CN109396576A - Stability of EDM and power consumption state Optimal Decision-making system and decision-making technique based on deep learning - Google Patents
Stability of EDM and power consumption state Optimal Decision-making system and decision-making technique based on deep learning Download PDFInfo
- Publication number
- CN109396576A CN109396576A CN201811145126.2A CN201811145126A CN109396576A CN 109396576 A CN109396576 A CN 109396576A CN 201811145126 A CN201811145126 A CN 201811145126A CN 109396576 A CN109396576 A CN 109396576A
- Authority
- CN
- China
- Prior art keywords
- data
- power consumption
- state
- value
- electrical discharge
- 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
- 238000000034 method Methods 0.000 title claims abstract description 97
- 238000013135 deep learning Methods 0.000 title claims abstract description 36
- 238000009760 electrical discharge machining Methods 0.000 claims abstract description 195
- 238000012545 processing Methods 0.000 claims abstract description 181
- 238000003754 machining Methods 0.000 claims abstract description 69
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 41
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 238000004134 energy conservation Methods 0.000 claims abstract description 20
- 238000009826 distribution Methods 0.000 claims abstract description 17
- 238000005457 optimization Methods 0.000 claims abstract description 17
- 230000000306 recurrent effect Effects 0.000 claims abstract description 13
- 238000004458 analytical method Methods 0.000 claims abstract description 10
- 238000005065 mining Methods 0.000 claims abstract description 4
- 238000010187 selection method Methods 0.000 claims abstract description 4
- 230000006641 stabilisation Effects 0.000 claims abstract description 3
- 238000011105 stabilization Methods 0.000 claims abstract description 3
- 230000008569 process Effects 0.000 claims description 71
- 210000001367 artery Anatomy 0.000 claims description 39
- 210000003462 vein Anatomy 0.000 claims description 39
- 238000005265 energy consumption Methods 0.000 claims description 33
- 238000007781 pre-processing Methods 0.000 claims description 24
- 238000007621 cluster analysis Methods 0.000 claims description 21
- 230000004927 fusion Effects 0.000 claims description 15
- 230000009471 action Effects 0.000 claims description 14
- 238000010892 electric spark Methods 0.000 claims description 14
- 230000005540 biological transmission Effects 0.000 claims description 13
- 238000007599 discharging Methods 0.000 claims description 10
- 238000010606 normalization Methods 0.000 claims description 9
- 238000012216 screening Methods 0.000 claims description 9
- 230000007246 mechanism Effects 0.000 claims description 8
- 230000005611 electricity Effects 0.000 claims description 7
- 230000015572 biosynthetic process Effects 0.000 claims description 5
- 230000008676 import Effects 0.000 claims description 5
- 238000003786 synthesis reaction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 230000001537 neural effect Effects 0.000 claims description 4
- 238000011946 reduction process Methods 0.000 claims description 4
- 230000001105 regulatory effect Effects 0.000 claims description 4
- 241000208340 Araliaceae Species 0.000 claims description 3
- 241000282461 Canis lupus Species 0.000 claims description 3
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims description 3
- 235000003140 Panax quinquefolius Nutrition 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 235000008434 ginseng Nutrition 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000005856 abnormality Effects 0.000 claims description 2
- 238000013506 data mapping Methods 0.000 claims description 2
- 238000003066 decision tree Methods 0.000 claims description 2
- 238000007637 random forest analysis Methods 0.000 claims description 2
- 238000012706 support-vector machine Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 abstract description 6
- 238000010586 diagram Methods 0.000 description 6
- 239000000463 material Substances 0.000 description 5
- 238000007418 data mining Methods 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000005520 cutting process Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 238000009412 basement excavation Methods 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 230000007797 corrosion Effects 0.000 description 2
- 238000005260 corrosion Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000009991 scouring Methods 0.000 description 2
- 230000003746 surface roughness Effects 0.000 description 2
- 241000209202 Bromus secalinus Species 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 241001269238 Data Species 0.000 description 1
- 229910001069 Ti alloy Inorganic materials 0.000 description 1
- 238000002679 ablation Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 238000002309 gasification Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000009916 joint effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B23—MACHINE TOOLS; METAL-WORKING NOT OTHERWISE PROVIDED FOR
- B23H—WORKING OF METAL BY THE ACTION OF A HIGH CONCENTRATION OF ELECTRIC CURRENT ON A WORKPIECE USING AN ELECTRODE WHICH TAKES THE PLACE OF A TOOL; SUCH WORKING COMBINED WITH OTHER FORMS OF WORKING OF METAL
- B23H1/00—Electrical discharge machining, i.e. removing metal with a series of rapidly recurring electrical discharges between an electrode and a workpiece in the presence of a fluid dielectric
- B23H1/02—Electric circuits specially adapted therefor, e.g. power supply, control, preventing short circuits or other abnormal discharges
-
- 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"
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2216/00—Indexing scheme relating to additional aspects of information retrieval not explicitly covered by G06F16/00 and subgroups
- G06F2216/03—Data mining
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
-
- 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)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Health & Medical Sciences (AREA)
- Manufacturing & Machinery (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Mechanical Engineering (AREA)
- Electrical Discharge Machining, Electrochemical Machining, And Combined Machining (AREA)
Abstract
The invention discloses a kind of stability of EDM based on deep learning and power consumption state Optimal Decision-making platform.With Feature Selection method mining analysis electrical discharge machining data, the optimizing index of processing stability and power consumption state is obtained；Optimizing index is clustered with K-medoids algorithm, obtains the distribution situation of processing stability and power consumption state, and constructs stable machining state database and energy saving discharge condition database；The predicted value of real-time electrical discharge machining state is obtained with history electrical discharge machining data training LSTM recurrent neural network deep learning, in conjunction with the statistical characteristics of optimizing index, when judgement is currently improper machining state, multiple-objection optimization is carried out to processing stability and power consumption state, stablized-comprehensive optimal objective the value of energy conservation machining state, and regulates and controls current machined parameters value accordingly.The present invention is based on deep learning, gives and stablize-comprehensive optimal electrical discharge machining parameter optimization the decision-making technique of energy conservation, run electrical discharge machining in stabilization and in the state of energy conservation.
Description
Technical field
The present invention relates to the electric machining field in special process field, in particular to a kind of electric spark based on deep learning
Processing stability and power consumption state Optimal Decision-making system.
Background technique
Electrical discharge machining refers in certain medium, passes through the pulsed discharge between tool-electrode and piece pole, shape
At instantaneous high-temperature by workpiece material local melting and gasification, to realize material ablation.This processing method does not generate cutting
Power is not limited by cutter material, can be processed ultrahigh hardness, brittleness and complex-shaped workpiece, is therefore widely used in
The multiple fields such as mold, aircraft industry, medical instrument.Electrical discharge machining is usually realized by electric spark machine tool.
The main technologic parameters for characterizing EDM Performance include that (peak point current, crest voltage, pulse are wide for electrical parameter
Degree, pulse spacing, processing polarity), the material parameter of non-electrical parameter (working fluid pressure washes away mode) and workpiece it is (specific heat, close
Degree, thermal conductivity coefficient, fusing point).In the prior art, many researchers from different perspectives, have studied above-mentioned technological parameter to electrical fire
The influence of flower processing performance, and try hard to establish accurate process optimization model.Wang Tong et al. has carried out electric spark in simple venation qi of chong channel ascending adversely
It processes galvanic corrosion and cheats pattern simulation analysis, obtain workpiece surface galvanic corrosion hole radius, depth, volume and aspect ratio with peak value electricity
The changing rule of stream and pulse width, and then the surface quality of predictable electrical discharge machining efficiency and workpiece.Yao Zhong etc. utilizes ash
Correlation analysis optimizes SKD11 electrical discharge machining electrical parameter, finds pulse discharge time and peak point current, inter-train pause time
With gap voltage to the affecting laws of surface roughness and material removing rate.Bright five is first-class thick by process velocity, three-dimensional surface
Rugosity analyzes the influence of duty ratio, peak point current, voltage to titanium alloy sinking EDM as evaluation index.Due to
Technological parameter is difficult to be indicated with accurate mathematical model to the influence degree of EDM Performance, therefore many scholars will
Artificial neural network, intelligent algorithm and fuzzy mathematics are introduced into electrical discharge machining process optimization process.Zhou Xiaoming etc. is with pulse
Width, pulse spacing, peak point current and processing establish the wire electric discharge based on BP neural network and cut with a thickness of technological parameter
Technic index prediction model is cut, keeps the cutting speed of wire cutting and surface roughness combination best.Sun Zhongfeng etc., which is proposed, to be based on
The Wire EDM technological parameter Fuzzy Optimal Method of neural net model establishing and genetic algorithm, so that it is global most to improve acquisition
The probability of excellent solution and be not easy to fall into local optimum.
However, research in the prior art is usually to be finished with the electric spark of middle-size and small-size workpiece for object, to optimize electricity
Spark processing efficiency and workpiece surface quality are target expansion.As global carbon emission amount is increased sharply, the energy is increasingly depleted, hair
Exhibition " green economy " has become global hot spot, and under such overall background, the capacity usage ratio of traditional electrical discharge machining is low, consumes
It can become increasingly conspicuous the drawbacks of height, non-steady state, especially for the electric spark roughing of large-scale workpiece, how make electric spark
Processing longtime running in the state of not only stable but also energy conservation becomes very urgent.But electrical discharge machining is a physics-change
The complicated random process for learning interaction, the characteristic parameter for evaluating its processing stability and power consumption state have interdisciplinary, strong coupling
How conjunction and nonlinear feature precisely establish the comprehensive optimal electrical discharge machining process model of stable-energy conservation and to its feature
Parameter optimizes control, it has also become traditional electrical discharge machining is urgently to be resolved into green, intelligence, sustainable transformation process
Critical issue.
Summary of the invention
That the purpose of the present invention is to provide a kind of energy utilization rates is relatively high, electrical discharge machining can be made not only stable but also
The long-term running stability of EDM based on deep learning and power consumption state Optimal Decision-making system energy savingly.
