, multiple performance parameters associated with the expected aero-engine can be input into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the expected aero-engine.

[0048] Figure 2 This is an example of an implementation process for estimating the amount of manufacturing resources for an aero-engine according to an embodiment of the present invention.

[0049] Step S1: Collection and preprocessing of aero-engine manufacturing resources (raw data). This involves collecting the technical parameters (e.g., performance parameters) of the aero-engine and the corresponding manufacturing resource usage. Preprocessing may include, for example, data cleaning, missing value handling, and data transformation. The collected manufacturing resource quantities can be organized into a dataset containing multiple data entries, each corresponding to an aero-engine, and including multiple performance parameters and manufacturing resource quantities for that aero-engine. The manufacturing resource quantities for each aero-engine can be known resource usage obtained through various channels, or estimated through other methods.

[0050] As an example, and not a limitation, specific technical parameters of an aero-engine may include: takeoff thrust, bypass ratio, overall pressure ratio, turbine inlet temperature, fuel consumption rate, airflow, fan diameter, overall engine length, and engine net weight. The manufacturing resources for an aero-engine may include one or more of the following: manufacturing time, number of parts, number of manufacturing processes, mold / tool ​​requirements, or a weighted average of two or more of the above.

[0051] Each data entry in this dataset can be represented as Data(Y) k ,X k ), where Y k The manufacturing resource quantity corresponding to the k-th engine or the k-th engine model, Y k =(y k1 ,y k2 ,…,y k9 Each sub-variable of ) corresponds to a quantified value of a manufacturing resource for that engine, where X k The technical parameters corresponding to the k-th engine or the k-th engine model, X k =(x k1 ,x k2 ,…,x k9 Each sub-variable in the table corresponds to the value of a technical parameter of the engine. Therefore, each data point can consist of the technical parameters of a known engine model and the corresponding manufacturing resources.

[0052] Step S2: Generate virtual data to expand the dataset, and divide the virtual data and the original data into training set and test set.

[0053] As an example and not a limitation, in one implementation, step S2 may include one or more of the following sub-steps:

[0054] Step S2.1: Randomly generate a set of performance parameters This can make The value of is between the upper and lower bounds of the corresponding technical parameter values ​​in the original dataset, that is...

[0055] Step S2.2: Transfer virtual data Input it into the original dataset, and then categorize it according to the correlation ρ between each parameter and the amount of manufacturing resources. yj The values ​​are sorted in ascending (or descending) order based on their magnitudes. As mentioned above, the correlation ρ between each performance parameter and the amount of manufacturing resources... yj It can be predetermined, or it can be determined using grey relational analysis based on historical datasets.

[0056] The following example demonstrates how grey relational analysis can be used to calculate the correlation ρ between the j-th performance parameter and a selected manufacturing resource quantity y. yj The method.

[0057] Take Y 0 =(y 0 (1),y 0 (2),…,y 0 (n)) and Y 0 and The starting point zero profile, where n is the number of known sample data points, then we have: So, Y 0 With X j The absolute correlation ε of (j=2,…,m) yj for: Where m is the total number of performance parameters;

[0058] Let X′ j =X j / x j (1) is X j The initial value image, Then the relative correlation r yj for:

[0059] Therefore, the correlation between the j-th performance parameter and the manufacturing resource quantity y can be a weighted average of the absolute and relative correlation, such as the average value: ρ yj =0.5(ε) yj +r yj In other implementations, other suitable algorithms can also be used to calculate the correlation degree ρ. yj .

[0060] Step S2.3: Interpolate the manufacturing resource quantities of two or more adjacent data points before and after the virtual data to obtain the resource quantity level corresponding to the virtual data. Interpolation may include the weighted values ​​of the manufacturing resource quantities of two or more data points. When each data point includes multiple manufacturing resources, interpolation can be performed for each manufacturing resource to obtain the corresponding multiple manufacturing resource quantity levels.

