CN112668227B - Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof - Google Patents
Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof Download PDFInfo
- Publication number
- CN112668227B CN112668227B CN202011626774.7A CN202011626774A CN112668227B CN 112668227 B CN112668227 B CN 112668227B CN 202011626774 A CN202011626774 A CN 202011626774A CN 112668227 B CN112668227 B CN 112668227B
- Authority
- CN
- China
- Prior art keywords
- cutter
- thin
- cutting force
- model
- deformation error
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000005520 cutting process Methods 0.000 claims abstract description 94
- 238000003754 machining Methods 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000012545 processing Methods 0.000 claims abstract description 36
- 238000012360 testing method Methods 0.000 claims abstract description 18
- 238000004519 manufacturing process Methods 0.000 claims abstract description 11
- 238000013528 artificial neural network Methods 0.000 claims abstract description 6
- 238000004458 analytical method Methods 0.000 claims abstract description 5
- 230000006870 function Effects 0.000 claims description 21
- 238000004088 simulation Methods 0.000 claims description 20
- 238000005259 measurement Methods 0.000 claims description 13
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 8
- 230000002708 enhancing effect Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000003860 storage Methods 0.000 claims description 6
- 238000007514 turning Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 13
- 230000000694 effects Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 3
- 230000003190 augmentative effect Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005459 micromachining Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000008485 antagonism Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Abstract
The invention discloses a thin-wall part cutter relieving deformation error prediction model establishment method and application thereof, belonging to the field of machining error prediction, and comprising the following steps: a cutting force model is established through a finite element analysis method, cutter parameters, machining process parameters and material cutting force parameters under different machining conditions of an actual production line are used as cutting force model input, cutting force and cutter yielding deformation errors corresponding to the parameters of each group are output, and a training data set is obtained; obtaining cutting force and corresponding cutter yielding errors of a plurality of groups of processing sites to obtain a test data set; a data enhancement module is added between an input layer and a first hidden layer of the neural network, and a small sample learning model is established and used for predicting cutter relieving deformation errors according to cutting force; and training and testing the small sample learning model by using the training data set and the testing data set respectively to obtain the thin-wall part cutter relieving deformation error prediction model. The method can improve the prediction precision of the deformation error of the cutter in the thin-wall part processing process.
Description
Technical Field
The invention belongs to the field of machining error prediction, and particularly relates to a method for establishing a thin-wall part cutter relieving deformation error prediction model and application thereof.
Background
In the thin-wall part processing process, a significant nonlinear relation is shown between the deformation error of the cutter and various measurable signals such as cutting force signals, vibration signals and the like. The thin-wall part is widely applied to the aerospace field, has various product types, is quick to update, and has high requirements on quick response in the processing process. The quality of the service performance of the thin-wall parts is greatly dependent on the processing precision. The aerospace parts are mostly free curved surfaces in space, are made of difficult-to-process materials, are large in load in the cutting process and poor in manufacturing process stability due to the fact that the machining is dependent on a multi-axis linkage numerical control machine tool, and the workpiece and the cutter inevitably generate machining deformation, so that machining errors are caused, and final machining precision and machining quality are affected. The final error value of the thin-walled workpiece is derived from the workpiece end and the cutter end. In the process of machining thin-wall parts, a cutter with a smaller diameter is often used for micro-machining, so that in the micro-machining of thin-wall parts, the error value of deformation of the cutter is one of the important causes of the overall machining error value. By predicting machining errors such as deformation errors of the cutter head, error control can be performed in the machining process, so that the machining errors are effectively reduced, and the machining precision is improved. Therefore, prediction of machining errors is particularly important.
With the rapid development of deep learning, the deep learning is applied to the mechanical field, but the current deep learning is mostly based on sufficient data samples, and in the mechanical field, particularly in an industrial scene facing the processing of aerospace complex thin-wall parts, the thin-wall parts belong to small batch processing, but the processing precision requirement is high. Therefore, there is generally less actual real process sample data available in the aerospace field. Meanwhile, thin-wall workpieces in the aerospace field are limited by experimental cost, so that the problem of sample sparseness in research becomes more remarkable. The data sample size is small, so that the model training times can be reduced, and the reliability of the machining prediction fitting result can be reduced. Therefore, the problem of small input sample size of the data driving model becomes one of difficulties in the current thin-walled workpiece processing error prediction.
In the traditional experimental method, finite element modeling is mainly carried out on a machined workpiece, the modeled finite element model is simulated to obtain a large amount of simulation data, and the obtained large amount of simulation data is used for replacing real machining sample data, so that the problem of few real machining sample data is solved. The finite element simulation method can effectively solve the problem of difficult data measurement and acquisition in actual processing, but has certain limitations in the aspects of boundary condition setting and the like. The final machining error prediction result has low accuracy.
Disclosure of Invention
Aiming at the defects and improvement demands of the prior art, the invention provides a method for establishing a thin-wall part cutter relieving deformation error prediction model and application thereof, and aims to improve the prediction precision of cutter relieving deformation errors in the thin-wall part processing process.
