CN110705650A - Metal plate layout method based on deep learning - Google Patents
Metal plate layout method based on deep learning Download PDFInfo
- Publication number
- CN110705650A CN110705650A CN201910972660.9A CN201910972660A CN110705650A CN 110705650 A CN110705650 A CN 110705650A CN 201910972660 A CN201910972660 A CN 201910972660A CN 110705650 A CN110705650 A CN 110705650A
- Authority
- CN
- China
- Prior art keywords
- sheet metal
- sample
- layout
- substrate
- metal part
- 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.)
- Granted
Links
- 239000002184 metal Substances 0.000 title claims abstract description 235
- 238000000034 method Methods 0.000 title claims abstract description 26
- 238000013135 deep learning Methods 0.000 title claims abstract description 21
- 230000000306 recurrent effect Effects 0.000 claims abstract description 37
- 238000013528 artificial neural network Methods 0.000 claims abstract description 29
- 238000010586 diagram Methods 0.000 claims abstract description 24
- 238000012549 training Methods 0.000 claims abstract description 17
- 238000013507 mapping Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000000758 substrate Substances 0.000 claims description 76
- 239000013598 vector Substances 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 20
- 238000003062 neural network model Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000002372 labelling Methods 0.000 claims description 3
- 239000000463 material Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
- G06Q10/043—Optimisation of two dimensional placement, e.g. cutting of clothes or wood
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Marketing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Tourism & Hospitality (AREA)
- Entrepreneurship & Innovation (AREA)
- Manufacturing & Machinery (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a deep learning-based metal plate layout method, which comprises the following steps: A. preprocessing an existing sheet metal layout; B. creating a sheet metal part dictionary; C. setting the end and empty mark of the sample; D. establishing numerical value mapping of sheet metal parts and position information in a sheet metal layout; E. determining the structure of a recurrent neural network; F. constructing a sheet metal layout diagram sample and a sample set; G. training the obtained sample data; H. predicting according to the given sheet metal part; I. the method can realize the automatic layout of the sheet metal by learning the sheet metal flattening graph, and even can achieve the optimal layout scheme along with continuous learning.
Description
Technical Field
The invention relates to the technical field of metal plate layout, in particular to a deep learning-based metal plate layout method.
Background
In the current background of large-scale customization, how to quickly respond to design requirements and timely complete order production tasks becomes an important problem in the manufacturing industry. The sheet metal part is used as an important part of a manufacturing enterprise, and how to effectively improve the sheet metal layout efficiency plays a significant role in the design and manufacturing aspects of industrial enterprises.
At present, in manufacturing enterprises, the layout mode of the flattening graph of the sheet metal part is mostly realized manually or by an algorithm based on certain rules.
In the traditional manual method, a large amount of labor is required to obtain a layout with a good substrate utilization rate; if the concept of the previous layout needs to be used for the subsequent layout, much time and energy are consumed, and the final effect is not ideal.
For an algorithm based on some rules, although automation of sheet metal layout can be achieved, manual participation is often required to further improve the utilization rate of the substrate, and due to the fact that the algorithm cannot be updated effectively, performance, efficiency and accuracy cannot be improved no matter how long the algorithm is adopted, and therefore the manual participation needs to be performed all the time.
Disclosure of Invention
The invention aims to provide a deep learning-based sheet metal layout method to solve the problems in the background art.
In order to achieve the purpose, the invention provides the following technical scheme:
a sheet metal layout method based on deep learning comprises the following steps:
A. preprocessing an existing sheet metal layout;
B. creating a sheet metal part dictionary;
C. setting the end and empty mark of the sample;
D. establishing numerical value mapping of sheet metal parts and position information in a sheet metal layout;
E. determining the structure of a recurrent neural network;
F. constructing a sheet metal layout diagram sample and a sample set;
G. training the obtained sample data;
H. predicting according to the given sheet metal part;
I. self-optimization and self-learning of the model.
As a further scheme of the invention: the step A specifically comprises the following steps: I) labeling the base plate and the sheet metal part in the layout by using the material code of the sheet metal part; II) digitizing the position information of the sheet metal part, and carrying out coordinate processing on the position of the sheet metal part in the layout. Establishing a global absolute coordinate system XOY by taking a certain point on the substrate as a coordinate origin (0, 0); marking the coordinate position of each sheet metal part by taking the upper left corner of the sheet metal part in the layout as a base point, namely (X)n,Yn) (ii) a III) digitizing the substrate utilization rate. To optimize the sheet metal layout using substrate utilization and better perform sample training, the substrate utilization is numerically processed here, denoted as (X,? "means that it can be any value; IV) performing matrixing on the sheet metal layout, and performing the above treatment on the sheet metal layout to obtain the matrixed sheet metal layout.
