Free-form surface machine tool contact type in-situ error compensation method

Abstract

The application discloses a free curved surface machine tool contact type in-situ detection error compensation method, comprising the following steps: S1, obtaining a measurement initial compensation value through a pre-compensation method; S2, obtaining an actual measurement error value and an actual probe posture; S3, training a convolutional neural network structure to obtain a curved surface measurement error prediction model; S4, performing measurement error compensation to obtain a final measurement coordinate value; and S5, verifying measurement precision through simulation measurement and actual measurement. The measurement error compensation method does not need to deeply analyze error sources, the error identification method is simple and has strong operability, complex error compensation modeling and error decoupling processes are avoided, measurement error compensation only needs a three-coordinate measuring device, does not need a large number of complex experiments and various expensive special measuring devices, and time and fund costs can be saved.

Description

TECHNICAL FIELD

[0001] The application relates to a free curved surface machine tool contact type in-situ detection error compensation method and belongs to the technical field of digitized manufacturing and geometric measurement. BACKGROUND

[0002] A free curved surface has a complex geometric shape, and it is difficult to guarantee machining precision in traditional machining, so in-situ detection is needed to obtain geometric measurement data in the machining process so as to subsequently carry out correction machining. An in-situ detection system adopting a contact type probe is an important means for obtaining geometric information in the numerical control machining process of a workpiece, and the measurement is achieved by installing the contact type probe on a main shaft of a machine tool. The in-situ detection method controls the multi-axis action of the machine tool to make the probe end of the probe contact the workpiece to obtain a trigger signal and record the spatial coordinates at the triggering time as the measurement result. Manufacturing and assembly errors of the axes of the machine tool and pre-travel errors generated by the measurement principle of the contact type probe will limit the final measurement precision to a certain extent.

[0003] A measurement error compensation method is a necessary means to improve the in-situ detection precision. In the existing measurement error compensation methods, some methods adopt a laser tracker, a laser interferometer, an R-test, a double ball bar instrument and the like to identify and compensate linear axis and rotary axis errors of a numerical control machine tool so as to improve the measurement precision; and some other methods control the probe to trigger measurement at positions of a standard part such as a standard ball, a standard block and the like, and obtain the measurement error condition by comparing the deviation between the measured size and the actual size of the standard part. The existing methods mainly start from the source of the measurement error, and realize the overall measurement error improvement by separately compensating. However, the deficiencies of the existing methods cannot be ignored, error compensation usually needs to carry out a large number of experiments, and needs to be completed in cooperation with a plurality of expensive special instruments and equipment, and needs to establish a complex measurement error compensation model and carry out corresponding calculation, in addition, the complexity of the measurement error sources determines that it is difficult to master the influence law of all error sources on the measurement error. SUMMARY

[0004] The application provides a free curved surface machine tool contact type in-situ detection error compensation method, and the in-situ detection measurement data is corrected based on three-coordinate measurement data, and measurement error compensation at the data level is realized.

[0005] TECHNICAL SOLUTION The application provides a free curved surface machine tool contact type in-situ detection error compensation method, and the in-situ detection measurement data is corrected based on three-coordinate measurement data, and measurement error compensation at the data level is realized.

[0006] S1: Obtain the initial measurement compensation value through the pre-compensation method: set sampling points on the standard sphere, calibrate and measure the sampling points using different probe postures, compare the measurement results with the actual size of the standard sphere to obtain the measurement error, and obtain the initial measurement error compensation value from the standard database. Input the initial measurement error compensation value into the probe system to form preliminary compensation;

[0007] S2: Obtain the actual measurement error value and actual probe posture;

[0008] S3: training a convolutional neural network structure based on the actual probe posture and actual measurement error value in S2 to obtain a surface measurement error prediction model, and using the surface measurement error prediction model to predict the measurement error of the surface to be measured;

[0009] S4: Compensate for the measurement error based on the measurement error prediction value obtained in S3 to obtain the final measurement coordinate value;

[0010] S5: Verify the measurement accuracy through simulation and actual measurement.

