Numerical control machine tool machining error analysis method

Abstract

The invention discloses a numerical control machine tool machining error analysis method, and belongs to the field of numerical control machine tools. The method comprises the following steps: measuring error data of each actual point on an actual contour of a processed workpiece, backtracking the error data to error data of a tool nose point in a working space of a machine tool, and normalizing process parameters, coordinates of actual measurement points and rotation angles a and c of a rotating shaft to form neural network input features; forming a data set by using the obtained pose error and the input feature; the method comprises the steps of dividing a data set into a training set, a verification set and a test set, constructing an improved Transformer model, training the model by adopting data of the training set, recording parameters when a loss function of the verification set reaches a minimum value, evaluating a pose error of a tool nose point on the test set by using the parameters, and improving the analysis efficiency and precision of a machining error. The reverse error of the machine tool is predicted in real time, and a basis is provided for timely adjustment of machining parameters.

Description

Technical Field

[0001] The invention relates to the field of numerically controlled machine tools, and in particular to a method for analyzing machining errors of numerically controlled machine tools. Background Art

[0002] In the field of modern machining, five-axis CNC machine tools play a vital role in modern manufacturing with their excellent flexibility and high precision. Such machine tools can achieve coordinated movement of five independent axes, allowing the tool to approach the workpiece in any direction in three-dimensional space, thereby efficiently processing various complex curved surface parts. Given the continuous improvement in the requirements for part accuracy and quality in industrial production, it is particularly important to accurately analyze the machining errors of five-axis CNC machine tools.

[0003] Traditional error analysis methods, such as empirical methods, statistical analysis, and physical modeling, are effective to a certain extent, but they have many limitations. On the one hand, these methods are difficult to accurately predict errors under complex processing conditions; on the other hand, their versatility and real-time performance are poor. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, one of the objectives of the present invention is to provide an analysis method that is highly versatile and can accurately predict machining errors of CNC machine tools under complex machining conditions.

[0005] One of the purposes of the present invention is achieved by the following technical solution: A method for analyzing machining errors of a numerical control machine tool comprises the following steps: Parts processing: CNC machine tools process parts according to processing parameters; Obtain error data: measure the position error of each actual point on the actual contour of the machined workpiece and normal error , and record the error data of each actual point ,in Represents the position data of the three coordinate directions of the actual measurement point, j Representative j workpiece, i represents the i Actual measurement points; Error data conversion: The error data of each actual point Error data traced back to the tool tip point in the machine tool workspace ,in is the position data of the tool tip in three coordinate directions, is the position error of the tool tip, is the normal error of the tool tip, Represents the position of the origin of the machine tool coordinate system in the coordinate system of the actual point being measured; Composition of data set: Processing parameters, The rotation angles a and c of the rotation axis are normalized to form the input features of the neural network, and the obtained posture error is And input features form a data set; Data training and error prediction: Divide the data set into training set, validation set and test set, build an improved Transformer model and use the data of the training set to train the Transformer model. When the loss function of the validation set reaches the minimum value, record the model parameters and use these parameters to evaluate the position error of the tool tip on the test set.

[0006] Furthermore, the processing parameters include the spindle speed S , tool diameter D , Feed speed F , Axial cutting depth A p , Radial cutting width A e .

[0007] Furthermore, in the step of forming the data set, the processing parameters, And the normalization processing of the rotation axis angles a and c is specifically as follows: ; in, w i For the dataset i Dimensional data, w min is the minimum value of the data in the dataset dimension, w max is the maximum value of the data set dimension. is the normalized data.

[0008] Furthermore, the CNC machine tool processing error analysis method also includes data optimization, and the data optimization step is located between the data set composition step and the data training and error prediction step. The data optimization step is specifically: Statistical data collection of a type of posture error data with the same process parameters, position, and rotation angle of the rotation axis ,in, k Represents the same parameters, position, and rotation angle of the rotation axis. k Error data; The truncated mean represents the pose error data with the same parameters, position, and rotation angle of the rotation axis.

[0009] Furthermore, the calculation formula for the truncated mean of the pose error is: ; in,n Represents the total amount of data, α It means to remove the proportion. α Take 3-7%.

[0010] Furthermore, in the data training and error prediction steps, the improved Transformer model is specifically: The embedding operation in the standard Transformer model is cancelled, and a 4-layer lightweight encoder and decoder are used. The final output is composed of the pose error prediction result of the fully connected layer and the probability information of the Sigmoid layer. The improved Transformer model includes a lightweight encoder, a lightweight decoder, and an output layer. The encoder includes a multi-head self-attention layer, a residual connection and layer normalization, and a feed-forward neural network layer; The lightweight decoder includes a multi-head self-attention layer, a masked multi-head self-attention layer, a cross self-attention layer, a residual connection and layer normalization, and a feedforward neural network layer; The output layer includes a fully connected layer and a Sigmoid layer.

