CN110488755A - A kind of conversion method of numerical control G code - Google Patents
A kind of conversion method of numerical control G code Download PDFInfo
- Publication number
- CN110488755A CN110488755A CN201910775600.8A CN201910775600A CN110488755A CN 110488755 A CN110488755 A CN 110488755A CN 201910775600 A CN201910775600 A CN 201910775600A CN 110488755 A CN110488755 A CN 110488755A
- Authority
- CN
- China
- Prior art keywords
- code
- control system
- row
- numerical control
- input
- 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
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/18—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form
- G05B19/4093—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by part programming, e.g. entry of geometrical information as taken from a technical drawing, combining this with machining and material information to obtain control information, named part programme, for the NC machine
- G05B19/40937—Numerical control [NC], i.e. automatically operating machines, in particular machine tools, e.g. in a manufacturing environment, so as to execute positioning, movement or co-ordinated operations by means of programme data in numerical form characterised by part programming, e.g. entry of geometrical information as taken from a technical drawing, combining this with machining and material information to obtain control information, named part programme, for the NC machine concerning programming of machining or material parameters, pocket machining
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32161—Object oriented control, programming
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Geometry (AREA)
- Human Computer Interaction (AREA)
- Manufacturing & Machinery (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of conversion methods of numerical control G code, steps are as follows: pre-processing first to the G code of input, then the G code type of current G code and generation target is determined, G code transfer framework is established based on neural network algorithm again, and G code transfer framework is trained, obtain G code transformation model, then G code is read in rows, G code is passed to G code transformation model, G code is translated, finally judge whether conversion is completed, it is unfinished then continue to convert.The characteristic for the different machine tool systems that method disclosed by the invention can adapt to very well has many advantages, such as that scalability is strong, G code conversion quality is high.
Description
Technical field
The present invention relates to a kind of conversion methods of numerical control G code.
Background technique
For numerical control industry, numerical control machine code relation the processing benefit of numerical control, quality the problems such as, with the epoch
Progress, different lathes is also required to be updated iteration, such as in the change-brain engineering that country advocates, some old numerically-controlled machine tools
System will change new digital control system into, but since the digital control system before most change-brain is different from the system after change-brain, cause
Some optimizations, cure parameter G code used in numerically-controlled machine tool processing before, can not new numerically-controlled machine tool after change-brain
Enterprising enforcement is used.Due to of the remote past, some even can not generate numerical control machining code by Geometric Modeling.Although big at present
Most lathes all use the universal standard, even if being the universal standard, still have some differences slightly on programming details, are
Understand never be badly in need of with code sharing problem between numerically-controlled machine tool it is a kind of can intelligently, the numerical control code that adapts between not homologous ray turns
Change method.
Summary of the invention
The numerical control G code that in order to solve the above technical problem, the present invention provides a kind of algorithms is simple, conversion accuracy is high turns
Change method.
Technical proposal that the invention solves the above-mentioned problems is: a kind of conversion method of numerical control G code, comprising the following steps:
Step 1: input G code is pre-processed:
Step 2: determining the G code type of current A digital control system and generates the class of the G code of target B digital control system
Type；
Step 3: G code transfer framework is established based on neural network algorithm, and G code transfer framework is trained, is obtained
To G code transformation model；
Step 4: reading in every row G code in rows, and the G code of A digital control system is passed to G code transformation model, G
Code conversion model encodes the G code of A digital control system, after attention mechanism, is decoded to A digital control system；
Deep learning G code model database trained in advance is called, G code is translated；
Step 5: judging whether the translation conversion of the row is completed, and then continues to translate as unfinished, continues to sentence if completing
Whether all disconnected translation terminates, and is such as not finished, then repeats step 4 to step 5, continue to read next line G code, if having tied
Beam then exports the G code for generating target B digital control system.
The conversion method of above-mentioned numerical control G code, the step 1 specific steps are as follows: to the G code of input according to line number into
Row is split, and every row is divided into a list, and to the element further division of every row.