As a further scheme of the invention: the step B is specifically as follows: according to the coding of sheet metal component, distinguish different sheet metal components to clustering is carried out to each different sheet metal component in the sample collection (constitute by all sheet metal layout maps), and put into the sheet metal component dictionary with it, the number of spare part is N in the sheet metal component dictionary.
As a further scheme of the invention: the step C is specifically defined in the following places: I) since the input of each sample is composed of a limited number of elements, after the last option is predicted, an end indication (EOP) needs to be given to represent the end of the sheet metal layout prediction; II) setting samples of all the sheet metal layout graphs to be the same length; III) filling samples with the length smaller than the set sample length by setting a null mark because the sample lengths in the sheet metal layout are not completely the same; IV) because the ending mark and the empty mark are set for the sample, the length of the sheet metal part dictionary is expanded to be N + 2.
As a further scheme of the invention: the step D is specifically as follows: according to the sheet metal part dictionary, creating a numerical value mapping-N-dimensional one-hot vector for each sheet metal part, wherein the dimensionality of the N-dimensional vector is determined by the size of the sheet metal part dictionary, namely NSheetMetal+ 2; because the position information of the sheet metal part is not identical in different sheet metal layout diagrams and the position information of the sheet metal part is set in the sheet metal layout diagram, the N-dimensional one-hot vector of the sheet metal part is expanded into NAll＝NSheetMetal+2+NPositionVector of dimensions, where NPosition2 (two coordinate positions, X-axis and Y-axis).
As a further scheme of the invention: in the step E, the number of neurons in the output layer of the recurrent neural network is the same as that in the input layer, i.e., N +1+1+2 neurons, and the output layer of the recurrent neural network includes both the output of the classification problem, i.e., N +1+1 part classification, and the output of the regression problem, i.e., the X-axis coordinate position (or the substrate utilization rate) and the Y-axis coordinate position; and then defining a time step of the recurrent neural network, wherein the time step is determined by the length of the longest training sample in the training sample set, and finally defining a recurrent neural network unit.
As a further scheme of the invention: the step F is specifically as follows: acquiring a target sheet metal layout diagram set, and converting all sample product information to be trained into respective sample matrixes, wherein the number of lines of the sample matrixes is the number of different product information, and the columns are the dimensionality N of vectorsAll＝NSheetMetal+2+NPositionThe matrix is a sample of a recurrent neural network, all sheet metal layout samples are subjected to matrixing processing, a sample matrix with the maximum line number is obtained, and the line number is Nmax(row)Adding a trailing flag, N, to the sample matrixmax(row)＝Nmax(row)+1, expanding the number of rows of all other sheet metal layout sample matrixes to N by adding ending marks and zero padding marksmax(row)Finally, a sample set of three-dimensional tensors (dimension N) is constructedsample×Nmax(row)×NAll，NsampleSample for representing sheet metal layoutThe number of the cells; n is a radical ofmax(row)Representing the maximum row number of a sample matrix of the sample set; n is a radical ofAllRepresenting the number of sheet metal parts in the sheet metal part dictionary + the sheet metal part coordinate dimension (x, y)).
As a further scheme of the invention: the step G is specifically as follows: putting the three-dimensional tensor sample set into a defined recurrent neural network, and selecting a proper activation function; because the output of the time step is an expanded one-hot vector, the output layer at each time step does not use a Softmax activation function any more, the corresponding weight matrix is obtained through back propagation calculation and MSE, and finally the network structure which is in line with the expectation is obtained.
