CN116012480A - Data-driven cutting surface morphology gray scale image generation method - Google Patents
Data-driven cutting surface morphology gray scale image generation method Download PDFInfo
- Publication number
- CN116012480A CN116012480A CN202310065585.4A CN202310065585A CN116012480A CN 116012480 A CN116012480 A CN 116012480A CN 202310065585 A CN202310065585 A CN 202310065585A CN 116012480 A CN116012480 A CN 116012480A
- Authority
- CN
- China
- Prior art keywords
- feature map
- data
- convolution operation
- size
- cutting surface
- 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.)
- Pending
Links
- 238000005520 cutting process Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000012545 processing Methods 0.000 claims abstract description 35
- 230000008569 process Effects 0.000 claims abstract description 5
- 238000012876 topography Methods 0.000 claims description 48
- 238000000605 extraction Methods 0.000 claims description 26
- 230000004913 activation Effects 0.000 claims description 25
- 238000010586 diagram Methods 0.000 claims description 25
- 230000008859 change Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 19
- 239000011159 matrix material Substances 0.000 claims description 14
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 10
- 238000001228 spectrum Methods 0.000 claims description 8
- 238000012986 modification Methods 0.000 claims description 7
- 230000004048 modification Effects 0.000 claims description 7
- 238000007781 pre-processing Methods 0.000 claims description 7
- 229910052799 carbon Inorganic materials 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 6
- 238000012544 monitoring process Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 claims description 3
- 238000005243 fluidization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 abstract description 11
- 230000009286 beneficial effect Effects 0.000 abstract description 5
- 238000003801 milling Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 2
- 238000003754 machining Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000007514 turning Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 238000007516 diamond turning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Image Analysis (AREA)
Abstract
Compared with the offline acquisition flow of the cutting surface morphology and the kinematic simulation flow of the cutting surface morphology, the data-driven generation method of the gray image of the cutting surface morphology greatly reduces the acquisition time and the image generation time of the cutting surface morphology image, directly converts the processing signal spectrogram and the tool rigidity data into the gray image data of the cutting surface morphology by using the generator in the generation countermeasure network, avoids the flow of the acquisition and the simulation of the surface morphology image data which takes longer time, is beneficial to generating global information between the processing signal data and the tool rigidity information of the countermeasure network learning, improves the generation accuracy of the cutting surface morphology image, can directly acquire the cutting surface morphology image of a processed workpiece in the processing process, can better reflect the depth, the density and the shape of a trace left on the processed surface, and is particularly beneficial to the rapid judgment of the product quality of a processed part by workers.
Description
Technical Field
The invention belongs to the technical field of cutting processing, and particularly relates to a data-driven method for generating a gray level image of a cutting processing surface morphology.
Background
The cutting process such as turning and milling is a basic manufacturing process in the field of numerical control cutting, the surface topography of a workpiece after cutting can influence the fatigue strength, the coordination performance, the friction performance, the corrosion resistance and the lubricity of a part, and the surface topography is a comprehensive index for evaluating the surface quality and is also a premise for calculating structural characteristic parameters such as surface shape errors, roughness and the like. The method for off-line measurement of the surface topography images by using expensive special instruments after the processing is finished consumes a great deal of manpower and financial resources, so that the three-dimensional surface topography simulation and the surface topography gray scale image generation method have more advantages. The existing surface topography simulation method is to solve the three-dimensional surface topography of a workpiece through an analytical model.
At present, generation of three-dimensional surface topography images for cutting processes has been partially studied. According to the surface morphology prediction method after the ultraprecise turning of the free-form surface disclosed in the Chinese patent with the publication number of CN112387995B, the area needing surface morphology simulation is divided into grids through the planned tool path, coordinate data of all grid points in the simulation area are calculated according to the geometric position relation between each grid point and a tool contact point on the planned tool path, and then the calculated coordinate data is used for reconstructing the surface, so that the surface morphology simulation modeling of the curved surface single-point diamond turning is realized. The Chinese patent with publication number of CN107577882B discloses a simulation method for modeling and shaping the surface morphology of a side milling straight line curved surface, a tool bit file is utilized to establish a main shaft coordinate system, a tool cutting edge is subjected to kinematic description, a series of discrete edge point space point cloud data are obtained, the point cloud data are subjected to grid division, a follow-up containing box is established, the lowest point representation surface morphology of the point cloud data in each containing box is extracted, all grid nodes are traversed, and the three-dimensional surface morphology of milling processing is obtained. Chinese patent application publication No. CN114858432a discloses a peripheral milling surface topography analysis and prediction method, in which the vibration differential equation of the workpiece during milling is deduced by energy method by equivalently converting the contact pattern between the workpiece and the fixture into a linear spring-damping system; a vibration differential equation modal analysis solving method with constraint conditions is provided through a coordinate conversion method, so that the prediction of the workpiece position deviation is realized; and then, establishing a motion trail equation of the cutting edge in the milling process, and combining the position deviation of the workpiece and the motion trail equation of the cutting edge to provide a surface topography discrete algorithm of the surface topography of the workpiece. Chinese patent publication No. CN108710339B discloses a method for rapidly modeling the surface morphology of peripheral milling, in which the side teeth of an end mill participate in cutting portion is evenly scattered into a series of discrete points according to radial position angles, and each discrete point corresponds to a discrete surface of a tool and a discrete surface of a workpiece. By utilizing the characteristic that the tooth angles of the side teeth of the milling cutter are equal, a simulation method which is far less than the calculation times of a normal method is established.