To achieve the above object, the present invention is based on the stability of EDM of deep learning and power consumption state optimization to determine
Plan system, it is characterised in that: excavate module including data preprocessing module, processing stability and power consumption state optimizing index, add
Apt qualitative and power consumption state optimizing index Cluster Analysis module stablizes machining state database and energy saving discharge condition data
Library constructs module, processing stability and power consumption state optimizing index statistical nature and obtains module, processing stability and power consumption state
Characteristic parameter deep learning module and processing stability and power consumption state Optimal Decision-making module.
Data preprocessing module is used to take out the electrical discharge machining data that the electric control gear of spark-erosion machine tool obtains
It takes, clean, merging and reduction process, being subsequent processing stability and power consumption state optimizing index excavates module, processing is stablized
Property with power consumption state optimizing index Cluster Analysis module, stablize machining state database and energy saving discharge condition database sharing
Module and processing stability and power consumption state characteristic parameter deep learning module provide the history electrical spark working number on basis
According to；
Processing stability and power consumption state optimizing index excavate module and use (Combining weights) Feature Selection method to data
The history electrical discharge machining data of preprocessing module transmission carry out mining analysis, and obtaining influences the quick of processing stability and energy consumption
Feel characteristic parameter, as processing stability and power consumption state optimizing index and is supplied to processing stability and power consumption state optimization
Cluster Analysis module；
Processing stability and power consumption state optimizing index Cluster Analysis module optimize according to processing stability and power consumption state
Index, to data preprocessing module transmission history electrical discharge machining data carry out clustering, mark stablize, meta-stable and
Unstable three processing stability status categories obtain each processing stability status categories in history electrical discharge machining data
Distribution situation；Mark energy conservation, energy consumption and three electric discharge power consumption state classifications of highly energy-consuming, obtain in history electrical discharge machining data
The distribution situation of each electric discharge power consumption state classification；
Stablize machining state database and energy saving discharge condition database sharing module according to processing stability and energy consumption shape
The cluster analysis result that state optimizing index Cluster Analysis module provides, to the history electrical spark working of data preprocessing module transmission
Number obtains stablizing machining state database according to stable processing classification screening is carried out；To the history of data preprocessing module transmission
Electrical discharge machining data carry out energy conservation electric discharge classification screening, obtain energy saving discharge condition database.
Processing stability and power consumption state optimizing index statistical nature obtain electric control gear of the module by spark-erosion machine tool
It obtains real-time electrical discharge machining data and calculates its statistical characteristics, as the foundation for judging current electrical discharge machining state.
Processing stability and power consumption state characteristic parameter deep learning module, using LSTM recurrent neural network to history
Electrical discharge machining data carry out deep learning and obtain prediction model, predict the variation tendency of real-time electrical discharge machining state, with pre-
The current electrical discharge machining state of measured value auxiliary identification；
Processing stability judges current electrical discharge machining state with power consumption state Optimal Decision-making module, works as judgement
When current electrical discharge machining state is improper, the comprehensive optimal target value of stable-energy saving machining state is calculated and passes through
The electric control gear of spark-erosion machine tool regulates and controls the parameter value of current electrical discharge machining parameter.
The invention also discloses a kind of stability of EDM based on deep learning and power consumption state Optimal Decision-making system
The decision-making technique of system, successively sequentially includes the following steps:
First step is to carry out data pick-up, data cleansing, data fusion and data regularization by data preprocessing module
Processing；
Second step is to excavate module by processing stability and power consumption state optimizing index to excavate processing stability and energy
Consume state optimization index；
Third step is to mark processing stability by processing stability and power consumption state optimizing index Cluster Analysis module
Status categories and electric discharge power consumption state classification；
Four steps is stablized by stablizing machining state database and energy saving discharge condition database sharing module building
Machining state database and energy saving discharge condition database；
5th step is to calculate real-time electric spark by processing stability and power consumption state optimizing index statistical nature module
The statistical characteristics of process data, as the foundation for judging current electrical discharge machining state；
6th step is using processing stability and power consumption state characteristic parameter deep learning module to history electrical spark working
Number obtains prediction model according to deep learning is carried out, and predicts the variation tendency of real-time electrical discharge machining state, is assisted with predicted value
Recognize current electrical discharge machining state；
7th step is that processing stability sentences current electrical discharge machining state with power consumption state Optimal Decision-making module
It is disconnected, when it is improper for determining current electrical discharge machining state, it is calculated and stablizes-comprehensive optimal the mesh of energy conservation machining state
Scale value, and regulate and control by the electric control gear of spark-erosion machine tool the parameter value of current electrical discharge machining parameter.
In the first step, it is by four data pick-up, data cleansing, data fusion and data regularization steps
Processing stability and power consumption state optimizing index excavate module, processing stability and power consumption state optimizing index clustering mould
Block stablizes machining state database and energy saving discharge condition database sharing module and processing stability and power consumption state spy
Levy the history electrical discharge machining data needed for parameter deep learning module provides.
In data pick-up, it is that electrical discharge machining data are obtained by the electric control gear of spark-erosion machine tool first, then exists
The principal element and correlation for influencing processing stability and energy consumption are found out in electrical discharge machining data, are formed on this basis
The parameter and parameter value to be extracted；Multiple parameters are contained in electrical discharge machining data (including such as spark-erosion machine tool high-frequency electrical
30 parameters that 30 measuring points such as source device, Wire-conveying mechanism of quick walk, SERVO CONTROL, scouring part, machining state obtain), it extracts wherein
8 major parameters, including electrode wire travelling speed, wire electrode tension, medium temperature, discharge current, electric discharge pulsewidth, electric discharge arteries and veins
Between, voltage across poles and feed speed；
Fault value, missing values, repetition values and noise figure existing for 8 major parameters are cleaned up, it is clear to complete data
It washes, improves the quality of data；
Data fusion is carried out after data cleansing；Data fusion is first to extract the meta-attribute of data to be fused, then analyze wait melt
The mapping relations for closing data and target data by data to be fused and target data mapping, integrate to form data bins
Library；
Data regularization is carried out after data fusion；Data regularization is for the data in data warehouse, advanced row data record
Statistics characteristic analysis, then remove attribute value and change small data record；Data record feature correlation analysis is carried out again, is removed
Feature strong correlation data record obtains the brief history electrical discharge machining data set of feature；
Wherein, attribute value changes small quantitative criteria are as follows: d (xi)≤0.01, d (xi) be by all data normalizations it
Afterwards, xiWith the minimum range of other data.If d (xi)≤0.01, then it is assumed that xiAttribute value variation it is small.
The quantitative criteria of strong correlation are as follows:
Wherein,It is xiThe mean value of middle all elements；It is using least square method to xi=f (x1,x2,…,xi-1,
xi+1,…,xn) fitting function,It is xiThe mean value of middle all elements, xijIt is xiJth each element.If R >=0.01,
Think data xiWith other element strong correlations.
After above-mentioned data preprocessing operation, the history electrical discharge machining data on the basis as subsequent operation are obtained,
The data set has p parameters；P is more than or equal to 6 and is less than or equal to 12.
The concrete operations of the second step are:
It is data acquisition movement first, times of collection M, M are natural number and its initial value is 0；In history electrical discharge machining
Electrical discharge machining data of data concentrated collection；
Followed by comparison compares the value of M and K, if M < K, executes the movement of calculating parameter score；If M >=
K is then jumped and is executed the movement of calculating parameter value；
The movement of calculating parameter score is to use nine kinds of algorithms respectively for the p item parameter in the secondary electrical discharge machining data
It is screened, calculates the score of each parameter；Nine kinds of algorithms are respectively single argument feature selecting algorithm, pearson correlation
Coefficient Algorithm, decision Tree algorithms, L1 regularization algorithm, L2 regularization algorithm, random forests algorithm, random LASSO algorithm, away from
From related coefficient algorithm and algorithm of support vector machine；The score is handled using extreme value standardized algorithm, result is limited
Make the scoring in [0,100] section, after being standardized；
Then the value of times of collection is made to add 1 (executing the increment operator of M++ herein)；The predetermined total degree of acquisition is K, K
For the natural number more than or equal to 3, after the value of times of collection adds 1, execution comparison is jumped；
When M >=K, is jumped from comparison and execute the movement of calculating parameter value；
The movement of calculating parameter value is the scoring mean value that each parameter (characteristic parameter i.e. in table 1) is calculated, scoring pole
Value and scoring variance；The different degree of each parameter is carried out according to the scoring mean value of each parameter, scoring extreme value and the variance that scores
Assessment and sequence classify scoring mean value (average i.e. in table 1) by the order of magnitude, and scoring mean value is units
Parameter is inessential parameter, and it is important parameter that scoring mean value, which is double-digit parameter, removes inessential parameter, retains important ginseng
Number and as processing stability and power consumption state optimizing index, completes to excavate processing stability to optimize with power consumption state and refers to
Target operation.
In second step, in terms of scoring event, electrode wire travelling speed, wire electrode tension, medium temperature mean value lower be
Units excludes the relatively low characteristic parameter of these scores.Several characteristic parameters of highest scoring, according to sequence from high to low
It is successively: between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles and feed speed, using these parameters as processing stabilization
Property with power consumption state optimizing index.Table 1 is that 8 characteristic parameters are carried out with the tables of data after specification scoring.