In order to achieve the above object, according to an aspect of the present invention, there is provided a thin-walled workpiece cutter relieving error prediction model building method, comprising:
establishing a cutting force model by a finite element analysis method, taking tool parameters, machining process parameters and material cutting force parameters of an actual production line of the thin-walled workpiece under different machining conditions as inputs of the cutting force model, and outputting cutting force and cutter yielding deformation errors corresponding to each group of parameters by the cutting force model to obtain a training data set;
obtaining a plurality of groups of cutting forces collected from a thin-wall part processing site and corresponding cutter yielding errors, and obtaining a test data set;
establishing a small sample learning model based on a neural network, wherein the small sample learning model is used for predicting a corresponding cutter relieving deformation error according to cutting force in the thin-wall part machining process; training and testing the small sample learning model by using the training data set and the test data set respectively, and taking the small sample learning model as a thin-wall part cutter relieving deformation error prediction model after the training and testing are finished;
the small sample learning model comprises a data enhancement module between an input layer and a first hidden layer, and the data enhancement module is used for expanding and enhancing input data.
According to the method, initial conditions of finite element simulation are set according to the machining tool parameters, the machining process parameters and the material cutting force coefficients which are given in the actual machining of the thin-wall part, and the cutting force obtained through simulation and the corresponding cutter yielding error can be well attached to the actual production condition of the aerospace field production line, so that the training effect of a model is ensured; because the training data set obtained by simulation is the same as the data set in the actual processing field in scale, the problem of fewer data samples exists, and the data enhancement module for expanding and enhancing the input data is added between the input layer and the first hidden layer on the basis of the neural network, so that a large amount of training data can be generated by using the obtained small sample data more efficiently, the condition of underdetermining the data under the small sample data is relieved, and the credibility of the model prediction result obtained by training is ensured; according to the invention, the trained model is tested by utilizing the real cutting force from the actual processing site and the corresponding cutter relieving deformation error framework test data set, so that the prediction accuracy of the model can be further improved.
Further, the data enhancement module augments and enhances the input data through the countermeasure generation network.
The standard data enhancement only generates limited seemingly-true alternative data, and the invention can generate more effective training samples by generating the countermeasure network to expand and enhance the input data, thereby ensuring the training effect of the model.
Further, the augmenting and enhancing of the input data by the generator in the countermeasure generation network includes:
(S1) regarding the original input data point, taking it as an original data point Q, and turning to step (S2);
(S2) according toGenerating a new data point omega, and updating the original data point Q according to Q=g (p, omega);
(S3) repeatedly executing the step (S2) until the preset iteration times are reached, and forming an enhanced data set by the original input data points and all the generated new data points, thereby realizing the expansion and enhancement of the input data;
wherein,the generator's generation function, p, is an implicit gaussian variable that varies the data generation, g is a function used to update the original data point Q.
According to the invention, through the iteration mode, the input data is expanded and enhanced, and before each iteration, the input of the generating function of the generator is updated by means of a Gaussian variable, so that the requirement on the generating function can be reduced, and more effective training samples can be ensured.
Further, in the small sample learning model, a generalization module is included between every two adjacent hidden layers;
and the generalization module is used for adjusting the parameters of the connected hidden layer through a fast and slow weight algorithm.
According to the invention, the generalization module is added between every two adjacent hidden layers in the small sample learning model, so that the hidden layer parameters connected with the generalization module are adjusted through the fast and slow weight algorithm, the generalization capability of the model can be increased, and the prediction precision of the model is further ensured.
Further, in the small sample learning model, an attention module is arranged between the last hidden layer and the output layer;
and the attention module is used for adjusting the probability of selecting each dimension element in the data output by the last hidden layer through an attention mechanism so as to screen out the element of interest.
According to the invention, the attention module is added between the last hidden layer and the output layer in the small sample learning model, so that the probability of selecting each dimension element in the data output by the last hidden layer is adjusted through an attention mechanism, and the adjustment result is a locally optimal solution which enables the machining error of the thin-wall part to be minimum, thereby accurately screening out the element of interest and further ensuring the prediction precision of the model.
Further, the tool parameters include: cutter diameter, cutting edge helix angle and cutter radial compliance;
the processing parameters include: spindle rotation speed, feeding quantity per tooth, normal cutting depth and row spacing;
the material cutting force parameter coefficients include: tangential shear force coefficient, radial shear force coefficient, and axial shear force coefficient.
According to the invention, the initial conditions of finite element simulation are set by the given machining tool parameters, machining process parameters and material cutting force coefficients in the actual machining of the thin-wall part, and the cutting force and cutter relieving deformation errors corresponding to each group of parameters can be accurately obtained by a finite element simulation method.
Further, the method for establishing the deformation error prediction model of the thin-walled workpiece cutter head provided by the invention further comprises the following steps:
using a cutter under laboratory conditions, and carrying out rigidity calibration measurement on the same material as the actually processed thin-wall piece to obtain a dynamic rigidity value DT of the cutter;
and adjusting the machining process parameters input into the cutting force model according to the dynamic stiffness value DT of the cutter obtained through stiffness calibration measurement so as to reduce the gap between the cutter deformation error calculated based on the dynamic stiffness value DT of the cutter and the cutter deformation error output by the cutting force model.
