Efficient prediction method for milling deformation of thin-walled workpiece

Abstract

The invention provides an efficient prediction method for milling deformation of a thin-walled workpiece, belongs to the field of numerical control machine tools, and aims to improve the prediction precision of the machining deformation of the thin-walled workpiece and reduce the calculation cost. Comprising the following steps that a cutter and a machining material are reasonably selected, and a thin-wall part milling finite element model is established based on Abaqus; designing an experimental scheme by using a Latin super method, and obtaining a sufficient low-credibility data set and a limited high-credibility data set through finite element software simulation; the difference between high and low credibility data is reduced by adopting an addition scale; and based on the variable credibility approximation model, fusing the high and low credibility data into a variable credibility data set, and training a machining deformation prediction model taking each milling parameter as input and machining deformation as output. According to the method, numerical simulation and the variable credibility model are combined, the prediction precision is improved through the addition scale, the efficient and accurate prediction method is provided for machining deformation of the thin-wall part, and the method has the practical application value.

Description

Technical Field

[0001] The invention belongs to the technical field of thin-walled part processing, and relates to a method for efficiently predicting the milling deformation of thin-walled parts. Background Art

[0002] Thin-walled parts have the characteristics of compact structure, high strength and light weight while meeting the strength requirements. Therefore, they are widely used in high-end manufacturing fields such as aerospace, national defense science and technology, and automobile manufacturing. However, due to the low yield strength and high elastic modulus of aviation thin-walled materials such as titanium alloy and aluminum alloy, such parts are very likely to cause the meshing boundary between the tool and the workpiece to deviate from the nominal position during the milling process, resulting in large elastic and plastic deformation. In addition, the workpiece has a high material removal rate during the milling process and is easily affected by the milling force and produces large machining deformation. Due to the low rigidity of thin-walled parts and the elastic recovery of deformation after the tool is passed, the actual milling width is not equal to the nominal value, which in turn affects the machining accuracy and final quality of the parts. Therefore, the research on milling deformation of thin-walled parts not only has theoretical value, but also can significantly improve processing efficiency and reduce production costs, thus having a profound impact on the high-end manufacturing industry.

[0003] At present, the research methods of deformation prediction technology for thin-walled parts milling are mainly experimental and simulation methods. The experimental method obtains deformation laws through actual processing experiments. Although this method is intuitive and reliable, it is difficult to meet the needs of rapid analysis under various working conditions due to the high experimental cost and long cycle. The finite element simulation method is based on numerical simulation. It can comprehensively consider the influence of tool parameters, material properties and processing conditions on deformation at a lower cost and predict deformation behavior. However, the existing finite element method has many limitations: on the one hand, the complexity of the three-dimensional simulation model leads to high computing resource requirements, and multiple calls to the simulation software to verify the process parameters are cumbersome and time-consuming; on the other hand, assumption errors may be introduced in the process of model simplification, thereby reducing the prediction accuracy.

[0004] Kurpiel et al. used tools with different properties to conduct adaptive face milling and adaptive cylindrical milling experiments, explored the influence of tool parameters and cutting parameters on the vibration and surface morphology of thin-walled parts, and intuitively explored the specific influence of machining parameters on machining deformation, but cost and time limit its practicality in multi-condition prediction. Based on finite element analysis software, Wang and Hao proposed a method to simultaneously iterate radial cutting depth and tool axis vector by considering the coupling effect between cutting force and tool / workpiece deformation, and then predict the deformation error caused by milling force. However, this method requires multiple calls to ANSYS to complete the iteration process, and the calculation efficiency is relatively low, which is not suitable for rapid prediction under multi-variable conditions.

[0005] The shortcomings of traditional experimental and simulation methods are particularly obvious in multi-working scenarios such as cutting parameters and tool-workpiece combination changes. Traditional experimental methods require a large number of repeated thin-walled parts milling experiments to establish and verify the machining deformation prediction model. This process will increase the experimental cost and reduce research efficiency. Although the finite element method can approximate the experimental results well, its high computational cost, complex model construction, and poor adaptability across working conditions limit its practical application. Especially in large-scale multi-working condition prediction, the single credibility data processing method lacks full utilization of the differences between data, further weakening the prediction accuracy and efficiency. Summary of the invention