The conversion method of above-mentioned numerical control G code, the step 2 specific steps are as follows:
2-1) pass through data prediction, after extracting to each G code of the every row of data, statistical method is based on, to defeated
The G code entered is judged: assigning lesser weight firstly for the code of standard universal, non-standard general code is assigned
Biggish weight is given, and possible generic is carried out to each G code and is given a mark, last score is counted, and root
It is ranked up according to weight to small；
Digital control system belonging to the G code for 2-2) estimating input according to last highest score is possible, provides judgement
As a result, being confirmed by client；
2-3) selection needs the G code classification exported.
The conversion method of above-mentioned numerical control G code, in the step 3, G code conversion frame is divided into coded portion and decoding
Part；
In code area part, G code passes through XiIt carries out after being input to hidden layer, is encoded,
ht=Φ (ht-1,xt)=f (W(hh)ht-1+W(hx)xt)
Wherein XtFor input layer, htFor hidden layer, connected each other between hidden layer, W is weight；
The prediction under the complete semanteme of every a line of attention mechanism help G code generates next G code for meeting semanteme,
Rather than one-to-one simple local code；When hidden layer is after attention mechanism, attention mechanism is according to the G code of every row
Whole sentence semantic information q, the hidden layer h of final each G codetThe weighted average c of output onet=∑tstht, wherein coefficient st
It indicates and most degree of correlation in every row G code q；
In decoded portion, wherein
ht'=ct=φ (ct-1)=f (W(hh)ct-1)
yt=softmax (W(s)ht′)
ht' indicate hidden layer, ytOutput is indicated, in forecast period, ytThe predicted value of G code after indicating conversion.
The conversion method of above-mentioned numerical control G code, the training process of G code transfer framework in the step 3 are as follows:
3-1) input G code is pre-processed: the G code of input being split by line number, every row is divided into a column
Table, and to the G code further division of every row；
3-2) pass through one sentence pair of G code of the G code of the A digital control system to same machining information and B digital control system
It should be used as training set, off-line training neural network；
It 3-3) is passed to G code transfer framework: after encoding using Recognition with Recurrent Neural Network to G code, passing through attention machine
It makes, finally G code to be transformed is predicted in decoding, reduces error by using minimum cross entropy in the training process, final to obtain
To trained G code transformation model.
The conversion method of above-mentioned numerical control G code, the step 3-3) in, using the formula for minimizing cross entropy reduction error
Are as follows:
The conversion method of above-mentioned numerical control G code in the step 4, passes through and calls trained G code transformation model, G
Code is after coding, attention mechanism, the value decoded, the target G code as to be converted.
The beneficial effects of the present invention are: the invention discloses a kind of conversion methods of numerical control G code, first to input
G code is pre-processed, and is then determined the G code type of current G code and generation target, is resettled G code modulus of conversion
Type then reads in G code in rows, and G code is passed to G code transformation model, translates to G code, finally judges
Whether conversion is completed, unfinished then continue to convert.The different machine tool systems that method disclosed by the invention can adapt to very well
Characteristic, have many advantages, such as that scalability is strong, G code conversion quality is high.
Detailed description of the invention
Fig. 1 is specific flow chart of the invention.
Fig. 2 is the structure chart of the G code transfer framework in the present invention.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples.
As shown in Figure 1 and Figure 2, a kind of conversion method of numerical control G code is converted into Fa Nake's with Central China numerical control G code
For G code, comprising the following steps:
Step 1: input G code is pre-processed.
The G code file of Central China numerical control is inputted, and the G code of input is split according to line number, every row is divided into one
List, and to the element further division of every row.
Step 2: determining the G code type of current A digital control system and generates the class of the G code of target B digital control system
Type.Specific steps are as follows:
2-1) in conversion, Central China numerical control G code file of input, by data prediction, to the every of the every row of data
After a G code extracts, it is based on statistical method, the G code of input is judged: assigned firstly for the code of standard universal
Lesser weight is given, if the G code of international regulations is meant that uniquely, smaller weight can be given.For non-standard general code
Biggish weight is assigned, such as: G92 is workpiece coordinate setting in Central China numerical control system, is followed in Fa Nake for the turning that is screwed
Ring gives big weight for this distinguishing code, and carries out possible generic to each G code and give a mark, to most
Score afterwards is counted, and is ranked up according to weight to small；
Digital control system belonging to the G code for 2-2) estimating input according to last highest score is possible, provides judgement
As a result, being confirmed by client；
The rough anticipation of this input, help the attribute for tentatively prejudging G code to be converted: Central China numerical control system reduces input
The input of G code to be converted, gets the wrong sow by the ear if provided, and can also voluntarily select classification；
2-3) selection needs the G code classification exported: Fa Nake.