[0061] For example, the manufacturing resource quantity Y of the previous data in the virtual data. lower As the lower bound, the manufacturing resource quantity Y in the next data point upper As the upper bound, the whitening parameter α is calculated based on these two data points to obtain the manufacturing resource value of the virtual data: Y 待预测 =αY lower +(1-α)Y upper The whitening parameter α can be a preset value, or it can be calculated as follows: assuming the technical parameter with the highest comprehensive correlation value is p, then the value of technical parameter p in the k-th data entry is marked as x. k,p Let Δx k,p =x k,p -x k-1,p (k=1,2,…,n), then the formula for calculating the whitening parameter α is: In other implementations, other suitable algorithms can also be used to calculate the whitening parameter α.

[0062] Step S2.4: Store the generated kth virtual data in the virtual dataset and remove it from the original dataset. Continue to generate the next virtual data based on the original dataset until the required data capacity N′ is reached.

[0063] Step S2.5: Divide the virtual data and the original data into training and test sets. For example, the training set contains at least some (or all) of the virtual data, and the test set contains at least some (or all) of the original data.

[0064] Step S3: Normalize the training set and the test set respectively.

[0065] Optionally, the training and test sets can be normalized separately, that is, each element of variable X is scaled to within [0,1]. The data normalization method is as follows: Where: X j,MAX and X j,MIN These are the maximum and minimum values ​​of the j-th column elements of variable X, respectively. Similarly, the dependent variable Y can be normalized to obtain... Among them, Y MAX and Y MIN These are the maximum and minimum values ​​of Y, respectively.

[0066] Step S4: Establish a manufacturing resource estimation model based on a neural network and train the model using a training set. The neural networks that can be used in this invention include feedforward neural networks, recurrent neural networks, or symmetric connection networks. For example, this invention can use a BP neural network (Back Propagation Neural Network), which is a subtype of feedforward neural networks and is currently the most widely used type of feedforward neural network. For different neural networks, appropriate model structures, required parameters, and network complexity can be constructed accordingly.

[0067] As an example, not a limitation. Figure 3 This is a schematic diagram of the basic architecture of a BP neural network according to an embodiment of the present invention. A BP neural network is initialized, and corresponding parameters are set. The hyperparameters (mainly network structure and optimization parameters) of the BP neural network can be optimized based on a grid search algorithm. The optimal structure of the BP neural network is: 1 input layer - 2 hidden layers - 1 output layer, with the number of nodes being 9-20-16-1 respectively. The optimal activation functions for the two hidden layers are tansig and purelin, respectively. The optimal training function is trainlm, the optimal learning rate is 0.01, and the optimal additional momentum factor is 0.95. Other parameter settings can be as follows: the maximum number of training epochs is 10,000, and the error value of the objective function is 1×10⁻⁶. -7 The minimum performance gradient is taken as 1×10. -7 The maximum number of failures allowed is 20.

[0068] It should be understood that Figure 3 The number of layers, nodes, and other parameters of the BP neural network shown are merely examples and not limitations. In other implementations, different neural network models, structures, and parameters may be used without departing from the scope of this invention. For example, when each data point includes multiple manufacturing resources, the output layer may include multiple output nodes corresponding to the multiple manufacturing resources, with each output node corresponding to one manufacturing resource.

[0069] Step S5: Use the test set to test the trained manufacturing resource quantity estimation model, calculate the goodness of fit between the predicted and actual values, and calculate the estimation accuracy.

[0070] Step S5.1: If normalization was performed in step S3, then the prediction result... The inverse normalization process is performed, and the specific calculation formula is as follows:

[0071] Step S5.2: Calculate the coefficient of determination R for goodness of fit. 2 , Root Mean Square Error (RMSE) and Estimation Accuracy (ε). The correlation coefficient is calculated using the following formula: Where y is the actual value, It is the predicted value, var(y) and They are y and standard deviation yes The covariance of y and the root mean square error (RMSE) are calculated using the following formula: The formula for calculating the estimation accuracy ε is: The R value corresponding to the manufacturing resource estimation model proposed in this invention is applied. 2 RMSE and ε are shown in Table 1 below. A comparison of the predicted and actual values ​​of engine resource quantities in the test set is shown below. Figure 5 As shown.