According to the invention, the dynamic stiffness value DT obtained by means of stiffness calibration measurement is used for adjusting the processing process parameters of the input cutting force model, so that the gap between the cutter yielding error calculated based on the dynamic stiffness value DT of the cutter and the cutter yielding error output by the cutting force model is reduced, the accuracy of a sample in training data set obtained by using the cutting force model is ensured, and therefore, the model has higher prediction accuracy after training is finished.
According to another aspect of the present invention, there is provided a thin-walled workpiece let-off deformation error prediction method, comprising:
the real cutting force measured on the thin-wall part processing site is input into the thin-wall part cutter relieving deformation error prediction model established by the thin-wall part cutter relieving deformation error prediction model establishment method provided by the invention, so that the corresponding cutter relieving deformation error is predicted by the thin-wall part cutter relieving deformation error prediction model.
The thin-wall part cutter relieving deformation error prediction model established by the thin-wall part cutter relieving deformation error prediction model establishment method provided by the invention has higher prediction precision, and the cutter relieving deformation error prediction method provided by the invention can accurately predict the cutter relieving deformation error in the thin-wall part processing process, so that clear indication is provided for subsequent error control, and the processing precision and the processing quality of the thin-wall part are effectively ensured.
According to yet another aspect of the present invention, there is provided a computer readable storage medium comprising a stored computer program; when the computer program is executed by the processor, the equipment where the computer readable storage medium is located is controlled to execute the thin-walled workpiece cutter relieving deformation error prediction model establishment method and/or the thin-walled workpiece cutter relieving deformation error prediction method.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
(1) According to the method, initial conditions of finite element simulation are set according to the machining tool parameters, the machining process parameters and the material cutting force coefficients which are given in the actual machining of the thin-wall part, and the cutting force obtained through simulation and the corresponding cutter yielding error can be well attached to the actual production condition of the aerospace field production line, so that the training effect of a model is ensured; on the basis of a neural network, a data enhancement module for expanding and enhancing input data is added between an input layer and a first hidden layer, so that a large amount of training data can be generated by using the obtained small sample data more efficiently, the condition of underdetermining data under the small sample data is relieved, and the credibility of a model prediction result obtained by training is ensured; the trained model is tested by utilizing the real cutting force from the actual processing site and the corresponding cutter relieving deformation error framework test data set, so that the prediction accuracy of the model can be further improved.
(2) According to the invention, the generalization module is added between every two adjacent hidden layers in the small sample learning model, so that the hidden layer parameters connected with the generalization module are adjusted through the fast and slow weight algorithm, the generalization capability of the model can be increased, and the prediction precision of the model is further ensured.
(3) According to the invention, the attention module is added between the last hidden layer and the output layer in the small sample learning model, so that the probability of selecting each dimension element in the data output by the last hidden layer is adjusted through an attention mechanism, and the adjustment result is the locally optimal solution which enables the machining error of the thin-wall part to be minimum, thereby accurately screening out the element of interest and further ensuring the prediction precision of the model.
Drawings
FIG. 1 is a flowchart of a method for establishing a deformation error prediction model of a thin-walled workpiece according to an embodiment of the present invention;
FIG. 2 is a schematic view of a cutting force model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a small sample learning model according to an embodiment of the present invention;
fig. 4 is a schematic diagram of a meta-network structure in a conventional fast and slow weighting algorithm.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
In the present invention, the terms "first," "second," and the like in the description and in the drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.
In order to solve the technical problems of low prediction accuracy due to less data of real processing samples in the existing thin-wall part processing error prediction method, the invention provides a thin-wall part yielding tool deformation error prediction model establishment method and application thereof, and the overall thought is as follows: establishing a cutting force model by using a finite element simulation method, taking machining tool parameters, machining process parameters and material cutting force coefficients which are given in actual machining as the cutting force model, obtaining cutting force and corresponding cutter yielding errors, obtaining a small sample training data set which is the same as the data scale in an actual machining site, and training a small sample learning model by using the training data set; in the small sample learning model, a data enhancement module is inserted between the input layer and the first hidden layer and used for expanding and enhancing input data so as to relieve the underdetermined condition of the data under the small sample data and ensure the training effect of the model; after model training is finished, a test data set is constructed by utilizing cutting force acquired from an actual processing site and corresponding cutter relieving deformation errors, and a small sample learning model is tested, so that the prediction accuracy of the model is further ensured.
The following are examples.