[0011] Preferably, the S1 includes:

[0012] S101: Set n sampling points on the standard sphere;

[0013] S102: Fit the sampling points in S101 to obtain the coordinates of the sphere center C (x0, y0, z0);

[0014] S103: Calibrate and measure the standard sphere using different probe positions and postures to obtain the measured coordinates of each sampling point;

[0015] S104: Correcting the ball center position C New ;

[0016] S105: Get the measurement error value Err i ;

[0017] S106: Get the error value Err in the standard database i The corresponding initial measurement error compensation values ​​in each direction are input into the probe system to form preliminary compensation.

[0018] Preferably, the S101 is specifically:

[0019] The contact point between the standard ball surface and the extended line of the probe axis is set as the initial sampling point. The angle between adjacent sampling points is θ, and the angle between each sampling point and the probe axis is Set the values ​​of m and n according to the calibration accuracy requirements. m represents the number of rings on the surface of the standard sphere.

[0020] Preferably, the S104 is specifically:

[0021] The distance from the initial sampling point coordinate value to the center of the sphere is the standard sphere radius R b The radius r of the probe p The sum d s :

[0022] d s =R b +r p

[0023] Assuming that the measured coordinate point obtained at the initial sampling point is P1(x1, y1, z1), the corrected sphere center position C New for:

[0024]

[0025] in, Represents the vector from point C to point P1.

[0026] Preferably, the S105 is specifically:

[0027] Calculate C New The coordinates of each sampling point are measured {P i (x i ,y i ,z i )|i=1,2,…,m*n} i , then the error value Err i for:

[0028]

[0029] Preferably, the S2 is specifically:

[0030] Sampling points are set on the surface to be measured, and the measurement data of the sampling points are measured by a three-dimensional coordinate measuring machine as the reference data, which is measured using a standard probe. The actual measurement data of the sampling points are measured using an in-situ measurement method, which is measured using the probe compensated in S1. The difference between the actual measurement data and the reference data is the measurement error value of each sampling point in the X, Y, and Z coordinate directions.

[0031] The measurement path of the surface sampling points is planned according to the interference between the probe and the workpiece and the sampling efficiency, and the measurement position information, namely the five-axis movement information of the machine tool X, Y, Z, B, and C and the I, J, and K direction information of the probe axis, is recorded to obtain the probe posture when measuring each sampling point.

[0032] Preferably, the S3 is specifically:

[0033] The convolutional neural network structure includes an input layer, a convolution layer, a ReLU activation function layer, a pooling layer, a fully connected layer and an output layer; the probe posture is used as an input parameter and the measurement error value is used as an output parameter to jointly train the convolutional neural network structure. After the training is completed, a surface measurement error prediction model is obtained, and the surface measurement error prediction model is used to measure and predict the surface to be measured to obtain the measurement errors △X, △Y, and △Z of the surface to be measured.

[0034] Preferably, the S4 is specifically:

[0035] The sum of the actual coordinate value XYZ of the sampling point obtained by in-situ measurement and the error prediction results △X, △Y, and △Z obtained in S3 is the final measured coordinate value.

[0036] Preferably, the simulation measurement of S5 is specifically as follows:

[0037] The method of equally dividing the surface parameter domain and the adaptive method are used to obtain dense point clouds and sparse point clouds respectively. The sampling results of the equal-parameter method are used as the training set, and the samples obtained by the adaptive sampling method are used as the validation set. The sampling points are obtained by the equal-parameter sampling method and the adaptive sampling method. The sampling points obtained by the equal-parameter sampling method are used for convolutional neural network structure training, thereby generating error distribution experience. The training process obtains measurement experience of in-situ detection in the machine tool workspace. The sampling points obtained by the feature-based sampling method are used as the validation set to test the prediction effect of the error compensation method on measurement errors.

[0038] Preferably, the actual measurement of S5 is specifically:

[0039] A three-coordinate measuring machine and in-situ measurement are used to measure the actual curved surface and obtain measurement results respectively. The curved surface measured by in-situ measurement is compared with the curved surface measured by three-coordinate measurement to obtain the in-situ detection error, and then the maximum error and average error of the in-situ measurement system before compensation are obtained; the maximum error and average error of the in-situ measurement can be calculated based on the measurement error predicted and compensated by the trained convolutional neural network structure.