Compared with the three-dimensional surface morphology image obtained by solving the cutting machining mechanism model by the method, the surface morphology gray level image can better reflect the depth, the density and the shape, particularly the texture characteristics of the trace left on the machined surface, and is time-consuming and complicated by solving the three-dimensional surface morphology and then converting the three-dimensional surface morphology into the surface morphology gray level image. Therefore, the direct generation of the gray scale image of the surface topography has great potential for surface quality monitoring.
Disclosure of Invention
In view of the above, the invention aims to provide a data-driven cutting surface topography gray scale image generation method, which aims at solving the problems of complex mathematical modeling and time-consuming iterative computation existing in the cutting surface topography based on the analytic modeling method at present, and realizes rapid and direct cutting surface topography gray scale image generation by establishing a mapping model from information such as processing signals, tool parameters and the like to the cutting surface topography gray scale image from end to end.
In order to achieve the above purpose, the present invention provides the following technical solutions:
a data-driven cutting surface topography gray scale image generation method comprises the following steps:
step one: a real cutting surface topography image in the training set is read and defined as a three-dimensional matrix of (H, W, C) according to the size of the image, expressed as: x is X t (H t ,W t ,C t )；t＝1；
Step two: x is to be t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 Outputting the probability that the cutting surface morphology image is true after the feature map is subjected to a classification function;
step three: calculating a loss value between the probability that the cutting surface morphology image obtained in the second step is true and the true probability 1 by using a binary cross entropy loss function, and carrying out reverse transfer so as to carry out iterative optimization on model parameters in a discriminant of iterative training t-1 times, thereby obtaining a discriminant model of iterative training t times;
step four: judging whether the iteration times T reach the set maximum iteration times T max : if yes, terminating iteration to obtain a generator with training completed, and executing a step ten; if not, t=t+1, executing the fifth step;
step five: the method comprises the steps of carrying out fluidization pretreatment on processing signal data, carrying out feature normalization and frequency spectrum transformation on the data by utilizing a short-time Fourier transformation technology by moving a short-time window along a tool path in the process of real-time monitoring of a processing surface, obtaining a processing signal spectrogram, defining the size of the spectrogram as a three-dimensional matrix of (H, W, C), and representing the size as X t1 (H t1 ,W t1 ,C t1 )；
Step six: acquiring stiffness data along the tool axis to obtain stiffness at a series of points along the tool axis, denoted as X t2 (1,W t2 1) let W be t1 ＝W t2 ；
Step seven: in the generator, a self-encoder network is constructed, and corresponding processing signal spectrogram and cutter rigidity data are converted into a five-channel identity matrix X 'through data conversion' t Expressed as: x'. t (H′ t ,W′ t ,C′ t )；
Step eight: constructing a generator network, adding the recursive residual block into the generator network to obtain an improved generator, and preprocessing the network characteristic diagram X' t Improved generator to obtain feature map M t8 ～M t13 ；
Step nine: map M of features t13 As the current X t Circularly executing the second step;
step ten: and inputting the processing signal spectrogram and the cutter rigidity data in the test set into a trained generator to generate a cutting surface topography image.