[0006] In order to solve the above problems, the purpose of the present invention is to provide an efficient prediction method for the milling deformation of thin-walled parts. The present invention aims at the influence of variable milling parameters on the milling deformation of thin-walled parts under different tool parameters, and the problem of limited data samples caused by this. The method intends to combine numerical simulation technology with a variable credibility approximation model, and on this basis introduces additive scaling to narrow the gap between data to improve prediction accuracy. It provides an efficient and accurate prediction method for the milling deformation of thin-walled parts, realizes the coordinated use of high and low credibility data, effectively improves the deformation prediction accuracy, reduces the calculation cost, and provides guidance for the milling process of high-precision thin-walled parts, which has practical application value.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] The present application provides a method for efficiently predicting deformation of thin-walled parts during milling, comprising the following steps:

[0009] S1: Select the cutting tools and materials for thin-walled parts milling, and establish a finite element simulation model for thin-walled parts milling based on the finite element simulation software Abaqus;

[0010] S2: Use the Latin hypercube method to design the experimental plan, conduct simulation experiments through finite element software, obtain sufficient simulation processing deformation data and construct the LF data set;

[0011] S3: Select different experimental schemes to construct a small sample data set, and perform simulation experiments using finite element software to construct a limited HF data set;

[0012] S4: Adopt additive scaling to narrow the gap between LF and HF data and generate an updated low-confidence dataset;

[0013] S5: The new LF data and HF data with different label rates are integrated as variable credibility data sets, and a variable credibility machining deformation prediction model with various milling parameters as input and machining deformation as output is trained to realize deformation prediction in the milling process of thin-walled parts.

[0014] Further, the step S1 is specifically as follows:

[0015] Based on the principle of reasonable tool and processing parameter selection, with reference to the Mechanical Processing Technology Manual, the tools, materials and milling parameter ranges for thin-walled parts milling were selected, and the finite element simulation model of thin-walled parts milling was established using the finite element simulation software Abaqus.

[0016] Further, the step S2 is specifically as follows:

[0017] S201: Designing an experimental scheme using the Latin hypercube method according to the milling processing parameter selection range of step S1;

[0018] S202: Determine a basic (source) tool, perform 64 corresponding simulation experiments through finite element software, obtain and organize sufficient simulation processing deformation data, and construct LF data set D_ with milling parameters and corresponding processing deformation. LF .

[0019] Further, the step S3 is specifically as follows:

[0020] Determine a new (target) tool, use the label sample formula to select experimental schemes with different proportions to construct a small sample data set, conduct finite element simulation experiments, obtain and organize the machining deformation data under the corresponding milling parameters of the target tool, and construct a limited HF data set D_ HF :

[0021] D_ HF =round(Num_data*ratio),ratio=0.1,0.2,…,0.8

[0022] In the formula, Num_data represents the number of target domain data samples, and ratio represents different ratios.

[0023] Further, the step S4 is specifically as follows:

[0024] According to steps S2 and S3, the LF and HF data sets are obtained, and the gap between the LF and HF data is reduced by using additive scaling, and the original LF data is updated to the new LF data set D_ LF’ :

[0025]

[0026] In the formula, δ(x i ) is the sample point x of the high-precision analysis model i The scaling factor at h (x i ) is the HF model, f l (x i ) is the LF model.

[0027] Further, the step S5 is specifically as follows:

[0028] S501: Fusion of new LF dataset D_ LF’ And HF datasets with different label rates D_ HF As the variable credibility data set, a Gaussian regression process variable credibility machining deformation prediction model is trained with various milling parameters as input and machining deformation as output;

[0029] S502: Introduce the mean absolute percentage error (MAPE) to evaluate the accuracy of the constructed prediction model to achieve deformation prediction in the milling process of thin-walled parts.

[0030]

[0031] Where D real and D predicted are the actual value and predicted value of machining deformation, respectively, and n is the number of samples in the test set;

[0032] The beneficial effects of the present invention are as follows: the present invention proposes an efficient prediction method for thin-walled parts milling deformation for CNC machine tool thin-walled parts milling process. When considering the influence of variable milling parameters on thin-walled parts milling deformation under different tool parameters in actual processing, the problem of limited data samples is often faced. A variable credibility machining deformation prediction model is established based on the Gaussian regression process GPR, and the difference between high-fidelity data and low-fidelity data is corrected by additive scaling, so as to improve the accuracy and reliability of thin-walled parts milling deformation prediction. The specific advantages are as follows:

[0033] ① The present invention establishes a variable credibility machining deformation prediction model based on the Gaussian regression process, which is helpful for effective modeling under limited data samples, avoids the limitations of small sample learning, and improves the accuracy of deformation prediction in thin-walled parts milling. At the same time, the model has better effect than the prediction model established directly using source data when the sample is limited, and can ensure the improvement of prediction accuracy while reducing the cost of simulation experiments.