[0072] The tested manufacturing resource estimation model can be stored, transmitted, and used later.

[0073] Step S6: Estimate manufacturing resources using the trained manufacturing resource estimation model.

[0074] The trained manufacturing resource quantity estimation model can be stored in the system. The standard format of its manufacturing resource quantity estimation input data is X. k =(x k1 ,x k2 ,…,x k9 ), where x k1 ,x k2 ,…,x k9 The corresponding engine parameters and their units are as follows: takeoff thrust (kN), bypass ratio, overall pressure ratio, turbine inlet temperature (°C), fuel consumption rate (kg / (kgf·h)), airflow (kg / s), fan diameter (m), engine overall length (m), and engine net weight (kg). As an example and not a limitation, when making predictions, you can input only one or more of the above performance parameters, instead of all of them.

[0075] Optionally, the sample X to be predicted k X is obtained by normalizing according to the method in step S3. * The input is fed into the stored model to estimate manufacturing resource quantities, and the output results are evaluated. The inverse normalization process is performed according to the method in S5.1, and the final estimated result Y is given. * It indicates that the desired engine performance X has been achieved. k Estimate the required levels of various manufacturing resources.

[0076] It should be noted that Figure 2The implementation process shown is merely an example and not a limitation. More steps may be added or one or more steps may be omitted. The training and usage processes may be implemented separately without having to implement both processes in one embodiment.

[0077] Figure 4 This is a schematic diagram illustrating the prediction results of an aero-engine manufacturing resource estimation method according to an embodiment of the present invention. The virtual data generation method proposed in this invention can be used to generate new data similar to the original dataset, thereby significantly expanding the sample size. Training the manufacturing resource estimation model based on the virtually generated dataset can effectively improve the accuracy of the estimation results. Figure 4 The horizontal axis represents the number of different engine samples, and the vertical axis represents the manufacturing resource quantity estimated using the manufacturing resource quantity estimation model according to the present invention, such as the weighted value of manufacturing time, number of parts, number of manufacturing processes, and mold / tool ​​requirements multiplied by the corresponding rates. This manufacturing resource quantity can also reflect the overall manufacturing resource requirement level of aero-engines, such as total manufacturing cost.

[0078] Compared with the three existing commonly used estimation models (multiple linear regression model, exponential regression model, and BP neural network), this invention combines the virtual data generation method with the neural network and optimizes the hyperparameters, resulting in a manufacturing resource quantity estimation model that can achieve higher estimation accuracy (see Table 1 for detailed comparison results).

[0079]

[0080] Table 1 Comparison of the estimation method of the present invention with three common estimation methods

[0081] Note:

[0082] (1) The example input value of the sample to be tested is X=(137.25,9.3,33.11,853.89,0.28,549.82,1.95,3.3,3852);

[0083] (2) The random number seed used in the BP neural network and the method proposed in this invention is the system default value;

[0084] (3) The larger the value, the better the fit to the dataset; RMSE data and ε data The smaller the value, the higher the accuracy of the estimation of the dataset.

[0085] Figure 5This is a comparative experiment on the predictive stability of an aero-engine manufacturing resource quantity estimation method according to an embodiment of the present invention. When the BP manufacturing resource quantity estimation model is trained using only the original dataset, as shown in Figure (a), the goodness of fit fluctuates significantly, failing to effectively predict manufacturing resource quantities.

[0086] Conversely, as shown in Figure (b), training a manufacturing resource estimation model based on generated virtual data according to the present invention can effectively improve the stability of the estimation results.

