Deterministic cutting compensation processing method for infrared thin-wall optical element

Abstract

The invention discloses a deterministic cutting compensation processing method for an infrared thin-wall optical element. The method comprises the following steps: S1, acquiring an initial surface shape of a workpiece; the method comprises the following steps: measuring an initial surface shape of a thin-wall optical element to be machined through an ultra-precision machine tool in-situ measurement technology to obtain a surface shape outline rho (x, y), and calculating a material removal amount deltaz (x, y) at each coordinate position according to a designed surface shape outline z (x, y), deltaz (x, y) = rho (x, y)-z (x, y); s2, elastic deformation prediction of clamping stress; based on a finite element simulation model, three-dimensional space numerical distribution simulation is conducted on clamping force distribution in the machining process, and element elastic deformation epsilon (x, y) caused by clamping stress is predicted; integrating the elastic deformation amount with the initial material removal amount to obtain a new material removal amount z '(x, y); s3, deformation prediction of the machining residual stress element; s4, generating a non-uniform material removal compensation track; and S5, non-uniform cutting compensation machining based on slow cutter servo is carried out.

Description

Technical Field

[0001] The invention belongs to the technical field of optical processing, and in particular relates to a deterministic cutting compensation processing method for infrared thin-walled optical elements. Background Art

[0002] Infrared thin-walled curved optical components are widely used in high-end optical instruments, infrared detection systems and other fields. The surface accuracy and surface quality of these components directly affect their optical performance, so their manufacturing accuracy is extremely high. Generally, the surface roughness of infrared thin-walled optical components is required to reach the nanometer level, and the surface accuracy must be controlled in the micron or even submicron range. During the processing, any slight error introduced by cutting, clamping and other processes will affect the final optical performance. In addition, no impurities, cracks or ripples can be introduced during the manufacturing process to ensure the accuracy and long-term stability of the optical system.

[0003] At present, ultra-precision single-point diamond cutting technology is the main means of manufacturing infrared thin-walled optical components. However, in the processing of thin-walled structures, the sources of surface errors include not only conventional processing errors, but also residual processing stress and elastic deformation caused by clamping. It is urgent to propose a non-uniform cutting compensation processing technology for infrared thin-walled optical components to compensate for surface errors in order to improve processing accuracy and meet micron or even sub-micron manufacturing requirements. This technology aims to solve the surface errors caused by residual stress and clamping stress in the ultra-precision cutting process of infrared thin-walled optical components, and ultimately achieve surface accuracy control and compensation.

[0004] Effect of residual stress on surface error: Residual stress is the internal stress in the processing process, which usually affects the surface shape of the component due to uneven deformation after material removal. Although the amplitude of residual stress can be reduced to a certain extent by optimizing the processing parameters, it is almost impossible to completely eliminate residual stress. In ultra-precision cutting, the slight deformation caused by residual stress may cause surface error, especially in thin-walled optical components. At present, research mainly focuses on reducing residual stress through process optimization, but with the continuous improvement of surface control accuracy requirements, the limitations of this method are becoming increasingly obvious.

[0005] Influence of clamping stress on surface error: In the processing of thin-walled components, clamping stress is another important source of error. Existing methods for controlling clamping stress usually reduce surface error by optimizing the fixture support structure and reducing the uneven distribution of fixture clamping force. However, the inevitable stress concentration during the clamping process will induce elastic deformation, resulting in changes in the surface shape of the component during processing. Although flexible support fixture design can be used to reduce this effect, it is still impossible to completely eliminate the influence of clamping stress on surface error.

[0006] In the manufacturing of high-precision infrared optical components, especially thin-walled components, how to effectively compensate for the residual stress and clamping stress introduced during the processing is a technical bottleneck that the industry needs to solve urgently. The generation of such surface errors greatly limits the processing accuracy of the components and hinders the performance improvement of infrared optical systems. Therefore, proposing a technical solution that can accurately compensate for these errors has important theoretical value and practical application significance. Summary of the invention

[0007] The purpose of the present invention is to solve the above problems and provide a deterministic cutting compensation processing method for infrared thin-walled optical elements that can achieve precise control of surface error by comprehensively analyzing and predicting residual stress and clamping stress during processing, establishing a corresponding compensation model, and combining deep learning and finite element simulation technology.

