Surface residual stress prediction method based on milling surface topography model and deep learning model
By combining milling surface morphology models with deep learning models, a multiple linear regression model was established, which solved the comprehensive problem of predicting surface morphology and residual stress in the milling process in the existing technology, realized high-precision three-dimensional residual stress distribution prediction, and optimized the control effect of the machining process.
Patent Information
- Application Number
- CN202411089113.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2026-02-10
AI Technical Summary
Existing technologies struggle to comprehensively predict surface morphology and residual stress during milling, resulting in low accuracy in residual stress prediction and an inability to fully predict the overall distribution of residual stress in machined parts.
By combining milling surface morphology models with deep learning models, and by acquiring the bottom edge size data of end mills and milling process parameters, we can perform single-pass surface morphology analysis prediction and residual stress simulation, extract key data, establish a multivariate linear regression model, and consider the superposition effect of multiple passes to achieve three-dimensional distribution prediction of surface residual stress.
It improves the prediction accuracy and response speed of residual stress during milling, optimizes the control and adjustment of machining quality, and realizes the prediction of residual stress distribution from two-dimensional to three-dimensional.
Smart Images
Figure CN121503186A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machining, and in particular to a method for predicting surface residual stress based on a milled surface morphology model and a deep learning model. Background Technology
[0002] Residual stress on machined surfaces is a crucial factor affecting material springback deformation and service life, especially for thin sheet metal workpieces. The distribution of surface residual stress significantly impacts material geometry and microstructure, necessitating accurate and efficient residual stress prediction methods. Song et al., in their study "Study on surface morphology and residual stress in inclined milling of titanium alloy TC11. (The International Journal of Advanced Manufacturing Technology (2022) 122: 3411–3423)," found that different milling speeds and feed rates in inclined milling experiments affect surface morphology and residual stress, with the feed mark morphology exhibiting periodic variations. Wang et al., in their study "Prediction of five-axis machining-induced residual stress based on cutting parameter identification. (I Journal of Manufacturing Processes 103 (2023) 320–336)," proposed a novel cutting parameter identification method to improve the prediction accuracy of residual stress induced by five-axis machining. Cai et al., in their paper "Modelling of end-milled floor surface topography considering system vibration and tool deflection. (Journal of Materials Processing Tech. 312(2023)117864.)", considered tool runout, vibration, and the overlap of tool deflection and trajectory, and established a predictive model for the surface topography of milled workpieces. Currently, many articles study the relationship between milling parameters and milled surface topography, as well as the predictive effect of different machining parameters on residual stress, but few studies the correlation between surface topography and milling residual stress.
[0003] Predicting residual stress in materials requires selecting different machining parameters in various cutting processes and combining them with the temperature field. The thermal stress generated during milling of difficult-to-machine materials can severely affect the accuracy of analytical predictions, leading to significant errors in residual stress prediction. Current residual stress prediction is limited to the depth direction and cannot comprehensively predict the overall residual stress distribution of the machined part. Achieving comprehensive prediction of surface morphology and residual stress, further optimizing part machining quality, and improving the response speed of residual stress prediction are all significant for enhancing the feedback control accuracy of residual stress during machining. Summary of the Invention
[0004] The purpose of this invention is to provide a method for predicting surface residual stress based on a milled surface topography model and a deep learning model.
[0005] The objective of this invention can be achieved through the following technical solutions:
[0006] A method for predicting surface residual stress based on a milled surface topography model and a deep learning model includes the following steps:
[0007] Step S1: Obtain the bottom edge dimension data and milling process parameters of the end mill, and perform analytical prediction of the surface morphology for a single pass;
[0008] Step S2: Perform simulation prediction of residual stress during a single milling pass;
[0009] Step S3: Based on the calculated surface morphology and the simulated residual stress cloud map, extract the key data of surface morphology geometric distribution and the key data of residual stress distribution;
[0010] Step S4: Based on key data of surface topography geometric distribution and key data of residual stress distribution, establish a deep learning-based residual stress prediction model for single-pass milling.
[0011] Step S5: Based on the residual stress prediction results output by the residual stress prediction model of single-pass milling, the influence of the superposition effect of multiple passes is considered to realize the prediction of residual stress of multi-pass milling under different superposition effects.
[0012] The key data for the geometric distribution of the surface morphology include surface roughness and cycle density.
[0013] The key data on residual stress distribution include the residual stress numerical distribution function and the depth of the residual stress-affected layer.
[0014] Step S4 specifically involves: using key data on process parameters and surface morphology geometric distribution as independent variables, and key data on residual stress distribution as dependent variables, and using the least squares method to fit a regression correlation model for predicting residual stress in a single pass.