[0087] Figure 6 This is a functional block diagram of an aero-engine manufacturing resource estimation device 600 according to an embodiment of the present invention. The aero-engine manufacturing resource estimation device 600 may include various components, such as a data receiving component 602, a virtual data generation component 604, a normalization component 606, a model training component 608, a prediction component 610, a storage component 612, etc., and the various components can communicate via a bus 620.

[0088] Data receiving component 602 can receive a historical dataset of aero-engines, which includes multiple performance parameters associated with each of the multiple aero-engines and the corresponding manufacturing resources for each aero-engine. Virtual data generation component 604 can generate virtual data based on the historical dataset, as described above. For example, virtual data generation component 604 can generate a set of random values ​​corresponding to multiple performance parameters, perform multi-key value sorting to determine the ranking position of the virtual data in the historical dataset, and generate interpolated manufacturing resources for the virtual data based on the manufacturing resources associated with two or more aero-engines adjacent to the virtual data in the historical dataset.

[0089] The model training component 608 can train an aero-engine manufacturing resource estimation model, at least in part, based on virtual data containing a set of random values ​​and interpolated manufacturing resource quantities. The normalization component 606 can optionally normalize the input data entering the model and denormalize the model output. The trained aero-engine manufacturing resource estimation model can be stored in memory 612.

[0090] The prediction component 610 can input multiple performance parameters associated with the expected aero-engine into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the expected aero-engine.

[0091] Compared with existing methods for estimating manufacturing resources, this invention has the following advantages:

[0092] (1) A virtual data generation method is proposed. Collecting manufacturing resources for aero-engines is difficult and cumbersome, resulting in generally small sample sizes. Fitting small sample data using parametric methods leads to poor stability and low accuracy in estimation results. This invention proposes a virtual data generation method to expand the sample size: First, by calculating the correlation between engine technical parameters and manufacturing resources in the original dataset, the sorting rules for the sample data are determined; then, a set of technical parameters is randomly generated and inserted into the original dataset, and the new dataset is reordered according to the correlation. Based on two or more adjacent data points of the inserted data, the manufacturing resource level value of the inserted data is estimated, thus obtaining a new dataset with similar characteristics to the original dataset. Repeating the above steps yields more virtual data. Expanding the sample size can significantly improve estimation accuracy.

[0093] (2) A manufacturing resource estimation model based on a neural network was established. When combined with a virtual data generation method, it can significantly improve the estimation accuracy and stability of the estimation results. In this invention, a neural network consisting of one input layer, two hidden layers, and one output layer was first established. Then, the hyperparameters (mainly network structure and optimization parameters) of the neural network were optimized using a grid search algorithm. Finally, a manufacturing resource estimation model was trained using a virtually generated dataset. Compared with data fitting based on parametric methods, the manufacturing resource estimation model based on a BP neural network in this invention can obtain more accurate estimation results. Specifically, the estimation model based solely on a BP neural network has an estimation error of approximately 21.87% on the training set and approximately 60.48% on the test set; the method proposed in this invention has an estimation error of approximately 2.4% on the training set and approximately 15.45% on the test set. A comparison of the stability of the estimation results of the two methods is shown in [link to relevant documentation]. Figure 1 Observations show that the combination of virtual data generation methods and BP neural networks can effectively reduce the problem of poor stability of estimation results caused by small sample size.

[0094] Therefore, this invention can expand the sample capacity based on a sample of aero-engine manufacturing resources using a virtual data generation method, and then use the virtual data to train an engine manufacturing resource estimation model based on a neural network, thereby improving the accuracy and stability of the estimation results. This invention is applicable to the estimation of manufacturing resource quantities for various types of aero-engines, such as civil aero-engines, military aero-engines, etc.