Example 1:
a method for establishing a thin-wall part cutter relieving deformation error prediction model is shown in fig. 1, and comprises the following steps:
establishing a cutting force model by a finite element analysis method, wherein the cutting force model is used for taking parameters in a machining process as input and outputting corresponding cutting force and cutter relieving deformation errors; taking cutter parameters, machining process parameters and material cutting force parameters of an actual production line of the thin-wall part under different machining conditions as inputs of a cutting force model, outputting cutting force and cutter yielding deformation errors corresponding to each group of parameters by the cutting force model, and obtaining a training data set;
obtaining a plurality of groups of cutting forces collected from a thin-wall part processing site and corresponding cutter yielding errors, and obtaining a test data set;
establishing a small sample learning model based on a neural network, wherein the small sample learning model is used for predicting a corresponding cutter relieving deformation error according to cutting force in the thin-wall part machining process; training and testing the small sample learning model by using the training data set and the test data set respectively, and taking the small sample learning model as a thin-wall part cutter relieving deformation error prediction model after the training and testing are finished;
the cutting force model established in this embodiment is shown in fig. 2, wherein the tool parameters include: cutter diameter D, cutting edge helix angle beta and cutter radial compliance p;
the processing parameters include: spindle rotation speed n, feeding quantity f of each tooth, normal cutting depth a and row spacing m;
the material cutting force parameter coefficients include: tangential shear coefficient K tc Coefficient of radial shear K rc And an axial shear coefficient K ac ；
Through the cutting force model, in the finite element simulation process, a cutter is discretized into a plurality of cutter microelements, each cutter microelement can be regarded as an independent cutting process, and the cutting force model expression of the cutter microelement is constructed as follows:
wherein, the cutting force infinitesimal value in three directions of tangential, radial and axial is dF t 、dF r 、dF a Representation, h a Indicating knifeThe corresponding instantaneous undeformed cutting thickness of the infinitesimal, db represents the width of the chip infinitesimal, and the cutting force coefficients in the tangential, radial and axial directions are K tc 、K rc 、K ac A representation; integrating the tangential, radial and axial cutting force microelements to obtain tangential, radial and axial cutting force values F t 、F r 、F a ；
The three directions of the tool coordinate system are tangential, radial and axial directions of the tool, and the three axial directions are orthogonal to each other; the workpiece coordinate system is a natural coordinate system and consists of three axes of x, y and z, wherein the three axes are orthogonal; the included angle between the workpiece coordinate system and the workpiece coordinate system is alpha, and the three-way cutting force F under the workpiece coordinate system can be obtained through coordinate transformation x 、F y 、F z The cutting force conversion equation under two types of coordinate systems is:
by inputting cutter parameters, namely a cutter diameter D, a cutting edge helix angle beta and cutter radial flexibility p, and inputting machining process parameters, namely a main shaft rotating speed n, a feeding amount f of each tooth, a normal cutting depth a and a line spacing m, the cutter for machining the thin-wall part can be used under the laboratory condition to carry out rigidity calibration measurement on the same material as the actually machined workpiece;
the specific measurement process of the rigidity calibration measurement is as follows: and (3) moving the processing machine tool to enable the point of the tool nose of the tool to contact the surface to be processed, enabling the tool to be just contacted with the workpiece, and adjusting the dynamometer to enable the reading to be zero. Moving the machine tool towards the direction of the workpiece, enabling the workpiece to be in contact with the cutter, recording the indication of the dynamometer after the workpiece moves equidistantly until the indication of the dynamometer is increased to about 3 newtons, and stopping moving the machine tool; drawing a relation curve between the contact force indication number and the moving distance under the dynamometer, and obtaining a dynamic stiffness value DT of the cutter according to the slope of the solving curve, and the machining process parameters including the spindle rotation speed n, the feeding quantity f of each tooth and the normal cutting depth a;
the cutter yielding error value e can be calculated by a cutting force F and a dynamic stiffness value DT of the cutter, wherein the cutting force F is obtained by decomposing a cutter microcell cutting force model along a workpiece coordinate system, and the calculation of the cutter yielding error value e can be simplified into the following equation expression:
in the error prediction of the thin-wall part processing cutter-back in the actual field, the dynamic stiffness of the corresponding cutter is difficult to directly obtain;
according to the method, initial conditions of finite element simulation are set according to machining tool parameters, machining process parameters and material cutting force coefficients which are given in actual machining of the thin-wall part, and cutting force obtained through simulation and corresponding cutter yielding errors can be well attached to actual production conditions of an aerospace field production line, so that training effects of a model are guaranteed;
in the embodiment, finite element simulation software is used for inputting a finite number of machining parameter conditions in an actual machining scene into the finite element simulation software, and the simulation software outputs tool bit rows, tool bit columns, simulation periods, tool path row numbers and tool path column numbers. The cutting force value can be calculated by finite element simulation software through the cutter point row, the cutter point column and the simulation period; the cutter deformation error value can be calculated by finite element simulation software through the number of rows and the number of columns of the cutter. Finally, the finite element simulation software can output cutting force and cutter relieving deformation error values, so that a data set D of a subsequent small and medium sample learning model is constructed FL The data set includes cutting force data F and cutter relieving deformation error data e output by the finite element simulation model, so that the data set D can be used for FL The writing is as follows: d (D) FL ＝{F、e}。
The small sample learning model established in this embodiment is shown in fig. 3, where a data enhancement module is included between the input layer and the first hidden layer, and is used to expand and enhance the input data;
because the standard data enhancement only generates limited plausible alternative data, in order to more efficiently use the obtained small sample data to generate a large amount of training data, thereby relieving the condition of underdetermining the data under the small sample data, ensuring the credibility of the model prediction result obtained by training, as a preferred implementation manner, in the embodiment, the data enhancement module expands and enhances the input data through an antagonism generation network;
optionally, in this embodiment, the augmenting and enhancing the input data by the generator in the countermeasure generation network includes:
(S1) regarding the original input data point, taking it as an original data point Q, and turning to step (S2);
(S2) according toGenerating a new data point omega, and updating the original data point Q according to Q=g (p, omega);
(S3) repeatedly executing the step (S2) until the preset iteration times are reached, and forming an enhanced data set by the original input data points and all the generated new data points, thereby realizing the expansion and enhancement of the input data;
wherein,the generator generates a function, p is an implicit Gaussian variable which causes the data to generate change, and g is a function for updating the original data point Q; in this embodiment, the gaussian model to which the gaussian variable p is subjected has a mean value of 0 and a variance of I, i.e., p=n (0,I);
according to the method, the input data are expanded and enhanced in the iteration mode, before each iteration is started, the input of the generating function of the generator is updated by means of a Gaussian variable, the requirement on the generating function can be reduced, and more effective training samples can be generated.