[0015] The residual stress prediction model for single-pass milling is specifically set as follows: The feature weights (where k) i Here, b is the weighting coefficient for each dependent variable, b is the intercept or bias of the model, F is the model response, i.e., each key data point of the residual stress distribution, and X is the input vector containing multiple geometric key data points of surface topography distribution. i (i.e., surface roughness, trajectory cycle density of the machined surface, etc.) are expressed as:
[0016] X = (x, 1) = [x1, x2, ..., x d ,1]
[0017]
[0018]
[0019] Where d represents the total number of surface morphology parameters input into this deep learning model.
[0020] Establish mean square error Minimizing the mean squared error can be expressed as:
[0021]
[0022] Where y represents key data on residual stress distribution, including simulation feature data such as the numerical distribution function of residual stress and the depth of the residual stress-affected layer.
[0023] The optimal solution is obtained by differentiating it to zero using the least squares method:
[0024]
[0025] The resulting multiple linear regression model for F is as follows:
[0026]
[0027] This yields a regression model between the input and output variables, completing the regression association for any dependent variable.
[0028] In step S5, the types of superposition include: any two passes are adjacent but do not overlap, any two passes are adjacent and overlap, the cutting width of any pass includes the boundary, and the milling range of any pass includes the intersection angle of two boundaries.
[0029] Step S5 specifically includes the following steps:
[0030] Step S51: Surface morphology prediction is performed based on multi-pass milling process parameters, and the multi-pass milling morphology is calculated.
[0031] Step S52: Based on the multi-pass milling process parameters, perform simulation prediction of residual stress to obtain the distribution results of residual stress after multi-pass milling;
[0032] Step S53: Extract key data related to the superposition effect based on the multi-pass milling morphology and residual stress after multi-pass milling, and perform data fitting. Determine whether the superposition effect is triggered and the type of triggering effect through the surface morphology results. Establish a multi-pass distribution function based on the data fitting results.
[0033] Step S54: Based on the established single-pass milling residual stress prediction model, using different combinations of end milling process parameters as input parameters, the single-pass milling residual stress prediction results are obtained and superimposed on all surfaces to be machined. For the superposition effect triggered during the superposition process, the overlapping part data between multiple passes is input into the multi-pass distribution function. The overlapping part data is updated according to the output of the multi-pass distribution function to obtain the three-dimensional distribution of multi-pass residual stress under different surface morphology distributions.
[0034] The bottom edge dimensions of the end mill include the tool body radius and the bottom edge transition radius.
[0035] The milling process parameters include feed rate, pass spacing, rotational speed, depth of feed, whether lubrication and cooling are used, number of feeds, and feed time.
[0036] The radius of the bottom cutting edge is expressed as:
[0037]
[0038] Where r(z) is the bottom cutting edge radius that varies with the distance from the bottom surface, R is the radius of the tool cylinder, and r e Let be the bottom edge transition radius, α and θ be the transition angle and the boundary angle, respectively, and z be the distance perpendicular to the machined bottom surface that varies with α.
[0039] Compared with the prior art, the present invention has the following beneficial effects:
[0040] (1) This invention predicts the distribution of residual stress on the surface during milling based on the milled surface morphology, and considers the superposition effect of residual stress prediction between different tool trajectories in multi-pass milling, thereby improving the prediction accuracy of residual stress in multi-pass milling workpieces.
[0041] (2) This invention regresses the residual stress depth with key data in the plane and surface topography geometric data, thus expanding the dimensions of residual stress calculation.
[0042] (3) Based on the distribution characteristics of the processed surface morphology, the present invention predicts the residual stress distribution of the current part in a timely manner, and optimizes the control and adjustment of residual stress during the processing. Attached Figure Description
[0043] Figure 1 This is a flowchart of the method of the present invention;
[0044] Figure 2 This is a schematic diagram of the end mill bottom cutting edge geometry parameters and end milling process in one embodiment;
[0045] Figure 3 This is a schematic diagram of the superposition of four types of multi-pass trajectories generated during end milling in one embodiment. In this diagram, I represents any two passes that are adjacent but do not overlap; II represents any two passes that are adjacent and overlap; III represents any pass whose cutting width includes the boundary; and IV represents any pass whose milling range includes the intersection angle of the two boundaries. Detailed Implementation
[0046] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.
[0047] This embodiment provides a method for predicting residual stress on the surface based on fitting milled surface topography data with a deep learning model. First, a residual stress distribution model of the milled surface is established using the toolpath of a single milling pass, and simulation is used to predict the residual stress on the single-pass milled surface. Then, based on two types of data, a correlation model between surface topography and residual stress is proposed using deep learning, allowing direct calculation of the residual stress distribution from the surface topography data. Finally, the residual stress distribution of the machined surface is obtained by superimposing the residual stresses across different passes. This invention achieves efficient prediction of residual stress distribution during multi-pass milling, extending residual stress prediction from two-dimensional depth to three-dimensional prediction, and improving the prediction accuracy of residual stress on the machined surface during milling.