[0008] In order to solve the above technical problems, the technical solution of the present invention is: a deterministic cutting compensation processing method for infrared thin-walled optical elements, comprising the following steps:

[0009] S1, obtaining the initial surface shape of the workpiece;

[0010] The initial surface shape of the thin-walled optical element to be processed is measured by ultra-precision machine tool in-situ measurement technology to obtain its surface profile ρ(x, y). Based on the designed surface profile z(x, y), the material removal amount Δz(x, y) at each coordinate position is calculated.

[0011] Δz(x,y)=ρ(x,y)-z(x,y);

[0012] S2, elastic deformation prediction of clamping stress;

[0013] Based on the finite element simulation model, the three-dimensional spatial numerical distribution simulation of the clamping force distribution during the processing is carried out, and the elastic deformation ε(x, y) of the component caused by the clamping stress is predicted; this elastic deformation is combined with the initial material removal amount to obtain a new material removal amount z′(x, y), thereby eliminating the influence of the clamping stress on the component processing surface error and providing a basis for the next step of compensation calculation:

[0014] z′(x,y)=Δz(x,y)+ε(x,y);

[0015] S3, deformation prediction of components with residual stress during processing;

[0016] S4, generation of compensation trajectory for non-uniform material removal;

[0017] S5. Non-uniform cutting compensation processing based on slow tool servo.

[0018] Furthermore, the step S3 includes the following sub-steps:

[0019] S31, constructing a database T of ultra-precision cutting parameters, tool geometric parameters, and residual stress; the cutting parameters include cutting linear speed, cutting depth, and feed rate, and the tool geometric parameters include tool nose radius, front angle, and back angle;

[0020] S32, based on the process parameter-residual stress association database T, using the database to train a deep learning model to predict the residual stress under different process parameters;

[0021] S33, using the residual stress prediction model trained in step S32, combined with a finite element analysis method, calculate the component deformation δ(x, y) caused by the machining residual stress under given machining parameter conditions.

[0022] Furthermore, the S32 further includes the following sub-steps:

[0023] S321. Data collection and preparation:

[0024] The database T should contain various parameters related to the cutting process, such as cutting speed, cutting depth, feed rate, and tool geometric parameters, including tool nose radius, front angle, and back angle; obtain the residual stress data of the material under these processing conditions through experiments, or obtain the corresponding relationship between these parameters and residual stress through theoretical model calculation;

[0025] S322. Selection of deep learning model:

[0026] Choose a deep learning model that is suitable for nonlinear mapping relationships, such as multi-layer perceptron (MLP), convolutional neural network (CNN), or long short-term memory network (LSTM), depending on the complexity of the data; the model input layer corresponds to process parameters, such as cutting speed, cutting depth, feed rate, and the output layer is the predicted residual stress;

[0027] S333, Model training:

[0028] When using the data in the database Τ for training, the data needs to be preprocessed first, including normalization and standardization operations, to improve the training efficiency of the model; select a loss function, such as mean square error (MSE), and use a back propagation algorithm, such as Adam or RMSprop, to optimize the model; the training process needs to use cross-validation technology to prevent overfitting, and the prediction performance of the model can be optimized by adjusting hyperparameters such as learning rate, number of layers, number of neurons, etc.

[0029] S334, Model Verification:

[0030] The model is verified using data that has not participated in the training to ensure the accuracy and robustness of the model in predicting residual stress; the performance of the model is evaluated by calculating the error between the predicted value and the actual residual stress value, such as the mean square error or the mean absolute error.