[0040] Beneficial effects: The measurement error compensation method of the present invention does not require in-depth investigation of the error source, the error identification method is simple and highly operational, and the complex error compensation modeling and error decoupling process is avoided; measurement error compensation only requires three-coordinate measuring equipment, and does not require a large number of complex experiments and the participation of multiple expensive special measuring equipment, which can save time and financial costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0042] Figure 1 Flow chart of the method of the present invention;

[0043] Figure 2 The whole process flow chart of the method of the present invention is as follows;

[0044] Figure 3 (a) A view perpendicular to the plane formed by the axis of the stylus and the axis of the standard ball shaft;

[0045] Figure 3 (b) is a schematic diagram of the angle θ between two calibration points on the ring on the surface of the standard sphere;

[0046] Figure 4 (ac) are schematic diagrams of probe calibration;

[0047] Figure 5 is the probe calibration result;

[0048] Figure 6 is the correspondence between the measurement position and the error;

[0049] Figure 7 It is a convolutional neural network structure;

[0050] Figure 8 Training samples and validation samples for the simulation process;

[0051] Figure 9 (ab) are the simulation results of measurement error compensation;

[0052] Figure 10 Training samples and validation samples for the actual measurement process;

[0053] Figure 11 Schematic diagram of free-form surface measurement;

[0054] Figure 12 is the measurement error diagram before compensation;

[0055] Figure 13 This is the measurement error diagram after compensation. DETAILED DESCRIPTION

[0056] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0057] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0058] In the present application, unless otherwise explicitly specified and limited, "on" or "under" of the first feature to the second feature can include that the first and second features are in direct contact, or that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, "on", "above" and "above" of the first feature to the second feature include that the first feature is directly above and obliquely above the second feature, or only indicates that the horizontal height of the first feature is higher than that of the second feature. "Below", "below" and "below" of the first feature to the second feature include that the first feature is directly below and obliquely below the second feature, or only indicates that the horizontal height of the first feature is less than that of the second feature.

[0059] The free-form surface in the embodiment refers to a surface that can be expressed by a B-spline, a non-uniform rational B-spline, a T-spline curve, and is not limited to the expression form of the surface, and can be a continuous surface or a surface represented by a discrete point cloud.

[0060] As shown in Figure 1 and Figure 2 A free-form surface machine tool contact type in-situ detection error compensation method, comprising the following steps:

[0061] S1: obtaining a measurement initial compensation value by a pre-compensation method: setting sampling points on a standard ball, using different probe attitudes to calibrate and measure the sampling points, comparing the measurement results with the actual size of the standard ball to obtain measurement errors, and obtaining an initial measurement error compensation value from the standard database; inputting the initial measurement error compensation value into the probe system to form a preliminary compensation;

[0062] As shown in Figure 3 S101: setting n sampling points on a standard ball;

[0063] The contact point between the standard ball surface and the extended line of the probe axis is set as the initial sampling point. The angle between adjacent sampling points is θ, and the angle between each sampling point and the probe axis is Set the values ​​of m and n according to the calibration accuracy requirements. m represents the number of rings on the surface of the standard sphere.

[0064] During sampling, the ruby ​​ball at the end of the probe approaches and touches the sampling point along the normal direction of the sampling point, and records the coordinate data measured at the point. Figure 3 (a) is a view perpendicular to the plane formed by the stylus axis and the standard ball shaft axis. A sampling point is set at the point closest to the stylus ruby ​​ball. B is the angle between the stylus axis and the standard ball shaft axis. There are n sampling points evenly distributed on the circular ring of the standard sphere surface, and the angle between any two sampling points is θ, such as Figure 3 (b) The values ​​of m and n are set according to the calibration accuracy requirements. The larger the values ​​of m and n are, the higher the calibration accuracy is but the longer the calibration time is. Figure 3 In the example: m = 3, n = 8, in actual implementation, m and n can select different values ​​according to the calibration accuracy requirements. Figure 4 (a-c) show the calibration of the probe using different probe postures. The purpose of this test is to verify the influence of the probe posture on the measurement error of the in-situ detection system during measurement.