Step 3: G code transfer framework is established based on neural network algorithm, and G code transfer framework is trained, is obtained
To G code transformation model.
For G code transfer framework in training and prediction, function has slightly difference, as shown in Fig. 2, G code converts frame
It is divided into coded portion and decoded portion；
In code area part, such as: when the G code of a line Central China numerical control system passes through XiIt carries out after being input to hidden layer,
It is encoded,
ht=Φ (ht-1,xt)=f (W(hh)ht-1+W(hx)xt)
Wherein XtFor input layer, htFor hidden layer, connected each other between hidden layer, W is weight；
The prediction under the complete semanteme of every a line of attention mechanism help G code generates next G code for meeting semanteme,
Rather than one-to-one simple local code；When hidden layer is after attention mechanism, attention mechanism is according to the G code of every row
Whole sentence semantic information q, the hidden layer h of final each G codetThe weighted average c of output onet=∑tstht, wherein coefficient st
It indicates and most degree of correlation in every row G code q；
In decoded portion, wherein
ht'=ct=φ (ct-1)=f (W(hh)ct-1)
yt=softmax (W(s)ht′)
ht' indicate hidden layer, ytThe G code for indicating output Fa Nake, in forecast period, ytIndicate the G code after converting
Predicted value, i.e., the G code for the Fa Nake being converted to by Central China numerical control system.But then indicate that the G of Fa Nake is true in the training stage
Code.
The training process of G code transfer framework are as follows:
3-1) input G code is pre-processed: the G code of input Central China numerical control and Fa Nake is torn open by line number
Point, every row is divided into a list, and to the G code further division of every row；
3-2) pass through one sentence pair of G code of the G code of the A digital control system to same machining information and B digital control system
It should be used as training set, off-line training neural network；Pay attention to the code and be the code that artificial optimization crosses, such as: in Central China numerical control system
Unite the G code of operation, after manually adjusting, can on Central China numerical control system perfect operation G code.
It 3-3) is passed to G code transfer framework: after encoding using Recognition with Recurrent Neural Network to G code, passing through attention machine
It makes, finally G code to be transformed is predicted in decoding, reduces error by using minimum cross entropy in the training process, final to obtain
To trained G code transformation model.Using the formula for minimizing cross entropy reduction error are as follows:
Deep neural network model can be RNN (Recognition with Recurrent Neural Network) or shot and long term Recognition with Recurrent Neural Network, and
Other neural networks.The model for reaching requirement by the final selection of multistage training is final mask, the G code which obtains
Transformation model be a kind of dual model, the G code that can not only be transformed into Fa Nake by setting Central China numerical control system can also pass through hair
That section is to Central China numerical control system.
Step 4: reading in every row G code in rows, and the G code of A digital control system is passed to G code transformation model, G
Code conversion model encodes the G code of A digital control system, after attention mechanism, is decoded to A digital control system；
Deep learning G code model database trained in advance is called, G code is translated；
Step 5: judging whether the translation conversion of the row is completed, and then continues to translate as unfinished, continues to sentence if completing
Whether all disconnected translation terminates, and is such as not finished, then repeats step 4 to step 5, continue to read next line G code, if having tied
Beam then exports the G code for generating target B digital control system.
Claims (7)
1. a kind of conversion method of numerical control G code, comprising the following steps:
Step 1: input G code is pre-processed:
Step 2: determining the G code type of current A digital control system and generates the type of the G code of target B digital control system；
Step 3: G code transfer framework is established based on neural network algorithm, and G code transfer framework is trained, obtains G
Code conversion model；
Step 4: reading in every row G code in rows, and the G code of A digital control system is passed to G code transformation model, G code
Transformation model encodes the G code of A digital control system, after attention mechanism, is decoded to A digital control system；It calls
Trained deep learning G code model database in advance, translates G code；
Step 5: judging whether the translation conversion of the row is completed, and then continues to translate as unfinished, and continuing judgement if completing is
No all translations terminate, and are such as not finished, then repeat step 4 to step 5, continue to read next line G code, if having terminated,
Output generates the G code of target B digital control system.