As a preferred implementation manner, as shown in fig. 3, in this embodiment, a generalization module is included between every two adjacent hidden layers in the small sample learning model;
the generalization module is used for adjusting the hidden layer parameters connected with the generalization module through a fast-slow weight algorithm, so that the generalization capability of the model can be increased, and the prediction precision of the model is further ensured;
the fast and slow weight algorithm is a generalized form based on a meta-learning algorithm: through the meta-network, a meta-level knowledge spanning task is learned. As shown in fig. 4, the meta network includes two learning sections: a Base learner (Base learner) and a Meta learner (Meta learner), and is provided with an external memory layer, the learning process taking place in two separate spaces, respectively. The base learner operates in the input task space, whereas the meta learner operates in a task agnostic meta space. The base learner first analyzes the input task while the base learner provides a higher order meta-information for meta-learning to state itself in the current task space. Based on meta information, the meta learner parameterizes itself as well as the base learner quickly so that the meta network model can recognize new concepts of the input task. The training of the meta-network mainly comprises three processes: (1) acquiring meta information; (2) generating a fast weight, wherein the fast weight is used for adjusting parameters in each hidden layer; (3) the slow weights are optimized and used to adjust the relationship between the number hidden layers. In the fast and slow weight algorithm, in order to realize the fast generalization of the slow weights in the base learner to the corresponding fast weight values, a layer enhancement method can be used. The input to the enhancement layer is first converted by fast and slow weights and is then converted by a nonlinear function. The nonlinear function used in the embodiment is a Sigmoid function whose basic expression for the input argument t is as follows:
and summing the fast and slow weight values of each layer after the Sigmoid function transformation, and distributing the fast and slow weight values of the next layer as new input. And by analogy, after summing the fast weight and the slow weight of the last layer, the predicted result can be output in a probability form through a Softmax function. If set SThe upper layer output of the oftmax function layer is l= { L 1 、l 2 、...、l j 、...、l J Wherein J is the number of output values of the last enhancement layer, l j Is one of the intermediate values of the output values. The Softmax function has the expression:
finally, after passing through the Softmax function, the output phi (l j ) Is a predicted quantity obtained in the form of a probability.
As a preferred implementation manner, as shown in fig. 3, in the present embodiment, in the small sample learning model, an attention module is included between the last hidden layer and the output layer;
the attention module is used for adjusting the probability of each dimension element in the data output by the last hidden layer through an attention mechanism so as to screen out the element of interest, namely an attention focus, and further carrying out more detailed and deep analysis on the part of data through subsequent processing precision prediction;
the attention focus factor in the attention mechanism is set as u, and the value range of the attention focus factor is u epsilon {0,1} which indicates whether the attention value of each dimension is selected or not is assumed that the original input of the attention mechanism has n dimensions; or u epsilon [0,1 ]]Representing the probability that each dimension of the attention value is selected. Accumulating the supervision constraint error and the reconstruction error, wherein the obtained error function expression is a cost loss function of the attention mechanism; solving the partial derivative of the cost function to obtain a local optimal solution u * The optimal solution corresponds to the condition that the machining error of the thin-wall part is minimum, so that the prediction of the cutter relieving error of the thin-wall part is realized.
In order to further ensure the prediction accuracy of the model, as a preferred embodiment, the present embodiment further includes:
using a cutter under laboratory conditions, and carrying out rigidity calibration measurement on the same material as the actually processed thin-wall piece to obtain a dynamic rigidity value DT of the cutter; specific stiffness calibration measurement procedures are referred to above and will not be repeated here;
and adjusting the machining process parameters input into the cutting force model according to the dynamic stiffness value DT of the cutter obtained through stiffness calibration measurement so as to reduce the gap between the cutter deformation error calculated based on the dynamic stiffness value DT of the cutter and the cutter deformation error output by the cutting force model.