[0031] Furthermore, the error between the predicted value and the actual residual stress value calculated in step S334 is specifically:

[0032] Assume that the trained deep learning model is represented by f θ , where θ is the parameters of the model (weights and biases), then for a given set of processing parameters (p1, p2, ..., p n ) (such as cutting speed, cutting depth), the residual stress prediction of the model can be expressed as:

[0033]

[0034] in, is the predicted residual stress;

[0035] The optimization goal during the training process is to minimize the loss function L, which is usually the mean square error (MSE) between the predicted value and the actual residual stress value:

[0036]

[0037] Among them, m is the number of training samples, R (i) is the actual residual stress of the i-th sample.

[0038] Furthermore, the step S33 also includes the following sub-steps:

[0039] S331, Residual stress prediction:

[0040] Based on the deep learning model obtained by S02, the cutting process parameters used in the actual processing are input to obtain the corresponding residual stress distribution

[0041]

[0042] in, is the predicted value of residual stress at the workpiece position (x, y); the distribution of residual stress may change at various positions of the workpiece, so it is necessary to calculate the corresponding residual stress at different coordinate positions (x, y);

[0043] S332, finite element analysis calculation:

[0044] The predicted residual stress is input into the finite element model as a boundary condition or load condition, and the influence of the residual stress on the deformation of the thin-walled optical component is calculated by finite element analysis. The finite element model should take into account the material properties (such as elastic modulus, Poisson's ratio, etc.) and geometric structure, and calculate the elastic deformation of the material by static analysis. Through the above method, the influence of residual stress on the surface deformation of the component under given processing parameters can be accurately predicted and calculated, providing basic data for subsequent compensation processing.

[0045] Furthermore, the finite element model can obtain the deformation δ(x, y) caused by residual stress at different positions (x, y) through finite element analysis, that is, the deformation of the material during the processing process. This process depends on the results of the finite element analysis software and is usually expressed as:

[0046]

[0047] Among them, FEA stands for Finite Element Analysis process, To predict the residual stress, the material properties include the elastic modulus E and Poisson's ratio ν.

[0048] Furthermore, the generation of the non-uniform material removal compensation trajectory in step S4 is based on the influence of the aforementioned clamping stress and residual stress, and the material removal amount is iteratively optimized based on the genetic algorithm to calculate the final compensated material removal amount Δz″(x, y), so that:

[0049] Δz″(x,y)=Δz′(x,y)+δ(x,y)

[0050] This result will be input into the processing equipment as a compensation trajectory for non-uniform cutting to ensure the realization of final surface accuracy.

[0051] Furthermore, the non-uniform cutting compensation processing based on slow tool servo in step S5 is to perform non-uniform cutting along the compensation trajectory calculated above through the free-form surface cutting technology of slow tool servo; the slow tool servo system can accurately control the motion trajectory of the tool, remove material layer by layer, and accurately realize the surface shape compensation of thin-walled optical elements.

[0052] The beneficial effects of the present invention are:

[0053] 1. The deterministic cutting compensation processing method for infrared thin-walled optical elements provided by the present invention first analyzes the main sources of surface errors of infrared thin-walled optical elements, including processing residual stress and clamping stress. By decomposing the surface error, these two types of stress are modeled and predicted respectively as the basis for the subsequent compensation process. Through comprehensive analysis and prediction of residual stress and clamping stress during processing, a corresponding compensation model is established, and combined with deep learning and finite element simulation technology, precise control of surface error is achieved.

[0054] 2. The present invention can effectively reduce the surface errors introduced in the processing process through accurate prediction and compensation of residual stress and clamping stress, so that the final surface accuracy of infrared thin-walled optical elements is significantly improved, meeting the accuracy requirements of micron or even sub-micron level.

[0055] 3. The present invention reduces the process of repeated measurement and correction of surface errors in traditional processes by establishing an accurate surface error compensation model, greatly improves processing efficiency and reduces production costs.

[0056] 4. The present invention can adapt to the requirements of complex surface shapes. The compensation processing method based on slow tool servo can adapt to the processing requirements of any complex surface shapes. It is not only suitable for regular surfaces, but also can be applied to ultra-precision processing of free-form surfaces.

[0057] 5. The present invention reduces costs. While achieving high-precision processing, the present invention reduces material waste and the working hours for multiple corrections and processing, further reducing production costs.