2. the conversion method of numerical control G code according to claim 1, which is characterized in that the step 1 specific steps are as follows:
The G code of input is split according to line number, every row is divided into a list, and to the element further division of every row.
3. the conversion method of numerical control G code according to claim 1, which is characterized in that the step 2 specific steps are as follows:
2-1) pass through data prediction, after extracting to each G code of the every row of data, statistical method is based on, to the G of input
Code is judged: assign lesser weight firstly for the code of standard universal, for non-standard general code assign compared with
Big weight, and possible generic is carried out to each G code and is given a mark, last score is counted, and according to power
Weight is ranked up to small；
Digital control system belonging to the G code for 2-2) estimating input according to last highest score is possible, provides judging result,
Confirmed by client；
2-3) selection needs the G code classification exported.
4. the conversion method of numerical control G code according to claim 1, which is characterized in that in the step 3, G code turns
Change frame and is divided into coded portion and decoded portion；
In code area part, G code passes through XiIt carries out after being input to hidden layer, is encoded,
ht=Φ (ht-1,xt)=f (W(hh)ht-1+W(hx)xt)
Wherein XtFor input layer, XiFor input signal, htFor hidden layer, connected each other between hidden layer, W is weight；
The prediction under the complete semanteme of every a line of attention mechanism help G code generates next G code for meeting semanteme, without
It is one-to-one simple local code；When hidden layer is after attention mechanism, attention mechanism is according to the whole sentence of G code of every row
Semantic information q, the hidden layer h of final each G codetThe weighted average c of output onet=∑tstht, wherein coefficient stIt indicates
With most degree of correlation in every row G code q；
In decoded portion, wherein
h′t=ct=φ (ct-1)=f (W(hh)ct-1)
yt=soft max (W(s)h′t)
h′tIndicate hidden layer, ytOutput is indicated, in forecast period, ytThe predicted value of G code after indicating conversion.
5. the conversion method of numerical control G code according to claim 4, which is characterized in that G code is converted in the step 3
The training process of frame are as follows:
3-1) input G code is pre-processed: the G code of input is split by line number, every row is divided into a list, and
And to the G code further division of every row；
It should 3-2) be made by the G code of the A digital control system to same machining information and one sentence pair of G code of B digital control system
For training set, off-line training neural network；
3-3) it is passed to G code transfer framework: after encoding using Recognition with Recurrent Neural Network to G code, by attention mechanism, most
G code to be transformed is predicted in decoding afterwards, in the training process by using cross entropy reduction error is minimized, finally obtains training
Good G code transformation model.
6. the conversion method of numerical control G code according to claim 5, which is characterized in that the step 3-3) in, using most
The formula of smallization cross entropy reduction error are as follows:
7. the conversion method of numerical control G code according to claim 6, which is characterized in that in the step 4, pass through calling
Trained G code transformation model, G code is after coding, attention mechanism, the value decoded, the mesh as to be converted
Mark G code.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910775600.8A CN110488755A (en) | 2019-08-21 | 2019-08-21 | A kind of conversion method of numerical control G code |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910775600.8A CN110488755A (en) | 2019-08-21 | 2019-08-21 | A kind of conversion method of numerical control G code |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110488755A true CN110488755A (en) | 2019-11-22 |
Family
ID=68552649
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910775600.8A Pending CN110488755A (en) | 2019-08-21 | 2019-08-21 | A kind of conversion method of numerical control G code |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110488755A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112346737A (en) * | 2021-01-08 | 2021-02-09 | 深圳壹账通智能科技有限公司 | Method, device and equipment for training programming language translation model and storage medium |
| CN113326581A (en) * | 2021-05-28 | 2021-08-31 | 江麓机电集团有限公司 | Genetic scheduling method based on combined production and equipment fault constraint |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102081378A (en) * | 2010-11-01 | 2011-06-01 | 北京数码大方科技有限公司 | Method and device for converting universal code based on macroprogram |
| CN109543667A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of text recognition method based on attention mechanism |
| CN109614111A (en) * | 2018-12-28 | 2019-04-12 | 北京百度网讯科技有限公司 | Method and apparatus for generating code |
| CN109766355A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | A kind of data query method and system for supporting natural language |
| CN109902862A (en) * | 2019-02-13 | 2019-06-18 | 北京航空航天大学 | A kind of time series forecasting system of time of fusion attention mechanism |
-
2019
- 2019-08-21 CN CN201910775600.8A patent/CN110488755A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102081378A (en) * | 2010-11-01 | 2011-06-01 | 北京数码大方科技有限公司 | Method and device for converting universal code based on macroprogram |
| CN109543667A (en) * | 2018-11-14 | 2019-03-29 | 北京工业大学 | A kind of text recognition method based on attention mechanism |
| CN109614111A (en) * | 2018-12-28 | 2019-04-12 | 北京百度网讯科技有限公司 | Method and apparatus for generating code |
| CN109766355A (en) * | 2018-12-28 | 2019-05-17 | 上海汇付数据服务有限公司 | A kind of data query method and system for supporting natural language |
| CN109902862A (en) * | 2019-02-13 | 2019-06-18 | 北京航空航天大学 | A kind of time series forecasting system of time of fusion attention mechanism |
Non-Patent Citations (1)
| Title |
|---|
| 李亚超 等: "神经机器翻译综述", 《计算机学报》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112346737A (en) * | 2021-01-08 | 2021-02-09 | 深圳壹账通智能科技有限公司 | Method, device and equipment for training programming language translation model and storage medium |
| CN112346737B (en) * | 2021-01-08 | 2021-04-13 | 深圳壹账通智能科技有限公司 | Method, device and equipment for training programming language translation model and storage medium |
| CN113326581A (en) * | 2021-05-28 | 2021-08-31 | 江麓机电集团有限公司 | Genetic scheduling method based on combined production and equipment fault constraint |
| CN113326581B (en) * | 2021-05-28 | 2022-07-19 | 江麓机电集团有限公司 | Genetic scheduling method based on combined production and equipment fault constraint |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110111399B (en) | Image text generation method based on visual attention | |
| CN110488755A (en) | A kind of conversion method of numerical control G code | |
| CN112818646B (en) | Method for editing pictures according to text based on generation of countermeasure network and dynamic editing module | |
| CN111652357B (en) | Method and system for solving video question-answer problem by using specific target network based on graph | |
| Zhao et al. | Gaussian mixture density modeling of non-Gaussian source for autoregressive process | |
| US5694128A (en) | Tree structured binary arithmetic coder | |
| EP1227595A3 (en) | System and method for encoding an input data stream by utilizing a predictive, look-ahead feature | |
| DE60316125D1 (en) | METHOD FOR OPERATING A LANGUAGE DIALOG SYSTEM | |
| CN108920472A (en) | A kind of emerging system and method for the machine translation system based on deep learning | |
| CN113591770A (en) | Multimode fusion obstacle detection method and device based on artificial intelligence blind guiding | |
| KR980011007A (en) | Speech recognition method, information formation method, speech recognition apparatus and recording medium | |
| CN108197123A (en) | A kind of cloud translation system and method based on smartwatch | |
| CN112634918B (en) | System and method for converting voice of any speaker based on acoustic posterior probability | |
| KR102061300B1 (en) | Linear predictive coding apparatus, linear predictive decoding apparatus, methods thereof, programs and recording media | |
| CN111898482A (en) | Face prediction method based on progressive generation confrontation network | |
| CN114168721A (en) | Method for constructing knowledge enhancement model for multi-sub-target dialogue recommendation system | |
| Nair et al. | State estimation under bit-rate constraints | |
| CN115346269A (en) | Gesture motion recognition method | |
| CN116486794A (en) | Chinese-English mixed speech recognition method | |
| CN114691858A (en) | Improved UNILM abstract generation method | |
| CN112348033B (en) | Collaborative saliency target detection method | |
| Chan et al. | Generalized product code vector quantization: A family of efficient techniques for signal compression | |
| CN101477686B (en) | Nonsupervision image segmentation process based on clone selection | |
| KR20190019460A (en) | Apparatus and method for discriminating vocoder type | |
| CN115407874B (en) | VR maintenance training operation proficiency prediction method based on neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191122 |