Parallel misalignment state mechanical transmission component wear amount probability prediction method

Abstract

The present application belongs to the technical field of health state evaluation and prediction of mechanical transmission system, and proposes a kind of parallel misalignment state mechanical transmission component wear amount probability prediction method.It includes the following steps: step S1: build the three-dimensional model of mechanical transmission component containing parallel misalignment error;Step S2: considering the random uncertainty of load profile, carry out wear simulation analysis based on grid node iteration;Step S3: considering the cognitive uncertainty of the model itself, train the Bayesian long short-term memory network model;Step S4: predict the expected value of wear and its uncertainty interval.The present application improves the description accuracy of traditional finite element method on mechanical transmission component wear behavior, breaks through the limitation of wear amount deterministic numerical prediction, provides high robustness and high reliability of probability prediction result, and provides scientific basis and data support for transmission system condition-based maintenance decision.

Description

Technical Field

[0001] This invention belongs to the field of health status assessment and prediction technology of mechanical transmission systems, specifically involving a probabilistic prediction method for the wear of mechanical transmission components under parallel misalignment conditions. Background Technology

[0002] In mechanical transmission systems, the health of key components such as gears, bearings, and couplings directly impacts the system's reliability and service life. However, due to manufacturing errors, installation deviations, and dynamic load variations during operation, mechanical transmission components often fail to maintain perfect alignment under actual working conditions. Parallel misalignment is a common and typical form of misalignment. This deviation leads to non-uniform contact and localized stress concentration between components, thereby exacerbating friction and wear and seriously threatening the normal operation of the system.

[0003] Currently, the prediction of wear on mechanical transmission components under parallel misalignment conditions mainly relies on traditional empirical models or data-driven methods. However, these methods all exhibit significant limitations under complex working conditions: traditional empirical models struggle to fully account for uncertainties, while data-driven methods have limited generalization ability in small sample situations; furthermore, although finite element-based simulation methods can provide a certain level of analytical accuracy, their computational efficiency is low due to the use of fixed mesh technology, and their accuracy is limited when dealing with complex working conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a probabilistic prediction method for the wear of mechanical transmission components under parallel misalignment conditions. This method aims to solve the problem that existing technologies struggle to balance prediction accuracy, computational efficiency, and uncertainty modeling under complex operating conditions. By integrating dynamic finite element analysis with Bayesian long short-term memory networks, this method can accurately predict the wear of mechanical transmission components and its uncertainty range, considering load randomness and system uncertainty, thus providing a scientific basis for equipment maintenance and life assessment.

[0005] The specific technical solution of the present invention is as follows:

[0006] S1: Based on the specific dimensional parameters and material properties of the mechanical transmission components, construct a three-dimensional model of the mechanical transmission components, including parallel misalignment errors;

[0007] S2: Conduct in-depth analysis of the working load of mechanical transmission components to determine the load profile; fully consider the random uncertainties in the load profile and use statistical methods to quantify the uncertainty of the load profile parameters; sample the load profile parameters; use all sampled load profile parameters as input conditions, and conduct wear simulation analysis based on mesh node iterative reconstruction in finite element analysis software to obtain a dataset of wear growth of mechanical transmission components under different load profiles;

[0008] S3: Using the wear growth dataset obtained in step S2, train the Bayesian long short-term memory network model for wear probability prediction, and use the Dropout technique to model the uncertainty of the weight parameters in the network.

[0009] S4: Using a trained Bayesian long short-term memory network model, predict the expected wear of mechanical transmission components and its uncertainty range in the short term.

[0010] Preferably, the wear simulation analysis method based on mesh node iterative reconstruction described in step S2 is as follows:

[0011] S2.1: In the transient simulation module of ANSYS software, define the initial load step increment Δt1; use the APDL command to integrate the Archard wear model into the simulation post-processing module; after completing the transient simulation for Δt1, obtain the contact stress and sliding velocity of all nodes on the contact surface of the mechanical transmission component, and calculate the wear depth Δh of all nodes. 1j Where j is the number of nodes on the contact surface;

[0012] S2.2: To simplify calculations and improve simulation efficiency, the wear depth of each node is set to change linearly with time over a short period, and the wear acceleration phenomenon during this period is ignored; therefore, the wear depth of each node within a time duration of N1×Δt1 can be expressed as N1×Δh 1j ;