According to the method, the dynamic stiffness value DT obtained by means of stiffness calibration measurement is used for adjusting the machining process parameters of the input cutting force model, so that the gap between the cutter yielding error calculated based on the dynamic stiffness value DT of the cutter and the cutter yielding error output by the cutting force model is reduced, the accuracy of samples in training data set obtained by the cutting force model is guaranteed, and therefore the model has higher prediction accuracy after training is finished.
Example 2:
a thin-wall part cutter relieving deformation error prediction method comprises the following steps:
the actual cutting force measured on the thin-walled workpiece processing site is input into the thin-walled workpiece cutter relieving deformation error prediction model established by the thin-walled workpiece cutter relieving deformation error prediction model establishment method provided by the embodiment 1, so that the corresponding cutter relieving deformation error is predicted by the thin-walled workpiece cutter relieving deformation error prediction model.
The thin-walled workpiece cutter relieving error prediction model established by the thin-walled workpiece cutter relieving error prediction model establishment method provided by the embodiment 1 has higher prediction precision, and the cutter relieving error prediction method provided by the embodiment can accurately predict the cutter relieving error in the thin-walled workpiece processing process, so that clear indication is provided for subsequent error control, and the processing precision and the processing quality of the thin-walled workpiece are effectively ensured.
Example 3:
a computer readable storage medium comprising a stored computer program; when the computer program is executed by the processor, the apparatus where the computer readable storage medium is located is controlled to execute the method for establishing the thin-walled workpiece let-down deformation error prediction model provided in the above embodiment 1 and/or the method for predicting the thin-walled workpiece let-down deformation error provided in the above embodiment 2.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (6)
1. A method for establishing a thin-wall part cutter relieving deformation error prediction model is characterized by comprising the following steps:
establishing a cutting force model by a finite element analysis method, taking tool parameters, machining process parameters and material cutting force parameters of an actual production line of the thin-walled workpiece under different machining conditions as inputs of the cutting force model, and outputting cutting force and cutter yielding deformation errors corresponding to each group of parameters by the cutting force model to obtain a training data set;
obtaining a plurality of groups of cutting forces collected from a thin-wall part processing site and corresponding cutter yielding errors, and obtaining a test data set;
establishing a small sample learning model based on a neural network, wherein the small sample learning model is used for predicting a corresponding cutter relieving deformation error according to cutting force in the thin-wall part machining process; training and testing the small sample learning model by using the training data set and the test data set respectively, and taking the small sample learning model as a thin-wall piece cutter relieving deformation error prediction model after the training and testing are finished;
the small sample learning model comprises a data enhancement module between an input layer and a first hidden layer, and is used for expanding and enhancing input data; the cutter parameters include: cutter diameter, cutting edge helix angle and cutter radial compliance; the processing parameters include: spindle rotation speed, feeding quantity per tooth, normal cutting depth and row spacing; the material cutting force parameters include: tangential shear force coefficient, radial shear force coefficient and axial shear force coefficient; outputting cutter point rows, cutter point columns, simulation periods, cutter line numbers and cutter line columns through finite element simulation software; calculating by finite element simulation software through the cutter point row, the cutter point column and the simulation period to obtain a cutting force value; calculating to obtain a cutter yielding error value by finite element simulation software through the number of cutter lines and the number of cutter lines; the finite element simulation software outputs a cutting force value and a cutter deformation error value, and a training data set of a small sample learning model is constructed, wherein the training data set comprises the cutting force value and the cutter deformation error value output by the finite element simulation model;
the data enhancement module expands and enhances the input data through an countermeasure generation type network; the generator in the countermeasure generation network expands and enhances the input data, including:
(S1) regarding the original input data point, taking it as an original data point Q, and turning to step (S2);
(S2) according toGenerating a new data point omega, and updating the original data point Q according to Q=g (p, omega);
(S3) repeatedly executing the step (S2) until the preset iteration times are reached, and forming an enhanced data set by the original input data points and all the generated new data points, thereby realizing the expansion and enhancement of the input data;
wherein,for the generator's generation function, p is an implicit gaussian variable that varies the data generation, g is a function used to update the original data point Q.
2. The method for building the deformation error prediction model of the thin-walled workpiece according to claim 1, wherein a generalization module is included between every two adjacent hidden layers in the small sample learning model;
the generalization module is used for adjusting the parameters of the connected hidden layer through a fast and slow weight algorithm.
3. The method for building the deformation error prediction model of the thin-walled workpiece according to claim 1, wherein the small sample learning model comprises an attention module between the last hidden layer and the output layer;
the attention module is used for adjusting the probability of selecting each dimension element in the data output by the last hidden layer through an attention mechanism so as to screen out the element of interest.
4. The thin-walled workpiece let-off deformation error prediction model building method according to claim 1, further comprising:
using the cutter under the laboratory condition to perform rigidity calibration measurement on the same material as the actually processed thin-wall piece to obtain a dynamic rigidity value DT of the cutter;
and adjusting the machining process parameters input into the cutting force model according to the dynamic stiffness value DT of the cutter, which is obtained through stiffness calibration measurement, so as to reduce the gap between the cutter deformation error calculated based on the dynamic stiffness value DT of the cutter and the cutter deformation error output by the cutting force model.
5. The method for predicting the deformation error of the cutter head of the thin-walled workpiece is characterized by comprising the following steps:
inputting the real cutting force measured on the thin-walled workpiece processing site into the thin-walled workpiece cutter relieving deformation error prediction model established by the thin-walled workpiece cutter relieving deformation error prediction model establishment method according to any one of claims 1 to 4 so as to predict the corresponding cutter relieving deformation error by the thin-walled workpiece cutter relieving deformation error prediction model.
6. A computer readable storage medium comprising a stored computer program; when the computer program is executed by a processor, the equipment where the computer readable storage medium is located is controlled to execute the method for establishing the thin-walled workpiece cutter relieving error prediction model according to any one of claims 1 to 4 and/or the method for predicting the thin-walled workpiece cutter relieving error according to claim 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011626774.7A CN112668227B (en) | 2020-12-31 | 2020-12-31 | Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011626774.7A CN112668227B (en) | 2020-12-31 | 2020-12-31 | Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112668227A CN112668227A (en) | 2021-04-16 |
| CN112668227B true CN112668227B (en) | 2024-02-02 |
Family
ID=75412589
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011626774.7A Active CN112668227B (en) | 2020-12-31 | 2020-12-31 | Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668227B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113592064B (en) * | 2021-07-05 | 2024-02-09 | 中国科学院上海光学精密机械研究所 | Ring polishing process parameter prediction method, system, application, terminal and medium |
| CN113570147B (en) * | 2021-07-30 | 2023-10-10 | 华中科技大学 | Method and equipment for rapidly predicting machining errors of thin-wall parts in multiple modes |
| WO2023077275A1 (en) * | 2021-11-02 | 2023-05-11 | 西门子股份公司 | Machine tool machining precision prediction method and apparatus, storage medium, and electronic device |
| CN113779895A (en) * | 2021-11-10 | 2021-12-10 | 深圳市裕展精密科技有限公司 | Data model establishing device and method, and machining quality predicting device and method |
| CN114925462B (en) * | 2022-04-11 | 2023-04-18 | 西北工业大学 | Thin-wall part machining deformation prediction method based on cutting force and rigidity correlation evolution |
| CN115422670B (en) * | 2022-08-09 | 2023-05-16 | 西北工业大学 | Thin-wall part machining deformation prediction method based on cutting force and clamping force space-time learning |
| CN116976018A (en) * | 2023-03-11 | 2023-10-31 | 哈尔滨理工大学 | Method for predicting relative position deviation of milling tool system |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006163472A (en) * | 2004-12-02 | 2006-06-22 | Toppan Printing Co Ltd | Equivalent young's modulus calculation method and device |
| UA99462U (en) * | 2014-11-17 | 2015-06-10 | Дніпропетровський Національний Університет Імені Олеся Гончара | Method for identification of parameters of damages in thin bodies |
| CN105160059A (en) * | 2015-07-11 | 2015-12-16 | 西安工业大学 | BP and GA based blade machining cutting quantity optimization selection method |
| CN107272580A (en) * | 2017-06-15 | 2017-10-20 | 大连理工大学 | A kind of hard brittle material thin-walled parts machining error compensating method |
| CN109977464A (en) * | 2019-02-18 | 2019-07-05 | 江苏科技大学 | A kind of prediction technique of the piston machining deflection based on BP neural network |
| CN111037365A (en) * | 2019-12-26 | 2020-04-21 | 大连理工大学 | Cutter state monitoring data set enhancing method based on generative countermeasure network |
| CN111159825A (en) * | 2019-12-16 | 2020-05-15 | 西北工业大学 | Thin-wall blade cutting track parameter optimization method |
| CN111783241A (en) * | 2020-06-16 | 2020-10-16 | 大连理工大学 | Prediction method for thin-wall micro-milling deformation |
| CN111797469A (en) * | 2020-06-24 | 2020-10-20 | 华中科技大学无锡研究院 | Aeroengine casing technological parameter optimization method based on machining cutter back-off deformation constraint |
| CN111832624A (en) * | 2020-06-12 | 2020-10-27 | 上海交通大学 | Tool remaining life prediction method based on anti-migration learning |
-
2020
- 2020-12-31 CN CN202011626774.7A patent/CN112668227B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006163472A (en) * | 2004-12-02 | 2006-06-22 | Toppan Printing Co Ltd | Equivalent young's modulus calculation method and device |