Further, in the second step, X t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 The method of (1) is as follows:
21 For X) t Feature extraction and modification of X by convolution operation t The number of channels is then normalized and activated to obtain a feature map M t1 (H′ t1 ，W′ t1 ，C′ t1 )；
22 For M) t1 Feature extraction and M change by convolution operation t1 The size of the feature map is then normalized and activated to obtain the feature map
23 For M) t2 Feature extraction and M change by convolution operation t2 The size of the feature map is then normalized and activated to obtain the feature map
24 For M) t3 Feature extraction and M change by convolution operation t3 The size of the feature map is then normalized and activated to obtain the feature map
25 For M) t4 Feature extraction and M change by convolution operation t4 The size of the feature map is then normalized and activated to obtain features
26 For M) t5 Feature extraction and M change by convolution operation t5 The size of the feature map is then normalized and activated to obtain features
27 For M) t6 Feature extraction and M change by convolution operation t6 The size of the feature map is then normalized and activated to obtain features
Further, the classification function is a Sigmoid function.
Further, in the seventh step, the corresponding processing signal spectrogram and tool stiffness data are converted into a five-channel identity matrix X' t The method of (1) is as follows:
71 For the processed signal spectrogram X t1 (H t1 ,W t1 ,C t1 ) Performing axis transposition, and reducing frequency dimension to 1/4 by (1×1) convolution to obtain
Then transpose the data channel and frequency axis of the processed signal spectrogram to obtainTo the point of
72 For tool stiffness data X) t2 (1,W t2 1) feature extraction, modification of X by convolution operation t2 Frequency dimension of
Obtain->
Then the frequency and the channel dimension axis are transposed to obtain +.>
73 Spectrum X of the processed signal t1 And tool stiffness data X t2 Splicing the channel dimensions to form a preprocessing network characteristic diagram
Further, in the step 8), the network characteristic diagram X is preprocessed t ′ Improved generator to obtain feature map M t8 ～M t13 The method of (1) is as follows:
81 For X' t Extracting features, and collecting X' t The size of the feature map and the number of channels are changed through convolution operation to obtain a feature map M t8 (H′ t ,W′ t ,C′ t )；
82 For M) t8 Extracting features, and extracting M t8 Changing the feature map M through N serial recursion residual units t8 Is the number of channels to obtain M t9 (H′ t ,W′ t ，N*C′ t )；
83 For M) t9 Extracting features, and extracting M t9 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t10 (H′ t ,W′ t ,2C′ t )；
84 For M) t10 Extracting features, and extracting M t10 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t11 (H′ t ,W′ t ,C′ t )；
85 (ii) M t8 And M t11 Direct addition by jump connection and change of the size of the feature map and the channel number by deconvolution operation
86 For M) t12 Extracting features, and extracting M t12 Changing the number of channels of the feature map through convolution operation, and obtaining a feature map M through data normalization and tanh activation operation t13 (2H′ t ，2W′ t ,1)。
Further, the recursive residual block comprises N recursive residual units, the recursive residual units being composed of a quadratic convolutional layer, a quadratic BN data normalization, and a quadratic linear activation function.
Further, the activation function is a LeakyReLU function.
The invention has the beneficial effects that:
the data-driven cutting surface morphology gray scale image generation method has the following advantages:
(1) Compared with the offline acquisition flow of the cutting surface topography and the kinematic simulation flow of the cutting surface topography, the acquisition time and the image generation time of the cutting surface topography image are greatly reduced, and the processing signal spectrogram and the tool rigidity data are directly converted into the gray image data of the cutting surface topography by using the generator in the generation countermeasure network, so that the flow of the acquisition and the simulation of the surface topography image data with long time consumption is avoided;
(2) The application of the generation countermeasure network realizes the conversion from the machining signal data and the tool rigidity information to the cutting surface topography image, is beneficial to generating global information between the processing signal data and the tool rigidity information of the countermeasure network, and improves the accuracy of the cutting surface topography image generation.
(3) The method can help a processing worker to directly acquire the appearance image of the cutting surface of the processed workpiece in the processing process, and the gray level image can better reflect the depth, the density and the shape of the trace left on the processing surface, particularly the texture characteristics, so that the method is beneficial for the worker to quickly judge the product quality of the processed part.
In summary, the data-driven method for generating the gray level image of the cutting surface topography can be used for quickly and directly generating the gray level image of the cutting surface topography by establishing the mapping model from the information such as the processing signals, the cutter parameters and the like to the gray level image of the cutting surface topography end to end.
Drawings
In order to make the objects, technical solutions and advantageous effects of the present invention more clear, the present invention provides the following drawings for description:
FIG. 1 is a schematic diagram of an embodiment of a data-driven machined surface topography gray scale image generation method of the present invention;
FIG. 2 is a schematic diagram of a discriminator network in the generation of an countermeasure network according to the present embodiment;
FIG. 3 is a schematic diagram of a generator network in the generation of an countermeasure network according to the present embodiment;
fig. 4 is a schematic structural diagram of a single recursive residual block in the present embodiment;
fig. 5 is an exemplary view of a gray scale image of the machined surface topography generated in this example.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and specific examples, which are not intended to limit the invention, so that those skilled in the art may better understand the invention and practice it.
As shown in fig. 1, the data-driven cutting surface topography gray scale image generation method of the present embodiment includes the following steps.
Step one: a real cutting surface topography image in the training set is read and defined as a three-dimensional matrix of (H, W, C) according to the size of the image, expressed as: x is X t (H t ,W t ,C t ) The method comprises the steps of carrying out a first treatment on the surface of the t=1. Specifically, in an embodiment, 1 real cut in the training set is readSurface topography image, three-dimensional matrix X for the read true cutting surface topography image t (640, 512,1) wherein the numbers 640, 512 of the first and second dimensions represent the width-to-height dimension of the true cutting surface topography image; the number 1 in the third dimension indicates that the number of channels of the true cut surface topography image is 1.
Step two: x is to be t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 And outputting the probability that the cutting surface topography image is true after the feature map is subjected to a classification function. As shown in fig. 2, for generating a schematic diagram of a discriminator network in the countermeasure network in this embodiment, each part of the discriminator is composed of a convolution operation, a data normalization operation, and an activation function. Specifically, in the present embodiment, X t Changing the number of channels of the feature map through convolution operation to obtain M t1 (H′ t1 ,W′ t1 ,C′ t1 ) The method comprises the steps of carrying out a first treatment on the surface of the Then the size of the feature map is changed through the size and the step length of the convolution kernel in the convolution layer to obtain a feature map M t2 ～M t7 . Feature map M t2 ～M t7 The sizes of (2) are respectively
Specifically, X t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 The method of (1) is as follows:
21 For X) t Feature extraction and modification of X by convolution operation t The number of channels is then normalized and activated to obtain a feature map M t1 (H′ t1 ,W′ t1 ,C′ t1 ). In this step, the number of convolution kernels in the convolution operation is 1, and the size is3×3, step size of 2, and padding of 0. Applying the convolution operation to X t Extracting features, and performing data standardization and activation operation on the feature map after the convolution operation to obtain a feature map M t1 (512,640,1)。
22 For M) t1 Feature extraction and M change by convolution operation t1 The size of the feature map is then normalized and activated to obtain the feature map
In this step, the convolution operation has a number of convolution kernels of 64, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t1 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t2 (256,320,64)。
23 For M) t2 Feature extraction and M change by convolution operation t2 The size of the feature map is then normalized and activated to obtain the feature map
In this step, the number of convolution kernels of the convolution operation is 128, the size is 3×3, the step size is 2, and the padding is 0. First apply the convolution operation to M t2 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t3 (128,160,128)。
24 For M) t3 Feature extraction and M change by convolution operation t3 The size of the feature map is then normalized and activated to obtain the feature map
In this step, the number of convolution kernels of the convolution operation is 256, the size is 3×3, the step size is 2, and the padding is 0. Applying the convolution operation to M t3 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t4 (64,80,256)。
25 For M) t4 Feature extraction and M change by convolution operation t4 The size of the feature map is then normalized and activated to obtain features
In this step, the convolution operation has 512 convolution kernels, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t4 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t5 (32,40,512)。
26 For M) t5 Feature extraction and M change by convolution operation t5 The size of the feature map is then normalized and activated to obtain features
In this step, the convolution operation has a convolution kernel number of 128, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t5 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t6 (32,40,128)。
27 For M) t6 Feature extraction and M change by convolution operation t6 The size of the feature map is then normalized and activated to obtain features
In this step, the convolution operation has a convolution kernel number of 1, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t6 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t7 (32,40,128)。/>
Thus, the classification function is applied to the feature map M t7 And (5) calculating, and outputting the probability of true cutting surface morphology online reconstructed image by using the classification function. In an embodiment, a classification function Sigmoid (Sigmoid growth curve) function versus signature M 7 Performing calculationAnd obtaining the probability that the online reconstructed image of the cutting surface morphology is true.
Step three: and (3) calculating a loss value between the probability that the cutting surface morphology image obtained in the step two is true and the true probability 1 by using a binary cross entropy loss function, and carrying out reverse transfer, so as to carry out iterative optimization on model parameters in the discriminant of iterative training t-1 times, and obtain a discriminant model of iterative training t times. In an embodiment, a binary cross entropy loss function is applied to calculate M t4 The loss value between the true probabilities with the same shape and the value of 1 is transmitted reversely in the discriminator, so that the model parameters in the discriminator are optimized iteratively. The binary cross entropy loss function is shown below.