1 processing stability of table and power consumption state characteristic parameter grade form
The concrete operations of the third step are:
Clustering carries out processing stability state using K-medoids algorithm, to history electrical discharge machining data
Classification mark and electric discharge power consumption state classification mark；
History electrical discharge machining data record is labeled as stable, meta-stable and unstable three shapes by processing stability
State classification.History electrical discharge machining data record is labeled as energy conservation, energy consumption and three state class of highly energy-consuming by electric discharge energy consumption
Not；
History electrical discharge machining data comprising above-mentioned five processing stabilities and power consumption state optimizing index are remembered
Record, marks status categories using K-Medoids algorithm；
Stable state classification is SA, and the value range of discharge current between 3~10A (including endpoints thereof, similarly hereinafter) is put
Electric pulsewidth is between 80~300 μ s, and between the arteries and veins that discharges between 20~150 μ s, voltage across poles is between 21~26V, feed speed
In 80~128mm2Between/min；
Metastable condition classification is SB, and the value range of discharge current is between 1~5A, and pulsewidth of discharging is in 30~100 μ s
Between, between the arteries and veins that discharges between 0~60 μ s, voltage across poles is between 20~22.5V, and feed speed is in 100~145.5mm2/
Between min；
Unstable state classification is SC；The value range of discharge current is between 0~1A, and pulsewidth of discharging is in 10~50 μ s
Between, between the arteries and veins that discharges between 0~35 μ s, voltage across poles is between 18.5~22V, and feed speed is in 116.5~150mm2/
Between min；Table 2 is that processing stability divides group's cluster centre table；
2 processing stability of table divides group's cluster centre
Power save mode classification is EA, and the value range of discharge current is between 6~10A, and pulsewidth of discharging is in 150~280 μ s
Between, between the arteries and veins that discharges between 50~160 μ s, voltage across poles is between 20.5~24V, and feed speed is in 90~135mm2/
Between min；
Energy consumption status categories be EB, the value range of discharge current between 3~5A, electric discharge pulsewidth 60~160 μ s it
Between, between the arteries and veins that discharges between 0~70 μ s, voltage across poles is between 19.5~22.5V, and feed speed is in 75~115.5mm2/
Between min；
Highly energy-consuming status categories are EC, and the value range of discharge current is between 0~2A, and pulsewidth of discharging is in 10~80 μ s
Between, between the arteries and veins that discharges between 0~45 μ s, voltage across poles is between 18.5~21V, and feed speed is in 20~90mm2/ min it
Between.Table 3 is that electric discharge energy consumption divides group's cluster centre table；
The electric discharge of table 3. energy consumption divides group's cluster centre
The operation of the four steps is to construct to stablize machining state database and energy saving discharge condition database.
Stablizing every electrical discharge machining data record in machining state database includes recording mechanism, parameter name (such as " electric discharge
Electric current ") and parameter value；In energy saving discharge condition database every electrical discharge machining data record include recording mechanism, parameter name and
Parameter value.
The concrete operations for constructing stable machining state database are:
It is first directed to and initialization action, is labeled with processing stability status categories after importing third step clustering
History electrical discharge machining data record, summary journal the quantity n, n calculated in history electrical discharge machining data is natural number；
It initializes integer variable i and is equal to 1；
Then read action is executed, the parameter value and this for reading i-th history electrical discharge machining data record process number
According to recording corresponding status categories (i.e. state tag in Figure 11)；Then judge the status categories of this process data record
Whether be SA, if it is judged that be it is no, then execute and skip the operation of this operation record, and jump execute i from increasing behaviour
Make；
If it is judged that be it is yes, then execute to calculate and this history electrical discharge machining data record and stablize machining state
The operation of the Euclidean distance s of existing each process data record in database；
The operation for calculating Euclidean distance s is:
If stablizing machining state database is empty database, this history electrical discharge machining data record is stored in
Stablize machining state database；
Process data record is stored in machining state database if stablized, by this history electrical discharge machining
The parameter value of data record with stablize in machining state database corresponding parameter value in existing each process data record
Carry out the calculating of Euclidean distance s；
If this history electrical discharge machining data record is processed with existing each in machining state database is stablized
Euclidean distance s between data record is all larger than 0.01, then illustrates this history electrical discharge machining data record and stablize to process
Existing each process data record is all different in slip condition database, then is deposited this history electrical discharge machining data record
Enter to machining state database is stablized, then executes the increment operator of i；
If this history electrical discharge machining data record is processed with existing each in machining state database is stablized
Euclidean distance s between data record is respectively less than 0.01, then executes the operation for skipping the operation record, and jumps oneself for executing i
Increase operation；
The increment operator of i is to increase the value of i by 1 (executing i=i+1 operation), then judges whether i is greater than n, such as
Fruit i is less than or equal to n, then jumps execution read action；If i is greater than n, terminate the behaviour for constructing stable machining state database
Make；
The concrete operations for constructing energy saving discharge condition database are:
It is first directed to and initialization action, imports after third step clustering and be labeled with the history of power save mode classification
Electrical discharge machining data record, summary journal the quantity m, m for calculating history electrical discharge machining data are natural number；It initializes whole
Number variable j is equal to 1；
Then read action is executed, the parameter value and this for reading j-th strip history electrical discharge machining data record process number
According to recording corresponding status categories (i.e. class label in Figure 12)；Then judge the status categories of this process data record
Whether be EA, if it is judged that be it is no, then execute skip this process data record operation, and jump execute j from increase
Operation；
If it is judged that be it is yes, then execute calculate this process data record in energy saving discharge condition database
The operation of the Euclidean distance s of some each process data records；
The operation for calculating Euclidean distance e is:
If energy saving discharge condition database is empty database, this history electrical discharge machining data record is stored in
Energy saving discharge condition database；
If process data record has been stored in energy saving discharge condition database, by this history electrical discharge machining
The parameter value of data record and corresponding parameter value in each process data record existing in energy saving discharge condition database
Carry out the calculating of Euclidean distance e；
If existing each is processed in this history electrical discharge machining data record and energy saving discharge condition database
Euclidean distance e between data record is all larger than 0.01, then illustrates this electrical discharge machining data record and energy saving discharge condition
Existing each process data record is all different in database, then this electrical discharge machining data record is stored in energy conservation electric discharge
Then slip condition database executes the increment operator of j；(judges that the meaning of Euclidean distance is duplicate removal, avoid being stored in database
Duplicate data, and then avoid the load for increasing memory and repeated data is avoided to reduce operational efficiency)
If existing each is processed in this history electrical discharge machining data record and energy saving discharge condition database
Euclidean distance e between data record is less than or equal to 0.01, then executes the operation for skipping the operation record, and jumps and execute j's
Increment operator；
The increment operator of j is to increase the value of j by 1 (executing j=j+1 operation), then judges whether j is greater than m, such as
Fruit j is less than or equal to m, then jumps execution read action；If j is greater than m, terminate the behaviour for constructing energy saving discharge condition database
Make.
In 5th step, processing stability and power consumption state optimizing index statistical nature obtain module and pass through electricity
The electric control gear of spark lathe obtains real-time electrical discharge machining data, calculates between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, interpolar
Mean value, extreme value, variance and fracture of wire frequency of abnormity of this five optimizing index of voltage, feed speed in window time, statistics
Characteristic value is as the foundation for recognizing current electrical discharge machining state.
Processing stability and power consumption state optimizing index statistical nature obtain module at work, import the K period (i.e. certain
One period) in real-time electrical discharge machining data record, processing summary journal quantity n (herein n be natural number) in the statistics K period,
And integer variable i is initialized equal to 1, initialization fracture of wire frequency of abnormity integer variable is equal to 0；
Then read action is executed, the process data of i-th operation record is read；
After reading, fracture of wire abnormality detection is carried out；In case of fracture of wire, then first make the value of fracture of wire frequency of abnormity integer variable
Add 1, then executes statistics movement；If fracture of wire does not occur, directly execution statistics movement；
Statistics movement is the fracture of wire number and discharge current, electric discharge pulsewidth, electric discharge counted in the K period in operation record
Between arteries and veins, mean value, extreme value and the variance of five parameters of voltage across poles and process velocity；
After statistics movement, so that the value of integer variable i is added 1 (i.e. execution i=i+1), then judge the processing note in the K period
Whether record, which reads, finishes, and specifically judges whether i value is greater than n value, if i value is less than or equal to n value, jumps and executes again
Read action；If i value is greater than n value, all statistical characteristics at this time are sent to processing stability and power consumption state
Optimal Decision-making module.
The operation of 6th step is processing stability and power consumption state characteristic parameter deep learning:
1) the history electrical discharge machining data for reading data preprocessing module transmission, with Z-score method by history electrical fire
Flower process data carries out feature normalization；
2) the history electrical discharge machining data after standardization are input in LSTM recurrent neural network and are trained, obtained
Obtain prediction model
The core of LSTM is neural network location mode, is transmitted backward by entire chain structure, and utilizes thresholding knot
The Sigmoid neural net layer and point multiplication operation of structure realize that the selectivity of information passes through；The calculation formula of LSTM algorithm is as follows:
ht=ot*tanh(ft*ct-1+it*c_int)
In prediction model, the output valve at current time is out gate ot, forget door ft, input gate itAnd last moment is single
First state ct-1Function；
3) with Z-score method by the discharge current acquired in real time, electric discharge pulsewidth, electric discharge arteries and veins between, voltage across poles, feeding
The value of this five characteristic parameters of speed is standardized；
4) by after standardization discharge current, electric discharge pulsewidth, between electric discharge arteries and veins, voltage across poles, feed speed this five features
The value of parameter is input to prediction model and is analyzed, and obtains predicted value by inverse feature normalization, predicted value is exported to processing
Stability and power consumption state Optimal Decision-making module.
Explanation is pre- using the training of LSTM recurrent neural network by taking voltage across poles and the real-time data collection of feed speed as an example
It surveys model and obtains the process of predicted value.
Processing stability and energy are carried out to the value of collected voltage across poles and feed speed in one section of continuous time first
State-detection is consumed, is divided between access 60 seconds, the value of continuous 60 voltage across poles and feed speed is as shown in figure 15.
As shown in figure 16, it using LSTM recurrent neural network training processing stability and energy consumption prediction model, is followed with using
Ring neural network RNN and multilayer perceptron MLP are compared, and find the prediction model obtained by LSTM recurrent neural network
Accuracy highest.
The relationship of voltage across poles, the predicted value of feed speed and actual value is as shown in figure 17.As can be seen from the figure it predicts
Error is lower, and predicted value is able to reflect the variation tendency of numerical value substantially, and the prediction effect of model is good.