As a further scheme of the invention: the step H is specifically as follows: firstly, determining a substrate; for a given m sheet metal parts to be typeset, a substrate for sheet metal layout needs to be determined. Finding all substrates in a dictionary of sheet metal parts, wherein the number of the substrates is recorded as mSubstrateCarrying out numerical mapping and vectorization on the vector, setting the utilization rate of the substrate to be 100 percent, and obtaining the vector of the substrate at the moment as [ one-hot ]Substrate,1，？]Wherein, one-hotSubstrateOne-hot vector of the substrate; 1, representing the target utilization rate of the substrate (100%); is there a The method has no practical significance, the method is put into the first time step of the trained recurrent neural network, the probability values and the coordinate values of all sheet metal parts at the moment can be predicted in the first time step respectively, the sheet metal part with the highest probability corresponding to each substrate at the moment is selected, whether the screened sheet metal parts are in m sheet metal parts to be typeset or not is judged, and the substrates corresponding to the absent sheet metal parts are excluded; finally, finding the sheet metal part with the highest probability and the corresponding substrate in the existing sheet metal parts, and finally determining the substrate; and inputting the substrate and the target utilization rate thereof as the first time step of the network again, and putting the trained recurrent neural network model into the network. In the first time step output probability of the model, searching the sheet metal part with the maximum probability value in the m sheet metal parts to be typeset, and taking the sheet metal part as the first part to be subsequently laid out in the substrate P0001; then, the predicted first sheet metal part and the coordinate value thereof are used as the input of a second time step, and the model meter is used for calculating the coordinate value of the first sheet metal partCalculating, in the output probability, searching the sheet metal part with the maximum probability value in the m-1 sheet metal parts to be typeset as a second part in the layout of the substrate P0001; and repeating the time steps until the end mark EOP in the prediction output sheet metal part dictionary or all m sheet metal parts are predicted, stopping prediction by the recurrent neural network model, and sequentially laying the predicted sheet metal parts in the substrate P0001 according to the coordinate positions of the predicted sheet metal parts to form a sheet metal layout.
As a further scheme of the invention: and (D) the user can adjust the sheet metal layout diagram finished in the step (H) to achieve the final expectation, and then the current scheme is used as a new sample to be put into a network together with other samples for training, so that the model can be further optimized according to the adjustment of the user, and the user is better guided to carry out the next sheet metal layout.
Compared with the prior art, the invention has the beneficial effects that: the method can realize the automatic layout of the sheet metal by the layout of the sheet metal flattening drawing by a learning engineer, and even can reach the optimal layout scheme along with continuous learning.
Drawings
FIG. 1 is a schematic diagram of a sheet metal layout diagram pretreatment.
Fig. 2 is a schematic diagram of numerical mapping of a sheet metal part and position information.
FIG. 3 is a schematic diagram of model training of a sheet metal layout diagram set based on deep learning.
Fig. 4 is a process diagram of generating a sheet metal layout diagram based on deep learning.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1-4, example 1: in the embodiment of the invention, a deep learning-based sheet metal layout method comprises the following steps:
A. preprocessing an existing sheet metal layout;
I) and labeling the base plate and the sheet metal part in the layout by using the material code of the sheet metal part.
II) digitizing the position information of the sheet metal part. And carrying out coordinate processing on the position of the sheet metal part in the layout. Taking a certain point on the substrate as a coordinate origin (0,0) to establish a global absolute coordinate system XOY; marking the coordinate position of each sheet metal part by taking the upper left corner of the sheet metal part in the layout as a base point, namely (X)n,Yn)。
III) digitizing the substrate utilization rate. To optimize sheet metal layout using substrate utilization and to better perform sample training. The substrate utilization rate is numerically represented as (X? "means that it may be any value.
IV) matrixing the sheet metal layout. After the sheet metal layout is processed, a matrix sheet metal layout is obtained, in the patent, matrix arrangement is performed along the X axis and then along the Y axis (or the matrix arrangement can be performed according to the Y axis and then the X axis, and only all samples need to be arranged in the same direction), as shown in fig. 1.
B. Creating a sheet metal part dictionary;
according to the coding of sheet metal component, distinguish different sheet metal components to clustering is carried out to each different sheet metal component in the sample collection (constitute by all sheet metal layout maps), and put into the sheet metal component dictionary with it, the number of spare part is N in the sheet metal component dictionary.