Where n represents the number of samples, y i A label representing the ith sample, z i Representing the probability that the i-th sample is predicted to be a positive example.
Step four: judging whether the iteration times T reach the set maximum iteration times T max : if yes, terminating iteration to obtain a generator with training completed, and executing a step ten; if not, t=t+1, step five is performed. In an embodiment, a maximum number of iterations T is set max 500 times when the number of iterations reaches the maximum number of iterations T max And obtaining a trained generator model.
Step five: the method comprises the steps of carrying out fluidization pretreatment on processing signal data, carrying out feature normalization and frequency spectrum transformation on the data by utilizing a short-time Fourier transformation technology by moving a short-time window along a tool path in the process of real-time monitoring of a processing surface, obtaining a processing signal spectrogram, defining the size of the spectrogram as a three-dimensional matrix of (H, W, C), and representing the size as X t1 (H t1 ,W t1 ,C t1 )。
In the embodiment, matlab is used for cutting force signals in the feeding direction, cutting force signals perpendicular to the feeding direction and vibration signals in the feeding direction in the processingVibration signals perpendicular to the feed direction are subjected to spectrogram conversion by using short-time Fourier transform STFT, a processing signal spectrogram is obtained, and the size of the spectrogram is defined as a three-dimensional matrix (512,160,3) which is expressed as X t1 (512,160,3)。
Step six: obtaining stiffness data along the axial direction of the tool by using ANSYS APDL simulation to obtain stiffness at a series of points along the axial direction of the tool, denoted as X t2 (1,W t2 1) let W be t1 ＝W t2 . In an embodiment, application ANSYS APDL obtains tool point stiffness data for 160 points in the tool axial direction, denoted as X t2 (1,160,1)。
Step seven: in the generator, a self-encoder network is constructed, and corresponding processing signal spectrogram and cutter rigidity data are converted into a five-channel identity matrix X 'through data conversion' t Expressed as: x'. t (H′ t ,W′ t ,C′ t ). Converting the corresponding processing signal spectrogram and tool rigidity data into a five-channel identity matrix X' t The method of (1) is as follows:
71 For the processed signal spectrogram X t1 (H t1 ,W t1 ,C t1 ) Performing axis transposition, and reducing frequency dimension to 1/4 by (1×1) convolution to obtain
Then transpose the data channel and frequency axis of the processed signal spectrogram to obtain
Specifically, in an embodiment, a self-encoder network is constructed, a PIL library is used to input a processing signal spectrogram and cutter stiffness data, X and y cutting forces and X and y vibration signals are spliced on a channel to form (512,160,4), and a spectrogram data axis is transposed to enable the channel and a frequency dimension to be interchanged to obtain frequency characteristics of a full frequency band, so that X is obtained t1 (4,160,512); the frequency dimension is reduced to 1/4 by convolution with (1X 1) to obtain X t1 (4,160,128) transpose the data channels and frequency axes of the processed signal spectrum to obtain X t1 (128,1604) for stitching with stiffness data.
72 For tool stiffness data X) t2 (1,W t2 1) feature extraction, modification of X by convolution operation t2 Frequency dimension of
Obtain->
Then the frequency and the channel dimension axis are transposed to obtain +.>
In the present embodiment, the tool stiffness data X t2 (1,160,1) by convolution with (1×1), frequency dimension 1 is lifted to 128 to obtain X t2 (1,160,128) and transposed with frequency and channel axes to obtain X t2 (128,160,1)。
73 Spectrum X of the processed signal t1 And tool stiffness data X t2 Splicing the channel dimensions to form a preprocessing network characteristic diagram
Signal spectrum X to be processed t1 And tool stiffness data X t2 Splicing the channel dimensions to form a preprocessing network characteristic diagram X' t (128,160,5)。
Step eight: constructing a generator network, adding the recursive residual block into the generator network to obtain an improved generator, and preprocessing the network characteristic diagram X' t Improved generator to obtain feature map M t8 ～M t13 . Specifically, as shown in fig. 3, the input data is firstly subjected to feature extraction by means of downsampling and then connecting with several recursive residual blocks, and then gradually upsampled to the dimension identical to the actual measured surface morphology, wherein each recursive residual block is composed of a convolution layer, a data standardization function and an activation function. As shown in fig. 4, the recursive residual block includes N recursive residual units consisting of a double convolution layer, a double BN data normalization, and a double linear activation function. In this embodiment, the activation function is LeakyReLU function.