The operation of 7th step is processing stability and power consumption state Optimal Decision-making:
1) processing stability and power consumption state Optimal Decision-making module receive processing stability and power consumption state optimizing index is united
Meter feature obtains the statistical characteristics for the real-time electrical discharge machining data that module transmits.If current electrical discharge machining state is sentenced
It is set to improper machining state, activates Optimal Decision-making program；
The standard of the improper machining state is: when processing stability state is in unsteady state (unsteady state
Refer to meta-stable or unstable state) either electric discharge power consumption state be in non-power save mode (non-power save mode refer to energy consumption or height
Energy consumption state).
2) multiple-objection optimization is carried out to processing stability and power consumption state using wolf pack algorithm, is stablized-energy conservation processing
State multiple-objection optimization disaggregation；
3) respectively from stablize search matching in machining state database and energy saving discharge condition database it is approximate stablize plus
Work parameter and energy saving discharge parameter are stablized the-optimal parameter objectives value of energy saving machining state synthesis；
4) according to the target value, the parameter of current electrical discharge machining parameter is regulated and controled by the electric control gear of spark-erosion machine tool
Value.
The present invention has the advantage that:
Beneficial effects of the present invention, which are mainly manifested in, can make electrical discharge machining not only stable but also energy saving ground longtime running, mention
High workpiece quality simultaneously reduces processing cost, guarantees that processing stability and energy consumption index are able to maintain in synthesis when operating condition changes
Optimum interval optimizes regulation to the characteristic parameter for influencing stability of EDM and energy consumption, reaches the mesh of green manufacturing
's.
The configuration of the present invention is simple, algorithm is succinct, finds out influence processing stability in numerous parameters in electrical discharge machining
With the key parameter of energy consumption, and stable machining state database and energy saving discharge condition database are constructed, succinctly promptly predicted
The parameter value of processing stability and energy consumption double goal is combined out, to efficiently adjust adding for electric spark machine tool
Work parameter guarantees the stability and energy saving of process.
Whether data mining process succeeds, and depends primarily on the quality of the quality of data.It is produced in spark-erosion machine tool operational process
Raw mass data, these data are not only multi-source heterogeneous, mode is multifarious, but also there is abnormal, gaps and omissions and repetition, it is difficult to prop up
Hold the work such as subsequent deep learning.Data prediction is exactly in data mining and before use, original electrical spark working to acquisition
Number is according to a series of processing work such as necessary extraction, cleaning, fusion and reduction are carried out, to improve quality of data sum number
According to the accuracy rate of analysis result, subsequent applications are preferably adapted to.
Detailed description of the invention
Fig. 1 is the flow chart of decision-making technique of the invention；
Fig. 2 is the flow chart of first step data prediction；
Fig. 3 is the excavation flow chart of processing stability and power consumption state optimizing index in second step；
Fig. 4 is the process of processing stability and power consumption state optimizing index K-medoids clustering in third step
Figure；
Fig. 5 is the parameter distribution probability density figure for stablizing classification (SA) in processing stability status categories；
Fig. 6 is the parameter distribution probability density figure that classification (SB) is stablized in the processing stability status categories Central Asia；
Fig. 7 is the parameter distribution probability density figure of unstable classification (SC) in processing stability status categories；
Fig. 8 is the parameter distribution probability density figure of power saving class (EA) in electric discharge power consumption state classification；
Fig. 9 is the parameter distribution probability density figure of energy consumption classification (EB) in electric discharge power consumption state classification；
Figure 10 is the parameter distribution probability density figure of highly energy-consuming classification (EC) in electric discharge power consumption state classification；
Figure 11 is the Establishing process figure for stablizing machining state database；
Figure 12 is the Establishing process figure of energy saving discharge condition database；
Figure 13 is the work flow diagram that processing stability and power consumption state optimizing index statistical nature obtain module；
Figure 14 is the work flow diagram of processing stability Yu power consumption state characteristic parameter deep learning module；
Figure 15 is the successive value figure of voltage across poles and feed speed；
Figure 16 is the accuracy comparison diagram of LSTM, RNN and MLP training prediction model；
Figure 17 is the relational graph of voltage across poles, the predicted value of feed speed and actual value；
Figure 18 is the work flow diagram of processing stability Yu power consumption state Optimal Decision-making module.
Specific embodiment
The present invention is further described below with reference to embodiment (attached drawing):
As shown in Fig. 1, stability of EDM and power consumption state Optimal Decision-making based on deep learning is proposed to put down
Logical relation between the function and each module of each module of platform.
Data preprocessing module is used to take out the electrical discharge machining data that the electric control gear of spark-erosion machine tool obtains
It takes, clean, merging and reduction process, being subsequent processing stability and power consumption state optimizing index excavates module, processing is stablized
Property with power consumption state optimizing index Cluster Analysis module, stablize machining state database and energy saving discharge condition database sharing
Module and processing stability and power consumption state characteristic parameter deep learning module provide the history electrical spark working number on basis
According to；
Processing stability excavates module with power consumption state optimizing index and uses the Feature Selection method of Combining weights pre- to data
The history electrical discharge machining data of processing module transmission carry out mining analysis, obtain the sensitivity for influencing processing stability and energy consumption
Characteristic parameter as processing stability and power consumption state optimizing index and is supplied to processing stability and refers to power consumption state optimization
Mark Cluster Analysis module；
Processing stability and power consumption state optimizing index Cluster Analysis module optimize according to processing stability and power consumption state
Index, to data preprocessing module transmission history electrical discharge machining data carry out clustering, mark stablize, meta-stable and
Unstable three processing stability status categories obtain each processing stability status categories in history electrical discharge machining data
Distribution situation；Mark energy conservation, energy consumption and three electric discharge power consumption state classifications of highly energy-consuming, obtain in history electrical discharge machining data
The distribution situation of each electric discharge power consumption state classification；
Stablize machining state database and energy saving discharge condition database sharing module according to processing stability and energy consumption shape
The cluster analysis result that state optimizing index Cluster Analysis module provides, to the history electrical spark working of data preprocessing module transmission
Number obtains stablizing machining state database according to stable processing classification screening is carried out；To the history of data preprocessing module transmission
Electrical discharge machining data carry out energy conservation electric discharge classification screening, obtain energy saving discharge condition database.
Processing stability and power consumption state optimizing index statistical nature obtain electric control gear of the module by spark-erosion machine tool
It obtains real-time electrical discharge machining data and calculates its statistical characteristics, mentioned as the foundation for judging current electrical discharge machining state
Supply processing stability and power consumption state Optimal Decision-making module.
Processing stability and power consumption state characteristic parameter deep learning module, using LSTM recurrent neural network to history
Electrical discharge machining data carry out deep learning and obtain prediction model, predict the variation tendency of real-time electrical discharge machining state, as
The predicted value of the current electrical discharge machining state of auxiliary identification is supplied to processing stability and power consumption state Optimal Decision-making module；
Processing stability is in conjunction with power consumption state Optimal Decision-making module from processing stability and power consumption state optimizing index
Statistical nature obtains the statistical characteristics that module provides and processing stability is mentioned with power consumption state characteristic parameter deep learning module
The predicted value of confession judges current electrical discharge machining state, when it is improper for determining current electrical discharge machining state,
The comprehensive optimal target value of stable-energy saving machining state is calculated and is regulated and controled currently by the electric control gear of spark-erosion machine tool
The parameter value of electrical discharge machining parameter.
It is illustrated in figure 2 data prediction flow chart.Whether data mining process succeeds, and depends primarily on the quality of data
Quality.Mass data is generated in spark-erosion machine tool operational process, these data are not only multi-source heterogeneous, mode is multifarious, and
And there is abnormal, gaps and omissions and repetition, it is difficult to support the work such as subsequent deep learning.Data prediction is exactly in data mining
With before use, to the initial data of acquisition carry out it is necessary extract, cleaning, fusion and a series of processing work of reduction, thus
The accuracy rate for improving the quality of data and data analysis result, preferably adapts to subsequent applications.
The present invention is achieved by the following technical solutions, the specific steps are as follows:
In the data preprocessing module, pass through data pick-up, data cleansing, data fusion and data regularization four
Step is that processing stability and power consumption state optimizing index excavate module, processing stability and power consumption state optimizing index and cluster
Analysis module stablizes machining state database and energy saving discharge condition database sharing module and processing stability and energy consumption
Characteristic condition parameter deep learning module provides required history electrical discharge machining data.
In data pick-up, the principal element and phase for influencing processing stability and energy consumption are found out in operation data first
Mutual relation forms the parameter to be extracted and parameter value on this basis.The electric discharge machine height of bed is contained in history data
The parameter that 30 measuring points such as frequency power equipment, Wire-conveying mechanism of quick walk, SERVO CONTROL, scouring part, machining state obtain, by extracting
Obtain 30 major parameters, including electrode wire travelling speed, wire electrode tension, medium temperature, discharge current, electric discharge pulsewidth, electric discharge
Between arteries and veins, voltage across poles, feed speed etc..
In the data preprocessing module, various operation datas are collected by various terminals, and there are many data matter
Amount problem, such as the missing, repetition, the exception that are worth etc..Data cleansing is exactly in 30 parameters for finding and handling after extracting
" dirty data ", it is by debugging value, missing values, repetition values, a series of processing of noise figure, undesirable data are " clear
Wash off ", further increase the quality of data.