| UA99462U (en) * | 2014-11-17 | 2015-06-10 | Дніпропетровський Національний Університет Імені Олеся Гончара | Method for identification of parameters of damages in thin bodies |
| CN105160059A (en) * | 2015-07-11 | 2015-12-16 | 西安工业大学 | BP and GA based blade machining cutting quantity optimization selection method |
| CN107272580A (en) * | 2017-06-15 | 2017-10-20 | 大连理工大学 | A kind of hard brittle material thin-walled parts machining error compensating method |
| CN109977464A (en) * | 2019-02-18 | 2019-07-05 | 江苏科技大学 | A kind of prediction technique of the piston machining deflection based on BP neural network |
| CN111159825A (en) * | 2019-12-16 | 2020-05-15 | 西北工业大学 | Thin-wall blade cutting track parameter optimization method |
| CN111037365A (en) * | 2019-12-26 | 2020-04-21 | 大连理工大学 | Cutter state monitoring data set enhancing method based on generative countermeasure network |
| CN111832624A (en) * | 2020-06-12 | 2020-10-27 | 上海交通大学 | Tool remaining life prediction method based on anti-migration learning |
| CN111783241A (en) * | 2020-06-16 | 2020-10-16 | 大连理工大学 | Prediction method for thin-wall micro-milling deformation |
| CN111797469A (en) * | 2020-06-24 | 2020-10-20 | 华中科技大学无锡研究院 | Aeroengine casing technological parameter optimization method based on machining cutter back-off deformation constraint |
Non-Patent Citations (7)
| Title |
|---|
| An Integrated Methodology for Workpiece-Fixture System Stiffness Calculation and Error Control;Zhang, FP等;《 2009 INTERNATIONAL CONFERENCE ON MEASURING TECHNOLOGY AND MECHATRONICS AUTOMATION》;全文 * |
| Deformation analysis and error prediction in machining of thin-walled honeycomb-core sandwich structural parts;Gaoqun Liu等;《Advanced Manufacturing Technology》;全文 * |
| 基于双三次B样条插值的薄壁件加工误差补偿方法;陈飞等;《航空制造技术》(第4期);全文 * |
| 基于神经网络的薄壁件加工变形预测方法;秦国华等;《兵工学报》;第34卷(第7期);全文 * |
| 基于神经网络的铣削复杂薄壁件受力变形分析和建模研究;刘新玲等;《机械制造》;第47卷(第535期);全文 * |
| 航空机匣工件车削加工变形量预测及误差补偿;陈维克等;制造技术与机床(第02期);全文 * |
| 薄壁件切削加工变形预测及装夹布局优化;王鸿;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112668227A (en) | 2021-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112668227B (en) | Thin-wall part cutter relieving deformation error prediction model establishment method and application thereof | |
| CN111366123B (en) | Part surface roughness and cutter wear prediction method based on multi-task learning | |
| Zimmermann et al. | Adaptive input selection for thermal error compensation models | |
| Zhu et al. | Sample augmentation for intelligent milling tool wear condition monitoring using numerical simulation and generative adversarial network | |
| CN111723440B (en) | Thin-wall part machining precision prediction hybrid modeling method | |
| CN110007645B (en) | Feeding system hybrid modeling method based on dynamics and deep neural network | |
| CN116448419A (en) | Zero sample bearing fault diagnosis method based on depth model high-dimensional parameter multi-target efficient optimization | |
| CN112100810B (en) | Flat-bottom milling cutter multi-axis milling force prediction method | |
| Kozłowski et al. | Integrating advanced measurement and signal processing for reliability decision-making | |
| Meng et al. | Tool wear prediction in milling based on a GSA-BP model with a multisensor fusion method | |
| CN108052963A (en) | The data screening method, apparatus and wind power generating set of wind power prediction modeling | |
| Ding et al. | Online monitoring model of micro-milling force incorporating tool wear prediction process | |
| CN113504768A (en) | High-precision product digital twin computability method for assembly quality prediction | |
| Ren et al. | Vibration monitoring with dependencies attention for gear hobbing quality evaluation | |
| Gao et al. | New tool wear estimation method of the milling process based on multisensor blind source separation | |
| Hu et al. | Reliability analysis of the chatter stability during milling using a neural network | |
| CN109446721B (en) | Machine tool process interaction algorithm based on identifier software thread execution sequence arrangement | |
| CN116952583A (en) | Rolling bearing fault diagnosis method of digital twin driving deep transfer learning model | |
| TWI802294B (en) | Test site recommendation device, test site recommendation method, and semiconductor device manufacturing system | |
| Soleymani Yazdi et al. | Analysis and estimation of state variables in CNC face milling of AL6061 | |
| Yuan et al. | A physics-assisted online learning method for tool wear prediction | |
| WO2021111936A1 (en) | Prediction system, information processing device, and information processing program | |
| Kaneko et al. | A Voxel-Based End Milling Simulation Method to Analyze the Elastic Deformation of a Workpiece | |
| Abdelwahab | Surface roughness modeling and prediction based on vibration signal analysis and machining parameters in milling of aluminum by artificial neural network | |
| Sulyukova et al. | Modeling and assessment of the energy state of the technological system of mechanical processing when creating digital twins |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |