CN117206063A - Grinding intelligent control system and method based on deep learning - Google Patents
Grinding intelligent control system and method based on deep learning Download PDFInfo
- Publication number
- CN117206063A CN117206063A CN202311419112.6A CN202311419112A CN117206063A CN 117206063 A CN117206063 A CN 117206063A CN 202311419112 A CN202311419112 A CN 202311419112A CN 117206063 A CN117206063 A CN 117206063A
- Authority
- CN
- China
- Prior art keywords
- grinding
- feature vector
- feature
- time
- sound
- 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
- 238000000227 grinding Methods 0.000 title claims abstract description 233
- 238000013135 deep learning Methods 0.000 title claims abstract description 49
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000012544 monitoring process Methods 0.000 claims abstract description 110
- 239000013598 vector Substances 0.000 claims description 190
- 239000011159 matrix material Substances 0.000 claims description 71
- 230000004927 fusion Effects 0.000 claims description 61
- 238000010586 diagram Methods 0.000 claims description 53
- 238000005498 polishing Methods 0.000 claims description 50
- 230000005236 sound signal Effects 0.000 claims description 45
- 238000013527 convolutional neural network Methods 0.000 claims description 36
- 230000008859 change Effects 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 20
- 238000003062 neural network model Methods 0.000 claims description 16
- 230000004043 responsiveness Effects 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 15
- 238000006243 chemical reaction Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 13
- 230000009466 transformation Effects 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 12
- 230000001965 increasing effect Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 229940050561 matrix product Drugs 0.000 claims description 9
- 230000004044 response Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 6
- 230000003247 decreasing effect Effects 0.000 claims description 5
- 238000012546 transfer Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 18
- 238000009826 distribution Methods 0.000 abstract description 10
- 230000008901 benefit Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 239000002245 particle Substances 0.000 description 3
- 239000002994 raw material Substances 0.000 description 3
- 239000007787 solid Substances 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005272 metallurgy Methods 0.000 description 1
- 238000003801 milling Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000000123 paper Substances 0.000 description 1
- 239000000049 pigment Substances 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Abstract
The application relates to the technical field of intelligent control, and particularly discloses an intelligent grinding control system and method based on deep learning. Therefore, the automatic monitoring of the grinding process and the accurate control of the grinding speed are realized, and the granularity distribution and the surface quality of the grinding product are ensured, so that the yield of the product is improved.
Description
Technical Field
The application relates to the technical field of intelligent control, in particular to an intelligent grinding control system and method based on deep learning.
Background
In various industrial fields, crushing and grinding are a very important step in the processing of solid raw materials, and grinding machines are also increasingly widely applied to industries such as metallurgy, ore, chemical industry, ceramics, pigment, paper making, food, medicine and the like.
In the fine grinding production of solid raw materials, different grinding speeds are generally used for different grinding stages, such as pre-grinding, primary grinding and re-grinding. The mode of controlling the polishing rate in stages is too rough, and the polishing rate is not adaptively adjusted based on the actual condition of the polishing product, resulting in low polishing efficiency and difficult guarantee of the yield of the polishing product. Moreover, the existing grinding equipment is controlled manually, so that the problems of low production efficiency, poor product consistency and the like exist, and the professional ability of operators directly influences the product quality.
Accordingly, a deep learning-based polishing intelligent control system and method are desired.
Disclosure of Invention
The present application has been made to solve the above-mentioned technical problems. The embodiment of the application provides an intelligent grinding control system and method based on deep learning, which adopts a machine learning technology to extract real-time change characteristics of granularity of an object to be ground and grinding sound time sequence characteristics from grinding monitoring videos of the object to be ground and sound signals generated by grinding equipment respectively, and carries out real-time control on the rotating speed of the grinding equipment based on the fusion characteristics of the real-time change characteristics and the grinding sound time sequence characteristics. Therefore, the automatic monitoring of the grinding process and the accurate control of the grinding speed are realized, and the granularity distribution and the surface quality of the grinding product are ensured, so that the yield of the product is improved.
Accordingly, according to one aspect of the present application, there is provided a deep learning-based intelligent control system for grinding, comprising:
the data acquisition module is used for acquiring a grinding monitoring video of an object to be ground in a preset time period and a sound signal generated by grinding equipment in the preset time period;
the S conversion module is used for carrying out S conversion on the sound signal to obtain an S conversion time-frequency diagram;
the sound feature extraction module is used for enabling the S-transformation time-frequency diagram to pass through a first convolution neural network model serving as a filter so as to obtain a sound feature vector;
the grinding state feature extraction module is used for enabling the grinding monitoring video to obtain a grinding state monitoring feature vector through a second convolution neural network model using a time attention mechanism;
the responsiveness estimation module is used for calculating responsiveness estimation of the grinding state monitoring feature vector relative to the sound feature vector so as to obtain a classification feature matrix;
the fusion module is used for fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fusion feature vector;
the optimization module is used for multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix;
And the control result generation module is used for passing the optimized classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the rotating speed value of the grinding equipment at the current time point is increased or reduced.
In the above-mentioned intelligent grinding control system based on deep learning, the S-conversion module is configured to: s-transforming the sound signal by using the following transformation formula to obtain the S-transformed time-frequency diagram;
wherein, the transformation formula is:
wherein S (f, τ) represents the S-transformed time-frequency diagram, τ is a time shift factor, x (t) represents the sound signal, f represents frequency, and t represents time.
In the above-mentioned grinding intelligent control system based on deep learning, the sound feature extraction module is configured to: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: performing convolution processing on the input data based on a two-dimensional convolution check to generate a convolution feature map; performing global average pooling processing on each feature matrix along the channel dimension on the convolution feature map to generate a pooled feature map; non-linear activation is carried out on the characteristic values of all the positions in the pooled characteristic map so as to generate an activated characteristic map; the output of the last layer of the first convolutional neural network model is the sound feature vector, the input from the second layer to the last layer of the first convolutional neural network model is the output of the last layer, and the input of the first layer of the first convolutional neural network model is the S-transformation time-frequency diagram.
In the above-mentioned intelligent grinding control system based on deep learning, the grinding state feature extraction module includes: the sampling unit is used for extracting a plurality of grinding monitoring key frames from the grinding monitoring video at a preset sampling frequency; the time sequence coding unit is used for enabling the plurality of grinding monitoring key frames to pass through the second convolution neural network model using the time attention mechanism so as to obtain a grinding state time sequence change characteristic diagram; and the pooling unit is used for carrying out global pooling processing on each characteristic matrix along the channel dimension on the grinding state time sequence change characteristic diagram so as to obtain the grinding state monitoring characteristic vector.
In the above-mentioned intelligent polishing control system based on deep learning, the time sequence encoding unit includes: an adjacent key frame extraction subunit, configured to extract adjacent first and second grinding monitor key frames from the plurality of grinding monitor key frames; the convolution coding subunit is used for enabling the first grinding monitoring key frame and the second grinding monitoring key frame to respectively pass through a first convolution layer and a second convolution layer of the second convolution neural network model so as to obtain a first convolution characteristic diagram and a second convolution characteristic diagram; a time attention subunit, configured to calculate a point-by-point multiplication between the first convolution feature map and the second convolution feature map to obtain a time attention map; an attention generation subunit, configured to input the time attention attempt to a Softmax activation function to obtain a time attention profile; an attention applying subunit, configured to calculate a multiplication by location point between the second convolution feature map and the time attention feature map to obtain a time enhancement feature map corresponding to the second grinding monitoring key frame; and the time enhancement feature map corresponding to the last grinding monitoring key frame in the plurality of grinding monitoring key frames is the grinding state time sequence change feature map.
In the above-mentioned intelligent control system for polishing based on deep learning, the responsiveness estimation module is configured to: calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain the classification feature matrix according to the following response formula;
wherein, the response formula is:
wherein V is 1 Representing the grinding state monitoring feature vector, V 2 The sound feature vector is represented by a vector of the sound,representing the matrix multiplied by the vector, M representing the classification feature matrix.
In the above-mentioned intelligent grinding control system based on deep learning, the fusion module includes: the per-position average value calculation unit is used for calculating a per-position average value vector between the grinding state monitoring feature vector and the sound feature vector to obtain a pivot feature vector; a first KL divergence value calculation unit configured to calculate a first KL divergence value between the grinding state monitoring feature vector and the pivot feature vector; a second KL-divergence value calculation unit configured to calculate a second KL-divergence value between the sound feature vector and the pivot feature vector; the weighted fusion unit is used for taking the first KL divergence value and the second KL divergence value as weights and fusing the grinding state monitoring feature vector and the sound feature vector by the following fusion formula to obtain the fused feature vector; wherein, the fusion formula is: v (V) i ＝αV 1 +βV 2 ，V 1 Representing the grinding state monitoring feature vector, V 2 Representing the sound feature vector, alpha representing the first KL divergence value, beta representing the second KL divergence value, V i Representing the fused feature vector.
According to another aspect of the present application, there is provided an intelligent control method for polishing based on deep learning, comprising:
acquiring a grinding monitoring video of an object to be ground in a preset time period and a sound signal generated by grinding equipment in the preset time period;
s-transforming the sound signal to obtain an S-transformed time-frequency diagram;
the S transformation time-frequency diagram is passed through a first convolution neural network model serving as a filter to obtain a sound feature vector;
the grinding monitoring video is processed through a second convolution neural network model using a time attention mechanism to obtain a grinding state monitoring feature vector;
calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix;
fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fused feature vector;
multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix;
And the optimized classification characteristic matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the grinding equipment at the current time point is increased or decreased.
In the above-mentioned intelligent polishing control method based on deep learning, performing S-conversion on the sound signal to obtain an S-conversion time-frequency diagram, including: s-transforming the sound signal by using the following transformation formula to obtain the S-transformed time-frequency diagram;
wherein, the transformation formula is:
wherein S (f, τ) represents the S-transformed time-frequency diagram, τ is a time shift factor, x (t) represents the sound signal, f represents frequency, and t represents time.
In the above-mentioned intelligent grinding control method based on deep learning, the step of passing the S-transformed time-frequency graph through a first convolutional neural network model as a filter to obtain a sound feature vector includes: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: performing convolution processing on the input data based on a two-dimensional convolution check to generate a convolution feature map; performing global average pooling processing on each feature matrix along the channel dimension on the convolution feature map to generate a pooled feature map; non-linear activation is carried out on the characteristic values of all the positions in the pooled characteristic map so as to generate an activated characteristic map; the output of the last layer of the first convolutional neural network model is the sound feature vector, the input from the second layer to the last layer of the first convolutional neural network model is the output of the last layer, and the input of the first layer of the first convolutional neural network model is the S-transformation time-frequency diagram.
Compared with the prior art, the intelligent grinding control system and method based on deep learning provided by the application adopt a machine learning technology, respectively extract the real-time change characteristic of granularity of an object to be ground and the time sequence characteristic of grinding sound from a grinding monitoring video of the object to be ground and a sound signal generated by grinding equipment, and perform real-time control on the rotating speed of the grinding equipment based on the fusion characteristic of the real-time change characteristic and the time sequence characteristic of the granularity of the object to be ground. Therefore, the automatic monitoring of the grinding process and the accurate control of the grinding speed are realized, and the granularity distribution and the surface quality of the grinding product are ensured, so that the yield of the product is improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing embodiments of the present application in more detail with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and together with the embodiments of the application, and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.
FIG. 1 is a block diagram of a deep learning based intelligent control system for grinding in accordance with an embodiment of the present application.
Fig. 2 is a schematic architecture diagram of a deep learning-based polishing intelligent control system according to an embodiment of the application.
Fig. 3 is a block diagram of a grinding status feature extraction module in a deep learning based grinding intelligent control system according to an embodiment of the application.
Fig. 4 is a block diagram of a timing encoding unit in a deep learning-based polishing intelligent control system according to an embodiment of the present application.
Fig. 5 is a flowchart of a method for intelligent control of polishing based on deep learning according to an embodiment of the present application.
Detailed Description
Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.
FIG. 1 is a block diagram of a deep learning based intelligent control system for grinding in accordance with an embodiment of the present application. As shown in fig. 1, a deep learning-based polishing intelligent control system 100 according to an embodiment of the present application includes: a data acquisition module 110, configured to acquire a grinding monitoring video of an object to be ground for a predetermined period of time and a sound signal generated by a grinding device for the predetermined period of time; the S conversion module 120 is configured to perform S conversion on the sound signal to obtain an S-converted time-frequency diagram; a sound feature extraction module 130, configured to pass the S-transformed time-frequency graph through a first convolutional neural network model serving as a filter to obtain a sound feature vector; the grinding state feature extraction module 140 is configured to obtain a grinding state monitoring feature vector by using the second convolutional neural network model of the time attention mechanism; a responsiveness estimation module 150 for calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix; a fusion module 160, configured to fuse the grinding state monitoring feature vector and the sound feature vector to obtain a fused feature vector; the optimizing module 170 is configured to multiply the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculate a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix; the control result generating module 180 is configured to pass the optimized classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the rotation speed value of the grinding apparatus at the current time point should be increased or decreased.
Fig. 2 is a schematic architecture diagram of a deep learning-based polishing intelligent control system according to an embodiment of the application. As shown in fig. 2, first, a grinding monitor video of an object to be ground for a predetermined period of time and a sound signal generated by a grinding apparatus for the predetermined period of time are acquired. And then, carrying out S-transformation on the sound signal to obtain an S-transformation time-frequency diagram. Then, the S-transformed time-frequency diagram is passed through a first convolutional neural network model as a filter to obtain a sound feature vector. And simultaneously, the grinding monitoring video is processed through a second convolution neural network model using a time attention mechanism to obtain a grinding state monitoring feature vector. Secondly, calculating the response estimation of the grinding state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix. And then fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fused feature vector. And then multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix. Finally, the optimized classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotation speed value of the grinding equipment at the current time point should be increased or decreased.
In the above-mentioned intelligent control system 100 for deep learning-based polishing, the data acquisition module 110 is configured to acquire a polishing monitoring video of an object to be polished for a predetermined period of time and a sound signal generated by a polishing device for the predetermined period of time. As noted above in the background art, in the fine grinding production of solid raw materials, different grinding speeds are generally employed for different grinding stages, such as pre-grinding, primary grinding and re-grinding. The mode of controlling the polishing rate in stages is too rough, and the polishing rate is not adaptively adjusted based on the actual condition of the polishing product, resulting in low polishing efficiency and difficult guarantee of the yield of the polishing product. Moreover, the existing grinding equipment is controlled manually, so that the problems of low production efficiency, poor product consistency and the like exist, and the professional ability of operators directly influences the product quality. Therefore, it is desirable to intelligently control the grinding speed in the grinding process, and to improve the production efficiency and the product quality.
Accordingly, the polishing rate should be adaptively adjusted according to the real-time change of the granularity of the object to be polished when the polishing rate is actually controlled. In addition, since the grinding device can generate a sound signal in the grinding process, the sound signal can reflect part of the grinding state characteristic information of the object to be ground, and therefore the expression of the grinding state change characteristic of the object to be ground is enhanced by utilizing the sound characteristic, and the accuracy of the grinding speed control can be obviously improved. Based on the above, in the technical scheme of the application, a machine learning technology is adopted to extract the real-time change characteristic of granularity of the object to be ground and the time sequence characteristic of grinding sound from the grinding monitoring video of the object to be ground and the sound signal generated by the grinding equipment respectively, and the real-time control of the rotating speed of the grinding equipment is performed based on the fusion characteristic of the real-time change characteristic and the time sequence characteristic of the granularity of the object to be ground. Therefore, the automatic monitoring of the grinding process and the accurate control of the grinding speed are realized, and the granularity distribution and the surface quality of the grinding product are ensured, so that the yield of the product is improved. Specifically, in the technical scheme of the application, firstly, a grinding monitoring video of an object to be ground in a preset time period is collected through a camera, and a sound signal generated by grinding equipment in the preset time period is collected through a sensor.
In the above-mentioned deep learning-based polishing intelligent control system 100, the S-transform module 120 is configured to perform S-transform on the sound signal to obtain an S-transformed time-frequency diagram. In order to convert the sound signal from the time domain to the time-frequency domain to obtain the energy distribution of the sound signal at different frequencies, thereby enhancing the expression of sound features, the sound signal is further subjected to an S-transform (also called short-time fourier transform) to obtain an S-transformed time-frequency diagram. It will be appreciated that the sound signal is typically composed of a superposition of frequency components, which may be represented as a time-varying spectrogram by S-transforming the sound signal to provide information about the frequency characteristics of the sound signal, including the time-varying behavior of the main frequency components in the sound signal, to provide data support for subsequent feature extraction and classification analysis.
Accordingly, in one specific example, the S transform module 120 is configured to: s-transforming the sound signal by using the following transformation formula to obtain the S-transformed time-frequency diagram;
wherein, the transformation formula is:
wherein S (f, τ) represents the S-transformed time-frequency diagram, τ is a time shift factor, x (t) represents the sound signal, f represents frequency, and t represents time.
In the above-mentioned deep learning-based polishing intelligent control system 100, the acoustic feature extraction module 130 is configured to pass the S-transformed time-frequency chart through a first convolutional neural network model as a filter to obtain an acoustic feature vector. It should be appreciated that convolutional neural networks (Convolutional Neural Network, CNN) are a deep learning model that is good at processing data having spatial structures, such as images and time-frequency diagrams. In the technical scheme of the application, the S-transformation time-frequency diagram is input into a first convolutional neural network model, local features in the S-transformation time-frequency diagram of the sound signal, such as the intensity of frequency components, the change mode of frequency spectrum and the like, are captured through convolution and pooling operations, and the features are gradually combined to obtain a higher-level representation. And the convolutional neural network model can automatically learn and extract the characteristics related to the grinding process without manually designing and selecting the characteristics, and the strong characterization capability of the convolutional neural network model can help to capture more complex and abstract sound characteristics, so that more accurate and effective characteristic representation is provided for the self-adaptive control of the grinding rotating speed in the grinding process.
Accordingly, in one specific example, the acoustic feature extraction module 130 is configured to: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: performing convolution processing on the input data based on a two-dimensional convolution check to generate a convolution feature map; performing global average pooling processing on each feature matrix along the channel dimension on the convolution feature map to generate a pooled feature map; non-linear activation is carried out on the characteristic values of all the positions in the pooled characteristic map so as to generate an activated characteristic map; the output of the last layer of the first convolutional neural network model is the sound feature vector, the input from the second layer to the last layer of the first convolutional neural network model is the output of the last layer, and the input of the first layer of the first convolutional neural network model is the S-transformation time-frequency diagram.
In the above-mentioned deep learning-based polishing intelligent control system 100, the polishing status feature extraction module 140 is configured to obtain a polishing status monitoring feature vector from the polishing monitoring video by using a second convolutional neural network model of a time attention mechanism. It should be appreciated that by video recording the milling process, important characteristic information of the particle distribution of the milled material may be obtained. In the technical scheme of the application, the grinding state change characteristic mining is performed on the grinding monitoring video by using a second convolution neural network model. The second convolutional neural network model may extract spatial features in the video image through convolution and pooling operations. Meanwhile, by applying a time attention mechanism on a time dimension, the video sequence can be weighted, so that a model can pay attention to and capture key frames related to the grinding state better, thereby capturing time sequence change characteristic information of particle distribution of grinding materials in the grinding process, providing more comprehensive and accurate information for analysis and control of the grinding process, and realizing self-adaptive control and optimization of the grinding process.
Fig. 3 is a block diagram of a grinding status feature extraction module in a deep learning based grinding intelligent control system according to an embodiment of the application. As shown in fig. 3, the grinding status feature extraction module 140 includes: a sampling unit 141, configured to extract a plurality of grinding monitoring key frames from the grinding monitoring video at a predetermined sampling frequency; a time sequence encoding unit 142, configured to pass the plurality of grinding monitoring key frames through the second convolutional neural network model using the time attention mechanism to obtain a grinding state time sequence variation feature map; and a pooling unit 143, configured to perform global pooling processing on each feature matrix along the channel dimension on the polishing state time sequence variation feature map to obtain the polishing state monitoring feature vector.
Fig. 4 is a block diagram of a timing encoding unit in a deep learning-based polishing intelligent control system according to an embodiment of the present application. As shown in fig. 4, the timing encoding unit 142 includes: an adjacent key frame extraction subunit 1421, configured to extract adjacent first and second grinding monitor key frames from the plurality of grinding monitor key frames; a convolutional encoding subunit 1422, configured to pass the first and second polishing monitoring key frames through a first and second convolutional layers of the second convolutional neural network model, respectively, to obtain a first convolutional feature map and a second convolutional feature map; a time attention subunit 1423 for calculating a point-by-point multiplication between the first convolution feature map and the second convolution feature map to obtain a time attention map; an attention generation subunit 1424 for inputting the time attention attempt to a Softmax activation function to obtain a time attention profile; an attention applying subunit 1425 configured to calculate a point-by-point multiplication between the second convolution feature map and the time attention feature map to obtain a time enhancement feature map corresponding to the second grinding monitor key frame; and the time enhancement feature map corresponding to the last grinding monitoring key frame in the plurality of grinding monitoring key frames is the grinding state time sequence change feature map.
In the above-mentioned deep learning-based polishing intelligent control system 100, the responsiveness estimation module 150 is configured to calculate a responsiveness estimation of the polishing state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix. It should be appreciated that the acoustic feature vector and the grinding status monitoring feature vector both provide important information about the grinding process. The sound feature vector reflects the frequency features of the sound signal during grinding, and the grinding state monitoring feature vector reflects the visual features during grinding. By calculating the response estimation of the grinding state monitoring feature vector relative to the sound feature vector, an index for measuring the association degree between the two features can be obtained, so that the grinding state monitoring feature vector and the sound feature vector are combined, visual and sound information in the grinding process is comprehensively utilized, and more comprehensive and accurate feature description is provided for grinding speed control in the grinding process.
Accordingly, in one specific example, the responsiveness estimation module 150 is configured to: calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain the classification feature matrix according to the following response formula;
Wherein, the response formula is:
wherein V is 1 Representing the grinding state monitoring feature vector, V 2 The sound feature vector is represented by a vector of the sound,representing the matrix multiplied by the vector, M representing the classification feature matrix.
In the above-mentioned deep learning-based polishing intelligent control system 100, the fusion module 160 is configured to fuse the polishing status monitoring feature vector and the sound feature vector to obtain a fused feature vector. Further, different features are extracted from the sound signal and the video image, respectively, taking into account the sound feature vector and the grinding state monitoring feature vector. The sound feature vector may contain information related to the sound spectral characteristics generated during the grinding process, and the grinding status monitoring feature vector may contain visual features related to the grinding status in the video image. By fusing these two vectors, the sound and image information can be combined, providing a more comprehensive representation of the features. Meanwhile, sound and image are different types of sensor data in the grinding process. The information they capture may be complementary in some way. For example, sound may be more sensitive to some subtle grinding state changes, while images may more particularly show the position and movement of the grinding tool. By fusing the information of the two sources, the limitation of each other can be overcome, and the understanding and judging ability of the grinding state can be improved.
In particular, in the technical solution of the present application, the grinding state monitoring feature vector and the sound feature vector respectively represent different high-dimensional feature manifolds in a high-dimensional feature space, but in a class probability tag domain, the grinding state monitoring feature vector and the sound feature vector respectively point to the same class probability tag, so that there is an implicit association between the grinding state monitoring feature vector and the high-dimensional feature manifold of the sound feature vector in a manifold expression level, that is, in the technical solution of the present application, the grinding state monitoring feature vector and the high-dimensional data manifold of the sound feature vector have smoothness and robustness in the manifold expression level.
Based on this, in the technical solution of the present application, first, a per-position mean value vector between the grinding state monitoring feature vector and the sound feature vector is calculated to obtain a pivot feature vector, where the pivot feature vector is used to represent a central high-dimensional feature manifold between the grinding state monitoring feature vector and the sound feature vector. Further, the geometrical distribution layer differences and consistencies between the grinding state monitoring feature vector and the sound feature vector and the high-dimensional feature manifold of the pivot feature vector in the high-dimensional feature space are represented by KL divergences with the pivot feature vector as a pivot, and after the first KL divergences and the second KL divergences are obtained, the grinding state monitoring feature vector and the sound feature vector are fused with both as weights and with the following formula to obtain the fused feature vector; wherein, the formula is: v (V) i ＝αV 1 +βV 2 ，V 1 Representing the grinding state monitoring feature vector, V 2 Representing the sound feature vector, alpha and beta representing the first KL divergence value and the second KL divergence value, V i Representing the fused feature vector. Thus doing soThe high-dimensional feature manifold of the fusion feature vector has collinearity with the grinding state monitoring feature vector and the sound feature vector at a geometric level, but the manifold model has difference with manifold measurement, and manifold transformation consistency exists at algebraic angles, so that the fusion feature vector can perform feature fusion by utilizing high-dimensional implicit association between the grinding state monitoring feature vector and the sound feature vector to improve smoothness and robustness of the fusion feature vector
Specifically, the fusion module 160 includes: the per-position average value calculation unit is used for calculating a per-position average value vector between the grinding state monitoring feature vector and the sound feature vector to obtain a pivot feature vector; a first KL divergence value calculation unit configured to calculate a first KL divergence value between the grinding state monitoring feature vector and the pivot feature vector; a second KL-divergence value calculation unit configured to calculate a second KL-divergence value between the sound feature vector and the pivot feature vector; the weighted fusion unit is used for taking the first KL divergence value and the second KL divergence value as weights and fusing the grinding state monitoring feature vector and the sound feature vector by the following fusion formula to obtain the fused feature vector; wherein, the fusion formula is: v (V) i ＝αV 1 +βV 2 ，V 1 Representing the grinding state monitoring feature vector, V 2 Representing the sound feature vector, alpha representing the first KL divergence value, beta representing the second KL divergence value, V i Representing the fused feature vector.
In the above-mentioned deep learning-based grinding intelligent control system 100, the optimization module 170 is configured to multiply the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculate a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix. It should be appreciated that the correlation between features may be increased by computing the fused feature vector multiplied by its transpose. Such an operation may help highlight the inherent relationships between features, better capturing the dependencies between features. By increasing the correlation between features, the discrimination capability and classification performance of the features can be improved. After the fusion feature matrix is obtained, calculating a matrix product between the fusion feature matrix and the classification feature matrix to map the classification feature matrix into a high-dimensional feature space where the fusion feature matrix is located, so that the high-dimensional implicit association between the grinding state monitoring feature vector and the sound feature vector is further utilized, and the classification judgment accuracy of the classification feature matrix through a classifier is improved.
In the above-mentioned deep learning-based polishing intelligent control system 100, the control result generating module 180 is configured to pass the optimized classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the rotation speed value of the polishing apparatus at the current time point should be increased or decreased. The classifier is a trained machine learning model that can determine, based on the input characteristics, whether the grinding apparatus should increase or decrease the rotational speed at the current point in time. Therefore, the rotating speed of the grinding equipment is adaptively adjusted based on the classification result, so that the intelligent control of the grinding speed in the grinding process is realized, the particle size distribution and the surface quality of a grinding product are ensured, and the grinding efficiency and the product quality are improved.
In summary, the intelligent polishing control system based on deep learning according to the embodiments of the present application is illustrated, which adopts a machine learning technique to extract real-time change characteristics of granularity and polishing sound time sequence characteristics of an object to be polished from a polishing monitoring video of the object to be polished and a sound signal generated by a polishing device, and performs real-time control of a rotation speed of the polishing device based on a fusion characteristic of the real-time change characteristics and the polishing sound time sequence characteristics. Therefore, the automatic monitoring of the grinding process and the accurate control of the grinding speed are realized, and the granularity distribution and the surface quality of the grinding product are ensured, so that the yield of the product is improved.
Fig. 5 is a flowchart of a method for intelligent control of polishing based on deep learning according to an embodiment of the present application. As shown in fig. 5, the intelligent control method for polishing based on deep learning according to the embodiment of the application comprises the following steps: s110, acquiring a grinding monitoring video of an object to be ground in a preset time period and a sound signal generated by grinding equipment in the preset time period; s120, carrying out S conversion on the sound signal to obtain an S conversion time-frequency diagram; s130, enabling the S-transformed time-frequency diagram to pass through a first convolution neural network model serving as a filter to obtain a sound feature vector; s140, the grinding monitoring video is processed through a second convolution neural network model using a time attention mechanism to obtain a grinding state monitoring feature vector; s150, calculating the response estimation of the grinding state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix; s160, fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fused feature vector; s170, multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix; and S180, the optimized classification characteristic matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the grinding equipment at the current time point is increased or reduced.
Here, it will be understood by those skilled in the art that the specific operations of the respective steps in the above-described deep learning-based grinding intelligent control method have been described in detail in the above description of the deep learning-based grinding intelligent control system with reference to fig. 1 to 4, and thus, repetitive descriptions thereof will be omitted.
The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not intended to be limiting, and these advantages, benefits, effects, etc. are not to be considered as essential to the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not necessarily limited to practice with the above described specific details.
The block diagrams of the devices, apparatuses, devices, systems referred to in the present application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.
It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent aspects of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.
Claims (10)
1. Grinding intelligent control system based on degree of depth study, characterized by comprising:
the data acquisition module is used for acquiring a grinding monitoring video of an object to be ground in a preset time period and a sound signal generated by grinding equipment in the preset time period;
The S conversion module is used for carrying out S conversion on the sound signal to obtain an S conversion time-frequency diagram;
the sound feature extraction module is used for enabling the S-transformation time-frequency diagram to pass through a first convolution neural network model serving as a filter so as to obtain a sound feature vector;
the grinding state feature extraction module is used for enabling the grinding monitoring video to obtain a grinding state monitoring feature vector through a second convolution neural network model using a time attention mechanism;
the responsiveness estimation module is used for calculating responsiveness estimation of the grinding state monitoring feature vector relative to the sound feature vector so as to obtain a classification feature matrix;
the fusion module is used for fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fusion feature vector;
the optimization module is used for multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix;
and the control result generation module is used for passing the optimized classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the rotating speed value of the grinding equipment at the current time point is increased or reduced.
2. The deep learning based lapping intelligent control system of claim 1, wherein the S-transformation module is configured to: s-transforming the sound signal by using the following transformation formula to obtain the S-transformed time-frequency diagram;
wherein, the transformation formula is:
wherein S (f, τ) represents the S-transformed time-frequency diagram, τ is a time shift factor, x (t) represents the sound signal, f represents frequency, and t represents time.
3. The deep learning based lapping intelligent control system of claim 2, wherein the acoustic feature extraction module is configured to: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer:
performing convolution processing on the input data based on a two-dimensional convolution check to generate a convolution feature map;
performing global average pooling processing on each feature matrix along the channel dimension on the convolution feature map to generate a pooled feature map;
non-linear activation is carried out on the characteristic values of all the positions in the pooled characteristic map so as to generate an activated characteristic map;
the output of the last layer of the first convolutional neural network model is the sound feature vector, the input from the second layer to the last layer of the first convolutional neural network model is the output of the last layer, and the input of the first layer of the first convolutional neural network model is the S-transformation time-frequency diagram.
4. The deep learning based lapping intelligent control system of claim 3, wherein the lapping status feature extraction module comprises:
the sampling unit is used for extracting a plurality of grinding monitoring key frames from the grinding monitoring video at a preset sampling frequency;
the time sequence coding unit is used for enabling the plurality of grinding monitoring key frames to pass through the second convolution neural network model using the time attention mechanism so as to obtain a grinding state time sequence change characteristic diagram;
and the pooling unit is used for carrying out global pooling processing on each characteristic matrix along the channel dimension on the grinding state time sequence change characteristic diagram so as to obtain the grinding state monitoring characteristic vector.
5. The intelligent control system for deep learning based polishing of claim 4, wherein the timing encoding unit comprises:
an adjacent key frame extraction subunit, configured to extract adjacent first and second grinding monitor key frames from the plurality of grinding monitor key frames;
the convolution coding subunit is used for enabling the first grinding monitoring key frame and the second grinding monitoring key frame to respectively pass through a first convolution layer and a second convolution layer of the second convolution neural network model so as to obtain a first convolution characteristic diagram and a second convolution characteristic diagram;
A time attention subunit, configured to calculate a point-by-point multiplication between the first convolution feature map and the second convolution feature map to obtain a time attention map;
an attention generation subunit, configured to input the time attention attempt to a Softmax activation function to obtain a time attention profile;
an attention applying subunit, configured to calculate a multiplication by location point between the second convolution feature map and the time attention feature map to obtain a time enhancement feature map corresponding to the second grinding monitoring key frame;
and the time enhancement feature map corresponding to the last grinding monitoring key frame in the plurality of grinding monitoring key frames is the grinding state time sequence change feature map.
6. The deep learning based lapping intelligent control system of claim 5, wherein the responsiveness estimation module is configured to: calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain the classification feature matrix according to the following response formula;
wherein, the response formula is:
wherein V is 1 Representing the grinding state monitoring feature vector, V 2 The sound feature vector is represented by a vector of the sound, Representing the matrix multiplied by the vector, M representing the classification feature matrix.
7. The deep learning based lapping intelligent control system of claim 6, wherein the fusion module comprises:
the per-position average value calculation unit is used for calculating a per-position average value vector between the grinding state monitoring feature vector and the sound feature vector to obtain a pivot feature vector;
a first KL divergence value calculation unit configured to calculate a first KL divergence value between the grinding state monitoring feature vector and the pivot feature vector;
a second KL-divergence value calculation unit configured to calculate a second KL-divergence value between the sound feature vector and the pivot feature vector;
the weighted fusion unit is used for taking the first KL divergence value and the second KL divergence value as weights and fusing the grinding state monitoring feature vector and the sound feature vector by the following fusion formula to obtain the fused feature vector; wherein, the fusion formula is: v (V) i ＝αV 1 +βV 2 ，V 1 Representing the grinding state monitoring feature vector, V 2 Representing the sound feature vector, alpha representing the first KL divergence value, beta representing the second KL divergence value, V i Representing the fused feature vector.