[0013] S2.3: Using the APDL command, move all nodes N1×Δh along the normal direction of the contact surface. 1j Simulates material loss during the wear process;

[0014] S2.4: Set the next load step increment Δt2, re-perform the wear simulation, and determine the wear depth Δh for each node. 2j Repeat steps S2 and S3;

[0015] S2.5: Finally, within the total operating time Δt, the cumulative wear depth Δh of each node. j and the maximum wear depth Δh of mechanical transmission components maxIt is calculated using the following formula:

[0016]

[0017] Δh max =max{Δh j}

[0018] Where k is the number of load steps.

[0019] Preferably, the training method for the Bayesian Long Short-Term Memory network model described in step S3 is as follows:

[0020] S3.1: Set the initial values ​​of network parameters; perform preliminary parameter range filtering based on grid search, and combine with Bayesian optimization to determine the current optimal number of long short-term memory layers, number of hidden layers, learning rate, and regularization coefficients of the network;

[0021] S3.2: The uncertain interval of the network output is evaluated based on the improved prediction interval coverage probability (IPICP) metric, which guides the adjustment of the dropout rate and the readjustment of the learning rate and regularization coefficient;

[0022] S3.3: Evaluate the network's performance under different amounts of training data by gradually increasing the amount of training data, and determine the shortest acceptable training data duration;

[0023] S3.4: Based on the above optimization steps, the training of the Bayesian Long Short-Term Memory Network is finally completed, and it is applied to the short-term probability prediction of wear of mechanical transmission components.

[0024] Preferably, the specific implementation steps for adjusting the Dropout rate, learning rate, and regularization coefficient based on the IPICP metric in step S3.2 are as follows:

[0025] S3.2.1: First, the coverage rate of observations within the uncertainty interval of the current network output is calculated using the IPICP metric. Simultaneously, an interval width penalty term is introduced to assess the reasonableness of the uncertainty range. The IPICP metric is:

[0026]

[0027] Where N is the number of predicted points output by the network, and c i This indicates whether the observed value at prediction point i falls within the uncertainty interval; if it does, then c... i =1, otherwise c i =0; and These represent the upper and lower bounds of the uncertainty interval, respectively; α is the penalty coefficient that controls the influence of the uncertainty interval width.

[0028] S3.2.2: Subsequently, the Dropout rate, learning rate, and regularization coefficient are iteratively adjusted to optimize the IPICP metric.

[0029] When the IPICP index is less than 0.9, the Dropout rate is reduced by 0.05 steps while the learning rate and regularization coefficient are increased by 20% to narrow the prediction interval; when the IPICP index is greater than or equal to 0.9, the Dropout rate, learning rate, and regularization coefficient remain unchanged.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] 1. This invention significantly enhances the accuracy and performance of the traditional finite element method in characterizing the wear behavior of mechanical transmission components by iteratively reconstructing the mesh distribution on the contact surface of mechanical transmission components during the simulation process.

[0032] 2. This invention innovatively combines Latin hypercube sampling with Bayesian long short-term memory networks, and by introducing Dropout probabilistic modeling and an improved dynamic parameter tuning mechanism for prediction interval coverage, it breaks through the limitations of traditional deterministic numerical prediction of wear amount, and can provide probabilistic prediction results with both high robustness and high reliability. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly introduced below. The features and advantages of the present invention can be more clearly understood by referring to the accompanying drawings. The accompanying drawings are schematic and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a flowchart of the probability prediction method for wear of mechanical transmission components under parallel misalignment conditions proposed in this invention.

[0035] Figure 2 This is a 3D model of the aviation spline coupling in Example 1 and a schematic diagram of its load profile.

[0036] Figure 3 This is a schematic diagram of the Bayesian Long Short-Term Memory network in Example 1.

[0037] Figure 4 This is a schematic diagram showing the results of predicting the probability of wear depth of an aerospace spline coupling using the present invention in Example 1. Detailed Implementation

[0038] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0039] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0040] like Figure 1 As shown, the wear probability prediction method for mechanical transmission components under parallel misalignment conditions proposed in this invention mainly includes four steps: three-dimensional model construction, dynamic wear simulation analysis considering random uncertainties, Bayesian long short-term memory network model training considering cognitive uncertainties, and wear probability prediction. The implementation process is described in detail below using an example of an aerospace spline coupling:

[0041] Example 1

[0042] Taking a typical splined coupling in the aerospace industry as the application object, the maximum wear depth of its tooth surface is probabilistically predicted. First, based on the key dimensional parameters of the splined coupling shown in Table 1, a three-dimensional finite element model containing a 0.06mm parallel misalignment is established in ANSYS, as follows: Figure 2 As shown in Table 2, material properties are defined. Sixty sets of tooth surface contact pairs are constructed using an enhanced Lagrange contact algorithm.

[0043] Table 1

[0044]

[0045] Table 2

[0046]

[0047] In the dynamic wear simulation phase, 100 load profiles with torque fluctuations were generated using Latin hypercube sampling technique, with a mean torque of 1.51 × 10⁻⁶. 5 N·mm, with a standard deviation of 5%, to characterize typical random load characteristics of aero-engines. An automated simulation process was constructed using APDL command flow. The internal spline was set to a constant speed of 1500 RPM, while the external spline was simultaneously subjected to a reverse torque load to reproduce the dynamic force transmission boundary conditions of the transmission system. In each simulation, the Arcard wear model was integrated into the transient dynamics solver via APDL command flow, constructing a mesh iterative reconstruction framework: with a load step size of 0.1 units of time, the contact stress and sliding distance of the contact surface nodes were extracted in each load step, based on the formula... Calculate the wear increment (N) ij =1000 is the set time amplification factor, k=51 is the set load step number); then call the APDL's *NMOVE command, based on the node's normal wear increment Δh j The geometry of the contact surface is dynamically reconstructed to simulate the material loss process. Each simulation lasts for 5.1 time units (equivalent to 85 minutes of actual operation), through 51 wear-geometry update cycles. Finally, 100 simulations are completed simultaneously using an HPC parallel computing cluster to construct a dataset of the maximum wear depth covering the random load spectrum.

[0048] During the network model training phase, 100 sets of simulation data were divided into a training set (80 sets), a test set (10 sets), and a validation set (10 sets) in an 8:1:1 ratio. Based on Figure 3 The architecture shown constructs a Bayesian neural network with three long short-term memory hidden layers. Initial parameters are set as follows: 64 hidden layer neurons, a dropout rate of 0.2, a learning rate of 0.001, and a regularization coefficient of 1e-3. A Bayesian optimization algorithm is used to perform a global search for key hyperparameters: traversing combinations of hidden layer neurons within the range of [32-128] and dropout rate within the range of [0.1-0.5], with the goal of minimizing the mean squared error (MSE) of the test set. An improved prediction interval coverage index, IPICP (α = 0.1, representing the interval width penalty coefficient), is further introduced for joint tuning. When IPICP falls below 0.9, a dynamic adjustment strategy is used to reduce the dropout rate by 0.05 steps and increase the learning rate by 20%, ultimately obtaining the optimal parameter combination: a learning rate of 0.0012, a dropout rate of 0.15, and a regularization coefficient of 1.2e-3.

[0049] After the model is trained, its probabilistic prediction performance is systematically evaluated on the validation set. For example... Figure 4 As shown, with a torque of 1.55 × 10 5 Taking N·mm wear data as an example, after inputting the data for the first 68 minutes (80% of the total data duration), the model can simultaneously output the mean wear depth prediction and uncertainty interval for the next 17 minutes (20% of the total data duration) with a 95% (2σ) confidence level. Full validation set testing shows that the model's prediction of the mean wear depth for the next 17 minutes achieves MSE = 0.0363 and a coefficient of determination R0.05. 2=0.9177; In terms of interval prediction, the 95% (2σ) confidence interval coverage reaches 94.7%, with an average bandwidth of ±0.011 mm. This method demonstrates excellent prediction accuracy and uncertainty quantification capability under typical torque fluctuation conditions of aero-engines, and can generate probabilistic wear prediction results 17 minutes in advance, providing high-confidence data support for condition-based maintenance decisions of transmission systems.

Claims

1. A method for predicting the amount of wear of a mechanical transmission component in a parallel misalignment condition, characterized in that The following steps are sequentially executed: S1: According to the specific size parameters and material characteristics of the mechanical transmission component, a three-dimensional mechanical transmission component model containing parallel misalignment errors is constructed; S2: The working condition load of the mechanical transmission component is analyzed to determine the load profile; considering the random uncertainty existing in the load profile, the load profile parameters are quantitatively processed using statistical methods; the load profile parameters are sampled; The load profile parameters obtained by all sampling are used as input conditions, and the wear simulation analysis based on grid node iterative reconstruction is carried out in the finite element analysis software to obtain the wear growth data set of the mechanical transmission component under different load profiles; S3: The wear growth data set obtained in step S2 is used to train the Bayesian long short-term memory network model for wear probability prediction, and the Dropout technology is used to model the uncertainty of the weight parameters in the network; S4: The trained Bayesian long short-term memory network model is used to predict the wear expectation value and its uncertainty interval of the mechanical transmission component in the future short period of time.

2. The method according to claim 1, wherein The wear simulation analysis method based on grid node iterative reconstruction in step S2 is specifically: S2.1: In the transient simulation module of ANSYS software, define the initial load step increment Δt1; use the APDL command to integrate the Archard wear model into the post-processing module of the simulation; after completing the transient simulation for the duration of Δt1, obtain the contact stress and sliding speed of all nodes on the contact surface of the mechanical transmission component, and calculate the wear depth Δh of all nodes 1j ; wherein j is the number of nodes on the contact surface; S2.2: In order to simplify the calculation and improve the efficiency of simulation, it is assumed that the wear depth of each node changes linearly with time in a short period of time, and the wear acceleration phenomenon during this period is ignored; therefore, the wear depth of each node can be expressed as N1 x Δh within the time period of N1 x Δt1 1j ; S2.3: Move all nodes along the contact normal direction by N1 x Ah through APDL commands 1j to simulate material loss during wear process; S2.4: Set the next load step increment At2, re-run the wear simulation, determine the wear depth Ah of each node 2j ; repeat steps S2 and S3; S2.5: Finally, the cumulative wear depth Δh of each node over the total running time Δt j and the maximum wear depth Δh of the mechanical transmission components max is calculated by the following equation: Δh max = max{Δh j} Wherein, k is the number of load steps.

3. The method of claim 1, wherein The training method of the Bayesian long short-term memory network model in step S3 is: S3.1: Set the initial value of the network parameter; based on grid search, preliminarily screen the parameter range, and combine Bayesian optimization to determine the current best long short-term memory layer number, hidden layer number, learning rate and regularization coefficient of the network; S3.2: Evaluate the uncertainty interval of the network output based on the improved prediction interval coverage probability (IPICP) index, guide the adjustment of the Dropout rate and the re-adjustment of the learning rate and the regularization coefficient; S3.3: Evaluate the performance of the network under different data amounts by gradually increasing the training data amount to determine the shortest acceptable training data length; S3.4: Based on the above steps S3.1-S3.3, the training of the Bayesian long short-term memory network is finally completed, and it is applied to the short-term probability prediction of the wear of the mechanical transmission component.

4. The method of claim 3, wherein The specific implementation steps of adjusting the Dropout rate, learning rate and regularization coefficient based on the IPICP index in step S3.2 are as follows: S3.2.1: First, calculate the coverage rate of the observed value in the uncertainty interval of the current network output through the IPICP index, and introduce an interval width penalty term to evaluate the rationality of the uncertainty range, and the IPICP index is: where N is the number of prediction points output by the network, c i represents whether the observation of prediction point i falls into the uncertainty interval, c i = 1 if it does, otherwise c i = 0; and represent the upper and lower bounds of the uncertainty interval, respectively; a is a penalty coefficient that controls the influence of the width of the uncertainty interval. S3.2.2: Then, iteratively adjust the Dropout rate, learning rate and regularization coefficient to make the IPICP index optimal: When the IPICP index is less than 0.9, reduce the Dropout rate by 0.05 steps and increase the learning rate by 20% and the regularization coefficient by 20% to narrow the prediction interval; when the IPICP index is greater than or equal to 0.9, the Dropout rate, learning rate and regularization coefficient remain unchanged.

Images