A synergic sensitivity analysis method for performance of three-split reducer

Abstract

The present application relates to a kind of collaborative sensitivity analysis method of three-ways flow reducer performance, belong to gear transmission performance analysis technical field.Aiming at the problems of low computational efficiency, insufficient high-dimensional parameter identification accuracy and unclear action path in the existing three-ways flow reducer tribology and dynamics coupling performance analysis, the present application first constructs a lubrication performance proxy prediction model based on convolutional neural network to replace high time-consuming physical simulation, then establishes a multi-source input system covering lubricating oil viscosity, tooth surface topography and gear macro-geometric parameters, and based on Sobol variance decomposition theory, global sensitivity analysis is performed on lubrication performance and dynamics performance respectively, and finally through the step-by-step mapping relationship between friction factor, lubrication characteristic parameter and dynamics performance index, collaborative sensitivity analysis is realized.The present application can accurately identify the key parameters affecting the service performance of reducer and their coupling mechanism, significantly improve the analysis efficiency and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of gear transmission performance analysis technology, and relates to a collaborative sensitivity analysis method for the performance of a three-way flow reducer. Background Technology

[0002] The three-way reducer is a core component of the power transmission system of underwater equipment, responsible for power transmission under high-speed and high-torque conditions. Its operational stability and reliability directly determine the overall service performance of the equipment. Under complex alternating loads, the helical gear pair of the three-way reducer is prone to four types of failure modes: fatigue fracture, pitting and spalling, tooth surface scuffing, and adhesive wear, which seriously affect the service life and operational safety of the equipment.

[0003] The lubrication and dynamic performance of a three-way flow reducer are strongly coupled. Lubrication parameters determine the quality of the oil film formation and anti-galling ability on the gear surface, while dynamic parameters determine the meshing contact state and dynamic response characteristics of the gear pair. These two types of parameters are transmitted and interact across levels, jointly determining the overall service performance of the reducer. To accurately identify key influencing parameters and quantify their driving effect on performance, a global sensitivity analysis is needed to clarify the transmission law of parameter fluctuations on output performance, providing support for reducer performance optimization and failure risk management.

[0004] In existing global sensitivity analysis, the Sobol method, based on variance decomposition theory, quantifies the contribution of a single parameter and parameter combinations to the output variance by calculating the first-order exponent, higher-order exponents, and the total effect exponent. It has been applied in parameter sensitivity identification in fields such as machinery, aerospace, and automotive. However, directly applying the Sobol method to the tribological-dynamic coupling performance analysis of a three-way flow reducer has three significant drawbacks. First, the input parameters involved in a three-way flow reducer cover multiple dimensions, including lubricating oil viscosity, tooth surface roughness, tooth surface hardness, gear module, pressure angle, helix angle, input speed, and input torque. Directly conducting full-parameter coupling analysis leads to an exponential increase in sampling scale, a surge in sampling costs, and a significant decrease in computational efficiency. Second, variance decomposition in high-dimensional parameter spaces is prone to accuracy distortion; the contribution of key parameters is easily masked by high-dimensional noise, resulting in insufficient reliability of the identification results. Third, traditional analysis requires repeated calls to the Thermoelastic-Hydrodynamic Lubrication (TEHL) numerical model and the multibody dynamics model to solve for parameter responses. Simulation of a single sample is time-consuming, and the total time cost under large-scale sampling is difficult to meet the needs of engineering applications. In addition, there are very few application cases of sensitivity analysis based on the thermo-fluid-structure interaction model combined with the Sobol method in the scenario of helical gears in three-way flow reducers, and there is a lack of targeted analysis methods to support it.

[0005] To improve computational efficiency, existing technologies attempt to introduce convolutional neural networks (CNNs) to construct surrogate models to replace high-precision physical simulations. However, existing solutions do not consider the cross-level transmission characteristics of tribological and dynamic performance of three-way flow reducers and still use a single-stage global sensitivity analysis framework. This cannot solve the problem of key parameter identification deviation under multi-parameter coupling, nor can it clarify the layer-by-layer action path of parameters on coupling performance, thus failing to provide effective guidance for the hierarchical optimization design of reducers.

[0006] In summary, to address the analytical needs of tribological-dynamic coupling performance of three-way flow reducers, there is an urgent need for a synergistic sensitivity analysis method that can balance computational efficiency, analytical accuracy, and action path identification capabilities, providing reliable methodological support for the control of lubrication status, improvement of dynamic performance, and prevention of failure risks in reducers. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a collaborative sensitivity analysis method for the performance of a three-way flow reducer.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A collaborative sensitivity analysis method for the performance of a three-way flow reducer includes the following steps: S1: Based on gear lubrication thermo-elasto-fluidic theory and engineering experience, determine the input and output parameters affecting the tooth surface lubrication performance of the three-way reducer. The input parameters include lubricating oil viscosity. Tooth surface roughness R q Tooth surface hardness H Input speed N in Input torque T in tooth surface normal load Q Gear module m n Pressure angle α and helix angle β The output parameters include oil film thickness. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c coefficient of friction f c and tooth surface temperature rise Δ TA sample database of lubrication performance of a three-way flow reducer is established based on the input and output parameters, and the sample database is divided into a training set and a test set according to a ratio of 80% and 20% respectively. S2: Establish a lubrication performance proxy prediction model for a three-way flow reducer based on a convolutional neural network (CNN). Use the training set and the test set to train and verify the lubrication performance proxy prediction model. When the lubrication performance proxy prediction model meets the preset prediction accuracy requirements, solidify it into a lubrication proxy prediction module. S3: Using the input layer feature parameters of the lubrication proxy prediction module as the input variables for lubrication performance sensitivity analysis, and using the output layer feature parameters of the lubrication proxy prediction module as the performance index for lubrication performance sensitivity analysis, a lubrication performance sensitivity analysis model is established. S4: Define the function based on the range of values ​​for each input variable. ,in X Let S1 be the vector consisting of the input parameters of the three-way flow reducer. Y The vector of output lubrication characteristic parameters determined by S1 is uniformly sampled within the range of values ​​of each input variable using the Monte Carlo method. N s Group data, in which N s ≥2000, forming sample points for Sobol sensitivity analysis. The sample points are then screened, and samples that exceed the reasonable range of parameters are removed to form a sample matrix of lubrication performance input parameters. S5: Import the lubrication performance input parameter sample matrix constructed in S4 into the lubrication proxy prediction module, extract parameter coupling features through convolution operation, perform inference and generate output parameter prediction values, obtain lubrication feature parameter prediction values, and form a lubrication performance input and output dataset; S6: Based on Sobol variance decomposition theory, a global sensitivity analysis is performed on the lubrication performance input-output dataset formed in S5 to determine the total variance of the lubrication performance indicators. V ( Y The input variables are decomposed into the independent contribution variances of each input variable and the contribution variances of the coupling effects between input variables, and the sensitivity index of each input variable to the lubrication performance index is calculated. S Ti According to the sensitivity index S Ti Identify the dominant and secondary parameters affecting lubrication performance, including the total variance. V ( Y The formula for calculating ) is: , For the first jPredicted values ​​of group lubrication characteristic parameters This represents the total number of sample points; S7: The oil film thickness from the lubrication characteristic parameters obtained in S5. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c As an intermediate variable, combined with the gear module Pressure angle helix angle To construct the input variables for dynamic performance sensitivity analysis, gear meshing stiffness is included. k m Gear transmission error δ Gear friction loss and reducer transmission efficiency η As an output index for sensitivity analysis of dynamic performance; S8: Based on the Sobol variance decomposition theory, a global sensitivity analysis of the dynamic performance of the three-way flow reducer is performed. The total variance of each dynamic performance index is decomposed into the independent contribution variance of the input variables and the contribution variance of the coupling effect between the input variables. The first-order sensitivity index and the total sensitivity index of each input variable to the dynamic performance index are calculated to determine the key parameters affecting the dynamic stability, transmission efficiency and load-bearing capacity of the three-way flow reducer. S9: Based on the sensitivity analysis results of lubrication performance obtained in S6 and the sensitivity analysis results of dynamic performance obtained in S8, and combined with the step-by-step mapping relationship between friction factors, lubrication characteristic parameters and dynamic performance indicators, the influence of each parameter on the tribological-dynamic coupling performance of the three-way flow reducer is comprehensively determined.

[0009] Furthermore, the convolutional neural network described in S2 includes an input layer, a one-dimensional convolutional layer, a flattening layer, a fully connected layer, and an output layer. The one-dimensional convolutional layer has 32 convolutional kernels with a kernel size of 3. The fully connected layer has 16 neurons. The output layer outputs the oil film thickness. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c coefficient of friction f c and tooth surface temperature rise Δ T .

[0010] Furthermore, the loss function of the convolutional neural network described in S2 is the Mean Squared Error (MSE), the optimizer is the Adaptive Moment Estimation (Adam) algorithm, the activation function is the Rectified Linear Unit (ReLU), and the model evaluation metrics include the Mean Absolute Error (MAE) and the coefficient of determination R. 2 .

[0011] Furthermore, the independent contribution variance described in S6 The calculation formula is: , Let f(X) be the mean of the function f(X) across the entire input parameter space. and For two different sets of sampling points, The calculation formula is: , x This is a sample of input parameters.

[0012] Furthermore, the coupling effect between input variables described in S6 contributes to the variance. The calculation formula is: , and For containing variables and Two different sets of sampling points.

[0013] Furthermore, the sensitivity index described in S6 The calculation formula is: The numerator is a variable containing variables. The sum of all variance contributions.

[0014] Furthermore, the formula for calculating the first-order sensitivity index mentioned in S8 is as follows: , The first-order sensitivity index, Input variables Independent contribution variance, This represents the total variance of the dynamic performance index.

[0015] Furthermore, the method for constructing the lubrication performance input parameter sample matrix described in S4 is as follows: construct two... random sample matrix A and B ,in d As the input parameter dimension, by swapping the first and second dimensions of matrix A and matrix B... i Generate a list of sample points. Matrix and matrix, .

[0016] Furthermore, the mathematical expression for the convolution operation described in S5 is: ,in These are the pre-activation feature values ​​obtained from the first layer convolution operation. For ReLU function, This is the trainable convolutional kernel weight matrix for the first one-dimensional convolutional layer. This is the bias term vector of the first one-dimensional convolutional layer. The kernel size is the convolution kernel size. The number of channels for the input feature. X The input parameter is the sample matrix.

[0017] Furthermore, the hierarchical mapping relationship described in S9 is a transmission path of "friction factor - lubrication characteristic parameter - dynamic performance index", where the friction factor is the input parameter defined in S1, and the lubrication characteristic parameter is the oil film thickness among the output parameters defined in S1. h Oil film stiffness Oil film damping Oil film bearing capacity Friction The dynamic performance index is the gear meshing stiffness defined by S7. Gear transmission error Gear friction loss and reducer transmission efficiency .

[0018] The beneficial effects of this invention are as follows: By constructing a convolutional neural network surrogate prediction model, this invention replaces the traditional time-consuming numerical simulation of thermo-elasto-fluidic lubrication, significantly improving the analysis efficiency of lubrication performance of three-way flow reducers. By establishing a hierarchical mapping relationship between lubrication performance and dynamic performance, cross-level correlation analysis from tribological parameters to dynamic response is achieved. Using the Sobol global sensitivity analysis method, the contribution of single parameters and multi-parameter coupling to lubrication and dynamic performance is simultaneously quantified, accurately identifying the dominant and secondary factors affecting the service performance of the reducer. This method effectively solves the problems of excessively high computational cost and accuracy distortion in sensitivity analysis under high-dimensional parameter space, clarifies the intrinsic mechanism of tribological-dynamic coupling, and provides a reliable theoretical basis for the lightweight design, lubrication parameter matching, and failure risk prevention of three-way flow reducers, significantly shortening the R&D cycle and reducing testing costs.

[0019] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a schematic diagram of a three-way flow reducer and its open-wire oblique meshing line and contact line. Figure 2 A schematic diagram of the collaborative sensitivity analysis of the three-way flow reducer; Figure 3 This is a schematic diagram of the CNN convolutional neural network architecture; Figure 4 Example graph showing the results of sensitivity analysis of lubrication performance; Figure 5 This is an example graph showing the results of the sensitivity analysis of dynamic performance. Detailed Implementation

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0023] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0024] The structural diagram of the three-way flow reducer is shown below. Figure 1 As shown in the diagram. The above problem can be solved using the collaborative sensitivity analysis method for a three-way flow reducer. Figure 2 express.

[0025] A collaborative sensitivity analysis method for the performance of a three-way flow reducer can be implemented through the following steps: S1: Based on gear lubrication thermo-elasto-fluidic theory and engineering experience, determine the input and output parameters affecting the tooth surface lubrication performance of the three-way reducer; wherein, the input parameters include lubricating oil viscosity. Tooth surface roughness R q Tooth surface hardness H Input speed N in Input torque T in tooth surface normal load Q Gear module m n Pressure angle α and helix angle β At least one of the following, the output parameter includes oil film thickness. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c coefficient of friction f c and tooth surface temperature rise Δ T At least one of the following; establish a sample database of lubrication performance of a three-way flow reducer based on the input parameters and output parameters, and divide the sample database into a training set and a test set according to a ratio of 80% and 20%; S2: Establish a lubrication proxy prediction model for a three-way flow reducer based on a CNN convolutional neural network (e.g., Figure 3As shown, this CNN convolutional neural network includes an input layer, a one-dimensional convolutional layer, a flattened layer, a fully connected layer, and an output layer. The one-dimensional convolutional layer has 32 convolutional kernels of size 3, used to extract the coupling features between lubrication performance input parameters. The fully connected layer has 16 neurons and is used to establish a nonlinear mapping relationship between the coupling features and lubrication feature parameters. The output layer outputs at least one lubrication feature parameter selected from oil film thickness, oil film stiffness, oil film damping, oil film load-bearing capacity, friction, friction coefficient, and tooth surface temperature rise. The loss function of the CNN is the mean squared error (MSE), the optimizer is the Adam optimizer, the activation function is the ReLU function, and the model evaluation metrics include the mean absolute error (MAE) and the coefficient of determination (R²).

[0026] The sample input is a one-dimensional parameter sequence, which is expanded into a tensor: (1) In the formula, N t The total number of training samples, d This represents the dimension of the input parameters; in this case, it represents the number of input parameters for the lubrication proxy prediction model.

[0027] One-dimensional convolutional layers are used to extract one-dimensional features of lubrication performance parameters. The mathematical expression is as follows: (2) In the formula, These are the pre-activation feature values ​​obtained from the convolution operation of the first layer; σ (·) represents the ReLU function. ; This is the trainable convolutional kernel weight matrix for the first one-dimensional convolutional layer. This is the bias term vector of the first one-dimensional convolutional layer; k 0 represents the kernel size; C 0 represents the number of channels for the input feature.

[0028] After passing through a one-dimensional convolutional layer, the data is fed into a flattening layer. The corresponding expression is:

[0029] (3) In the formula, This is the one-dimensional feature vector output by the flattening layer, and the matrix corresponds to the lubrication feature input parameters.

[0030] After passing through a fully connected layer, the data is imported into the output layer (the fully connected layer and the output layer have the same formula structure), and the corresponding expression is: (4) In the formula, The output vector of the output layer corresponds to the output parameters of lubrication characteristics such as oil film thickness, oil film stiffness, and oil film damping.

[0031] Training until convergence yields the CNN surrogate model for lubrication performance.

[0032] S3: Using the input layer feature parameters of the lubrication proxy prediction module established in S2 as the input variables for lubrication performance sensitivity analysis, and using the output layer feature parameters of the lubrication proxy prediction module as the performance index for lubrication performance sensitivity analysis, a lubrication performance sensitivity analysis model is established.

[0033] S4: Using the Monte Carlo method, uniformly sample at least 2000 sets of values ​​for each input variable within its range (i.e., ... N s ≥2000) data points are used to cover the entire input parameter space to form sample points for Sobol sensitivity analysis. The sample points are then screened for reasonableness, and samples that exceed the reasonable range of parameters are removed to form a sample matrix of lubrication performance input parameters.

[0034] Its mathematical expression is as follows: (5) In the formula, For neural input parameters; x i The vector consists of the input parameters of the three-way flow reducer determined by S1.

[0035] S5: Import the input parameter sample matrix constructed in S4 into the lubrication proxy prediction module, extract parameter coupling features through convolution operation, perform fast inference and generate output parameter prediction values ​​in batches, obtain the corresponding lubrication feature parameter prediction values ​​in batches, and form a lubrication performance input-output dataset. The mathematical expression is as follows: (6) In the formula, f (·) represents the CNN lubrication prediction model; y i The lubrication performance features output by the CNN surrogate model; Lubrication performance characteristics obtained under different input parameters y i The output dataset is composed of these components.

[0036] S6: Based on the Sobol variance decomposition theory, a global sensitivity analysis is performed on the lubrication performance input-output dataset described in S5. The total variance of the lubrication performance index is decomposed into the independent contribution variance of each input parameter and the contribution variance of the coupling effect between parameters, and the sensitivity index of each input parameter to the lubrication performance index is calculated.

[0037] The output of the Sobol model is decomposed into: (7) In the formula, f 0 is a constant; f i yes X i The function; f ij yes X i and X j The function; f t For all k The components of the input parameters are coupled together; the functions are mutually orthogonal, and the expected value of the product of different cross terms is 0. Therefore, squaring and integrating both sides of the above equation, we get: (8) The results were: (9) In the formula, V This represents the total variance. V i For variables X i Independent contributions (main effects), partial variance; V ij For variables X i and X j Interaction contribution; V t Contributes to the higher-order interactions of all variables, with partial variance. V i 、V ij The formula is as follows: (10) Estimating the Sobol index using Monte Carlo integration: (11) In the formula, f 0 represents the entire x The mean of the domain; N s The total number of samples for Monte Carlo sampling; k i For the sample index, that is, from the first sample to the second sample. N s The subscript of each sample.

[0038] (12a) (12b) Two sets of samples are randomly generated. X i Fixed, everything else changes.

[0039] (13) In this embodiment, lubricating oil viscosity, tooth surface roughness, tooth surface hardness, input torque, input speed, and tooth surface normal load are selected as exemplary input variables, i.e., the number of independent variables. D =9; In other implementations, the input variables can be expanded to structural parameters such as gear module, pressure angle, and helix angle, depending on the analysis requirements. Number of samples N s By constructing two N s A random sample matrix of ×D A and B : (14a) (14b) Matrix A and B are obtained by exchanging their first and second parts. i Columns x i The sample points are generated N s AB with 9 dimensions i Matrix and BA i matrix( i =1,2,…D): (15a) (15b) The corresponding function values ​​are calculated using the sample matrix: (16a) (16b) (16c) After constructing the sample values, the Monte Carlo estimate is calculated using the matrix as follows:

[0040] Lubrication performance sensitivity index ( S Ti ): (17) In the formula, X -i In addition to X iAll other input parameters besides V -i Remove all X i After this parameter, the output variance that all remaining parameters can contribute.

[0041] In this example, the sensitivity index result obtained from calculating the input parameters to the lubrication performance parameters is as follows: Figure 4 As shown, the key parameters affecting lubrication performance are analyzed.

[0042] S7: The oil film thickness from the lubrication characteristic parameters obtained in S5. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F Friction F c Some lubrication characteristic parameters are used as intermediate transfer variables, and combined with the gear module. m n Pressure angle helix angle etc., construct the input variables for dynamic performance sensitivity analysis; include gear meshing stiffness k m Gear transmission error δ Gear friction loss and reducer transmission efficiency η These are used as output indicators for sensitivity analysis of dynamic performance; S8: Based on Sobol variance decomposition theory, a global sensitivity analysis of the dynamic performance of a three-way flow reducer is performed. The total variance of each dynamic performance index is decomposed into the independent contribution variance of the input variables and the contribution variance of the coupling effect between the input variables. The first-order sensitivity index of each input variable to the dynamic performance index and the total sensitivity index are calculated, and the key parameters affecting the dynamic stability, transmission efficiency, and load-bearing capacity of the three-way flow reducer are determined. The results in this example are as follows: Figure 5 As shown.

[0043] S9: Based on the lubrication performance sensitivity analysis results obtained in S6 and the dynamic performance sensitivity analysis results obtained in S8, and combined with the step-by-step mapping relationship between friction factors, lubrication characteristic parameters and dynamic performance indicators, the influence of each parameter on the tribological-dynamic coupling performance of the three-way flow reducer is comprehensively determined.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention. For example, the example diagrams of lubrication performance and dynamic performance results in S6 and S8 are only used as an example of a typical working condition of a three-way flow reducer, and some influencing factors are selected to demonstrate the method of the present invention. Those skilled in the art should understand that the factors that actually affect lubrication performance and dynamic performance are not limited to the parameters shown in the example diagrams, and may include, but are not limited to, material Poisson's ratio, density, specific heat capacity, tooth flank clearance, elastic modulus, surface roughness, rotational speed, load, and lubrication parameters. Accordingly, the horizontal axis parameters in the example diagrams can be replaced or expanded according to specific application scenarios and analysis needs. Due to differences in working conditions, structural parameters, and operating parameters, the degree of contribution and ranking of each influencing factor to lubrication performance and dynamic performance may change. Therefore, the results reflected in the example diagrams are only used to illustrate the implementation process and application effect of the method of the present invention and should not be construed as a limitation on the scope of protection of the present invention.

Claims

1. A collaborative sensitivity analysis method for the performance of a three-way flow reducer, characterized in that: Includes the following steps: S1: Based on gear lubrication thermo-elasto-fluidic theory and engineering experience, determine the input and output parameters affecting the tooth surface lubrication performance of the three-way reducer. The input parameters include lubricating oil viscosity. Tooth surface roughness R q Tooth surface hardness H Input speed N in Input torque T in tooth surface normal load Q Gear module m n Pressure angle α and helix angle β The output parameters include oil film thickness. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c coefficient of friction f c and tooth surface temperature rise Δ T A sample database of lubrication performance of a three-way flow reducer is established based on the input and output parameters, and the sample database is divided into a training set and a test set according to a ratio of 80% and 20% respectively. S2: Establish a lubrication performance proxy prediction model for a three-way flow reducer based on a convolutional neural network (CNN). Use the training set and the test set to train and verify the lubrication performance proxy prediction model. Once the lubrication performance proxy prediction model meets the preset prediction accuracy requirements, solidify it into a lubrication proxy prediction module. S3: Using the input layer feature parameters of the lubrication proxy prediction module as the input variables for lubrication performance sensitivity analysis, and using the output layer feature parameters of the lubrication proxy prediction module as the performance index for lubrication performance sensitivity analysis, a lubrication performance sensitivity analysis model is established. S4: Define the function based on the range of values ​​for each input variable. ,in X Let S1 be the vector consisting of the input parameters of the three-way flow reducer. Y The vector of output lubrication characteristic parameters determined by S1 is uniformly sampled within the range of values ​​of each input variable using the Monte Carlo method. N s Group data, in which N s ≥2000, forming sample points for Sobol sensitivity analysis. The sample points are then screened, and samples that exceed the reasonable range of parameters are removed to form a sample matrix of lubrication performance input parameters. S5: Import the lubrication performance input parameter sample matrix constructed in S4 into the lubrication proxy prediction module, extract parameter coupling features through convolution operation, perform inference and generate output parameter prediction values, obtain lubrication feature parameter prediction values, and form a lubrication performance input and output dataset; S6: Based on Sobol variance decomposition theory, a global sensitivity analysis is performed on the lubrication performance input-output dataset formed in S5 to determine the total variance of the lubrication performance indicators. V ( Y The input variables are decomposed into the independent contribution variances of each input variable and the contribution variances of the coupling effects between input variables, and the sensitivity index of each input variable to the lubrication performance index is calculated. S Ti According to the sensitivity index S Ti Identify the dominant and secondary parameters affecting lubrication performance, including the total variance. V ( Y The formula for calculating ) is: , For the first j Predicted values ​​of group lubrication characteristic parameters This represents the total number of sample points; S7: The oil film thickness from the lubrication characteristic parameters obtained in S5. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c As an intermediate variable, combined with the gear module Pressure angle helix angle To construct the input variables for dynamic performance sensitivity analysis, gear meshing stiffness is included. k m Gear transmission error δ Gear friction loss and reducer transmission efficiency η As an output index for sensitivity analysis of dynamic performance; S8: Based on the Sobol variance decomposition theory, a global sensitivity analysis of the dynamic performance of the three-way flow reducer is performed. The total variance of each dynamic performance index is decomposed into the independent contribution variance of the input variables and the contribution variance of the coupling effect between the input variables. The first-order sensitivity index and the total sensitivity index of each input variable to the dynamic performance index are calculated to determine the key parameters affecting the dynamic stability, transmission efficiency and load-bearing capacity of the three-way flow reducer. S9: Based on the sensitivity analysis results of lubrication performance obtained in S6 and the sensitivity analysis results of dynamic performance obtained in S8, and combined with the step-by-step mapping relationship between friction factors, lubrication characteristic parameters and dynamic performance indicators, the influence of each parameter on the tribological-dynamic coupling performance of the three-way flow reducer is comprehensively determined.

2. A collaborative sensitivity analysis method for the performance of a three-way flow reducer according to claim 1, characterized in that: The convolutional neural network described in S2 includes an input layer, a one-dimensional convolutional layer, a flattening layer, a fully connected layer, and an output layer. The one-dimensional convolutional layer has 32 convolutional kernels with a kernel size of 3. The fully connected layer has 16 neurons. The output layer outputs the oil film thickness. h Oil film stiffness k h Oil film damping c h Oil film bearing capacity F h Friction F c coefficient of friction f c and tooth surface temperature rise Δ T .

3. A collaborative sensitivity analysis method for the performance of a three-way flow reducer according to claim 2, characterized in that: The loss function of the convolutional neural network described in S2 is the mean squared error (MSE), the optimizer is the Adam adaptive moment estimation algorithm, the activation function is the ReLU linear rectified function, and the model evaluation metrics include the mean absolute error (MAE) and the coefficient of determination (R²). 2 .

4. A collaborative sensitivity analysis method for the performance of a three-way flow reducer according to claim 1, characterized in that: Independent contribution variances as described in S6 The calculation formula is: , Let f(X) be the mean of the function f(X) across the entire input parameter space. and For two different sets of sampling points, The calculation formula is: , x This is a sample of input parameters.

5. A method for synergistic sensitivity analysis of the performance of a three-way flow reducer according to claim 4, characterized in that: The contribution variance of the coupling effect between input variables described in S6 The calculation formula is: , and For containing variables and Two different sets of sampling points.

6. A method for synergistic sensitivity analysis of the performance of a three-way flow reducer according to claim 5, characterized in that: The sensitivity index described in S6 The calculation formula is: The numerator is a variable containing variables. The sum of all variance contributions.

7. A method for synergistic sensitivity analysis of the performance of a three-way flow reducer according to claim 1, characterized in that: The formula for calculating the first-order sensitivity index in S8 is as follows: , The first-order sensitivity index, Input variables Independent contribution variance, This represents the total variance of the dynamic performance index.

8. A collaborative sensitivity analysis method for the performance of a three-way flow reducer according to claim 1, characterized in that: The method for constructing the sample matrix of lubrication performance input parameters described in S4 is as follows: construct two... random sample matrix A and B ,in d As the input parameter dimension, by swapping the first and second dimensions of matrix A and matrix B... i Generate a list of sample points. Matrix and matrix, .

9. A method for synergistic sensitivity analysis of the performance of a three-way flow reducer according to claim 1, characterized in that: The mathematical expression for the convolution operation described in S5 is: ,in These are the pre-activation feature values ​​obtained from the first layer convolution operation. For ReLU function, This is the trainable convolutional kernel weight matrix for the first one-dimensional convolutional layer. This is the bias term vector of the first one-dimensional convolutional layer. The kernel size is the convolution kernel size. The number of channels for the input feature. X The input parameter is the sample matrix.

10. A method for synergistic sensitivity analysis of the performance of a three-way flow reducer according to claim 1, characterized in that: The hierarchical mapping relationship described in S9 is the transmission path of "friction factor - lubrication characteristic parameter - dynamic performance index", where the friction factor is the input parameter defined in S1, and the lubrication characteristic parameter is the oil film thickness among the output parameters defined in S1. h Oil film stiffness Oil film damping Oil film bearing capacity Friction The dynamic performance index is the gear meshing stiffness defined by S7. Gear transmission error Gear friction loss and reducer transmission efficiency .

Images