C. The end of sample and empty flag are set.
I) Since the input of each sample is composed of a limited number of elements, after the last option is predicted, an end indication (EOP) needs to be given to represent the end of the sheet metal layout prediction;
II) in the method, the lengths of the samples of all the sheet metal layout diagrams are set to be the same and cannot be smaller than the length of the longest sample;
III) because the sample lengths in the sheet metal layout are not completely the same, filling processing is carried out on the samples with the lengths smaller than the set sample length by setting the empty marks, and the filling processing is realized by setting the empty marks.
IV) because the ending mark and the empty mark are set for the sample, the length of the sheet metal part dictionary is expanded to be N + 2.
D. And creating numerical value mapping of the sheet metal parts and the position information in the sheet metal layout.
And according to the sheet metal part dictionary, creating a numerical value mapping-an N-dimensional one-hot vector for each sheet metal part. Wherein the dimension of the N-dimensional vector is determined by the size of the sheet metal part dictionary (i.e., N in this patent)SheetMetal+2)；
In the patent, the position information of the sheet metal part is not identical in different sheet metal layout diagrams, and the position information of the sheet metal part needs to be set in the sheet metal layout diagram, so that the N-dimensional one-hot vector of the sheet metal part needs to be expanded into NAll＝NSheetMetal+2+NPositionVector of dimensions, where NPosition＝2(NPositionTwo coordinate positions, X-axis and Y-axis, are 2).
The numerical value mapping of the sheet metal part and the position information in the sheet metal layout diagram is as shown in figure 2;
E. determining the structure of a recurrent neural network
The number of the neurons of the output layer of the recurrent neural network is the same as that of the neurons of the input layer, namely N +1+1+ 2. Here, unlike the output of a normal recurrent neural network, a normal recurrent neural network is often used to solve the classification problem, i.e., the probability that a prediction is of a certain class. In the present invention, the output layer of the recurrent neural network includes both the output of the classification problem, i.e., N +1+1 part classifications, and the output of the regression problem, i.e., the X-axis coordinate position (or substrate utilization rate) and the Y-axis coordinate position.
A time step of the recurrent neural network is defined. The time step is determined based on the length of the longest training sample in the set of training samples.
Finally, the recurrent neural network unit is defined to be the same as the common recurrent neural network (such as BasicRNNCell, BasicLSTMCell, LSTMCell, GRUCell and the like).
F. Sheet metal layout sample and sample set construction
Acquiring a target sheet metal layout diagram set, and converting all sample product information to be trained into respective sample matrixes (the number of lines of the sample matrixes is the number of different product information, and the columns are the dimension N of vectorsAll＝NSheetMetal+2+NPosition) The matrix is a sample of the recurrent neural network;
after all the sheet metal layout samples are subjected to matrixing treatment, the number of rows of each sample matrix is different, and a sample matrix with the maximum number of rows (the number of rows is N) is obtainedmax(row)) Add a trailing flag (i.e., N) to the sample matrixmax(row)＝Nmax(row)+1)；
Then expanding the line number of the sample matrix of all other sheet metal layouts to the line number (N) by adding ending marks and zero padding marksRow+1)；
Finally, a sample set of three-dimensional tensors is constructed, namely the dimensionality of the three-dimensional tensor is Nsample×Nmax(row)×NAll
NSample: the number of the sheet metal layout pattern books;
Nmax(row): the maximum row number of the sample matrix of the sample set;
NAll: number of sheet Metal parts in sheet Metal dictionary + dimension of sheet Metal coordinates (x, y)
G. Training the obtained sample data.
The three-dimensional tensor sample set is put into a well-defined recurrent neural network, a proper activation function (such as Tanh, Sigmoid, ReLu and the like), namely, because the output of a time step is an expanded one-hot vector (including x and y position information), the Softmax activation functions of different recurrent neural networks are not used in the output layer of each time step, the corresponding weight matrix is obtained through back propagation calculation and MSE (mean square error loss function), and finally the network structure which is in line with expectation is obtained.
The training process of the sheet metal layout based on deep learning is shown in fig. 3;
H. and predicting according to the given sheet metal part.
I) Determining a substrate; for a given m sheet metal parts to be typeset, a substrate for sheet metal layout needs to be determined. Finding all substrates in a dictionary of sheet metal parts, wherein the number of the substrates is recorded as mSubstrateCarrying out numerical mapping and vectorization on the vector, setting the utilization rate of the substrate to be 100 percent, and obtaining the vector of the substrate at the moment as [ one-hot ]Substrate,1，？]
one-hotSubstrateOne-hot vector of the substrate;
1, representing the target utilization rate of the substrate (100%);
is there a It is meant to be of no practical significance here.