In an embodiment, the network profile X 'is preprocessed' t Obtaining a characteristic diagram M through a generator t8 ～M t13 The method of (1) is as follows: first X 'is added' t Changing the number of channels of the feature map through up-sampling operation to obtain a feature map M t8 (H′ t ,W′ t ,C′ t ) The method comprises the steps of carrying out a first treatment on the surface of the Then M is added t8 M through N recursive residual block changes t8 The size of the feature map, M t9 (H′ t ,W′ t ，N*C′ t ) Then M is added with t9 Obtaining a characteristic diagram M through convolution operation t10 (H′ t ,W′ t ,2C′ t ) Then M is taken up t10 Obtaining a characteristic diagram through convolution operation to obtain a characteristic diagram M t11 (H′ t ,W′ t ,C′ t ) Then M is taken t8 And M t11 Direct addition through jump connection and feature map is obtained through deconvolution operation
Finally will M t12 Obtaining a characteristic diagram M through convolution operation, data standardization and tanh activation operation t13 (2H′ t ，2W′ t ,1)。
Specifically, in this embodiment, the network feature map X will be preprocessed t ′ Improved generator to obtain feature map M t8 ～M t13 The method of (1) is as follows:
81 For X' t Extracting features, and collecting X' t The size of the feature map and the number of channels are changed through convolution operation to obtain a feature map M t8 (H′ t ,W′ t ,C′ t ). In this step, the convolution operation has a number of convolution kernels of 64, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to X' t Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t8 (256,320,64)。
82 For M) t8 Extracting features, and extracting M t8 Changing bits through N serial recursive residual unitsSign map M t8 Is the number of channels to obtain M t9 (H′ t ,W′ t ，N*C′ t ). In this step, the convolution operation has a number of convolution kernels of 64, a size of 3×3, a step size of 2, and a padding of 0. Will M t8 Feature extraction is carried out through N serial recursion residual units, and then data standardization and activation operation are carried out on the feature map obtained after convolution operation, so as to obtain a feature map M t9 (256,320,64*N)。
83 For M) t9 Extracting features, and extracting M t9 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t10 (H′ t ,W′ t ,2C′ t ). In this step, the convolution operation has a convolution kernel number of 128, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t9 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t10 (256,320,128)。
84 For M) t10 Extracting features, and extracting M t10 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t11 (H′ t ,W′ t ,C′ t ). In this step, the convolution operation has a number of convolution kernels of 64, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t10 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t11 (256,320,64)。
85 (ii) M t8 And M t11 Direct addition by jump connection and change of the size of the feature map and the channel number by deconvolution operation
In this step, the number of convolution kernels of the convolution operation is 32, the size is 3×3, the step size is 2, and the padding is 0. First M is t8 (256,320,64) and M t11 (256,320,64) direct addition by jump ligation to give M' t12 (256,320,64) then applying the deconvolution operation to M' t12 Feature extractionThen, carrying out data standardization and activation operation on the feature map obtained after deconvolution operation to obtain a feature map M t12 (512,640,32)。
86 For M) t12 Extracting features, and extracting M t12 Changing the number of channels of the feature map through convolution operation, and obtaining a feature map M through data normalization and tanh activation operation t13 (2H′ t ，2W′ t ,1). In this step, the convolution operation has a convolution kernel number of 1, a size of 3×3, a step size of 2, and a padding of 0. Applying the convolution operation to M t12 Extracting features, and performing data standardization and activation operation on the feature map obtained after convolution operation to obtain a feature map M t13 (512,640,1)。
Step nine: map M of features t13 As the current X t And (3) circularly executing the step two.
Step ten: and inputting the processing signal spectrogram and the cutter rigidity data in the test set into a trained generator to generate a cutting surface topography image.
In the embodiment, the processing signal spectrogram and the cutter rigidity data in the test set are input into a trained generator, so that the generation of the cutting surface morphology image is realized. The resulting cutting surface topography image is shown in fig. 5.
The above-described embodiments are merely preferred embodiments for fully explaining the present invention, and the scope of the present invention is not limited thereto. Equivalent substitutions and modifications will occur to those skilled in the art based on the present invention, and are intended to be within the scope of the present invention. The protection scope of the invention is subject to the claims.
Claims (7)
1. A data-driven cutting surface morphology gray scale image generation method is characterized in that: the method comprises the following steps:
step one: a real cutting surface topography image in the training set is read and defined as a three-dimensional matrix of (H, W, C) according to the size of the image, expressed as: x is X t (H t ，W t ，C t )；t＝1；
Step two: x is to be t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 Outputting the probability that the cutting surface morphology image is true after the feature map is subjected to a classification function;
step three: calculating a loss value between the probability that the cutting surface morphology image obtained in the second step is true and the true probability 1 by using a binary cross entropy loss function, and carrying out reverse transfer so as to carry out iterative optimization on model parameters in a discriminant of iterative training t-1 times, thereby obtaining a discriminant model of iterative training t times;
step four: judging whether the iteration times T reach the set maximum iteration times T max : if yes, terminating iteration to obtain a generator with training completed, and executing a step ten; if not, t=t+1, executing the fifth step;
step five: the method comprises the steps of carrying out fluidization pretreatment on processing signal data, carrying out feature normalization and frequency spectrum transformation on the data by utilizing a short-time Fourier transformation technology by moving a short-time window along a tool path in the process of real-time monitoring of a processing surface, obtaining a processing signal spectrogram, defining the size of the spectrogram as a three-dimensional matrix of (H, W, C), and representing the size as X t1 (H t1 ，W t1 ，C t1 )；
Step six: acquiring stiffness data along the tool axis to obtain stiffness at a series of points along the tool axis, denoted as X t2 (1，W t2 1) let W be t1 ＝W t2 ；
Step seven: in the generator, a self-encoder network is constructed, and corresponding processing signal spectrogram and cutter rigidity data are converted into a five-channel identity matrix X 'through data conversion' t Expressed as: x'. t (H′ t ，W′ t ，C′ t )；
Step eight: constructing a generator network, adding the recursive residual block into the generator network to obtain an improved generator, and preprocessing the network characteristic diagram X' t Improved generator to obtain feature map M t8 ～M t13 ；
Step nine: will beFeature map M t13 As the current X t Circularly executing the second step;
step ten: and inputting the processing signal spectrogram and the cutter rigidity data in the test set into a trained generator to generate a cutting surface topography image.
2. The data-driven machined surface topography gray scale image generation method of claim 1, wherein: in the second step, x t Obtaining a characteristic diagram M through iterative training of a discriminator for t-1 times t1 ～M t7 The method of (1) is as follows:
21 For X) t Feature extraction and modification of X by convolution operation t The number of channels is then normalized and activated to obtain a feature map M t1 (H′ t1 ，W′ t1 ，C′ t1 )；
22 For M) t1 Feature extraction and M change by convolution operation t1 The size of the feature map is then normalized and activated to obtain the feature map
23 For M) t2 Feature extraction and M change by convolution operation t2 The size of the feature map is then normalized and activated to obtain the feature map
24 For M) t3 Feature extraction and M change by convolution operation t3 The size of the feature map is then normalized and activated to obtain the feature map
25 For M) t4 Feature extraction and M change by convolution operation t4 The size of the feature map is then normalized and activated to obtain features
26 For M) t5 Feature extraction and M change by convolution operation t5 The size of the feature map is then normalized and activated to obtain features
/>
27 For M) t6 Feature extraction and M change by convolution operation t6 The size of the feature map is then normalized and activated to obtain features
3. The data-driven machined surface topography gray scale image generation method of claim 1, wherein: the classification function is a Sigmoid function.
4. The data-driven machined surface topography gray scale image generation method of claim 1, wherein: in the seventh step, the corresponding processing signal spectrogram and the cutter rigidity data are converted into a five-channel identity matrix X' t The method of (1) is as follows:
71 For the processed signal spectrogram X t1 (H t1 ，W t1 ，C t1 ) Performing axis transposition, and reducing frequency dimension to 1/4 by (1×1) convolution to obtain
Then transpose the data channel and frequency axis of the processed signal spectrogram to obtain
72 For tool stiffness data X) t2 (1，W t2 1) feature extraction, modification of X by convolution operation t2 Frequency dimension of
Obtaining
Then the frequency and the channel dimension axis are transposed to obtain +.>
73 Spectrum X of the processed signal t1 And tool stiffness data X t2 Splicing the channel dimensions to form a preprocessing network characteristic diagram
5. The data-driven machined surface topography gray scale image generation method of claim 1, wherein: in the step 8), the network characteristic diagram X 'is preprocessed' t Improved generator to obtain feature map M t8 ～M t13 The method of (1) is as follows:
81 For X' t Extracting features, and collecting X' t The size of the feature map and the number of channels are changed through convolution operation to obtain a feature map M t8 (H′ t ，W′ t ，C′ t )；
82 For M) t8 Extracting features, and extracting M t8 Changing the feature map M through N serial recursion residual units t8 Is the number of channels to obtain M t9 (H′ t ，W′ t ，N*C′ t )；
83 For M) t9 Extracting features, and extracting M t9 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t10 (H′ t ，W′ t ，2C′ t )；
84 For M) t10 Extracting features, and extracting M t10 Changing the number of channels of the feature map through convolution operation to obtain a feature map M t11 (H′ t ，W′ t ，C′ t )；
85 (ii) M t8 And M t11 Direct addition by jump connection and change of the size of the feature map and the channel number by deconvolution operation
86 For M) t12 Extracting features, and extracting M t12 Changing the number of channels of the feature map through convolution operation, and obtaining a feature map M through data normalization and tanh activation operation t13 (2H′ t ，2W′ t ，1)。
6. The data-driven machined surface topography gray scale image generation method of claim 1, wherein: the recursive residual block comprises N recursive residual units, the recursive residual units being composed of a quadratic convolution layer, a quadratic BN data normalization and a quadratic linear activation function.
7. The data-driven machined surface topography gray scale image generation method of claim 6, wherein: the activation function is a LeakyReLU function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310065585.4A CN116012480A (en) | 2023-01-13 | 2023-01-13 | Data-driven cutting surface morphology gray scale image generation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310065585.4A CN116012480A (en) | 2023-01-13 | 2023-01-13 | Data-driven cutting surface morphology gray scale image generation method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116012480A true CN116012480A (en) | 2023-04-25 |
Family
ID=86037187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310065585.4A Pending CN116012480A (en) | 2023-01-13 | 2023-01-13 | Data-driven cutting surface morphology gray scale image generation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116012480A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116977652A (en) * | 2023-09-22 | 2023-10-31 | 之江实验室 | Workpiece surface morphology generation method and device based on multi-mode image generation |
-
2023
- 2023-01-13 CN CN202310065585.4A patent/CN116012480A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116977652A (en) * | 2023-09-22 | 2023-10-31 | 之江实验室 | Workpiece surface morphology generation method and device based on multi-mode image generation |
| CN116977652B (en) * | 2023-09-22 | 2023-12-22 | 之江实验室 | Workpiece surface morphology generation method and device based on multi-mode image generation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107877262B (en) | A kind of numerical control machine tool wear monitoring method based on deep learning | |
| CN113414638B (en) | Variable working condition milling cutter wear state prediction method based on milling force time sequence diagram deep learning | |
| CN111633467B (en) | Cutter wear state monitoring method based on one-dimensional depth convolution automatic encoder | |
| CN112247674B (en) | Cutter wear prediction method | |
| CN111914480B (en) | Processing feature intelligent recognition method based on point cloud semantic segmentation | |
| CN113780153B (en) | Cutter wear monitoring and predicting method | |
| CN109434564A (en) | A kind of cutter wear state monitoring method based on deep neural network | |
| CN112435363A (en) | Cutter wear state real-time monitoring method | |
| CN113458873B (en) | Method for predicting wear loss and residual life of cutter | |
| CN114297912B (en) | Cutter abrasion prediction method based on deep learning | |
| CN113798920A (en) | Cutter wear state monitoring method based on variational automatic encoder and extreme learning machine | |
| Zhang et al. | Tool wear online monitoring method based on DT and SSAE-PHMM | |
| CN116012480A (en) | Data-driven cutting surface morphology gray scale image generation method | |
| CN113504768B (en) | High-precision product digital twin computability method for assembly quality prediction | |
| Li et al. | A data and knowledge-driven cutting parameter adaptive optimization method considering dynamic tool wear | |
| CN113485244A (en) | Numerical control machine tool control system and method based on cutter wear prediction | |
| CN111666632A (en) | Hull profile curve reverse construction method based on geometric characteristic parameters | |
| CN113664626A (en) | Spiral groove grinding process system establishing method based on discrete point cloud principle | |
| CN115186574A (en) | Tool residual life prediction method based on gated cyclic residual error network | |
| CN114860709A (en) | Bi-GAN-based power system missing value filling method | |
| CN112846938B (en) | Main shaft rotation precision degradation traceability system under cutting working condition | |
| CN106709198A (en) | Method for carrying out transform from bent pipe numerical control digital analogy to CAD (Computer-Aided Design) digital analogy | |
| CN117066972A (en) | Turbine blade root slot milling outlet burr prediction method based on data reasoning | |
| Li et al. | A signal-to-image fault classification method based on multi-sensor data for robotic grinding monitoring | |
| CN110175372B (en) | Envelope surface characterization method based on mother surface characteristic parameters |
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 |