[0011] Furthermore, in the data training and error prediction steps, these parameters are used to evaluate the position and posture error of the tool tip on the test set: The normalized data spindle speed , tool diameter , Feed speed , Axial cutting depth , Radial cutting width , spatial location , Rotation angle of the rotation axis , Input feature input encoder, perform multi-head self-attention calculation, feedforward neural network, residual connection and layer normalization operation to obtain a vector containing machine tool status information γ ; The tool tip position error Input decoder layer to calculate mask self-attention and compare with vector γ Perform cross-attention calculation, feedforward neural network, residual connection and layer normalization operations to obtain the next prediction sequence; After the prediction sequence passes through the fully connected layer, the posture error prediction result of the tool tip point is output, and then through the Sigmoid function, the probability information of the posture error prediction result is output.

[0012] Furthermore, the multi-head self-attention calculation is specifically as follows: The input feature data is first processed by position encoding in the following way: ; in, POS is the location of the current input data, i is a natural number, d is the dimension of a single input sample; After the self-attention layer obtains the input, different linear transformations are performed on it to generate Q (Query)、 K (Key)、 V (Value) three matrices, the calculation method is as follows: ; in, is a learnable weight parameter, is the input; The dot product operation is performed on the query vector, keyword vector, and value vector of the input variable set with position information to obtain the output of a single self-attention layer, which is expressed as: ; in, for The split, The number of columns for keywords; The output of a single self-attention head is calculated in parallel to obtain the final output of the multi-head self-attention layer, which is expressed as:

[0013] in, is the final output of the multi-head self-attention layer, To parallelize the symbolic computation, h is the number of heads in the multi-head self-attention layer, is the trainable projection matrix of the multi-head attention layer.

[0014] Furthermore, the CNC machine tool processing error analysis method also includes a verification step on a test set, which is located after the data training and error prediction steps. The verification step on the test set is specifically: combining the position error of the tool tip point predicted by the improved Transformer model with the true value in the test set, and using MSE, RMSE and R 2 The evaluation parameters evaluate the test results, and the formula is as follows: ; ; in, For the i The true value of the samples, For the i The model prediction value of samples, is the mean of the true values.

[0015] Compared with the prior art, the CNC machine tool processing error analysis method of the present invention processes the parts according to the processing parameters of the CNC machine tool; measures the position error of each actual point on the actual contour of the processed workpiece and normal error , and record the error data of each actual point , the error data of each actual point Error data traced back to the tool tip point in the machine tool workspace , the processing parameters, The rotation angles a and c of the rotation axis are normalized to form the neural network input features, and the obtained posture error and input features to form a data set; the data set is divided into a training set, a validation set and a test set, an improved Transformer model is constructed and the Transformer model is trained with the data of the training set, the model parameters are recorded when the loss function of the validation set reaches the minimum value, and these parameters are used to evaluate the position and posture error of the tool tip on the test set, which improves the analysis efficiency and accuracy, provides a basis for realizing real-time prediction of machine tool reverse error, and timely adjusts the processing parameters, thereby ensuring the stability of the processing quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flow chart of the CNC machine tool processing error analysis method of the present invention; Figure 2 Improved Transformer model structure diagram for the present invention; Figure 3 This is an error prediction effect diagram of the CNC machine tool processing error analysis method of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0019] See also Figure 1 The present invention provides a method for analyzing machining errors of numerically controlled machine tools, comprising the following steps: Parts processing: CNC machine tools process parts according to processing parameters; Obtain error data: measure the position error of each actual point on the actual contour of the machined workpiece and normal error , and record the error data of each actual point ,in Represents the position data of the three coordinate directions of the actual measurement point, j Representative j workpiece, i represents the i Actual measurement points; Error data conversion: The error data of each actual point Error data traced back to the tool tip point in the machine tool workspace ,in is the position data of the tool tip in three coordinate directions, is the position error of the tool tip, is the normal error of the tool tip, Represents the position of the origin of the machine tool coordinate system in the coordinate system of the actual point being measured; Composition of data set: Processing parameters, The rotation angles a and c of the rotation axis are normalized to form the input features of the neural network, and the obtained posture error is And input features form a data set; Data training and error prediction: Divide the data set into training set, validation set and test set, build an improved Transformer model and use the data of the training set to train the Transformer model. When the loss function of the validation set reaches the minimum value, record the model parameters and use these parameters to evaluate the position error of the tool tip on the test set.

[0020] Parts processing is as follows: The CNC machine tool is a five-axis linkage CNC machine tool. By adjusting the spindle speed on the five-axis linkage CNC machine tool S , tool diameter D , Feed speed F , Axial cutting depth A p , Radial cutting width A e Various complex curved surface parts can be processed according to different process parameters according to CNC codes.

[0021] The specific method for obtaining error data is: Use a three-dimensional coordinate measuring machine to measure the position error of each actual point on the actual contour of the machined workpiece based on the ideal three-dimensional model of various complex curved surface parts and normal error , and record the error data of each actual point .in, Represents the position data of the actual measurement point, j Representative j workpiece, i represents the i actual measurement points.

[0022] The error data conversion is specifically: The point error data measured by the three-dimensional coordinate measuring machine is traced back to the error data of the tool tip point in the machine tool workspace .

[0023] Tool tip error data The calculation formula is as follows: ; in, Represents the position of the origin of the machine tool coordinate system in the coordinate measuring machine coordinate system.

[0024] The composition of the dataset is as follows: Determine the rotation angle of the rotary axis based on the CNC code for machining various complex curved surface parts a , c ; Calculate spindle speed S , tool diameter D , Feed speed F , Axial cutting depth A p , Radial cutting width A e , spatial location , Rotation angle of the rotation axis a , c Normalization of parameters; ; in, w i For the dataset i Dimensional data, w min is the minimum value of the data in the dataset dimension, w max is the maximum value of the data set dimension. is the normalized data.

[0025] The initialization range of the training set data is shown in Table 1 Table 1 ; The obtained pose error The input features after standardization constitute the data set, as shown in Table 2: Table 2 ; The specific steps of data optimization are: cleaning and optimizing the data in the data set based on statistical methods to improve the quality of the data set. The specific steps are as follows: Statistical data collection of a type of posture error data with the same process parameters, position, and rotation angle of the rotation axis ;in, k Represents the same parameters, position, and rotation angle of the rotation axis. k Error data; The truncated mean represents the pose error data with the same parameters, position, and rotation angle of the rotation axis.

[0026] The calculation formula for the truncated mean of the pose error is as follows: ; in, n Represents the total amount of data, α It means to remove the proportion. α Take 3-7%, preferably, α Take 5%.

[0027] Data training and error prediction are specifically as follows: The optimized data set is divided into training set, validation set and test set according to 8:1:1 using Latin cube sampling method; An improved Transformer model was constructed for parameter training and the pose error of the tool tip was evaluated on the test set.

[0028] Please continue reading Figure 2 , the improved Transformer model is as follows: The embedding operation in the standard Transformer model is cancelled, and a 4-layer lightweight encoder and decoder are used. The final output is composed of the pose error prediction result of the fully connected layer and the probability information of the Sigmoid layer. The improved Transformer model includes a lightweight encoder, a lightweight decoder, and an output layer. The encoder includes multi-head self-attention layers, residual connections and layer normalization, and feed-forward neural network layers; The decoder includes a multi-head self-attention layer, a masked multi-head self-attention layer, a cross self-attention layer, a residual connection and layer normalization, and a feedforward neural network layer; The output layer includes a fully connected layer and a Sigmoid layer.

[0029] The process of using the improved Transformer model is as follows: The spindle speed after normalization and truncated mean , tool diameter , Feed speed , Axial cutting depth , Radial cutting width , spatial location , Rotation angle of the rotation axis , Input feature input encoder, perform multi-head self-attention calculation, feedforward neural network, residual connection and layer normalization operation to obtain a vector containing machine tool status information γ ; The tool tip position error Input decoder layer to calculate mask self-attention and compare with vector γ Perform cross-attention calculation, feedforward neural network, residual connection and layer normalization operations to obtain the next prediction sequence; After the prediction sequence passes through the fully connected layer, the posture error prediction result of the tool tip point is output, and then after passing through the Sigmoid function, the probability information of the posture error prediction result can be output.

[0030] Specifically, in the improved Transformer model, the input feature data is first processed by position encoding in the following way: ; in, POS is the location of the current input data, i is a natural number, d is the dimension of a single input sample, d =8.

[0031] The attention mechanism imitates the attention characteristics of humans, allowing the network to process a large amount of input information in parallel, while selecting more critical information as the input of the next layer of the network and ignoring unimportant information. After the self-attention layer obtains the input, it will perform different linear transformations on it to generate Q (Query)、 K (Key)、 V (Value) three matrices, the calculation method is as follows: ; in, is a learnable weight parameter, is the input; The dot product operation is performed on the query vector, keyword vector, and value vector of the input variable set with position information to obtain the output of a single self-attention layer, which is expressed as: ; in, for The split, The number of columns for keywords.

[0032] The output of a single self-attention head is calculated in parallel to obtain the final output of the multi-head self-attention layer, which is expressed as: ; in, is the final output of the multi-head self-attention layer, To parallelize the symbolic computation, h is the number of heads in the multi-head self-attention layer, is the trainable projection matrix of the multi-head attention layer.

[0033] The output of the self-attention layer is processed with residual connection and layer normalization to obtain the output of the improved Transformer single-layer encoder, as shown below: ; The output after residual connection and layer normalization is sent to the feedforward neural network layer. The feedforward neural network contains two linear layers and a GELU activation function. The input y L Perform feature transformation and nonlinear mapping to obtain the output of the feedforward neural network y F .

[0034] ; Among them, the GELU function is approximately expressed as: ; Tool tip position error The relative position information is calculated in a similar way to the encoder input feature position encoding; The decoder layer mask self-attention calculation method is similar to the encoder. The difference is that the input vector is now a vector after position masking, making the current tool tip point pose error The input can only accept the error information before and after this moment, and will not be affected by future input. The calculation method is as follows: ; in, Q comes from the representation of the decoder's current position, and K and V Representations from all positions of the encoder, Output is the decoder’s input layer output sequence, Indicates that the attention matrix is ​​masked using an upper triangular matrix.

[0035] The cross-attention layer receives part of the output from the encoder and the output from the first attention module of the current decoder, so that each decoding module can use all the information in the encoder. Its calculation method refers to the self-attention mechanism.

[0036] The output of the cross attention is used as the current prediction result and added to the input tail of the decoder to obtain a new sequence, which is used as the output of the next prediction, and then autoregressively obtains the prediction sequence with the specified error.

[0037] After the prediction sequence passes through the fully connected layer, the posture error prediction result of the tool tip point is output, and then after passing through the Sigmoid function, the probability information of the posture error prediction result can be output.

[0038] The data set is input into the improved Transformer model for training. The training process uses the hyperparameter settings as shown in Table 3: Table 3

[0039] The steps for verifying the trained parameters on the test set are as follows: The trained improved Transformer model is used to detect the samples in the test set to obtain the tool tip position error based on the improved Transformer. The test results, such as Figure 3 As shown, black is the actual error and red is the predicted error.

[0040] Tool tip position error obtained based on the improved Transformer prediction model The detection results of MSE, RMSE and R 2 The evaluation parameters evaluate the test results, and the formula is as follows: ; in, For the i The true value of the samples, For the i The model prediction value of samples, is the mean of the true values.

[0041] Compared with the prior art, the CNC machine tool processing error analysis method of the present invention processes the parts according to the processing parameters of the CNC machine tool; measures the position error of each actual point on the actual contour of the processed workpiece and normal error , and record the error data of each actual point , the error data of each actual point Error data traced back to the tool tip point in the machine tool workspace , the processing parameters, The rotation angles a and c of the rotation axis are normalized to form the neural network input features, and the obtained posture error and input features to form a data set; the data set is divided into training set, validation set and test set, an improved Transformer model is constructed and the Transformer model is trained with the data of the training set, the model parameters are recorded when the loss function of the validation set reaches the minimum value, and these parameters are used to evaluate the position and posture error of the tool tip on the test set, which improves the analysis efficiency and accuracy, provides a basis for realizing real-time prediction of machine tool reverse error, and provides a basis for timely adjustment of processing parameters, thereby ensuring the stability of processing quality.

[0042] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which are equivalent modifications and improvements made to the above embodiments based on the essential technology of the present invention, and all of them belong to the protection scope of the present invention.

Claims

1. A method for analyzing machining errors of CNC machine tools, characterized in that: The following steps are involved: Parts processing: CNC machine tools process parts according to processing parameters; Obtain error data: measure the position error of each actual point on the actual contour of the machined workpiece and normal error , and record the error data of each actual point ,in Represents the position data of the three coordinate directions of the actual measurement point, j Representative j workpiece, i represents the i Actual measurement points; Error data conversion: The error data of each actual point Error data traced back to the tool tip point in the machine tool workspace ,in is the position data of the tool tip in three coordinate directions, is the position error of the tool tip, is the normal error of the tool tip, Represents the position of the origin of the machine tool coordinate system in the coordinate system of the actual point being measured; Composition of data set: Processing parameters, The rotation angles a and c of the rotation axis are normalized to form the neural network input features, and the obtained posture error And input features form a data set; Data training and error prediction: Divide the data set into training set, validation set and test set, build an improved Transformer model and use the data of the training set to train the Transformer model. When the loss function of the validation set reaches the minimum value, record the model parameters and use these parameters to evaluate the position error of the tool tip on the test set.

2. The method for analyzing machining errors of CNC machine tools according to claim 1, characterized in that: The processing parameters include spindle speed S , tool diameter D , Feed speed F , Axial cutting depth A p , Radial cutting width A e .

3. The method for analyzing machining errors of CNC machine tools according to claim 1, characterized in that: In the step of forming the data set, the processing parameters, And the normalization processing of the rotation axis angles a and c is specifically as follows: ； in, w i For the dataset i Dimensional data, w min is the minimum value of the data in the dataset dimension, w max is the maximum value of the data set dimension. is the normalized data.

4. The method for analyzing machining errors of numerically controlled machine tools according to claim 1, characterized in that: The CNC machine tool processing error analysis method further includes data optimization, wherein the data optimization step is located between the data set composition step and the data training and error prediction step, and the data optimization step is specifically as follows: Statistical data collection of a type of posture error data with the same process parameters, position, and rotation angle of the rotation axis ,in, k Represents the same parameters, position, and rotation angle of the rotation axis. k Error data; The truncated mean represents the pose error data with the same parameters, position, and rotation angle of the rotation axis.

5. The method for analyzing machining errors of CNC machine tools according to claim 4, characterized in that: The calculation formula for the truncated mean of the pose error is: ； in, n Represents the total amount of data, α It means to remove the proportion. α Take 3-7%.

6. The method for analyzing machining errors of CNC machine tools according to claim 1, characterized in that: In the data training and error prediction steps, the improved Transformer model is specifically: The embedding operation in the standard Transformer model is cancelled, and a 4-layer lightweight encoder and decoder are used. The final output is composed of the pose error prediction result of the fully connected layer and the probability information of the Sigmoid layer. The improved Transformer model includes a lightweight encoder, a lightweight decoder, and an output layer. The encoder includes a multi-head self-attention layer, a residual connection and layer normalization, and a feed-forward neural network layer; The lightweight decoder includes a multi-head self-attention layer, a masked multi-head self-attention layer, a cross self-attention layer, a residual connection and layer normalization, and a feedforward neural network layer; The output layer includes a fully connected layer and a Sigmoid layer.

7. The method for analyzing machining errors of CNC machine tools according to claim 6, characterized in that: In the data training and error prediction steps, these parameters are used to evaluate the position and posture error of the tool tip on the test set: The normalized data spindle speed , tool diameter , Feed speed , Axial cutting depth , Radial cutting width , spatial location , Rotation angle of the rotation axis , Input feature input encoder, perform multi-head self-attention calculation, feedforward neural network, residual connection and layer normalization operation to obtain a vector containing machine tool status information γ ; The tool tip position error Input decoder layer to calculate mask self-attention and compare with vector γ Perform cross-attention calculation, feedforward neural network, residual connection and layer normalization operations to obtain the next prediction sequence; After the prediction sequence passes through the fully connected layer, the posture error prediction result of the tool tip point is output, and then through the Sigmoid function, the probability information of the posture error prediction result is output.

8. The method for analyzing machining errors of CNC machine tools according to claim 7, characterized in that: The multi-head self-attention calculation is specifically as follows: The input feature data is first processed by position encoding in the following way: ； in, POS is the location of the current input data, i is a natural number, d is the dimension of a single input sample; After the self-attention layer obtains the input, different linear transformations are performed on it to generate Q (Query)、 K (Key)、 V (Value) three matrices, the calculation method is as follows: ； in, is a learnable weight parameter, is the input; The dot product operation is performed on the query vector, keyword vector, and value vector of the input variable set with position information to obtain the output of a single self-attention layer, which is expressed as: ； in, for The split, The number of columns for keywords; The output of a single self-attention head is calculated in parallel to obtain the final output of the multi-head self-attention layer, which is expressed as: ； in, is the final output of the multi-head self-attention layer, To parallelize the symbolic computation, h is the number of heads in the multi-head self-attention layer, is the trainable projection matrix of the multi-head attention layer.

9. The method for analyzing machining errors of numerically controlled machine tools according to claim 6, characterized in that: The CNC machine tool processing error analysis method also includes a verification step on a test set, which is located after the data training and error prediction steps. The verification step on the test set is specifically: combining the position error of the tool tip point predicted by the improved Transformer model with the real value in the test set, and using MSE, RMSE and R 2 The evaluation parameters evaluate the test results, and the formula is as follows: ； in, For the i The true value of the samples, For the i The model prediction value of samples, is the mean of the true values.

Images