[0095] The various steps and modules of the methods and apparatus described above can be implemented in hardware, software, or a combination thereof. If implemented in hardware, the various illustrative steps, modules, and circuits described in connection with this disclosure can be implemented or executed using a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic components, hardware components, or any combination thereof. A general-purpose processor can be a processor, microprocessor, controller, microcontroller, or state machine, etc. If implemented in software, the various illustrative steps and modules described in connection with this disclosure can be stored as one or more instructions or codes on a computer-readable medium or transmitted. Software modules implementing the various operations of this disclosure can reside in a storage medium, such as RAM, flash memory, ROM, EPROM, EEPROM, registers, hard disk, removable disk, CD-ROM, cloud storage, etc. The storage medium can be coupled to a processor so that the processor can read and write information from / to the storage medium and execute corresponding program modules to implement the various steps of this disclosure. Moreover, software-based embodiments can be uploaded, downloaded, or remotely accessed through appropriate communication means. Such appropriate means of communication include, for example, the Internet, the World Wide Web, intranets, software applications, cables (including fiber optic cables), magnetic communication, electromagnetic communication (including RF, microwave and infrared communication), electronic communication, or other such means of communication.

[0096] The numerical values ​​given in the various embodiments are merely examples and are not intended to limit the scope of the invention. Furthermore, as a whole, there are other components or steps not listed in the claims or specification of this invention. Moreover, a single name for a component does not preclude other names for that component.

[0097] It should also be noted that these embodiments may be described as processes depicted as flowcharts, flow diagrams, structure diagrams, or block diagrams. Although a flowchart may describe the operations as a sequential process, many of these operations can be executed in parallel or concurrently. Furthermore, the order of these operations can be rearranged.

[0098] The disclosed methods, apparatuses, and systems should not be limited in any way. Rather, this disclosure covers all novel and non-obvious features and aspects of the various disclosed embodiments (individually and in various combinations and sub-combinations of each other). The disclosed methods, apparatuses, and systems are not limited to any particular aspect or feature or combination thereof, and no disclosed embodiment is required to have any one or more specific advantages or to solve any particular or all technical problems.

[0099] This invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other modifications based on the teachings of this invention without departing from the spirit and scope of the claims. All of these modifications are within the scope of protection of this invention.

Claims

1. A method for estimating the manufacturing resources of an aero-engine, characterized in that, include: Receive historical datasets of aero engines, which include multiple performance parameters associated with each of the multiple aero engines and the amount of manufacturing resources used by each aero engine; Generating virtual data based on the historical dataset, wherein generating the virtual data includes: Generate a set of random values ​​corresponding to the plurality of performance parameters. Based on the correlation between each of the multiple performance parameters and the amount of manufacturing resources, the virtual data and each aero-engine in the historical dataset are sorted by multiple key values ​​for each of the multiple performance parameters to determine the ranking position of the virtual data in the historical dataset. The interpolated manufacturing resource quantity of the virtual data is generated based on the manufacturing resource quantity associated with two or more aero engines adjacent to the virtual data in the historical dataset. And at least in part, the aircraft engine manufacturing resource estimation model is trained based on the virtual data containing the set of random values ​​and the interpolated manufacturing resource quantities.

2. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The aircraft engine's performance parameters include two or more of the following: Takeoff thrust, bypass ratio, overall pressure ratio, turbine inlet temperature, fuel consumption rate, airflow, fan diameter, engine length, and engine net weight.

3. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The amount of aero-engine manufacturing resources includes one or more of the following: Manufacturing time, number of parts, number of manufacturing processes, mold / tool ​​requirements, or a weighted average of two or more of the above.

4. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The aero-engine manufacturing resources include at least a first manufacturing resource and a second manufacturing resource, and generating the virtual data also includes: The first sorting position of the virtual data in the historical dataset is determined based on the correlation between each of the plurality of performance parameters and the first manufacturing resource quantity, and a first interpolated manufacturing resource quantity of the virtual data is generated based on the first manufacturing resource quantities associated with two or more aero-engines adjacent to the virtual data in the historical dataset; and The second sorting position of the virtual data in the historical dataset is determined based on the correlation between each of the plurality of performance parameters and the second manufacturing resource quantity, and the second interpolated manufacturing resource quantity of the virtual data is generated based on the second manufacturing resource quantity associated with two or more aero engines adjacent to the virtual data in the historical dataset.

5. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The correlation between each performance parameter and the amount of manufacturing resources is predetermined, or determined using grey relational analysis based on the historical dataset.

6. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The range of random values ​​in the virtual data corresponding to the plurality of performance parameters is selected from at least one of the following: Within the predetermined range of the corresponding performance parameters; and The values ​​of the corresponding performance parameters in the historical dataset are between the upper and lower bounds.

7. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, Also includes: The trained aero-engine manufacturing resource estimation model is tested based on at least a portion of the historical dataset.

8. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, Also includes: Multiple performance parameters associated with the anticipated aero-engine are input into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the anticipated aero-engine.

9. The method for estimating the manufacturing resources of an aero-engine as described in claim 1, characterized in that, The model for estimating the manufacturing resources of aero-engines includes feedforward neural networks, recurrent neural networks, or symmetric connected neural networks.

10. A computer-readable medium storing a computer program, which, when executed by a processor, causes the processor to perform the method as described in any one of claims 1-9.

11. A device for estimating the manufacturing resources of an aero-engine, characterized in that, include: A data receiving component receives a historical dataset of aero engines, the historical dataset including multiple performance parameters associated with each of the multiple aero engines and the amount of manufacturing resources used by each aero engine; A virtual data generation component generates virtual data based on the historical dataset. The generation of the virtual data includes: generating a set of random values ​​corresponding to the plurality of performance parameters; sorting each aero-engine in the historical dataset and the virtual data for the plurality of performance parameters based on the correlation between each performance parameter and the manufacturing resource quantity to determine the sorting position of the virtual data in the historical dataset; and generating an interpolated manufacturing resource quantity of the virtual data based on the manufacturing resource quantities associated with two or more aero-engines adjacent to the virtual data in the historical dataset. as well as A model training component that trains an aero-engine manufacturing resource estimation model based at least in part on the virtual data containing the set of random values ​​and the interpolated manufacturing resource quantities.

12. The aero-engine manufacturing resource estimation device as described in claim 11, characterized in that, The aircraft engine's performance parameters include two or more of the following: Takeoff thrust, bypass ratio, overall pressure ratio, turbine inlet temperature, fuel consumption rate, airflow, fan diameter, engine length, and engine net weight.

13. The aero-engine manufacturing resource estimation device as described in claim 11, characterized in that, The amount of aero-engine manufacturing resources includes one or more of the following: Manufacturing time, number of parts, number of manufacturing processes, mold / tool ​​requirements, or a weighted average of two or more of the above.

14. The aero-engine manufacturing resource estimation device as described in claim 11, characterized in that, The aero-engine manufacturing resources include at least a first manufacturing resource and a second manufacturing resource, and the virtual data generation component is configured to: The first sorting position of the virtual data in the historical dataset is determined based on the correlation between each of the plurality of performance parameters and the first manufacturing resource quantity, and the first interpolated manufacturing resource quantity of the virtual data is generated based on the first manufacturing resource quantity associated with two or more aero engines adjacent to the virtual data in the historical dataset. as well as The second sorting position of the virtual data in the historical dataset is determined based on the correlation between each of the plurality of performance parameters and the second manufacturing resource quantity, and the second interpolated manufacturing resource quantity of the virtual data is generated based on the second manufacturing resource quantity associated with two or more aero engines adjacent to the virtual data in the historical dataset.

15. The aero-engine manufacturing resource estimation device as described in claim 11, characterized in that, The correlation between each performance parameter and the amount of manufacturing resources is predetermined, or determined using grey relational analysis based on the historical dataset.

16. The aero-engine manufacturing resource estimation device as described in claim 11, characterized in that, Also includes: The prediction component inputs multiple performance parameters associated with the expected aero-engine into a trained aero-engine manufacturing resource estimation model to generate an estimated manufacturing resource quantity for the expected aero-engine.

Images