In the data preprocessing module, data volume is simplified through data fusion, data regularization realization: passing through number
According to mapping, 30 parameters of multi-source, isomery are integrated, fusion is uniformly processed in a unified data warehouse so as to subsequent；
Reduction is carried out to data, exactly comprehensively considers distribution characteristics, human factor and controllable factor of electrical discharge machining data etc., from
Key parameter that is certain amount of, being capable of accurate description processing stability and power consumption state is selected in 30 parameter sets, to reduce
The dimension of data, saves data processing time, and characteristic parameter collection to be selected in this way is simplified to 8 characteristic parameters: electrode wire speed
Between degree, wire electrode tension, medium temperature, discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles and feed speed；
It is illustrated in figure 3 processing stability and power consumption state optimizing index excavates flow chart, specific excavation step is as follows:
1) using processing stability and the optimizing index of power consumption state as output variable Yi, it is with characteristic parameter to be selected
Input variable Xi, characteristic parameter to be selected is screened using nine kinds of algorithms respectively, calculates the score of each characteristic parameter；
2) it is handled using every kind scores of the extreme value normalization method to nine kinds of algorithms, result is limited
In [0,100] section, the mean value, extreme value and variance of each characteristic parameter are then repeatedly calculated.Using the mean value of scores,
Extreme value and variance are assessed and are sorted to the different degree of characteristic parameter, and processing stability and power consumption state optimizing index are carried out
Screening.It is applied in electrical discharge machining data, it is as shown in table 1 below that result is obtained after algorithm process.
Table 1 is using nine kinds of algorithms to the feature score after characteristic parameter to be selected screening
3) comprehensive score of characteristic parameter is analyzed, determining pair of the controllability and physical meaning of binding characteristic parameter
Processing stability and energy consumption have the sensitive features parameter of larger impact.In terms of scoring event, electrode wire travelling speed, wire electrode
Power, the mean value of medium temperature are lower, exclude the relatively low characteristic parameter of these scores.Several characteristic parameters of highest scoring, according to
Sequence from high to low is successively: discharge current, electric discharge pulsewidth, electric discharge arteries and veins between, voltage across poles, feed speed.
4) according to the analysis in step 2) and step 3), the selection result of characteristic parameter is assessed.Five higher spies of score
Levy in parameter, between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles, feed speed this five characteristic parameters belong to shape
State variable, the value of parameter are the results obtained under other controlled variable combined influences.
In summary it analyzes, finally determines between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles, feed speed five
Optimizing index of a characteristic parameter as processing stability and power consumption state.
It is illustrated in figure 4 the flow chart of processing stability Yu power consumption state optimizing index K-medoids clustering.Knot
Close practical production experience and history data distribution, between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles, feeding
Five optimizing index of speed carry out data prediction, input of the processing result as clustering.
Clustering finds the state class in history electrical discharge machining data set, cluster point using K-medoids algorithm
The purpose of analysis is that history electrical discharge machining data are carried out with processing stability status categories mark and electric discharge power consumption state classification mark
Note.
Processing stability cluster result is as follows, data point number such as table when choosing n=3, in cluster centre and each class
Shown in 2, divide the parameter distribution probability density figure of group as shown in Fig. 5,6,7.
2 processing stability of table divides group's cluster centre table
As can be seen from Figure 5:
Classification SA feature: the value range of discharge current between 3~10A (including endpoints thereof, similarly hereinafter), pulsewidth of discharging
Between 80~300 μ s, between the arteries and veins that discharges between 20~150 μ s, voltage across poles between 21~26V, feed speed 80~
128mm2Between/min.
As can be seen from Figure 6:
Classification SB feature: the value range of discharge current between 1~5A, put between 30~100 μ s by pulsewidth of discharging
Between electric arteries and veins between 0~60 μ s, voltage across poles is between 20~22.5V, and feed speed is in 100~145.5mm2Between/min；
As can be seen from Figure 7:
Classification SC feature: the value range of discharge current is between 0~1A, and pulsewidth of discharging is between 10~50 μ s, electric discharge
Between arteries and veins between 0~35 μ s, voltage across poles is between 18.5~22V, and feed speed is in 116.5~150mm2Between/min；
In conjunction with circumferential edge source electric spark production status, the record mark of the classification SA obtained when taking three cluster centres
It is set to stable state, the record of classification SB is demarcated as metastable condition, the record in classification SC is demarcated as unstable state.
Energy consumption cluster result is as follows, when choosing n=3, the data point number in cluster centre and each class as shown in table 3,
Divide the parameter distribution probability density figure of group as shown in Figure 8,9, 10.
3 energy consumption of table divides group's cluster centre table
As can be seen from Figure 8:
Classification EA feature: the value range of discharge current between 6~10A, discharge pulsewidth between 150~280 μ s,
Between electric discharge arteries and veins between 50~160 μ s, voltage across poles is between 20.5~24V, and feed speed is in 90~135mm2/ min it
Between；
As can be seen from Figure 9:
Classification EB feature: the value range of discharge current between 3~5A, put between 60~160 μ s by pulsewidth of discharging
Between electric arteries and veins between 0~70 μ s, voltage across poles is between 19.5~22.5V, and feed speed is in 75~115.5mm2/ min it
Between；
As can be seen from Figure 10:
Classification EC feature: the value range of discharge current is between 0~2A, and pulsewidth of discharging is between 10~80 μ s, electric discharge
Between arteries and veins between 0~45 μ s, voltage across poles is between 18.5~21V, and feed speed is in 20~90mm2Between/min.
In conjunction with the electric spark production status suggestion of circumferential edge source, the note of the classification EA obtained when taking three cluster centres
Record is demarcated as power save mode, and the record of classification EB is demarcated as energy consumption state, the record in classification EC is demarcated as highly energy-consuming state.
According to the calibration to data mode in cluster grouping, complete steady to being processed in existing history electrical discharge machining data
The classification of qualitative state marks, and the classification of stable state is labeled as SA, the classification of metastable condition is labeled as SB, unstable
The classification of state is labeled as SC；It completes to mark the classification of power consumption state in existing history data, the class of power save mode
It is not labeled as EA, the classification for the state that consumes energy is labeled as EB, and the classification of highly energy-consuming state is labeled as EC.
It is as shown in figure 11 to stablize machining state database sharing process.It is wrapped in one history electrical discharge machining data record
It is the history electrical discharge machining data record of SA to every class label containing recording mechanism, parameter name, parameter value and class label,
Calculate it and stablize the Euclidean distance between the record of each process data in machining state database, if distance less than 0.01,
Then think to stablize existing process data record in machining state database, is not repeated to record.Otherwise, by history electricity
Spark process data record, which is stored in, stablizes machining state database.
Energy saving discharge condition database sharing process is as shown in figure 12.It is wrapped in one history electrical discharge machining data record
It is the history electrical discharge machining data record of EA to every class label containing recording mechanism, parameter name, parameter value and class label,
Calculate itself and in energy saving discharge condition database each process data record between Euclidean distance, if distance less than 0.01,
Then think existing process data record in energy saving discharge condition database, is not repeated to record.Otherwise, by history electricity
Spark process data record is stored in energy saving discharge condition database.
The workflow of module is obtained for processing stability and power consumption state optimizing index statistical nature as shown in figure 13
Figure.Specific statistical flowsheet is as follows:
1) the processing on real-time data in the K period are imported, and count operation record number n in the period.
2) fracture of wire rejecting outliers are carried out one by one to n data record.It is in this way then the fracture of wire in data record is different
Normal number is added up.
The judgment basis of fracture of wire exception is referring to from stablizing parameters value range obtained in cooked mode library, when adopting
The parameter collected exceeds normal range (NR), then it is assumed that the data at the moment are fracture of wire exceptional value.
3) statistics current data record in discharge current, electric discharge pulsewidth, electric discharge arteries and veins between, voltage across poles, feed speed this five
Mean value, extreme value and the variance of a parameter, finally obtain mean value, extreme value and variance that each parameter obtains within the sampling period and
4 dimensions of fracture of wire frequency of abnormity characteristic value of totally 20 numerical value as judgement, to sentencing for processing stability and power consumption state
It is disconnected.
4) continue to read next data record and be handled, repeated the above process until the last item data record.
It is as shown in figure 14 the work flow diagram of processing stability and power consumption state characteristic parameter deep learning module.
1) history data is read, history data is subjected to feature normalization with Z-score method.
2) history data after standardization is input in LSTM recurrent neural network and is trained, predicted
Model.
The core of LSTM is neural network location mode, is transmitted backward by entire chain structure, and utilizes " thresholding "
The Sigmoid neural net layer and point multiplication operation of structure realize that the selectivity of information passes through.The calculation formula of LSTM algorithm is such as
Under:
ht=ot*tanh(ft*ct-1+it*c_int)
The model thinks that the output valve at current time is out gate ot, forget door ft, input gate itAnd last moment is single
First state ct-1Function.LSTM thinks historical series value and location mode parameter joint effect model prediction htValue.
3) with Z-score method by the discharge current acquired in real time, electric discharge pulsewidth, electric discharge arteries and veins between, voltage across poles, feeding
The value of this five characteristic parameters of speed is standardized；
4) by after standardization discharge current, electric discharge pulsewidth, between electric discharge arteries and veins, voltage across poles, feed speed this five features
The value of parameter is input to prediction model and is analyzed, and obtains predicted value by inverse feature normalization, predicted value is exported to processing
Stability and power consumption state Optimal Decision-making module.
With voltage across poles and the acquisition data instance of feed speed explanation using LSTM recurrent neural network training prediction mould
Type and the process for obtaining predicted value.
Processing stability and energy are carried out to the value of collected voltage across poles and feed speed in one section of continuous time first
State-detection is consumed, is divided between access 60 seconds, the value of continuous 60 voltage across poles and feed speed is as shown in figure 15.
As shown in figure 16, it using LSTM recurrent neural network training processing stability and energy consumption prediction model, is followed with using
Ring neural network RNN and multilayer perceptron MLP are compared, and find the prediction model obtained by LSTM recurrent neural network
Accuracy highest.
The relationship of voltage across poles, the predicted value of feed speed and actual value is as shown in figure 17.As can be seen from the figure it predicts
Error is lower, and predicted value is able to reflect the variation tendency of numerical value substantially, and the prediction effect of model is good.
It is as shown in figure 18 the work flow diagram of processing stability and power consumption state Optimal Decision-making module.Specific decision
Method is as follows:
1) processing stability and power consumption state Optimal Decision-making module receive processing stability and power consumption state optimizing index is united
Meter feature obtains the statistical characteristics for the real-time electrical discharge machining data that module transmits.If current electrical discharge machining state is sentenced
It is set to improper machining state, activates Optimal Decision-making program；
The standard of the improper machining state is: when processing stability state be in unsteady state (meta-stable or
Unstable state) either electric discharge power consumption state be in non-power save mode (consume energy or highly energy-consuming state).