[0065] S102: Fit the sampling points in S101 to obtain the coordinates of the sphere center C (x0, y0, z0);

[0066] S103: Calibrate and measure the standard sphere using different probe positions and postures to obtain the measured coordinates of each sampling point. Since the coordinates of the sphere center are known, the obtained measured coordinates have higher accuracy.

[0067] S104: Correcting the ball center position C New :

[0068] When the probe contacts the standard ball in the axial direction, the deformation of the stylus and the pre-travel error are very small, and the measurement error is the smallest compared to other trigger directions. Therefore, when the probe measures in the axial direction, its measurement error is ignored. Figure 3 The distance from the initial sampling point coordinates shown in (a) to the center of the sphere is the radius R of the standard sphere. b The radius r of the probe p The sum d s :

[0069] d s =R b +r p

[0070] The new sphere center position C can be inferred based on the measured coordinate values ​​obtained at the sampling point. New (x New0 ,y New0 ,z New0 ), assuming that the measured coordinate point obtained at the initial sampling point is P1(x1,y1,z1), then the corrected sphere center position C New for:

[0071]

[0072] in, Represents the vector from point C to point P1.

[0073] S105: Get the measurement error value Err i :

[0074] Calculate C New The coordinates of each sampling point are measured {P i (x i ,y i ,z i )|i=1,2,…,m*n} i , then the error value Err i for:

[0075]

[0076] S106: Get the error value Err in the standard database i The corresponding initial measurement error compensation values ​​in each direction are input into the probe system to form preliminary compensation. After the pre-compensation method is processed, the corresponding Figure 4 The calibration results, such as Figure 5 shown.

[0077] S2: Get the actual measurement error value and actual probe posture:

[0078] Sampling points are set on the surface to be measured, and the measurement data of the sampling points are measured by a three-dimensional coordinate measuring machine as the reference data, which is measured using a standard probe. The actual measurement data of the sampling points are measured using an in-situ measurement method, which is measured using the probe compensated in S1. The difference between the actual measurement data and the reference data is the measurement error value of each sampling point in the X, Y, and Z coordinate directions.

[0079] The measurement path of the surface sampling points is planned according to the interference between the probe and the workpiece and the sampling efficiency, and the measurement position information, namely the five-axis movement information of the machine tool X, Y, Z, B, and C and the I, J, and K direction information of the probe axis, is recorded to obtain the probe posture when measuring each sampling point.

[0080] like Figure 6 As shown in the figure, assuming that different probe postures are used at different locations on the surface to be measured, different measurement errors will also occur, and there is a complex corresponding relationship between the two. Under this assumption, the equal parameter method is first used to relatively densely plan a certain number of sampling points on the free-form surface of the workpiece clamped at a specific position on the machine tool. Then, the measurement path is planned, and the planned sampling points are measured respectively using in-situ detection and three-dimensional coordinate measuring machine. After measurement, the data are recorded. Figure 6 The input and output data shown are as follows: the input data, namely the five-axis movement information of the machine tool X, Y, Z, B, and C and the I, J, and K direction information of the probe axis, are read directly from the planned measurement path; the output data are the errors △X, △Y, and △Z in the three coordinate axis directions when measuring the same sampling point between the in-situ inspection and the three-coordinate measuring machine.

[0081] S3: Based on the actual probe posture and actual measurement error value in S2, the convolutional neural network structure is trained to obtain the surface measurement error prediction model, and the surface measurement error is predicted using the surface measurement error prediction model:

[0082] The convolutional neural network structure includes an input layer, a convolution layer, a ReLU activation function layer, a pooling layer, a fully connected layer and an output layer; the probe posture in S2 is used as an input parameter, and the measurement error value in S2 is used as an output parameter. The convolutional neural network structure is trained together. After the training is completed, a surface measurement error prediction model is obtained. The surface measurement error prediction model is used to measure and predict the surface to be measured, and the measurement errors △X, △Y, and △Z of the surface to be measured are obtained.