[0058] 6. The present invention can improve product reliability. The compensation technology of the present invention reduces the surface deformation caused by residual stress and clamping stress, so that the finally processed infrared thin-walled optical element has higher surface quality and more stable optical performance, and is suitable for key components in high-precision optical systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of deterministic compensating cutting processing of the deterministic cutting compensating processing method of the infrared thin-walled optical element of the present invention;

[0060] Figure 2 It is an actual surface error contour map obtained by the present invention based on the measured surface error and simulated deformation of the plane workpiece;

[0061] Figure 3 It is a comparison diagram of surface error results of machining the same workpiece using the original tool trajectory and the compensated tool trajectory according to the present invention. DETAILED DESCRIPTION

[0062] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments:

[0063] like Figure 1 , Figure 2 and Figure 3 As shown, the deterministic cutting compensation processing method for infrared thin-walled optical elements provided by the present invention comprises the following steps:

[0064] S1. Obtaining the initial surface shape of the workpiece.

[0065] The initial surface shape of the thin-walled optical element to be processed is measured by ultra-precision machine tool in-situ measurement technology to obtain its surface profile ρ(x, y), where x and y represent the coordinate positions in the xOy plane in the Cartesian coordinate system. According to the designed surface profile z(x, y), the material removal amount Δz(x, y) at each coordinate position is calculated.

[0066] Δz(x,y)=ρ(x,y)-z(x,y).

[0067] In actual use, the present invention measures the surface error of a φ38×2mm single-crystal silicon thin-walled planar optical element in situ to be ρ(x, y) after adsorption and clamping by a vacuum suction cup fixture.

[0068] S2. Prediction of elastic deformation of clamping stress.

[0069] Based on the finite element simulation model, the three-dimensional spatial numerical distribution simulation of the clamping force distribution during the processing is carried out, and the elastic deformation ε(x, y) of the component caused by the clamping stress is predicted; this elastic deformation is combined with the initial material removal amount to obtain a new material removal amount z′(x, y), thereby eliminating the influence of the clamping stress on the component processing surface error and providing a basis for the next step of compensation calculation:

[0070] z′(x,y)=Δz(x,y)+ε(x,y).

[0071] Based on the workpiece size, suction cup geometry and suction pressure value, a static model of the clamping force deformation of single-crystal silicon optical components was constructed using finite element analysis software, and the clamping deformation of the planar component was accurately calculated as ε(x, y).

[0072] Calculated The actual surface error of the single crystal silicon planar element before processing is:

[0073] z′(x,y)=ρ(x,y)+ε(x,y).

[0074] The calculated actual surface error point cloud data is plotted as a three-dimensional picture, and the result is as follows: Figure 2 shown.

[0075] S3. Prediction of deformation of components subjected to machining residual stress.

[0076] Step S3 includes the following sub-steps:

[0077] S31, constructing a database T of the association between ultra-precision cutting parameters, tool geometric parameters, and residual stress; the cutting parameters include cutting linear speed, cutting depth, and feed rate, and the tool geometric parameters include tool nose radius, front angle, and back angle.

[0078] S32. Based on the process parameter-residual stress association database Τ, a deep learning model is trained using the database to predict the residual stress under different process parameters.

[0079] Step S32 also includes the following sub-steps:

[0080] S321. Data collection and preparation:

[0081] The database T should contain various parameters related to the cutting process, such as cutting speed, cutting depth, feed rate, and tool geometry parameters, including tool nose radius, rake angle, and clearance angle. The residual stress data of the material under these processing conditions can be obtained through experiments, or the corresponding relationship between these parameters and residual stress can be obtained through theoretical model calculation.

[0082] S322. Selection of deep learning model:

[0083] Select a deep learning model suitable for nonlinear mapping relationships, such as multilayer perceptron (MLP), convolutional neural network (CNN) or long short-term memory network (LSTM), depending on the complexity of the data; the model input layer corresponds to process parameters, such as cutting speed, cutting depth, feed rate, and the output layer is the predicted residual stress.