2) multiple-objection optimization is carried out to processing stability and power consumption state using wolf pack algorithm, is stablized-energy conservation processing
State multiple-objection optimization disaggregation；
3) respectively from stablize search matching in machining state database and energy saving discharge condition database it is approximate stablize plus
Work parameter and energy saving discharge parameter are stablized the-optimal parameter objectives value of energy saving machining state synthesis；
4) according to the target value, the parameter of current electrical discharge machining parameter is regulated and controled by the electric control gear of spark-erosion machine tool
Value.
The above embodiments are only used to illustrate and not limit the technical solutions of the present invention, although referring to above-described embodiment to this
Invention is described in detail, those skilled in the art should understand that: can still modify to the present invention or
Equivalent replacement should all cover of the invention without departing from the spirit or scope of the invention, or any substitutions
In scope of the claims.
Claims (10)
1. stability of EDM and power consumption state Optimal Decision-making system based on deep learning, it is characterised in that: including number
Data preprocess module, processing stability and power consumption state optimizing index excavate module, processing stability and power consumption state optimization and refer to
It marks Cluster Analysis module, stablize machining state database and energy saving discharge condition database sharing module, processing stability and energy
State optimization indicator-specific statistics feature is consumed to obtain module, processing stability and power consumption state characteristic parameter deep learning module and add
Apt qualitative and power consumption state Optimal Decision-making module；
Data preprocessing module is used to extract, clearly the electrical discharge machining data that the electric control gear of spark-erosion machine tool obtains
It washes, merge and reduction process, be that subsequent processing stability and power consumption state optimizing index excavate module, processing stability and energy
Consume state optimization Cluster Analysis module, stablize machining state database and energy saving discharge condition database sharing module and
Processing stability and power consumption state characteristic parameter deep learning module provide the history electrical discharge machining data on basis；
Processing stability and power consumption state optimizing index excavate what module transmitted data preprocessing module using Feature Selection method
History electrical discharge machining data carry out mining analysis, the sensitive features parameter for influencing processing stability and energy consumption are obtained, as adding
Apt qualitative and power consumption state optimizing index is simultaneously supplied to processing stability and power consumption state optimizing index Cluster Analysis module；
Processing stability and power consumption state optimizing index Cluster Analysis module according to processing stability and power consumption state optimizing index,
Clustering carried out to the history electrical discharge machining data of data preprocessing module transmission, mark stablizes, meta-stable and unstable
Three processing stability status categories obtain the distribution feelings of each processing stability status categories in history electrical discharge machining data
Condition；Mark energy conservation, energy consumption and three electric discharge power consumption state classifications of highly energy-consuming, obtain the energy that respectively discharges in history electrical discharge machining data
Consume the distribution situation of status categories；
Stablize machining state database and energy saving discharge condition database sharing module is excellent with power consumption state according to processing stability
Change the cluster analysis result that Cluster Analysis module provides, to the history electrical discharge machining data of data preprocessing module transmission
It carries out stablizing processing classification screening, obtains stablizing machining state database；To the history electric spark of data preprocessing module transmission
Process data carries out energy conservation electric discharge classification screening, obtains energy saving discharge condition database；
Processing stability and power consumption state optimizing index statistical nature obtain module and are obtained by the electric control gear of spark-erosion machine tool
Real-time electrical discharge machining data simultaneously calculate its statistical characteristics, as the foundation for judging current electrical discharge machining state；
Processing stability and power consumption state characteristic parameter deep learning module, using LSTM recurrent neural network to history electric spark
Process data carries out deep learning and obtains prediction model, predicts the variation tendency of real-time electrical discharge machining state, auxiliary with predicted value
Help the current electrical discharge machining state of identification；
Processing stability judges current electrical discharge machining state with power consumption state Optimal Decision-making module, current electric when determining
When spark machining state is improper, the comprehensive optimal target value of stable-energy saving machining state is calculated and passes through electric spark
The electric control gear of lathe regulates and controls the parameter value of current electrical discharge machining parameter.
2. using the stability of EDM described in claim 1 based on deep learning and power consumption state Optimal Decision-making system
Decision-making technique, it is characterised in that successively sequentially include the following steps:
First step is to carry out data pick-up, data cleansing, data fusion and data reduction process by data preprocessing module；
Second step is to excavate module by processing stability and power consumption state optimizing index to excavate processing stability and energy consumption shape
State optimizing index；
Third step is to mark processing stability state by processing stability and power consumption state optimizing index Cluster Analysis module
Classification and electric discharge power consumption state classification；
Four steps is to stablize processing by stablizing machining state database and energy saving discharge condition database sharing module building
Slip condition database and energy saving discharge condition database；
5th step is to calculate real-time electrical discharge machining by processing stability and power consumption state optimizing index statistical nature module
The statistical characteristics of data, as the foundation for judging current electrical discharge machining state；
6th step is using processing stability and power consumption state characteristic parameter deep learning module to history electrical spark working number
Prediction model is obtained according to deep learning is carried out, the variation tendency of real-time electrical discharge machining state is predicted, with predicted value auxiliary identification
Current electrical discharge machining state；
7th step is that processing stability judges current electrical discharge machining state with power consumption state Optimal Decision-making module, when
When determining that current electrical discharge machining state is improper, it is calculated and stablizes-optimal the target value of energy saving machining state synthesis, and
Regulate and control the parameter value of current electrical discharge machining parameter by the electric control gear of spark-erosion machine tool.
3. decision-making technique according to claim 2, it is characterised in that:
It is that processing is steady by four data pick-up, data cleansing, data fusion and data regularization steps in the first step
It is qualitative to excavate module, processing stability and power consumption state optimizing index Cluster Analysis module, stabilization with power consumption state optimizing index
Machining state database and energy saving discharge condition database sharing module and processing stability and power consumption state characteristic parameter are deep
Spend the history electrical discharge machining data needed for study module provides；
In data pick-up, it is that electrical discharge machining data are obtained by the electric control gear of spark-erosion machine tool first, then extracts it
In 8 major parameters, including electrode wire travelling speed, wire electrode tension, medium temperature, discharge current, electric discharge pulsewidth, electric discharge arteries and veins
Between, voltage across poles and feed speed；
Fault value, missing values, repetition values and noise figure existing for 8 major parameters are cleaned up, data cleansing is completed, is improved
The quality of data；
Data fusion is carried out after data cleansing；Data fusion is first to extract the meta-attribute of data to be fused, then analyze number to be fused
According to the mapping relations with target data, by data to be fused and target data mapping, integrate to form data warehouse；
Data regularization is carried out after data fusion；Data regularization is for the data in data warehouse, advanced row data record statistics
Signature analysis, then remove attribute value and change small data record；Data record feature correlation analysis is carried out again, and removal feature is strong
Related data record, obtains history electrical discharge machining data set；
Wherein, attribute value changes small quantitative criteria are as follows: d (xi)≤0.01, d (xi) it is the x by after all data normalizationsiWith
The minimum range of other data；If d (xi)≤0.01, then it is assumed that xiAttribute value variation it is small；
The quantitative criteria of strong correlation are as follows: R >=0.01,
Wherein,It is xiThe mean value of middle all elements；It is using least square method to xi=f (x1,x2,…,xi-1,xi+1,…,
xn) fitting function,It is xiThe mean value of middle all elements, xijIt is xiJth each element；If R >=0.01, then it is assumed that data
xiWith other element strong correlations；
After above-mentioned data preprocessing operation, the history electrical discharge machining data on the basis as subsequent operation, the number are obtained
There are p parameters according to collection；P is more than or equal to 6 and is less than or equal to 12.
4. decision-making technique according to claim 2, it is characterised in that: the concrete operations of the second step are:
It is data acquisition movement first, times of collection M, M are natural number and its initial value is 0；In history electrical discharge machining data
Electrical discharge machining data of concentrated collection；
Followed by comparison compares the value of M and K, if M < K, executes the movement of calculating parameter score；If M >=K is jumped
Turn to execute the movement of calculating parameter value；
The movement of calculating parameter score is that the p item parameter in the secondary electrical discharge machining data is carried out using nine kinds of algorithms respectively
Screening, calculates the score of each parameter；Nine kinds of algorithms are respectively single argument feature selecting algorithm, Pearson correlation coefficients calculation
Method, decision Tree algorithms, L1 regularization algorithm, L2 regularization algorithm, random forests algorithm, random LASSO algorithm, apart from phase relation
Figure method and algorithm of support vector machine；The score is handled using extreme value standardized algorithm, result is limited in [0,
100] section, the scoring after being standardized；
Then the value of times of collection is made to add 1；The predetermined total degree of acquisition is K, and K is the natural number more than or equal to 3, times of collection
After value plus 1, execution comparison is jumped；
When M >=K, is jumped from comparison and execute the movement of calculating parameter value；
The movement of calculating parameter value is the scoring mean value that each parameter is calculated, scoring extreme value and scoring variance；According to each ginseng
Several scoring mean value, scoring extreme value and scoring variances are assessed and are sorted to the different degree of each parameter, and scoring mean value is pressed
The order of magnitude is classified, and scoring mean value is that the parameter of units is inessential parameter, and scoring mean value is that double-digit parameter is attached most importance to
Parameter is wanted, inessential parameter is removed, retains important parameter and as processing stability and power consumption state optimizing index, completion
Excavate the operation of processing stability and power consumption state optimizing index.
5. decision-making technique according to claim 2, it is characterised in that: in second step, to score mean value from high to low
Sequentially, using between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, five parameters of voltage across poles and feed speed as processing stability and
Power consumption state optimizing index.