[0083] The model established is as Figure 7 As shown, the input layer is an 8*1 image, which is Figure 6 The input shown is a 3*1 image, which is the output layer. Figure 6 The output is shown. Following the input layer is a convolutional layer, which is grouped with a ReLU activation function layer. Three pairs of convolutional layers and ReLU activation function layers are used, with the number of convolution kernels in each layer being 128, 256, and 512, respectively. This layer is followed by a fully connected layer that outputs the regression prediction value. This architecture does not include a pooling layer because its primary function is to reduce the size and complexity of features, thereby reducing computational effort. The feature maps in this model are not complex, so this layer is not required. Figure 7 This is only an implementation case, and the present invention is not limited to the structure shown in this embodiment. For specific measurement error compensation problems, the error prediction model established by selecting different layers and optimizing the convolutional neural network structure is also within the framework of the present invention.

[0084] S4: Compensate for the measurement error based on the measurement error prediction value obtained in S3 to obtain the final measurement coordinate value:

[0085] The sum of the actual coordinate values ​​X, Y, and Z of the sampling point obtained by in-situ measurement and the error prediction results △X, △Y, and △Z obtained in S3 is the final measured coordinate value.

[0086] S5: Verify measurement accuracy through simulation and actual measurement:

[0087] The simulation measurements are as follows:

[0088] like Figure 8 As shown in the figure, the method of equally dividing the surface parameter domain and the adaptive method are used to obtain dense point clouds and sparse point clouds respectively. The sampling results of the equal parameter method are used as the training set, and the samples obtained by the adaptive sampling method are used as the validation set. There are 30*30=900 sampling points obtained by the equal parameter sampling method, and the number of sampling points obtained by the adaptive sampling method is 57. The 900 sampling points obtained by the equal parameter sampling method are used for CNN training, thereby generating error distribution experience. The training process obtains measurement experience of in-situ detection in the workspace of the machine tool. The 57 sampling points obtained by the feature-based sampling method are used as a validation set to test the prediction effect of the error compensation method on the measurement error. The experimental results are shown in Figure 9 As shown in (ab), the error compensation method proposed in the present invention can effectively predict the measurement errors in the three coordinate directions of X, Y, and Z. Figure 9 (b) It is easy to observe that the values ​​of measurement accuracy improvement after compensation are all positive, that is, the measurement accuracy at all sampling points is improved after compensation. Through calculation, it can be obtained that the average measurement error after compensation is reduced by 0.0316mm relative to that before compensation. The average value of the original measurement error is 0.0337mm. The average error after compensation is reduced by 93.8% relative to the original measurement error, and the measurement accuracy is significantly improved.

[0089] The actual measurements are as follows:

[0090] like Figure 10 The training set and validation set were generated by a similar method to the simulation experiment. Figure 11 The three-coordinate measuring machine and the in-situ detection respectively measure the actual surface and obtain the measurement results. The in-situ detection error can be measured by comparing the surface measured by the in-situ detection with the surface measured by the three-coordinate measurement. The distribution of the in-situ detection error is shown in the figure. Figure 12 As shown in the figure, the maximum error of the in-situ detection system before compensation is 0.0615mm, and the average error is 0.0226mm. The measurement error after prediction and compensation using the trained convolutional neural network is shown in the figure. Figure 13 As shown in the figure, the maximum error of in-situ detection is calculated to be 0.0277mm and the average error is 0.0075mm. Figure 12As shown in the error distribution, the in-situ detection measurement accuracy after compensation is improved by 55.0% in terms of maximum error and 64.8% in terms of average error.

[0091] It should be noted that in simulation and actual verification, the sampling method uses the equal parameter method to generate the training set, while the adaptive method generates the verification set. This does not represent the sampling requirements of the present invention. In fact, the data for generating the training set only needs to be densely distributed on the surface. The degree of density depends on the accuracy requirements of the application. The adaptive method can also adopt different methods.