[0048] Specifically, such as Figure 1 As shown, the method includes the following steps:
[0049] Step S1: Obtain the bottom edge dimension data and milling process parameters of the end mill, and perform analytical prediction of the surface morphology for a single pass.
[0050] This embodiment implements the residual stress prediction method using end milling experiments on titanium alloy TC4. The obtained end mill bottom edge dimension data includes the tool body radius R and the bottom edge transition radius r. e Milling process parameters include feed rate v, pass spacing, spindle speed N, depth of feed, whether lubrication and cooling are used, number of feeds and feed time.
[0051] The tool path during the milling process is calculated, and a surface topography prediction model for single-pass end milling (climb milling) is established. End milling experiments and dimensional data are as follows: Figure 2 As shown, where α and θ are the transition angle and the boundary angle, respectively, z is the distance perpendicular to the machined bottom surface that varies with α, and r(z) is the bottom cutting tool radius that varies with the distance from the bottom surface, which can be expressed as:
[0052]
[0053] Step S2: Based on the same milling process, set the loading process, constraint unloading and cooling, and perform simulation prediction of residual stress during a single milling pass.
[0054] First, the loading process is set up by inputting the JC constitutive model and damage model of TC4, fixing the bottom surface of the workpiece, setting the load rotation speed to 500 rad / min, feed speed to 30 m / min and other process parameters, completing the assembly between the workpiece and the tool, setting the ambient temperature to 20℃, and using a temperature-displacement coupled explicit analysis step. Based on the milling depth of 0.5 mm, different transition mesh sizes are divided, with the minimum mesh size for the milling layer being 0.03 mm. Then, the simulation results of the loading process are reanalyzed, entering the second stage of the simulation: constraint unloading and cooling. The boundary constraints are removed and the workpiece thermal radiation and thermal convection are set. The workpiece is waited to cool down to 20℃, and the residual stress distribution on the material surface is obtained through post-processing.
[0055] Step S3: Based on the calculated surface morphology and the simulated residual stress cloud map, extract the key data of surface morphology geometric distribution and the key data of residual stress distribution.
[0056] In this embodiment, the key data of surface topography geometric distribution extracted include surface roughness Ra, cycle density m (circular track distribution density of a single milling pass), etc., and the key data of residual stress distribution extracted include residual stress numerical distribution function H(x,z) and residual stress influence layer depth.
[0057] Step S4: Based on key data of surface topography geometric distribution and key data of residual stress distribution, establish a deep learning-based residual stress prediction model for single-pass milling.
[0058] Specifically, key data on process parameters and surface morphology geometric distribution are used as independent variables, and key data on residual stress distribution are used as dependent variables. The least squares method is used to fit and obtain a regression correlation model for predicting residual stress in a single pass.
[0059] set up The feature weights (where k) iHere, b is the weighting coefficient for each dependent variable, b is the intercept or bias of the model, F is the model response, i.e., each key data point of the residual stress distribution, and X is the input vector containing multiple geometric key data points of surface topography distribution. i (i.e., surface roughness, trajectory cycle density of the machined surface, etc.), expressed as:
[0060] X = (x, 1) = [x1, x2, ..., x d ,1]
[0061]
[0062]
[0063] Where d represents the total number of surface morphology parameters input into this deep learning model.
[0064] Establish mean square error Minimizing the mean squared error can be expressed as:
[0065]
[0066] Where y represents key data on residual stress distribution, including simulation feature data such as the numerical distribution function of residual stress and the depth of the residual stress-affected layer.
[0067] The optimal solution is obtained by differentiating it to zero using the least squares method:
[0068]
[0069] The resulting multiple linear regression model for F is as follows:
[0070]
[0071] This yields a regression model between the input and output variables, completing the regression association for any dependent variable.
[0072] Step S5: Based on the residual stress prediction results output by the residual stress prediction model of single-pass milling, the influence of the superposition effect of multiple passes is considered to realize the prediction of residual stress of multi-pass milling under different superposition effects.
[0073] The types of superposition effects include: any two passes are adjacent but do not overlap; any two passes are adjacent and overlap; the cutting width of any pass includes the boundary; and the milling range of any pass includes the intersection angle of two boundaries. All four effects can be achieved through methods such as... Figure 3 The surface morphology distribution shown is defined by its range and triggering conditions.
[0074] Step S5 specifically includes the following steps:
[0075] Step S51: Surface morphology prediction is performed based on multi-pass milling process parameters, and the multi-pass milling morphology is calculated.
[0076] Step S52: Based on the multi-pass milling process parameters, perform simulation prediction of residual stress to obtain the distribution results of residual stress after multi-pass milling;
[0077] Step S53: Extract key data related to the superposition effect based on the multi-pass milling morphology and residual stress after multi-pass milling, and perform data fitting. Determine whether the superposition effect is triggered and the type of triggering effect through the surface morphology results. Establish a multi-pass distribution function based on the data fitting results.