6. decision-making technique according to claim 2, it is characterised in that: the concrete operations of the third step are:
Clustering carries out processing stability status categories using K-medoids algorithm, to history electrical discharge machining data
Mark and electric discharge power consumption state classification mark；
History electrical discharge machining data record is labeled as stable, meta-stable and unstable three state class by processing stability
Not；History electrical discharge machining data record is labeled as energy conservation, energy consumption and three status categories of highly energy-consuming by electric discharge energy consumption；
For the history electrical discharge machining data record comprising above-mentioned five processing stabilities Yu power consumption state optimizing index, use
K-medoids algorithm marks status categories；
Stable state classification be SA, the value range of discharge current between 3~10A, discharge pulsewidth between 80~300 μ s,
Between electric discharge arteries and veins between 20~150 μ s, voltage across poles is between 21~26V, and feed speed is in 80~128mm2Between/min；
Metastable condition classification be SB, the value range of discharge current between 1~5A, discharge pulsewidth between 30~100 μ s,
Between electric discharge arteries and veins between 0~60 μ s, voltage across poles is between 20~22.5V, and feed speed is in 100~145.5mm2/ min it
Between；
Unstable state classification is SC；The value range of discharge current between 0~1A, discharge pulsewidth between 10~50 μ s,
Between electric discharge arteries and veins between 0~35 μ s, voltage across poles is between 18.5~22V, and feed speed is in 116.5~150mm2/ min it
Between；
Power save mode classification be EA, the value range of discharge current between 6~10A, discharge pulsewidth between 150~280 μ s,
Between electric discharge arteries and veins between 50~160 μ s, voltage across poles is between 20.5~24V, and feed speed is in 90~135mm2Between/min；
Energy consumption status categories are EB, and the value range of discharge current between 3~5A, put between 60~160 μ s by pulsewidth of discharging
Between electric arteries and veins between 0~70 μ s, voltage across poles is between 19.5~22.5V, and feed speed is in 75~115.5mm2Between/min；
Highly energy-consuming status categories be EC, the value range of discharge current between 0~2A, discharge pulsewidth between 10~80 μ s,
Between electric discharge arteries and veins between 0~45 μ s, voltage across poles is between 18.5~21V, and feed speed is in 20~90mm2Between/min.
7. decision-making technique according to claim 2, it is characterised in that: the operation of the four steps is that processing is stablized in building
Slip condition database and energy saving discharge condition database；
Stablizing every electrical discharge machining data record in machining state database includes recording mechanism, parameter name and parameter value；Energy conservation
Every electrical discharge machining data record includes recording mechanism, parameter name and parameter value in discharge condition database；
The concrete operations for constructing stable machining state database are:
It is first directed to and initialization action, imports after third step clustering and be labeled with the history of processing stability status categories
Electrical discharge machining data record, summary journal the quantity n, n calculated in history electrical discharge machining data is natural number；It initializes whole
Number variable i is equal to 1；
Then read action is executed, the parameter value and this process data note of i-th history electrical discharge machining data record are read
Record corresponding status categories；Then judge this process data record status categories whether be SA, if it is judged that be it is no,
The operation for skipping this operation record is then executed, and jumps the increment operator for executing i；
If it is judged that be it is yes, then execute to calculate and this history electrical discharge machining data record and stablize machining state database
In existing each process data record Euclidean distance s operation；
The operation for calculating Euclidean distance s is:
If stablizing machining state database is empty database, the deposit of this history electrical discharge machining data record is stablized and is added
Work slip condition database；
It has been stored with process data record in machining state database if stablized, this history electrical discharge machining data have been remembered
The parameter value of record with stablize in machining state database that corresponding parameter value carries out Europe in existing each process data record
The calculating of family name's distance s；
If this history electrical discharge machining data record and stablizing existing each process data in machining state database
Euclidean distance s between record is all larger than 0.01, then illustrates this history electrical discharge machining data record and stablize machining state
Existing each process data record is all different in database, then is deposited into this history electrical discharge machining data record surely
Determine machining state database, then executes the increment operator of i；
If this history electrical discharge machining data record and stablizing existing each process data in machining state database
Euclidean distance s between record is respectively less than 0.01, then executes the operation for skipping the operation record, and jumps and execute grasping from increasing for i
Make；
The increment operator of i is that the value of i is made to increase by 1, then judges whether i is greater than n, if i is less than or equal to n, jumps and execute reading
Take movement；If i is greater than n, terminate the operation for constructing stable machining state database；
The concrete operations for constructing energy saving discharge condition database are:
It is first directed to and initialization action, imports after third step clustering and be labeled with the history electric spark of power save mode classification
Process data record, summary journal the quantity m, m for calculating history electrical discharge machining data are natural number；Initialize integer variable j
Equal to 1；
Then read action is executed, the parameter value and this process data note of j-th strip history electrical discharge machining data record are read
Record corresponding status categories；Then judge this process data record status categories whether be EA, if it is judged that be it is no,
The operation for skipping this process data record is then executed, and jumps the increment operator for executing j；
If it is judged that be it is yes, then execute calculate this process data record with it is existing every in energy saving discharge condition database
The operation of the Euclidean distance s of one process data record；
The operation for calculating Euclidean distance e is:
If energy saving discharge condition database is empty database, the deposit energy conservation of this history electrical discharge machining data record is put
Electricity condition database；
If being stored with process data record in energy saving discharge condition database, this history electrical discharge machining data are remembered
The parameter value of record carries out Europe with corresponding parameter value in each process data record existing in energy saving discharge condition database
The calculating of family name's distance e；
If existing each process data in this history electrical discharge machining data record and energy saving discharge condition database
Euclidean distance e between record is all larger than 0.01, then illustrates this electrical discharge machining data record and energy saving discharge condition data
Existing each process data record is all different in library, then this electrical discharge machining data record is stored in energy saving discharge condition number
According to library, the increment operator of j is then executed；
If existing each process data in this history electrical discharge machining data record and energy saving discharge condition database
Euclidean distance e between record is less than or equal to 0.01, then executes the operation for skipping the operation record, and jumps and execute increasing certainly for j
Operation；
The increment operator of j is that the value of j is made to increase by 1, then judges whether j is greater than m, if j is less than or equal to m, jumps and execute reading
Take movement；If j is greater than m, terminate the operation for constructing energy saving discharge condition database.
8. decision-making technique according to claim 2, it is characterised in that: in the 5th step, processing stability with
Power consumption state optimizing index statistical nature obtains module and obtains real-time electrical spark working number by the electric control gear of spark-erosion machine tool
According to calculating between discharge current, electric discharge pulsewidth, electric discharge arteries and veins, voltage across poles, feed speed this five optimizing index are in window time
Mean value, extreme value, variance and fracture of wire frequency of abnormity, using statistical characteristics as the foundation for recognizing current electrical discharge machining state；
Processing stability and power consumption state optimizing index statistical nature obtain module at work, import the real-time electricity in the K period
Spark process data record, processes summary journal quantity n in the statistics K period, and initializes integer variable i equal to 1, initializes fracture of wire
Frequency of abnormity integer variable is equal to 0；
Then read action is executed, the process data of i-th operation record is read；
After reading, fracture of wire abnormality detection is carried out；In case of fracture of wire, then the value of fracture of wire frequency of abnormity integer variable is first made to add 1, then
Execute statistics movement；If fracture of wire does not occur, directly execution statistics movement；
Statistics movement is between the fracture of wire number and discharge current, electric discharge pulsewidth, electric discharge arteries and veins counted in the K period in operation record,
Mean value, extreme value and the variance of five parameters of voltage across poles and process velocity；
After statistics movement, so that the value of integer variable i is added 1, then judge whether the operation record in the K period reads and finish, specifically
It is to judge whether i value is greater than n value, if i value is less than or equal to n value, jumps and execute read action again；If i value is greater than n
Value, then send all statistical characteristics at this time to processing stability and power consumption state Optimal Decision-making module.
9. decision-making technique according to claim 2, it is characterised in that: the operation of the 6th step be processing stability with
Power consumption state characteristic parameter deep learning:
1) the history electrical discharge machining data for reading data preprocessing module transmission, with Z-score method by history electrical spark working
Number is according to progress feature normalization；
2) the history electrical discharge machining data after standardization are input in LSTM recurrent neural network and are trained, predicted
Model
The core of LSTM is neural network location mode, is transmitted backward by entire chain structure, and utilizes threshold structure
Sigmoid neural net layer and point multiplication operation realize that the selectivity of information passes through；The calculation formula of LSTM algorithm is as follows:
ht=ot*tanh(ft*ct-1+it*c_int)
In prediction model, the output valve at current time is out gate ot, forget door ft, input gate itAnd last moment cell-like
State ct-1Function；
3) with Z-score method by the discharge current acquired in real time, electric discharge pulsewidth, electric discharge arteries and veins between, voltage across poles, feed speed this
The value of five characteristic parameters is standardized；
4) by after standardization discharge current, electric discharge pulsewidth, between electric discharge arteries and veins, voltage across poles, feed speed this five characteristic parameters
Value be input to prediction model and analyzed, predicted value is obtained by inverse feature normalization, predicted value is exported and is stablized to processing
Property with power consumption state Optimal Decision-making module.
10. decision-making technique according to claim 2, it is characterised in that: the operation of the 7th step is processing stability
With power consumption state Optimal Decision-making:
1) processing stability and power consumption state Optimal Decision-making module receive processing stability and power consumption state optimizing index statistics is special
Sign obtains the statistical characteristics for the real-time electrical discharge machining data that module transmits；If it is determined that current electrical discharge machining state is non-
Normal process state activates Optimal Decision-making program；
The standard of the improper machining state is: when processing stability state is in unsteady state or electric discharge power consumption state
In non-power save mode；
2) multiple-objection optimization is carried out to processing stability and power consumption state using wolf pack algorithm, is stablized-energy conservation machining state
Multiple-objection optimization disaggregation；
3) from stablizing in machining state database and energy saving discharge condition database, search matching is approximate to stablize processing ginseng respectively
Several and energy saving discharge parameter is stablized the-optimal parameter objectives value of energy saving machining state synthesis；
4) according to the target value, the parameter value of current electrical discharge machining parameter is regulated and controled by the electric control gear of spark-erosion machine tool.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811145126.2A CN109396576B (en) | 2018-09-29 | 2018-09-29 | Deep learning-based electric spark machining stability and energy consumption state optimization decision system and decision method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811145126.2A CN109396576B (en) | 2018-09-29 | 2018-09-29 | Deep learning-based electric spark machining stability and energy consumption state optimization decision system and decision method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109396576A true CN109396576A (en) | 2019-03-01 |
| CN109396576B CN109396576B (en) | 2020-02-04 |
Family
ID=65466543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811145126.2A Active CN109396576B (en) | 2018-09-29 | 2018-09-29 | Deep learning-based electric spark machining stability and energy consumption state optimization decision system and decision method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109396576B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070145A (en) * | 2019-04-30 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | LSTM wheel hub single-item energy consumption prediction based on increment cluster |
| CN110280852A (en) * | 2019-07-01 | 2019-09-27 | 浙江科技学院 | A kind of wire cutting electrical parameter control strategy based on energy consumption prediction model |
| CN110442099A (en) * | 2019-08-05 | 2019-11-12 | 湘潭大学 | A kind of numerical control processing parameter optimizing method based on shot and long term memory |
| CN110544111A (en) * | 2019-08-05 | 2019-12-06 | 北京市天元网络技术股份有限公司 | ETC customer obtaining method and device based on operator big data |
| CN111008919A (en) * | 2019-12-19 | 2020-04-14 | 国家电网有限公司 | Anti-electricity-stealing system based on artificial intelligence |
| CN111966045A (en) * | 2020-07-08 | 2020-11-20 | 航天科工深圳（集团）有限公司 | Machine tool crash monitoring method and device, terminal equipment and storage medium |
| CN112987666A (en) * | 2021-02-09 | 2021-06-18 | 浙大城市学院 | Power plant unit operation optimization regulation and control method and system |
| CN113703396A (en) * | 2021-07-26 | 2021-11-26 | 北京市机械施工集团有限公司 | Intelligent upgrading method of numerical control cutting equipment based on intelligent terminal |
| CN115167279A (en) * | 2022-09-07 | 2022-10-11 | 中科航迈数控软件(深圳)有限公司 | Energy consumption prediction method and system for numerical control machine tool and related equipment |
| CN115840431A (en) * | 2023-02-27 | 2023-03-24 | 一夫科技股份有限公司 | Production control method and system for II-type anhydrous gypsum |
| CN116307938A (en) * | 2023-05-17 | 2023-06-23 | 成都瑞雪丰泰精密电子股份有限公司 | Health state assessment method for feeding system of machining center |
| CN116842415A (en) * | 2023-09-01 | 2023-10-03 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020074238A1 (en) * | 1998-10-26 | 2002-06-20 | Mayer Steven T. | Method and apparatus for uniform electropolishing of damascene ic structures by selective agitation |
| CN1985013A (en) * | 2004-04-29 | 2007-06-20 | U.I.T.有限责任公司 | Method for modifying or producing materials and joints with specific properties by generating and applying adaptive impulses, normalizing energy thereof and pauses therebetween |
| CN106354106A (en) * | 2016-08-19 | 2017-01-25 | 广东省自动化研究所 | Data processing system based on MES |
| CN107220734A (en) * | 2017-06-26 | 2017-09-29 | 江南大学 | CNC Lathe Turning process Energy Consumption Prediction System based on decision tree |
| CN107817890A (en) * | 2017-10-31 | 2018-03-20 | 郑州云海信息技术有限公司 | A kind of high density rack load linkage energy efficiency management design method based on BP algorithm |
| CN107891199A (en) * | 2017-12-01 | 2018-04-10 | 中州大学 | Spark discharge condition checkout gear and recognition methods |
-
2018
- 2018-09-29 CN CN201811145126.2A patent/CN109396576B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020074238A1 (en) * | 1998-10-26 | 2002-06-20 | Mayer Steven T. | Method and apparatus for uniform electropolishing of damascene ic structures by selective agitation |
| CN1985013A (en) * | 2004-04-29 | 2007-06-20 | U.I.T.有限责任公司 | Method for modifying or producing materials and joints with specific properties by generating and applying adaptive impulses, normalizing energy thereof and pauses therebetween |
| CN106354106A (en) * | 2016-08-19 | 2017-01-25 | 广东省自动化研究所 | Data processing system based on MES |
| CN107220734A (en) * | 2017-06-26 | 2017-09-29 | 江南大学 | CNC Lathe Turning process Energy Consumption Prediction System based on decision tree |
| CN107817890A (en) * | 2017-10-31 | 2018-03-20 | 郑州云海信息技术有限公司 | A kind of high density rack load linkage energy efficiency management design method based on BP algorithm |
| CN107891199A (en) * | 2017-12-01 | 2018-04-10 | 中州大学 | Spark discharge condition checkout gear and recognition methods |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070145B (en) * | 2019-04-30 | 2021-04-27 | 天津开发区精诺瀚海数据科技有限公司 | LSTM hub single-product energy consumption prediction based on incremental clustering |
| CN110070145A (en) * | 2019-04-30 | 2019-07-30 | 天津开发区精诺瀚海数据科技有限公司 | LSTM wheel hub single-item energy consumption prediction based on increment cluster |
| CN110280852A (en) * | 2019-07-01 | 2019-09-27 | 浙江科技学院 | A kind of wire cutting electrical parameter control strategy based on energy consumption prediction model |
| CN110442099A (en) * | 2019-08-05 | 2019-11-12 | 湘潭大学 | A kind of numerical control processing parameter optimizing method based on shot and long term memory |
| CN110544111A (en) * | 2019-08-05 | 2019-12-06 | 北京市天元网络技术股份有限公司 | ETC customer obtaining method and device based on operator big data |
| CN111008919A (en) * | 2019-12-19 | 2020-04-14 | 国家电网有限公司 | Anti-electricity-stealing system based on artificial intelligence |
| CN111966045A (en) * | 2020-07-08 | 2020-11-20 | 航天科工深圳（集团）有限公司 | Machine tool crash monitoring method and device, terminal equipment and storage medium |
| CN112987666B (en) * | 2021-02-09 | 2022-05-20 | 浙大城市学院 | Power plant unit operation optimization regulation and control method and system |
| CN112987666A (en) * | 2021-02-09 | 2021-06-18 | 浙大城市学院 | Power plant unit operation optimization regulation and control method and system |
| CN113703396A (en) * | 2021-07-26 | 2021-11-26 | 北京市机械施工集团有限公司 | Intelligent upgrading method of numerical control cutting equipment based on intelligent terminal |
| CN115167279A (en) * | 2022-09-07 | 2022-10-11 | 中科航迈数控软件(深圳)有限公司 | Energy consumption prediction method and system for numerical control machine tool and related equipment |
| CN115840431A (en) * | 2023-02-27 | 2023-03-24 | 一夫科技股份有限公司 | Production control method and system for II-type anhydrous gypsum |
| CN115840431B (en) * | 2023-02-27 | 2023-05-16 | 一夫科技股份有限公司 | Production control method and system for II-type anhydrous gypsum |
| CN116307938A (en) * | 2023-05-17 | 2023-06-23 | 成都瑞雪丰泰精密电子股份有限公司 | Health state assessment method for feeding system of machining center |
| CN116842415A (en) * | 2023-09-01 | 2023-10-03 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
| CN116842415B (en) * | 2023-09-01 | 2023-12-26 | 广东台正精密机械有限公司 | Remote monitoring method, system and medium for mirror surface electric discharge machine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109396576B (en) | 2020-02-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109396576A (en) | Stability of EDM and power consumption state Optimal Decision-making system and decision-making technique based on deep learning | |
| Cheng et al. | Data and knowledge mining with big data towards smart production | |
| Addeh et al. | Control chart pattern recognition using RBF neural network with new training algorithm and practical features | |
| Kuo et al. | Application of a hybrid of genetic algorithm and particle swarm optimization algorithm for order clustering | |
| Chang | A comparative study of artificial neural networks, and decision trees for digital game content stocks price prediction | |
| Mori | State-of-the-art overview on data mining in power systems | |
| CN112987666B (en) | Power plant unit operation optimization regulation and control method and system | |
| CN101477374B (en) | Continuous casting bleed-out time sequence spacing combined diagnosis prediction method based on fuzzy neural network | |
| CN115630839B (en) | Intelligent feedback production regulation and control system based on data mining | |
| Ishibashi et al. | GFRBS-PHM: A genetic fuzzy rule-based system for phm with improved interpretability | |
| Liu et al. | An overview of decision tree applied to power systems | |
| Li et al. | A new method of identification of complex lithologies and reservoirs: task-driven data mining | |
| CN106780018A (en) | A kind of stock certificate data mining model reappeared based on history and system | |
| CN108985455A (en) | A kind of computer application neural net prediction method and system | |
| CN109726740A (en) | A kind of trade power consumption behavior analysis method based on clustering | |
| Sun et al. | A novel Day-ahead electricity price forecasting using multi-modal combined integration via stacked pruning sparse denoising auto encoder | |
| CN115238197A (en) | Expert thinking model-based field business auxiliary analysis method | |
| Qin et al. | A novel online tool condition monitoring method for milling titanium alloy with consideration of tool wear law | |
| Wang et al. | A spatiotemporal feature learning-based RUL estimation method for predictive maintenance | |
| Zegarra et al. | Comparison of CNN and CNN-LSTM architectures for tool wear estimation | |
| CN117725537A (en) | Real-time metering data processing platform | |
| Kutschenreiter-Praszkiewicz | Application of artificial neural network for determination of standard time in machining | |
| Saravanakumar et al. | An efficient ensemble model for forecasting time series analysis of crypto currency data using machine learning | |
| Hallman | A comparative study on Linear Regression and Neural Networks for estimating order quantities of powder blends | |
| Zhang et al. | Machine learning applications in Cyber-Physical Production Systems: a survey |
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 |