[0092] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0093] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A contact-type in-situ detection error compensation method for a free-form surface machine tool, characterized by: The following steps are involved: S1: Obtain the initial measurement compensation value through the pre-compensation method: set sampling points on the standard sphere, calibrate and measure the sampling points using different probe postures, compare the measurement results with the actual size of the standard sphere to obtain the measurement error, and obtain the initial measurement error compensation value from the standard database. Input the initial measurement error compensation value into the probe system to form preliminary compensation; S2: Obtain the actual measurement error value and actual probe posture; Set sampling points on the surface to be measured, use a three-dimensional coordinate measuring machine to measure the measurement data of the sampling points as the benchmark data, and use a standard probe for measurement; The actual measurement data of the sampling points are measured using the in-situ measurement method, which uses the compensated probe in S1 for measurement; the difference between the actual measurement data and the reference data is the difference between the actual measurement data and the reference data for each sampling point. Measurement error values ​​in three coordinate directions; According to the interference between the probe and the workpiece and the sampling efficiency, the measurement path of the surface sampling point is planned and the measurement position information is recorded. Five-axis movement information and probe axis Direction information, and then obtain the probe posture when measuring each sampling point; S3: training a convolutional neural network structure based on the actual probe posture and actual measurement error value in S2 to obtain a surface measurement error prediction model, and using the surface measurement error prediction model to predict the measurement error of the surface to be measured; The convolutional neural network structure includes an input layer, a convolution layer, a ReLU activation function layer, a pooling layer, a fully connected layer and an output layer; the probe posture is used as an input parameter and the measurement error value is used as an output parameter to jointly train the convolutional neural network structure. After the training is completed, a surface measurement error prediction model is obtained, and the surface measurement error prediction model is used to measure and predict the surface to be measured to obtain the measurement error of the surface to be measured. 、 、 ; S4: Compensate for the measurement error based on the measurement error prediction value obtained in S3 to obtain the final measurement coordinate value; S5: Verify the measurement accuracy through simulation and actual measurement.

2. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 1, characterized in that: Said S1 comprises: S101: Setting on a standard ball sampling points; S102: Fit the sampling points in S101 to obtain the coordinates of the sphere center ; S103: Calibrate and measure the standard sphere using different probe positions and postures to obtain the measured coordinates of each sampling point; S104: Correct the center position of the ball ; S105: Get the measurement error value ; S106: Obtain error values ​​in the standard database The corresponding initial measurement error compensation values ​​in each direction are input into the probe system to form preliminary compensation.

3. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 2, characterized in that: The S101 is specifically as follows: The contact point between the standard ball surface and the extended line of the probe axis is set as the initial sampling point. , ,…, The standard sphere surface is evenly distributed on the ring sampling points, and the angle between any two sampling points on each ring is , set according to calibration accuracy requirements and The value of , Indicates the number of rings on the surface of a standard sphere.

4. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 2, characterized in that: The S104 is specifically as follows: The distance from the initial sampling point coordinate value to the center of the sphere is the radius of the standard sphere Radius of the probe sum : ； Assume that the measurement coordinate point obtained at the initial sampling point is , then the corrected sphere center position is for: ； in, Indicates a point Arrive vector.

5. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 2, characterized in that: The S105 is specifically as follows: calculate Measure coordinates of each sampling point The distance between , then the error value for: ； 6. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 1, characterized in that: The S4 is specifically: The actual coordinate values ​​of the sampling points obtained by in-situ measurement And the error prediction results obtained in S3 、 、 The sum of the two is the final measured coordinate value.

7. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 1, characterized in that: The simulation measurement of S5 is specifically as follows: The method of equally dividing the surface parameter domain and the adaptive method are used to obtain dense point clouds and sparse point clouds respectively. The sampling results of the equal-parameter method are used as the training set, and the samples obtained by the adaptive sampling method are used as the validation set. The sampling points are obtained by the equal-parameter sampling method and the adaptive sampling method. The sampling points obtained by the equal-parameter sampling method are used for convolutional neural network structure training, thereby generating error distribution experience. The training process obtains measurement experience of in-situ detection in the machine tool workspace. The sampling points obtained by the feature-based sampling method are used as the validation set to test the prediction effect of the error compensation method on measurement errors.

8. The contact-type in-situ detection error compensation method for free-form surface machine tools according to claim 1, characterized in that: The actual measurement of S5 is specifically as follows: A three-coordinate measuring machine and in-situ measurement are used to measure the actual curved surface and obtain measurement results respectively. The curved surface measured by in-situ measurement is compared with the curved surface measured by three-coordinate measurement to obtain the in-situ detection error, and then the maximum error and average error of the in-situ measurement system before compensation are obtained; the maximum error and average error of the in-situ measurement can be calculated based on the measurement error predicted and compensated by the trained convolutional neural network structure.

Images