[0034] ② The present invention uses an additive scaling algorithm to reduce the difference between high-credibility data and low-credibility data, which can further improve the accuracy of the thin-walled part milling deformation prediction model based on the variable credibility approximate model. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following figure for explanation:

[0036] Figure 1A schematic diagram of the process framework of an efficient prediction method for milling deformation of thin-walled parts;

[0037] Figure 2 This is a schematic diagram of finite element simulation of milling of thin-walled parts;

[0038] Figure 3 Schematic diagram of the locations for extracting deformation data. DETAILED DESCRIPTION

[0039] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0040] An efficient prediction method for deformation of thin-walled parts during milling Figure 1 As shown in Table 1, when selecting tools and processing parameters, we followed reasonable principles and referred to the "Machining Technology Manual". Two carbide end mills with different geometric parameters were selected for case analysis. The specific parameters of the tools are shown in Table 1. The workpiece is made of TC4 titanium alloy, and its material parameter performance is density: 4430kg / m 3 , elastic modulus: 113000Mpa, Poisson's ratio: 0.342, specific heat 546J / kg·℃, thermal conductivity: 7.0W / m·℃; the relevant model attribute parameters of the workpiece material are shown in Table 2 and Table 3, and the variation range of each milling processing parameter is shown in Table 4.

[0041] Based on the finite element simulation software Abaqus, Figure 2 The finite element simulation model of thin-walled parts milling shown in the figure has a workpiece size of 50mm×4mm×25mm for titanium alloy TC4. The boundary conditions, contact relationship and cutting force loading mode of the workpiece are fully considered in the model, and the mesh division is based on the principle of accurate output of deformation data of units and nodes.

[0042] Table 1 Parameter information of different end mills

[0043]

[0044] Table 2 JC constitutive model parameter information of workpiece material

[0045]

[0046] Table 3 JC failure model parameter information of workpiece materials

[0047]

[0048] Table 4 Cutting parameters range

[0049]

[0050] The working condition corresponding to tool T1 is selected as the source domain, and the working condition corresponding to the other tool is selected as the target tool T2. According to the milling parameter range, the Latin hypercube method is used to design and plan 64 groups of experimental schemes. The experimental schemes are shown in Table 5.

[0051] Table 5 Experimental plan

[0052]

[0053] Replace the tool according to Table 1, perform finite element simulation of thin-walled part milling according to Table 5, and collect Figure 3 The machining deformation data of the 9 points shown, the machining deformation data of the 2 tools and the 9 deformation points corresponding to each other can obtain 2×9×64=1152 sets of deformation data.

[0054] Select the deformation data of tool T1 to construct the low-confidence data set D_ LF , according to the formula D_ HF =round(Num_data*ratio), ratio=0.1,0.2,…,0.8 Planning tool T2 simulation experiment deformation data of different ratios to build label sample data set D_ HF ; Adopt additive scaling to narrow the gap between LF and HF data and update the original LF data into a new LF dataset D_ LF’ .

[0055] Gaussian process is a random process based on Bayesian statistical data theory. In kernel machine learning, it has a strong probabilistic learning method. When considering the influence of variable milling parameters on the milling deformation of thin-walled parts under different tool parameters in actual processing, we often face the problem of limited data samples. Gaussian process regression (GPR) has good adaptability in small sample problem learning and can effectively model with limited data. Therefore, based on the Gaussian regression process, the 64 sets of data D_ of tool T1 are fused. LF’ And the HF data D_ of tool T2 at different label rates HF As the training set, each milling processing parameter {n,ae,ap,f z} is used as input, and a variable credibility prediction model for tool T2 machining deformation is established. The deformation prediction results of the corresponding 9 points are calculated, among which the GPR hyperparameter optimization sets the mean mean=10, the covariance function cov=

[000] , and the likelihood function lik=-2.5.

[0056] In order to verify the effectiveness of the method of the present invention, the machining deformation prediction model that does not use additive scaling to narrow the gap between high and low confidence data under the method of the present invention is compared, and the accuracy of the constructed prediction model is evaluated using MAPE, and the formula is as follows:

[0057]

[0058] Where D real and D predicted are the actual value and predicted value of machining deformation respectively, and n is the number of samples in the test set.

[0059] The prediction results of the corresponding 9 points before and after the additive scaling change are shown in Table 6.

[0060] Table 6 Prediction results of 9 points before and after additive scaling changes

[0061]

[0062] Note: GPR refers to the variable credibility prediction model without additive scaling transformation, and GPR-AS refers to the variable credibility prediction model after additive scaling transformation.

[0063] The model was tested at different label rates. The results show that the prediction accuracy of the variable credibility model (GPR-AS) adjusted by additive scaling is significantly improved. When the label rate is 0.1, the maximum average prediction error MAPE of GPR is 25.19%; when the label rate is 0.4, the average prediction error of GPR is 16.85%; when the label rate is 0.8, the minimum average prediction error MAPE is 5.38%. When the label rate is 0.1, the maximum average prediction error MAPE of GPR-AS is 21.05%; when the label rate is 0.4, the average prediction error of GPR-AS is 12.27%; when the label rate is higher than 0.4, the prediction error of GPR-AS is further reduced, and the minimum average prediction error MAPE is 3.39%. Therefore, the additive scaling method can narrow the gap between high- and low-credibility data, and is more conducive to assisting high-credibility data modeling. When the label rate of the proposed method GPR-AS is 0.5, the average MAPE drops below 10%, showing the best balance effect. It can not only save the finite element simulation calculation cost, but also achieve efficient deformation prediction, verifying the effectiveness of the efficient deformation prediction method for thin-walled parts milling based on the variable credibility approximate model proposed in this invention.

[0064] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. An efficient prediction method for milling deformation of thin-walled parts, characterized in that: The method comprises the following steps: S1: By rationally selecting cutting tools and processing materials, a finite element model for thin-walled parts milling is established based on the finite element simulation software Abaqus; S2: Use the Latin hypercube method to design the experimental plan, conduct simulation experiments through finite element software, obtain sufficient simulation processing deformation data and construct the LF data set; S3: Select different experimental schemes to construct a small sample data set, and perform simulation experiments using finite element software to construct a limited HF data set; S4: Adopt additive scaling to narrow the gap between LF and HF data and generate an updated low-confidence dataset; S5: The new LF data and HF data with different label rates are integrated as variable credibility data sets, and a variable credibility machining deformation prediction model with various milling parameters as input and machining deformation as output is trained to realize deformation prediction in the milling process of thin-walled parts.

2. The method for efficiently predicting deformation of thin-walled parts during milling according to claim 1, characterized in that: The step S1 is specifically as follows: Based on the principle of reasonable tool and processing parameter selection, with reference to the Mechanical Processing Technology Manual, the tools, materials and milling parameter ranges for thin-walled parts milling were selected, and the finite element simulation model of thin-walled parts milling was established using the finite element simulation software Abaqus.

3. The method for efficiently predicting deformation of thin-walled parts during milling as claimed in claim 1, characterized in that: The step S2 is specifically as follows: S201: Designing an experimental scheme using the Latin hypercube method according to the milling processing parameter selection range of step S1; S202: Determine a basic (source) tool, perform 64 corresponding simulation experiments through finite element software, obtain and organize sufficient simulation processing deformation data, and construct LF data set D_ with milling parameters and corresponding processing deformation. LF .

4. The method for efficiently predicting deformation of thin-walled parts during milling as claimed in claim 1, characterized in that: The step S3 is specifically as follows: Determine a new (target) tool, use the label sample formula to select experimental schemes with different proportions to construct a small sample data set, conduct finite element simulation experiments, obtain and organize the machining deformation data under the corresponding milling parameters of the target tool, and construct a limited HF data set D_ HF : D_ HF =round(Num_data*ratio),ratio=0.1,0.2,…,0.8 In the formula, Num_data represents the number of target domain data samples, and ratio represents different ratios.

5. The method for efficiently predicting deformation of thin-walled parts during milling as claimed in claim 1, characterized in that: The step S4 is specifically as follows: According to steps S2 and S3, the LF and HF data sets are obtained, and the gap between the LF and HF data is reduced by using additive scaling, and the original LF data is updated to the new LF data set D_ LF’ , In the formula, δ(x i ) is the sample point x of the high-precision analysis model i The scaling factor at h (x i ) is the HF model, f l (x i ) is the LF model.

6. The method for efficiently predicting deformation of thin-walled parts during milling as claimed in claim 1, characterized in that: The step S5 is specifically as follows: S501: Fusion of new LF dataset D_ LF’ And HF datasets with different label rates D_ HF As the variable credibility data set, a Gaussian regression process variable credibility machining deformation prediction model is trained with various milling parameters as input and machining deformation as output; S502: Introduce the mean absolute percentage error (MAPE) to evaluate the accuracy of the constructed prediction model to achieve deformation prediction in the milling process of thin-walled parts. Where D real and D predicted are the actual value and predicted value of machining deformation respectively, and n is the number of samples in the test set.

Images