[0078] Step S54: Based on the established single-pass milling residual stress prediction model, using different combinations of end milling process parameters as input parameters, the single-pass milling residual stress prediction results are obtained and superimposed on all surfaces to be machined. For the superposition effect triggered during the superposition process, the overlapping part data between multiple passes is input into the multi-pass distribution function. The overlapping part data is updated according to the output of the multi-pass distribution function to obtain the three-dimensional distribution of multi-pass residual stress under different surface morphology distributions.
[0079] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.
Claims
1. A method for predicting surface residual stress based on a milled surface topography model and a deep learning model, characterized in that, Includes the following steps: Step S1: Obtain the bottom edge dimension data and milling process parameters of the end mill, and perform analytical prediction of the surface morphology for a single pass; Step S2: Perform simulation prediction of residual stress during a single milling pass; Step S3: Based on the calculated surface morphology and the simulated residual stress cloud map, extract the key data of surface morphology geometric distribution and the key data of residual stress distribution; Step S4: Based on key data of surface topography geometric distribution and key data of residual stress distribution, establish a deep learning-based residual stress prediction model for single-pass milling. Step S5: Based on the residual stress prediction results output by the residual stress prediction model of single-pass milling, the influence of the superposition effect of multiple passes is considered to realize the prediction of residual stress of multi-pass milling under different superposition effects.
2. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, The key data for the geometric distribution of the surface morphology include surface roughness and cycle density.
3. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, The key data on residual stress distribution include the residual stress numerical distribution function and the depth of the residual stress-affected layer.
4. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, Step S4 specifically involves: using key data on process parameters and surface morphology geometric distribution as independent variables, and key data on residual stress distribution as dependent variables, and using the least squares method to fit a regression correlation model for predicting residual stress in a single pass.
5. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 4, characterized in that, The residual stress prediction model for single-pass milling is specifically set as follows: Let k be the feature weights, where k is the feature weight. i Here, b is the weighting coefficient for each dependent variable, b is the intercept or bias of the model, F is the model response, i.e., each key data point of the residual stress distribution, and X is the input vector containing multiple geometric key data points of surface topography distribution. i , is represented as: X=(x,1)=[x1,x2,…,x d ,1] Where d is the total number of surface morphology parameters input into this deep learning model; Establish mean square error Minimizing the mean squared error can be expressed as: Where y represents key data on residual stress distribution; The optimal solution is obtained by differentiating it to zero using the least squares method: The resulting multiple linear regression model for F is as follows: This yields a regression model between the input and output variables, completing the regression association for any dependent variable.
6. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, In step S5, the types of superposition include: any two passes are adjacent but do not overlap, any two passes are adjacent and overlap, the cutting width of any pass includes the boundary, and the milling range of any pass includes the intersection angle of two boundaries.
7. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, Step S5 specifically includes the following steps: Step S51: Surface morphology prediction is performed based on multi-pass milling process parameters, and the multi-pass milling morphology is calculated. Step S52: Based on the multi-pass milling process parameters, perform simulation prediction of residual stress to obtain the distribution results of residual stress after multi-pass milling; Step S53: Extract key data related to the superposition effect based on the multi-pass milling morphology and residual stress after multi-pass milling, and perform data fitting. Determine whether the superposition effect is triggered and the type of triggering effect through the surface morphology results. Establish a multi-pass distribution function based on the data fitting results. Step S54: Based on the established single-pass milling residual stress prediction model, using different combinations of end milling process parameters as input parameters, the single-pass milling residual stress prediction results are obtained and superimposed on all surfaces to be machined. For the superposition effect triggered during the superposition process, the overlapping part data between multiple passes is input into the multi-pass distribution function. The overlapping part data is updated according to the output of the multi-pass distribution function to obtain the three-dimensional distribution of multi-pass residual stress under different surface morphology distributions.
8. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, The bottom edge dimensions of the end mill include the tool body radius and the bottom edge transition radius.
9. The method for predicting surface residual stress based on a milled surface topography model and a deep learning model according to claim 1, characterized in that, The milling process parameters include feed rate, pass spacing, rotational speed, depth of feed, whether lubrication and cooling are used, number of feeds, and feed time.
10. The surface residual stress prediction method based on a milled surface topography model and a deep learning model according to claim 8, characterized in that, The radius of the bottom cutting edge is expressed as: Where r(z) is the bottom cutting edge radius that varies with the distance from the bottom surface, R is the radius of the tool cylinder, and r e Let be the bottom edge transition radius, α and θ be the transition angle and the boundary angle, respectively, and z be the distance perpendicular to the machined bottom surface that varies with α.