Putting the metal plate parts into a trained Recurrent Neural Network (RNN) in a first time step, predicting probability values and coordinate values of all metal plate parts in the first time step respectively, selecting the metal plate part with the highest probability corresponding to each substrate, judging whether the screened metal plate parts are in m metal plate parts to be typeset, and excluding the substrates corresponding to the absent metal plate parts; finally, the sheet metal part with the highest probability and the corresponding substrate are found from the existing sheet metal parts, and the substrate is finally determined (taking P0001 as an example).
And inputting the substrate and the target utilization rate thereof as the first time step of the network again, and putting the trained recurrent neural network model into the network. In the first time step output probability of the model, searching the sheet metal part with the maximum probability value in the m sheet metal parts to be typeset, and taking the sheet metal part as the first part to be subsequently laid out in the substrate P0001; and then, taking the predicted first sheet metal part and the coordinate value thereof as the input of a second time step, and searching the sheet metal part with the maximum probability value in m-1 sheet metal parts to be typeset as a second part in the layout in the substrate P0001 in the output probability through model calculation.
And repeating the time steps until the end mark EOP in the prediction output sheet metal part dictionary or all m sheet metal parts are predicted, stopping prediction by the recurrent neural network model, and sequentially laying the predicted sheet metal parts in the substrate P0001 according to the coordinate positions of the predicted sheet metal parts to form a sheet metal layout.
The prediction process of the sheet metal layout based on deep learning is shown in fig. 4.
I. Self-optimization and self-learning of the model.
Example 2: on the basis of embodiment 1, the user can adjust the sheet metal layout diagram completed in step H to achieve the final expectation. And then, the current scheme is used as a new sample and is put into a network together with other samples for training, so that the model can be further optimized according to the adjustment of the user, and the user is better guided to carry out the next sheet metal layout.
Because the increase of the data set is performed under the condition of user intervention, effective samples can be continuously increased, then the recurrent neural network model is continuously trained, the generalization capability of the recurrent neural network model is continuously improved, and finally the purpose of automatically performing the metal plate layout instead of an engineer can be achieved.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
Claims (10)
1. A sheet metal layout method based on deep learning is characterized by comprising the following steps:
A. preprocessing an existing sheet metal layout;
B. creating a sheet metal part dictionary;
C. setting the end and empty mark of the sample;
D. establishing numerical value mapping of sheet metal parts and position information in a sheet metal layout;
E. determining the structure of a recurrent neural network;
F. constructing a sheet metal layout diagram sample and a sample set;
G. training the obtained sample data;
H. predicting according to the given sheet metal part;
I. self-optimization and self-learning of the model.
2. The sheet metal layout method based on deep learning according to claim 1, wherein the step A specifically comprises the following steps: I) labeling the base plate and the sheet metal part in the layout by using the material code of the sheet metal part; II) digitizing the position information of the sheet metal part, carrying out coordinate processing on the position of the sheet metal part in the layout drawing, and taking a certain point on the substrate as a coordinate origin (0,0) so as to establish a global absolute coordinate system XOY; marking the coordinate position of each sheet metal part by taking the upper left corner of the sheet metal part in the layout as a base point, namely (X)n,Yn) (ii) a III) substrate utilization digitization, where substrate utilization is numerically processed, denoted as (X,? ) Where X denotes the utilization of the substrate, "? "means that it can be any value; IV) performing matrixing on the sheet metal layout, and performing the above treatment on the sheet metal layout to obtain the matrixed sheet metal layout.
3. The sheet metal layout method based on deep learning according to claim 1, wherein the step B specifically comprises: according to the coding of sheet metal component, distinguish different sheet metal components to concentrate each different sheet metal component to the sample and carry out cluster processing, put into the sheet metal component dictionary with it, the number of spare part is N in the sheet metal component dictionary.
4. The sheet metal layout method based on deep learning according to claim 1, wherein the step C is specifically defined in the following places: I) since the input of each sample is composed of a limited number of elements, after the last option is predicted, an end indication (EOP) needs to be given to represent the end of the sheet metal layout prediction; II) setting samples of all the sheet metal layout graphs to be the same length; III) filling samples with the length smaller than the set sample length by setting a null mark because the sample lengths in the sheet metal layout are not completely the same; IV) because the ending mark and the empty mark are set for the sample, the length of the sheet metal part dictionary is expanded to be N + 2.