8. The intelligent grinding control method based on deep learning is characterized by comprising the following steps of:
acquiring a grinding monitoring video of an object to be ground in a preset time period and a sound signal generated by grinding equipment in the preset time period;
s-transforming the sound signal to obtain an S-transformed time-frequency diagram;
the S transformation time-frequency diagram is passed through a first convolution neural network model serving as a filter to obtain a sound feature vector;
the grinding monitoring video is processed through a second convolution neural network model using a time attention mechanism to obtain a grinding state monitoring feature vector;
calculating a responsiveness estimate of the grinding state monitoring feature vector relative to the sound feature vector to obtain a classification feature matrix;
fusing the grinding state monitoring feature vector and the sound feature vector to obtain a fused feature vector;
multiplying the fusion feature vector by a transpose of the fusion feature vector to obtain a fusion feature matrix, and calculating a matrix product between the fusion feature matrix and the classification feature matrix to obtain an optimized classification feature matrix;
and the optimized classification characteristic matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the grinding equipment at the current time point is increased or decreased.
9. The intelligent control method for deep learning based grinding of claim 8, wherein S-transforming the sound signal to obtain an S-transformed time-frequency diagram comprises: s-transforming the sound signal by using the following transformation formula to obtain the S-transformed time-frequency diagram;
wherein, the transformation formula is:
wherein S (f, τ) represents the S-transformed time-frequency diagram, τ is a time shift factor, x (t) represents the sound signal, f represents frequency, and t represents time.
10. The intelligent control method for deep learning based grinding of claim 9, wherein passing the S-transformed time-frequency graph through a first convolutional neural network model as a filter to obtain a sound feature vector, comprises: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer:
performing convolution processing on the input data based on a two-dimensional convolution check to generate a convolution feature map;
performing global average pooling processing on each feature matrix along the channel dimension on the convolution feature map to generate a pooled feature map;
non-linear activation is carried out on the characteristic values of all the positions in the pooled characteristic map so as to generate an activated characteristic map;
The output of the last layer of the first convolutional neural network model is the sound feature vector, the input from the second layer to the last layer of the first convolutional neural network model is the output of the last layer, and the input of the first layer of the first convolutional neural network model is the S-transformation time-frequency diagram.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311419112.6A CN117206063A (en) | 2023-10-30 | 2023-10-30 | Grinding intelligent control system and method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311419112.6A CN117206063A (en) | 2023-10-30 | 2023-10-30 | Grinding intelligent control system and method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117206063A true CN117206063A (en) | 2023-12-12 |
Family
ID=89046481
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311419112.6A Pending CN117206063A (en) | 2023-10-30 | 2023-10-30 | Grinding intelligent control system and method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117206063A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117618708A (en) * | 2024-01-26 | 2024-03-01 | 吉林大学 | Intelligent monitoring system and method for intravenous infusion treatment |
-
2023
- 2023-10-30 CN CN202311419112.6A patent/CN117206063A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117618708A (en) * | 2024-01-26 | 2024-03-01 | 吉林大学 | Intelligent monitoring system and method for intravenous infusion treatment |
| CN117618708B (en) * | 2024-01-26 | 2024-04-05 | 吉林大学 | Intelligent monitoring system and method for intravenous infusion treatment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117206063A (en) | Grinding intelligent control system and method based on deep learning | |
| TWI761787B (en) | Method and system for predicting dynamical flows from control inputs and limited observations | |
| CN116048031B (en) | Control system and method for petroleum auxiliary production | |
| CN116036639B (en) | Control system and method for ultrasonic wave induced hexafluorophosphate crystallization | |
| CN115647819A (en) | Turning and grinding integrated compound machine and control method thereof | |
| CN114871486B (en) | Planer and processing control method thereof | |
| CN112418387A (en) | Quantum data processing method and apparatus | |
| Pan et al. | Cognitive acoustic analytics service for Internet of Things | |
| Meng et al. | Bearing Fault Diagnosis under Multi-sensor Fusion Based on Modal Analysis and Graph Attention Network | |
| Xue et al. | A novel framework for motor bearing fault diagnosis based on multi-transformation domain and multi-source data | |
| KR102289212B1 (en) | Fault diagnosis apparatus and method based on artificial intelligence technology | |
| Choi et al. | Appearance-based gaze estimation using kinect | |
| Hyndman et al. | Higher-order Autoregressive Models for Dynamic Textures. | |
| CN108965885B (en) | Video online reconstruction and moving target detection method based on frame compression measurement | |
| CN117404853B (en) | External circulating water cooling system and method for tunnel boring machine | |
| Zhu et al. | Image scaling factor estimation based on normalized energy density and learning to rank | |
| CN117454301B (en) | Method and system for displaying change of PV value tracked by nuclear fusion test | |
| WO2015161490A1 (en) | Target motion detection method for water surface polarization imaging based on compound eyes simulation | |
| CN117636908B (en) | Digital mine production management and control system | |
| CN117649597B (en) | Underwater three-dimensional hand gesture estimation method and system based on event camera | |
| US20230222322A1 (en) | Vibration data analysis with functional neural network for predictive maintenance | |
| CN117067004B (en) | Flat grinder and control method thereof | |
| CN115741475A (en) | Hardware polishing device and polishing method thereof | |
| Kuhlmann et al. | Integrating Feedforward Design into a Generative Network to Synthesize Supplementary Training Data for Object Classification | |
| CN117351385A (en) | Welding method and system for file cabinet production |
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 |