[0084] S333, Model training:

[0085] When using the data in the database Τ for training, the data needs to be preprocessed first, including normalization and standardization operations to improve the training efficiency of the model; select a loss function, such as the mean square error (MSE), and use a back propagation algorithm, such as Adam or RMSprop, to optimize the model; the training process needs to use cross-validation technology to prevent overfitting, and the prediction performance of the model can be optimized by adjusting hyperparameters such as the learning rate, number of layers, and number of neurons.

[0086] S334, Model Verification:

[0087] The model is verified using data that has not participated in the training to ensure the accuracy and robustness of the model in predicting residual stress; the performance of the model is evaluated by calculating the error between the predicted value and the actual residual stress value, such as the mean square error or the mean absolute error.

[0088] The error between the predicted value and the actual residual stress value calculated in step S334 is specifically:

[0089] Assume that the trained deep learning model is represented by f θ , where θ is the parameters of the model (weights and biases), then for a given set of processing parameters (p1, p2, ..., p n ) (such as cutting speed, cutting depth), the residual stress prediction of the model can be expressed as:

[0090]

[0091] in, is the predicted residual stress.

[0092] The optimization goal during the training process is to minimize the loss function L, which is usually the mean square error (MSE) between the predicted value and the actual residual stress value:

[0093]

[0094] Among them, m is the number of training samples, R (i) is the actual residual stress of the i-th sample.

[0095] S33, using the residual stress prediction model trained in step S32, combined with a finite element analysis method, calculate the component deformation δ(x, y) caused by the machining residual stress under given machining parameter conditions.

[0096] Step S33 also includes the following sub-steps:

[0097] S331, Residual stress prediction:

[0098] Based on the deep learning model obtained by S02, the cutting process parameters used in the actual processing are input to obtain the corresponding residual stress distribution

[0099]

[0100] in, is the predicted value of the residual stress at the workpiece position (x, y); the distribution of residual stress may change at various positions of the workpiece, so it is necessary to calculate the corresponding residual stress at different coordinate positions (x, y).

[0101] S332, finite element analysis calculation:

[0102] The predicted residual stress is input into the finite element model as a boundary condition or load condition, and the influence of the residual stress on the deformation of the thin-walled optical component is calculated by finite element analysis. The finite element model should take into account the material properties (such as elastic modulus, Poisson's ratio, etc.) and geometric structure, and calculate the elastic deformation of the material by static analysis. Through the above method, the influence of residual stress on the surface deformation of the component under given processing parameters can be accurately predicted and calculated, providing basic data for subsequent compensation processing.

[0103] Finite element model Through finite element analysis, the deformation δ(x,y) caused by residual stress at different positions (x,y) can be obtained, that is, the deformation of the material during the processing process. This process depends on the results of the finite element analysis software and is usually expressed as:

[0104]

[0105] Among them, FEA stands for Finite Element Analysis process, To predict the residual stress, the material properties include the elastic modulus E and Poisson's ratio ν.

[0106] S4. Generation of compensation trajectory for non-uniform material removal.

[0107] The generation of the non-uniform material removal compensation trajectory in step S4 is based on the influence of the aforementioned clamping stress and residual stress, and the material removal amount is iteratively optimized based on the genetic algorithm to calculate the final compensated material removal amount Δz″(x,y), so that:

[0108] Δz″(x,y)=Δz′(x,y)+δ(x,y)

[0109] This result will be input into the processing equipment as a compensation trajectory for non-uniform cutting to ensure the realization of final surface accuracy.

[0110] S5. Non-uniform cutting compensation processing based on slow tool servo.

[0111] The non-uniform cutting compensation processing based on slow tool servo in step S5 is to perform non-uniform cutting along the above-calculated compensation trajectory through the free-form surface cutting technology of slow tool servo; the slow tool servo system can accurately control the motion trajectory of the tool, remove material layer by layer, and accurately realize the surface shape compensation of thin-walled optical elements.

[0112] In actual use, the present invention constructs an association database T between ultra-precision cutting parameters (workpiece speed = 2500RPM; cutting depth = 4μm; feed rate = 1μm / rev) and tool geometric parameters (knife nose radius = 0.5mm; front angle = -25°; back angle = 10°) and residual stress for single-crystal silicon planar optical elements. And based on the process parameter-residual stress association database, a residual stress prediction model for ultra-precision cutting of single-crystal silicon is obtained by deep learning network training. Furthermore, the finite element analysis method is used to calculate the deformation of the element under the processing residual stress condition as δ(x, y).

[0113] If the single crystal silicon plane is directly compensated according to the actual surface error z′(x, y) of the component, then the theoretical surface error of the processed single crystal silicon flat piece should be δ(x, y). The processing results in this embodiment are as follows: Figure 3As shown in the left figure, the machined surface error PV is 1.21μm.

[0114] If the residual stress and deformation are directly compensated and the material removal amount is set to z′(x,y)+δ(x,y), then the residual stress value and deformation caused by the change of cutting depth processing parameters will change accordingly, and the residual stress and deformation cannot be effectively eliminated.

[0115] Therefore, based on the residual stress prediction model of ultra-precision cutting of single crystal silicon obtained above, the genetic algorithm is used to iteratively optimize the material removal amount in different regions so that the final compensated material removal amount Δz″(x, y) satisfies:

[0116] Δz″(x,y)=z′(x,y)+δ′(x,y)

[0117] Where δ′(x, y) is the residual stress deformation under the current cutting amount Δz″(x, y).

[0118] Based on the above analysis and calculation results, a deterministic compensation cutting trajectory is obtained. The machining trajectory is input into the machining equipment. The slow tool servo turning method is used to perform non-uniform cutting along the above calculated compensation trajectory to accurately achieve the surface shape compensation of thin-walled optical components. The surface shape measurement results of the workpiece machined by this method are as follows: Figure 3 As shown in the figure on the right, the machined surface shape error PV is 0.39μm.

[0119] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A deterministic cutting compensation processing method for infrared thin-walled optical elements, characterized in that: The following steps are involved: S1, obtaining the initial surface shape of the workpiece; The initial surface shape of the thin-walled optical element to be processed is measured by ultra-precision machine tool in-situ measurement technology to obtain its surface profile ρ(x, y). Based on the designed surface profile z(x, y), the material removal amount Δz(x, y) at each coordinate position is calculated. Δz(x,y)=ρ(x,y)-z(x,y); S2, elastic deformation prediction of clamping stress; Based on the finite element simulation model, the three-dimensional spatial numerical distribution simulation of the clamping force distribution during the processing is carried out, and the elastic deformation ε(x, y) of the component caused by the clamping stress is predicted; this elastic deformation is combined with the initial material removal amount to obtain a new material removal amount z′(x, y), thereby eliminating the influence of the clamping stress on the component processing surface error and providing a basis for the next step of compensation calculation: z′(x,y)=Δz(x,y)+ε(x,y); S3, deformation prediction of components with residual stress during processing; S4, generation of compensation trajectory for non-uniform material removal; S5. Non-uniform cutting compensation processing based on slow tool servo.

2. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 1 is characterized in that: The step S3 comprises the following sub-steps: S31, constructing a database T of ultra-precision cutting parameters, tool geometric parameters, and residual stress; the cutting parameters include cutting linear speed, cutting depth, and feed rate, and the tool geometric parameters include tool nose radius, front angle, and back angle; S32, based on the process parameter-residual stress association database T, using the database to train a deep learning model to predict the residual stress under different process parameters; S33, using the residual stress prediction model trained in step S32, combined with a finite element analysis method, calculate the component deformation δ(x, y) caused by the machining residual stress under given machining parameter conditions.

3. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 1, characterized in that: The S32 further comprises the following sub-steps: S321. Data collection and preparation: The database T should contain various parameters related to the cutting process, such as cutting speed, cutting depth, feed rate, and tool geometric parameters, including tool nose radius, front angle, and back angle; obtain the residual stress data of the material under these processing conditions through experiments, or obtain the corresponding relationship between these parameters and residual stress through theoretical model calculation; S322. Selection of deep learning model: Choose a deep learning model that is suitable for nonlinear mapping relationships, such as multi-layer perceptron (MLP), convolutional neural network (CNN), or long short-term memory network (LSTM), depending on the complexity of the data; the model input layer corresponds to process parameters, such as cutting speed, cutting depth, feed rate, and the output layer is the predicted residual stress; S333, Model training: When using the data in the database Τ for training, the data needs to be preprocessed first, including normalization and standardization operations, to improve the training efficiency of the model; select a loss function, such as mean square error (MSE), and use a back propagation algorithm, such as Adam or RMSprop, to optimize the model; the training process needs to use cross-validation technology to prevent overfitting, and the prediction performance of the model can be optimized by adjusting hyperparameters such as learning rate, number of layers, number of neurons, etc. S334, Model Verification: The model is verified using data that has not participated in the training to ensure the accuracy and robustness of the model in predicting residual stress; the performance of the model is evaluated by calculating the error between the predicted value and the actual residual stress value, such as the mean square error or the mean absolute error.

4. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 1, characterized in that: The error between the predicted value and the actual residual stress value calculated in step S334 is specifically: Assume that the trained deep learning model is represented by f θ , where θ is the parameters of the model (weights and biases), then for a given set of processing parameters (p1, p2, ..., p n ) (such as cutting speed, cutting depth), the residual stress prediction of the model can be expressed as: in, is the predicted residual stress; The optimization goal during the training process is to minimize the loss function L, which is usually the mean square error (MSE) between the predicted value and the actual residual stress value: Among them, m is the number of training samples, R (i) is the actual residual stress of the i-th sample.

5. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 2, characterized in that: The step S33 also includes the following sub-steps: S331, Residual stress prediction: Based on the deep learning model obtained by S02, the cutting process parameters used in the actual processing are input to obtain the corresponding residual stress distribution in, is the predicted value of residual stress at the workpiece position (x, y); the distribution of residual stress may change at various positions of the workpiece, so it is necessary to calculate the corresponding residual stress at different coordinate positions (x, y); S332, finite element analysis calculation: The predicted residual stress is input into the finite element model as a boundary condition or load condition, and the influence of the residual stress on the deformation of the thin-walled optical component is calculated by finite element analysis. The finite element model should take into account the material properties (such as elastic modulus, Poisson's ratio, etc.) and geometric structure, and calculate the elastic deformation of the material by static analysis. Through the above method, the influence of residual stress on the surface deformation of the component under given processing parameters can be accurately predicted and calculated, providing basic data for subsequent compensation processing.

6. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 5, characterized in that: The finite element model can obtain the deformation δ(x, y) caused by residual stress at different positions (x, y) through finite element analysis, that is, the deformation of the material during the processing. This process depends on the results of the finite element analysis software and is usually expressed as: Among them, FEA stands for Finite Element Analysis process, To predict the residual stress, the material properties include the elastic modulus E and Poisson's ratio ν.

7. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 1, characterized in that: The generation of the non-uniform material removal compensation trajectory in step S4 is based on the influence of the aforementioned clamping stress and residual stress, and the material removal amount is iteratively optimized based on the genetic algorithm to calculate the final compensated material removal amount Δz″(x,y), so that: Δz″(x,y)=Δz′(x,y)+δ(x,y) This result will be input into the processing equipment as a compensation trajectory for non-uniform cutting to ensure the realization of final surface accuracy.

8. The deterministic cutting compensation processing method for infrared thin-walled optical elements according to claim 1, characterized in that: The non-uniform cutting compensation processing based on slow tool servo in step S5 is to perform non-uniform cutting along the above-calculated compensation trajectory through the free-form surface cutting technology of slow tool servo; the slow tool servo system can accurately control the motion trajectory of the tool, remove material layer by layer, and accurately realize the surface shape compensation of thin-walled optical elements.

Images