5. The sheet metal layout method based on deep learning according to claim 4, wherein the step D specifically comprises: according to the sheet metal part dictionary, creating a numerical value mapping-N-dimensional one-hot vector for each sheet metal part, wherein the dimensionality of the N-dimensional vector is determined by the size of the sheet metal part dictionary, namely NSheetMetal+ 2; because the position information of the sheet metal part is not identical in different sheet metal layout diagrams and the position information of the sheet metal part is set in the sheet metal layout diagram, the N-dimensional one-hot vector of the sheet metal part is expanded into NAll＝NSheetMetal+2+NPositionVector of dimensions, where NPosition2 (two coordinate positions, X-axis and Y-axis).
6. The sheet metal layout method based on deep learning of claim 1, wherein in step E, the number of neurons in the output layer of the recurrent neural network is the same as that in the input layer, i.e., N +1+1+2, and the output layer of the recurrent neural network includes both the output of the classification problem, i.e., N +1+1 part classification, and the output of the regression problem, i.e., X-axis coordinate position/substrate utilization rate and Y-axis coordinate position; and then defining a time step of the recurrent neural network, wherein the time step is determined by the length of the longest training sample in the training sample set, and finally defining a recurrent neural network unit.
7. The sheet metal layout method based on deep learning according to claim 1, wherein the step F specifically comprises: acquiring a target sheet metal layout diagram set, and converting all sample product information to be trained into respective sample matrixes, wherein the number of lines of the sample matrixes is the number of different product information, and the columns are the dimensionality N of vectorsAll＝NSheetMetal+2+NPositionThe matrix is a sample of a recurrent neural network, all sheet metal layout samples are subjected to matrixing processing, a sample matrix with the maximum line number is obtained, and the line number is Nmax(row)Adding a trailing flag, N, to the sample matrixmax(row)＝Nmax(row)+1, expanding the number of rows of all other sheet metal layout sample matrixes to N by adding ending marks and zero padding marksmax(row)(ii) a Finally, a sample set of three-dimensional tensors is constructed, and the dimensionality is Nsample×Nmax(row)×NAll，NsampleRepresenting the number of samples of the sheet metal layout; n is a radical ofmax(row)Representing the maximum row number of a sample matrix of the sample set; n is a radical ofAllAnd (4) representing the number of the sheet metal parts in the sheet metal part dictionary plus the coordinate dimension (x, y) of the sheet metal parts.
8. The sheet metal layout method based on deep learning according to claim 1, wherein the step G specifically comprises: putting the three-dimensional tensor sample set into a defined recurrent neural network, and selecting a proper activation function; because the output of the time step is an expanded one-hot vector, the output layer at each time step does not use a Softmax activation function any more, the corresponding weight matrix is obtained through back propagation calculation and MSE, and finally the network structure which is in line with the expectation is obtained.
9. According to claim1, the deep learning-based sheet metal layout method is characterized in that the step H specifically comprises the following steps: firstly, determining a substrate; for given m sheet metal parts to be typeset, substrates for sheet metal layout need to be determined first, all the substrates are found in a sheet metal part dictionary, and the number of the substrates is recorded as mSubstrateCarrying out numerical mapping and vectorization on the vector, setting the utilization rate of the substrate to be 100 percent, and obtaining the vector of the substrate at the moment as [ one-hot ]Substrate,1，？]Wherein, one-hotSubstrateOne-hot vector of the substrate; 1, representing the target utilization rate of the substrate (100%); is there a It means no practical meaning; putting the metal plate parts into a first time step of a trained recurrent neural network, predicting probability values and coordinate values of all metal plate parts in the first time step respectively, selecting the metal plate part with the highest probability corresponding to each substrate, judging whether the screened metal plate parts are in m metal plate parts to be typeset, and excluding the substrates corresponding to the absent metal plate parts; finally, finding the sheet metal part with the highest probability and the corresponding substrate in the existing sheet metal parts, and finally determining the substrate; inputting the substrate and the target utilization rate of the substrate as the first time step of the network again, putting the substrate and the target utilization rate into the trained recurrent neural network model, searching the sheet metal part with the maximum probability value in the m sheet metal parts to be typeset in the first time step output probability of the model, and taking the sheet metal part as the first part to be subsequently laid out in the substrate P0001; then, the predicted first sheet metal part and the coordinate value of the first sheet metal part are used as input of a second time step, and in the output probability, the sheet metal part with the maximum probability value in m-1 sheet metal parts to be typeset is searched and used as a second part which is laid out in the substrate P0001; and repeating the time steps until the end mark EOP in the prediction output sheet metal part dictionary or all m sheet metal parts are predicted, stopping prediction by the recurrent neural network model, and sequentially laying the predicted sheet metal parts in the substrate P0001 according to the coordinate positions of the predicted sheet metal parts to form a sheet metal layout.
10. The sheet metal layout method based on deep learning of claim 1, wherein a user can adjust the sheet metal layout diagram in step H to a final expectation, and then put the current scheme as a new sample into a network together with other samples for training, so that the model can be further optimized according to the adjustment of the user, and better guide the user to perform the next sheet metal layout.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910972660.9A CN110705650B (en) | 2019-10-14 | 2019-10-14 | Sheet metal layout method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910972660.9A CN110705650B (en) | 2019-10-14 | 2019-10-14 | Sheet metal layout method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110705650A true CN110705650A (en) | 2020-01-17 |
| CN110705650B CN110705650B (en) | 2023-10-24 |
Family
ID=69200203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910972660.9A Active CN110705650B (en) | 2019-10-14 | 2019-10-14 | Sheet metal layout method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110705650B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112232423A (en) * | 2020-10-21 | 2021-01-15 | 深制科技(苏州)有限公司 | Part2Vec Part vectorization processing method based on deep learning |
| CN113705111A (en) * | 2021-09-22 | 2021-11-26 | 百安居信息技术(上海)有限公司 | Fitment furniture automatic layout method and system based on deep learning |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003178081A (en) * | 2001-12-04 | 2003-06-27 | Matsushita Electric Ind Co Ltd | Document classification and labeling method using layout graph matching |
| CN106840869A (en) * | 2016-12-15 | 2017-06-13 | 北京航空航天大学 | A kind of hole-edge crack diagnostic method for being based on fiber grating spectral image analysis under two kinds of cloth patch modes |
| CN206614331U (en) * | 2017-02-14 | 2017-11-07 | 铁王数控机床（苏州）有限公司 | A kind of lathe sliding door modular structure |
| CN108230121A (en) * | 2018-02-09 | 2018-06-29 | 艾凯克斯（嘉兴）信息科技有限公司 | A kind of product design method based on Recognition with Recurrent Neural Network |
| CN108280746A (en) * | 2018-02-09 | 2018-07-13 | 艾凯克斯（嘉兴）信息科技有限公司 | A kind of product design method based on bidirectional circulating neural network |
| CN108875651A (en) * | 2018-06-22 | 2018-11-23 | 深圳市易成自动驾驶技术有限公司 | Laying for goods appraisal procedure, device and computer readable storage medium |
| CN109102493A (en) * | 2018-07-03 | 2018-12-28 | 柳州市木子科技有限公司 | A kind of automobile metal plate work flaw detection system based on CNN and LR |
| WO2019085329A1 (en) * | 2017-11-02 | 2019-05-09 | 平安科技（深圳）有限公司 | Recurrent neural network-based personal character analysis method, device, and storage medium |
| CN110321846A (en) * | 2019-07-04 | 2019-10-11 | 上海融客软件科技有限公司 | 3D graphic processing method, device, processing method and electric terminal |
-
2019
- 2019-10-14 CN CN201910972660.9A patent/CN110705650B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003178081A (en) * | 2001-12-04 | 2003-06-27 | Matsushita Electric Ind Co Ltd | Document classification and labeling method using layout graph matching |
| CN106840869A (en) * | 2016-12-15 | 2017-06-13 | 北京航空航天大学 | A kind of hole-edge crack diagnostic method for being based on fiber grating spectral image analysis under two kinds of cloth patch modes |
| CN206614331U (en) * | 2017-02-14 | 2017-11-07 | 铁王数控机床（苏州）有限公司 | A kind of lathe sliding door modular structure |
| WO2019085329A1 (en) * | 2017-11-02 | 2019-05-09 | 平安科技（深圳）有限公司 | Recurrent neural network-based personal character analysis method, device, and storage medium |
| CN108230121A (en) * | 2018-02-09 | 2018-06-29 | 艾凯克斯（嘉兴）信息科技有限公司 | A kind of product design method based on Recognition with Recurrent Neural Network |
| CN108280746A (en) * | 2018-02-09 | 2018-07-13 | 艾凯克斯（嘉兴）信息科技有限公司 | A kind of product design method based on bidirectional circulating neural network |
| CN108875651A (en) * | 2018-06-22 | 2018-11-23 | 深圳市易成自动驾驶技术有限公司 | Laying for goods appraisal procedure, device and computer readable storage medium |
| CN109102493A (en) * | 2018-07-03 | 2018-12-28 | 柳州市木子科技有限公司 | A kind of automobile metal plate work flaw detection system based on CNN and LR |
| CN110321846A (en) * | 2019-07-04 | 2019-10-11 | 上海融客软件科技有限公司 | 3D graphic processing method, device, processing method and electric terminal |
Non-Patent Citations (2)
| Title |
|---|
| 李斌;: "钣金加工工艺中数控冲剪复合机床的应用" * |
| 王磊;张振东;: "汽车线束设计思路及零件选型" * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112232423A (en) * | 2020-10-21 | 2021-01-15 | 深制科技(苏州)有限公司 | Part2Vec Part vectorization processing method based on deep learning |
| CN112232423B (en) * | 2020-10-21 | 2024-05-28 | 深制科技(苏州)有限公司 | Part2Vec Part vectorization processing method based on deep learning |
| CN113705111A (en) * | 2021-09-22 | 2021-11-26 | 百安居信息技术(上海)有限公司 | Fitment furniture automatic layout method and system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110705650B (en) | 2023-10-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111199016B (en) | Daily load curve clustering method for improving K-means based on DTW | |
| CN113034026B (en) | Q-learning and GA-based multi-target flexible job shop scheduling self-learning method | |
| CN108694502B (en) | Self-adaptive scheduling method for robot manufacturing unit based on XGboost algorithm | |
| CN110488810B (en) | Optimal path planning method for welding robot based on improved particle swarm optimization | |
| CN107609694B (en) | Structure optimization method for offshore wind power cluster power transmission system and storage medium | |
| CN112947300A (en) | Virtual measuring method, system, medium and equipment for processing quality | |
| CN110705650B (en) | Sheet metal layout method based on deep learning | |
| CN112308298B (en) | Multi-scenario performance index prediction method and system for semiconductor production line | |
| CN105844334B (en) | A kind of temperature interpolation method based on radial base neural net | |
| CN115130749A (en) | NSGA-III and TOPSIS fused data-driven multi-objective optimization method | |
| CN111597943B (en) | Table structure identification method based on graph neural network | |
| CN112907150A (en) | Production scheduling method based on genetic algorithm | |
| Liu et al. | Illustration design model with clustering optimization genetic algorithm | |
| Mishra et al. | An improved hybrid flower pollination algorithm for assembly sequence optimization | |
| Zhang et al. | End‐to‐end generation of structural topology for complex architectural layouts with graph neural networks | |
| CN109116300A (en) | A kind of limit learning position method based on non-abundant finger print information | |
| CN110688722B (en) | Automatic generation method of part attribute matrix based on deep learning | |
| CN112257202A (en) | Neural network-based two-dimensional structure grid automatic decomposition method for multi-inner-hole part | |
| CN111161094A (en) | Electric power work order demand point identification method based on deep learning | |
| CN115908697A (en) | Generation model based on point cloud probability distribution learning and method thereof | |
| Kontolatis et al. | Optimisation of press-brake bending operations in 3D space | |
| CN115082726A (en) | Ceramic biscuit product classification method for toilet based on PointNet optimization | |
| CN114781013A (en) | Method for realizing avoidance and arrangement of labeled characters in design drawing | |
| CN107767035A (en) | A kind of electric energy meter detection mixed production line dispatching method based on genetic algorithm | |
| CN108153254B (en) | A kind of part based on glowworm swarm algorithm is clustered to process route optimization method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |