CN105912773A - Novel intelligent stamping process design method based on data mining technology - Google Patents

Abstract

The present invention discloses a novel intelligent stamping process design method based on a data mining technology, relates to data mining, knowledge engineering and stamping process technologies, and belongs to the multidisciplinary crossed fields. The method comprises six processes of (1) establishing a uniform feature information model supporting the system construction; (2) establishing a relevance knowledge representation model; (3) constructing an inference system for data mining modeling; (4) researching an evaluation technology of a data mining model; (5) performing intelligent integration of local processes; and (6) performing general integration, test and improvement of the system. The stamping process design method is mainly used for transforming and upgrading the system, and accuracy and efficiency of stamping process design are improved.

Description

A kind of new based on data mining technology intelligent punching press process design method

Technical field

This method (technology) relates to the technology such as data mining, knowledge engineering, Sheet Metal Forming Technology, belongs to multi-crossed disciplines field, It is mainly used in promoting traditional stamping process design method, improves the accuracy and efficiency of stamping process design.

Background technology

Along with developing rapidly of information technology, CAPP (Computer Aided Process Planning, advise by CAP Draw) application in manufacturing enterprise is more and more extensive, more and more deep, new knowledge technology such as KBE (Knowledge Based Engineering, knowledge based engineering) technology, data mining technology etc. and new information technology such as ontology, Network technologies etc. continue to bring out and fast development.Under the background that global economic integration trend strengthens day by day, all trades and professions are compeled Highly necessary ask and constantly absorb new knowledge technology, information technology is accelerated to promote conventional industries, it is achieved conventional industries More educated, information-based and intelligent, thus establish oneself in an unassailable position under violent global competition.

Stamping technology is as the manufacturing a kind of staple product processing method of Pillar Industry, and especially urgent needs utilizes Knowledge technology, information technology upgrade promote traditional means and technology, it is achieved highly intelligent production and processing.Will the warp of expert Test knowledge and expert people itself separate so that it is can independent Decision-making Function, reach to break away from the dependence to technique expert, Be easy to process knowledge inherit and integrated, have the target of higher big ability to solve problem.

At present, data mining technology has become the focus of stamping parts Intelligent art design studies, has the most carried out numerous studies And application practice, also going out many achievements, data mining technology the most fully shows its superiority obtaining process knowledge: only Thering is provided data complete, accurate, reliable, data mining technology just can obtain the knowledge of needs, not only avoids neck The heavy dependence of territory expert, and the knowledge obtained is than traditional knowledge more high-quality, it is easier to and automated system linking.Cause It is the things internal relation (knowledge inferred from data (phenomenon of things) by algorithm strict, science for these knowledge Essence), the most accurately, it is easy to change into quantization knowledge, thus realize, as controlling knowledge, the automation process that knowledge based drives CAD design；But these researchs and application practice display that, data mining technology is in the application in stamping parts technological design field Still suffer from wretched insufficiency: stamping parts technological design is local intelligence state, and overall intelligence, automaticity are the most relatively low, Causing system accuracy, efficiency and problem-solving ability the most limited, the target that distance academia and industry are expected is still There is the biggest distance.It is in particular in two aspects:

The first, mostly current data digging technology applied research in stamping parts technological design is locality applied research, or The excavation that data mining technology is used for technological design data individually designs to complete initial process, or is used separately for place Reason emulation data realize the correction to first design to obtain knowledge, the data mining in two stages are not integrated formation One organic unity process, thus cause system intelligent, automaticity limited, and then the two organic cooperation of impact, harm Hinder the raising of the accuracy and efficiency of technological design.

The second, the carrying out of the data mining process still strong participation relying on veteran expert, not only hampers system Intelligent, automaticity, and seriously undermined system problem-solving ability.

Summary of the invention

Intelligent for stamping parts technological design field, automaticity is relatively low, hinders the accuracy and efficiency of technological design, Weakening the problems such as system problem-solving ability, the present invention proposes a kind of new based on data mining technology intelligent punching press simultaneously Process design method.

The technical solution used in the present invention is: a kind of new based on data mining technology intelligent punching press process design method, bag Include procedure below:

Process one: set up the unified feature information model supporting that native system builds；

On the basis of research is currently based on the stamping process design specialist system characteristic information expression model of data mining, research is melted Enter KBE technology, characteristic information that ontology is increased newly, set up the Feature concept that whole Process Planning is complete, unified Definition.

Process two: set up relatedness knowledge representation model；

Analyzing in stamping process design involved by CAD process based on data mining and CAE process based on data mining Function, data, characteristic parameter (parameter influential on forming quality object function) and shape facility and technique are set On the basis of the cause effect relation of knowledge, knowledge representation based on Ontology is used to set up the relatedness knowledge table of two processes Reach model.

Process three: build the inference system for Modeling of Data Mining；

Process four: data mining model assessment technique；

The evaluation index of data mining model, builds accurate, complete assessment indicator system；The meter of agriculture products weight Calculation method and the computational methods of quantification of targets value；Set up and comprehensively utilize these index quantification metric data mining model performances Method.

Process five: the intelligent integration of local process；

This process comprises two aspects: one is initial designs process collection based on Intelligent data mining technology and association knowledge model Becoming, two is that simulation feedback design process based on Intelligent data mining technology and association knowledge model is integrated.

Process six: the overall assembly of system, test, perfect.

System is divided into four layers: user interface layer, functional module layer, technology platform layer, data Layer.

Further, the described feature information model in process one includes: part feature information model, manufacture resource feature are believed Breath model, process planning feature information model.

Further, described process three is divided into again three steps:

Step one: the decomposition technique of design objective and determine the associated data set for excavating.Utilize the RBR of KBE technology Way of realization completes complex task resolves into some enforceability subtasks, and determines the data set that this subtask is relevant；

Step 2: build the data mining case searching based on Ontology towards stamping process design: first, from field Collect abundant vocabulary, term, and using the vocabulary being widely recognized as in the industry, term as the concept of body, simultaneously that body is general Thought is categorized as background concepts and result concept；Secondly, connecting each other between concept in analysis field, accurately obtain field originally The various relations of body；Finally, all concepts of body are attached by the relation between them, form a tree network Network structure, is i.e. the data structure of body example knowledge base；

Step 3: determine the knowledge reasoning scheme of Modeling of Data Mining subsystem: first, determines the overall retrieval scheme of system, This project uses the 2-level search scheme that knowledge guides and nearest neighbor algorithm combines；Secondly, the similarity algorithm of example is determined； Finally, it is verified that the effect of similarity algorithm and the effect of retrieval scheme.

Further, the functional module layer in described process six includes 6 basic functional modules: knowledge retrieval module, data Excavate Case retrieval course module, data mining case retrieving module, data mining and Knowledge Discovery module, process knowledge study mould Block and data mining event selection module.

The invention have the benefit that and significantly improve the intellectuality of stamping process design process, automaticity, overcome setting The dependence of meter person's experience, is greatly improved the ability of specialist system solving practical problems and designing quality and efficiency, greatly promotees Enter the manufacturing development of particularly diel of China's process industry；It addition, the present invention is substantially reduced answering of data mining technology With threshold, make data mining technology be easier to concrete application and combine, thus promote that it extensively and profoundly should in every field With.

Accompanying drawing explanation

A kind of new based on data mining technology intelligent punching press process design method flow chart of Fig. 1；

Fig. 2 part feature information model；

Fig. 3 manufacture resource feature information model；

Fig. 4 processing technique feature information model；

Fig. 5 data mining based on body case searching；

The relation of Fig. 6 concept Ci ', Ci and leaves (Ci^)；

The body construction system of Fig. 7 data mining model Performance Evaluation index；

Fig. 8 data mining model based on body and CBR evaluation procedure false code；

Fig. 9 system general frame；

Figure 10 knowledge retrieval module；

Figure 11 data mining Case retrieval course module；

Figure 12 data mining case retrieving module；

Figure 13 data-mining module；

Figure 14 process knowledge study module；

Figure 15 data mining event selection module.

Detailed description of the invention

The invention will be further described with embodiment below in conjunction with the accompanying drawings.

Fig. 1 is method for designing flow chart of the present invention, and specifically details are as follows:

(1) the unified feature information model supporting that native system builds is set up；

On the basis of research is currently based on the stamping process design specialist system characteristic information expression model of data mining, research is melted Enter KBE technology, characteristic information that ontology is increased newly, set up the Feature concept that whole Process Planning is complete, unified Definition.

According to feature modeling thought, by Feature Extraction Technology, set up the part feature information model supporting native system construction, Manufacture resource feature information model, process planning feature information model.In each feature information model building process, adopt With successively decomposition strategy, feature is specifically dissected expression step by step from being abstracted into.In part feature information model, the most abstract Feature such as shape facility, accuracy characteristic, material feature, performance characteristic etc., these abstract features can specifically resolve into very Many subcharacters, by that analogy, as shown in Figure 2.In like manner, in manufacture resource feature information model, the most abstract feature Being lathe, fixture, measurer, cutter etc., these abstract features can specifically resolve into a lot of subcharacter, by that analogy, As shown in Figure 3；In process planning feature information model, the most abstract feature is that processing method selects, process equipment selects Select, part processing sequence, work step etc., these abstract features can specifically resolve into a lot of subcharacter, by that analogy, as Shown in Fig. 4.

(2) relatedness knowledge representation model is set up；

Analyzing in stamping process design involved by CAD process based on data mining and CAE process based on data mining Function, data, characteristic parameter (parameter influential on forming quality object function) and shape facility and technique are set On the basis of the cause effect relation of knowledge, knowledge representation based on Ontology is used to set up the relatedness knowledge table of two processes Reach model.

(3) inference system for Modeling of Data Mining is built；This link includes three steps:

Step one: the decomposition technique of design objective and determine the associated data set for excavating.Utilize the RBR of KBE technology Way of realization completes complex task resolves into some enforceability subtasks, and determines the data set that this subtask is relevant.

Step 2: structure is towards the data mining case searching based on Ontology of stamping process design:

First, from field, collect abundant vocabulary, term, and using the vocabulary being widely recognized as in the industry, term as body Concept, is categorized as background concepts and result concept simultaneously by Ontological concept；

Secondly, connecting each other between concept in analysis field, accurately obtain the various relations of domain body；

Finally, all concepts of body are attached by the relation between them, form a tree network structure, i.e. Being the data structure of body example knowledge base, concept is the most abstract, and its position is more on the top of tree.The top of tree is the most abstract The most general concept " Process plan data excavation example ", the tip that lowermost end is i.e. set be the such as material of concept most specifically " PS ", " PPS " etc., in body tree structure, the concept of leaves part (leaf node) is also the label of a data base simultaneously, The inside can store the eigenvalue of corresponding concepts.These leaf nodes can regard a set as, and each of real world has Body example, actual is exactly that this group leaf node is carried out assignment, thus becomes an example of whole body Case, such as Fig. 5 institute Show.So, the data base that the specific features value of the example of a large amount of real worlds is stored in correspondence just constitutes case searching, stores Domain knowledge.

Step 3: determine the knowledge reasoning scheme of Modeling of Data Mining subsystem:

First, determining the overall retrieval scheme of system, Case retrieval course is to utilize retrieval information retrieve from the case searching of source and select to dive At available source example, and the similarity between new example and source example being made rationally judge, its core technology includes retrieval Strategy and the design of Similarity algorithm and selection.The strategy generally used is nearest neighbor method and the combination of knowledge daoyin technique.The present invention Using the 2-level search scheme that knowledge guides and nearest neighbor algorithm combines, first order retrieval is semantic understanding based on body inspection Rope, its function is equivalent to knowledge daoyin technique, first the solution of problem narrows down to a suitable solution space；The second level is to utilize The numerical computations of nearest neighbor algorithm, and the result example number returned by arranging certain threshold value to control；How to access this The calculating of body case searching and semantic similarity is two core links understanding retrieval based on Ontology, and it decides retrieval Win or lose.

Secondly, the similarity algorithm of example is determined；

The Arithmetic of Semantic Similarity that the present invention builds is as follows:

Consider ND (Node Distance) and IC (Information Content) similarity, tie in a proper manner Close, Semantic Similarity Measurement accuracy can be effectively improved.Under the inspiration of this theory, the present invention proposes new ND-IC Similarity Measure method i.e. W-IC-ND (Weighted Information Content and Node Distance).Assuming that concept bunch C '=[C1 ', C2 ' ..., Ci ' ..., Cn '] come from user's query, concept bunch C=[C1, C2 ..., Ci ..., Cn] come from coupling Body lexicon.The overall similarity of concept bunch C ' and C is expressed as SimIC-ND (C ', C).For calculate SimIC-ND (C ', C), first calculate each concept similarity to (Ci ', Ci) in these two groups of concepts bunch, be designated as SimIC-ND (Ci ', Ci). When calculating SimIC-ND (Ci ', Ci), first we calculate by IC Similarity Measure method, is designated as SimIC (Ci ', Ci), then Calculate by ND method, be designated as SimND (Ci ', Ci), the most again will the two weighted sum.Its detailed process is described as follows:

1) value of SimIC (Ci ', Ci) is calculated

The IC Similarity value of two concepts refers to that it has the degree of common information.Assuming that concept Ci^ be concept Ci ' and Ci Nearly common ancestor.In body construction, all concepts from concept Ci^ and classification (including concept Ci^) are defined as The leaf of concept Ci^, is denoted as leaves (Ci^).Relation such as Fig. 6 of concept Ci ', Ci, Ci^ and leaves (Ci^) Shown in.

Obviously, the leaf of a concept comprises more rich and comprehensive semantic content than a simple concept, more can distinguish it Difference with other concept.So, in a domain body, a concept leaf more can one concept of explication.Then The present invention with concept leaf as definition IC similarity sole indicator.

At present, the frequency that the value of IC similarity occurs in archives by estimating concept obtains, according to the reason in information theory Reading, the IC value of concept C can calculate so by formula, if we calculate IC similarity with concept leaf, and formula phase It is expressed as with answering,

$IC\left({C_i}^{^}\right)=log\frac{1}{P\left(leaves\right({C_i}^{^}\left)\right)}---\left(1\right)$

Wherein P (leaves (Ci^)) is the probability that in concept leaf, any one example occurs.

Then, the value computing formula of similarity SimIC (Ci ', Ci) is:

${\mathrm{Sim}}_{IC}(C_i^{\′},C_i)=IC\left({C_i}^{^}\right)=log\frac{1}{P\left(leaves\right({C_i}^{^}\left)\right)}---\left(2\right)$

Finally, the value of standardization IC similarity, as shown in formula (3).

${\stackrel{\‾}{Sim}}_{IC}(C_i^{\′},C_i)=\frac{{\mathrm{Sim}}_{IC}(C_i^{\′},C_i)}{\underset{i=1}{\overset{n}{\Σ}}{\mathrm{Sim}}_{IC}(C_i^{\′},C_i)}---\left(3\right)$

2) value of SimND (Ci ', Ci) is calculated

Definition is 1 in body layer aggregated(particle) structure, if there being two concept node Ci ', Ci, note len (Ci ', Ci) be Ci ', Ci it Between shortest path.

Definition 2, in body layer aggregated(particle) structure, if there being a node Ci, remembers that its degree of depth is depth (Ci)=len (root, Ci), wherein Root is the root node in structure.

So, in body layer aggregated(particle) structure, to any two node Ci ', Ci, then defining its ND similarity is:

${\mathrm{Sim}}_{ND}(C_i^{\′},C_i)=\frac{2\×depth\left({C_i}^{^}\right)}{len(C_i,{C_i}^{^})+len(C_i^{\′},{C_i}^{^})+2\×depth\left({C_i}^{^}\right)}---\left(4\right)$

After utilizing formula (4) to calculate the ND similarity of all of concept pair, formula (5) is utilized to be standardized.

${\stackrel{\‾}{Sim}}_{ND}(C_i^{\′},C_i)=\frac{{\mathrm{Sim}}_{ND}(C_i^{\′},C_i)}{\underset{i=1}{\overset{n}{\Σ}}{\mathrm{Sim}}_{ND}(C_i^{\′},C_i)}---\left(5\right)$

3) SimIC-ND (Ci ', Ci) and the value of SimIC-ND (C ', C) are calculated

When the value of SimIC (Ci ', Ci) and SimND (Ci ', Ci) is obtained and after standardization, obtained by formula (6) SimIC-ND (Ci ', Ci), obtains the value of SimIC-ND (C ', C) by formula (7).

${\mathrm{Sim}}_{IC-ND}(C_i^{\′},C_i)=k_1{\stackrel{\‾}{Sim}}_{IC}(C_i^{\′},C_i)+k_2{\stackrel{\‾}{Sim}}_{NC}(C_i^{\′},C_i)---\left(6\right)$

${\mathrm{Sim}}_{IC-ND}(C^{\′},C)=\underset{i=1}{\overset{n}{\Σ}}{w}_i{\mathrm{Sim}}_{IC-ND}(C_i^{\′},C_i)---\left(7\right)$

Finally, it is verified that the effect of similarity algorithm and the effect of retrieval scheme.

(4) data mining model assessment technique

First, the evaluation index of data mining model, build accurate, complete assessment indicator system.

Data mining model index is only the feature having quantified mining model from different perspectives, and these indexs pair the clearest and the most definite In the influence degree of model, the most how integrated treatment various features, it is thus achieved that the tolerance of model the strengths and weaknesses can be embodied, need Want the overall evaluation system of model.Concrete analysis according to data mining model performance evaluation and set up evaluation index system institute The principle followed and process, the present invention establishes the body construction system of mining model Performance Evaluation index, as shown in Figure 7.

The body coding of the present invention uses OWL language.OWL is applicable to such application: in such applications, not only Need only provide for the document content readable to user, and wish to process document content information.OWL can be used clearly The relation between implication and these entries of the entry (term) in vocabulary is expressed on ground.And it is this to entry and they it Between the expression of relation be referred to as body.OWL has more mechanism relative to XML, RDF and RDFSchema to be come Express semanteme, thus OWL has surmounted XML, RDF and RDFSchema and has been merely capable of expressing the most machine-readable The ability of document content.Ontology development instrument uses the protege2000 of Stanford.

Data mining model is evaluated each concept class of body and is included that 9 underlying attribute describe: ClassName, Weight, Haschild, Value Type, Effect Type, EvaluateMethod, EstimateFunction, NodeValue, Unit.

ClassName is the title of this concept class, with concept name as unique designation, is not allow for weight between each evaluation points Name.

Weight is the weights of evaluation points, and evaluate the weight of node on body has following constraints simultaneously:

The weight of root node is 1；

The weight of any one evaluation points node is the summation of its all child node weights；

Whether Haschild indicates has child node, if there being child node, itself does not has independent desired value, by its common table of sub-index Levy.

Value Type represents value type (numeric type, interval type, language type, Boolean type etc.).

Effect Type represents the quality impact of the factor, wherein efficient element such as availability, indicates that desired value is the bigger the better, And cost element such as price, desired value is the smaller the better.

EvaluateMethod indicates index obtaining value method (fixed pattern, statistical, calculation type, setting type).

EstimateFunction is the evaluation function of computational index.

NodeValue is the value of evaluation points.

Unit is the unit of index value.

Utilizing OWL language to realize data mining performance indications appraisement system herein, partial code shows as follows:

Then, it is determined that the computational methods of index weights and the computational methods of quantification of targets value.

Seeking weight is the key of overall merit.Data mining technology is application-oriented field, in same application, different Evaluation points the influence degree of data mining model performance is also differed；For same evaluation index, digging of application Pick business is different, and the degree that may pay close attention to also can difference.So, index weights need according to applied business and evaluate because of The feature of son itself obtains, and needs to consider objective and subjective two aspects.Therefore, the present invention uses subjective and objective comprehensive integration Body construction is calculated the method for weighting by enabling legislation and AHP method combines.

The application characteristic of combining assessment system, selects step analysis (Analytic Hierarchy Process, AHP) method as supplementing. Analytic hierarchy process (AHP) (Analytic Hierarchy Process, AHP), is to the one of quantitative analysis comprehensive integration from qualitative analysis Typical systematic approach, the qualitative analysis that subjective judgment is main, by the thinking process mathematicization of complication system, is carried out by it Quantification, by the values of disparity between various judgement key elements, thus, keep the concordance of thinking process, it is adaptable to complicated Fuzzy overall evaluation application.So, additional upper layer fractional analysis, we can be external expert to evaluation index importance Subjective assessment be used as to weight important adjustment factor, thus substantially increase the accuracy of result.

Os-Ahp (the Ontology structure-Analytic hierarchy process) method that the present invention proposes, i.e. weights bulk junction Structure method and analytic hierarchy process (AHP) process are as follows: calculate the weight of each data mining evaluation index first with body construction method, Recycling analytic hierarchy process (AHP) (Analytic Hierarchy Process-AHP) calculates the weight of each index, and finally weighting is obtained Final index weights.So, in terms of objective factor from the point of view of, i.e. determine data mining evaluation from the immanent structure of body The weight of index；From the point of view of in terms of subjective factors, i.e. by external expert, the subjective assessment of evaluation index importance is come really Its weight fixed.

Detailed process is as follows:

1) body construction method agriculture products weight is utilized

If the weight that body construction method determines is denoted as Wo, it is divided into two parts i.e. Wo=Wm+Ws；Wm represents main portion Dividing (Main part), Ws represents secondary part (Secondary part).

The computational methods of Wm:

Rule 1: if the weight of father's concept A is a, and have n sub-concept, the major part of the most every sub-concept weight Wm=a/m.Definition body root node weight is 1 i.e. W (root)=1

The computational methods of Ws: Wm=β Sim (. .)；

Sim (. .) represent that the similarity of this concept and his father's concept, its computational methods are shown in that chapter 3 two concept similarity seeks method herein； β represents regulation coefficient, the most empirically determined.

2) AHP method is utilized to seek weight

The way of Weight of Coefficient through Analytic Hierarchy Process is as follows: 1. set up multi-level hierarchical structure.Difference according to target, realize merit The difference of energy, is divided into recursive hierarchy structure system by system.2. Judgement Matricies.Setting up Hierarchical structural system After, compared two-by-two by element in each layer, construct multilevel iudge matrix, determine next layer phase for certain factor of last layer time To importance, and give certain score value.The scale criterion generally used is the scale table that T L professor Saaty proposes, such as table 1 Shown in.

The comparison scale table of index judgment matrix and implication thereof

The comparison scale table of table 1 index judgment matrix and implication thereof

Scale value	Implication
		1	Factor ui and uj compare, of equal importance
3	Factor ui and uj compare, and ui is more important than uj
		5	Factor ui and uj compare, and ui is more even more important than uj
7	Factor ui and uj compare, and ui is strong more important than uj
		9	Factor ui and uj compare, and ui is more extremely important than uj
2,4,6,8	Represent adjacent judgement respectively, take 1～3,3～5 respectively ..., intermediate value

Reciprocal: factor ui compares to judge uij with uj, then ui with uj compares to judge uji=1/uij, can structure according to scale table Make judgment matrix T:

$T=\left[\begin{array}{cccc}{U}_{11}& U_{12}& \mathrm{...}& U_{1m}\\ U_{21}& U_{22}& \mathrm{...}& U_{2m}\\ .& .& & .\\ .& .& & .\\ .& .& \mathrm{...}& .\\ U_{m1}& U_{m2}& \mathrm{...}& U_{mm}\end{array}\right]$

3. parameter weight.According to discrimination matrix T, utilize linear algebra knowledge, can accurately obtain its maximum feature Root and characteristic of correspondence vector.Characteristic vector normalized is i.e. obtained this Hierarchy Evaluation factor to father's factor influence degree Size.Maximum vector solved multiple method such as and area method, root method of approximation etc., owing to root method of approximation is more commonly used, Therefore use root method of approximation to solve herein, step is as follows:

Step 1 calculates the product Mi of each row element of judgment matrix,

M_i=IIu_ij, (i, j=1,2 ... m)

Step 2 calculates the m th Root of Mi

$W_i=\sqrt[m]{M_i}$

Step 3 to vector W '=(W1, W2 ..., Wm) normalized.

$w_i=W_i/\[\underset{i=1}{\overset{m}{\Σ}}{W}_i\]$

WA=(w1, w2 ..., wm) it is the weight of required index

Step 4 consistency check

After obtaining weight, needing discrimination matrix is carried out consistency check, formula is as follows:

CR=CI/RI

CR is the random Consistency Ratio of judgment matrix, and CI is judgment matrix approach index, and computing formula is as follows:

CI=(λ_max-m)/(m-1)

The value of RI can calculate according to the value of Saaty scale result and CI.

In formula, λ max is the Maximum characteristic root of judgment matrix.

The two weighted sum i.e. obtains final weight W

W=k1Wo+k2WA

(k1, k2 are the two weight, and k1+k2=1)

Finally, the method comprehensively utilizing these index quantification metric data mining model performances is set up.

Specifically comprise the following steps that

Step 1 sets up mining model evaluation index collection

Set up rational data mining model index evaluation system, be the basis effectively assessed of mining model performance, be also to close most The problem of key.Not having the evaluation index system of science, evaluation work just cannot correctly be carried out.DMME-OAF metrics evaluation Following basic principle is followed in the design of system:

(1) comprehensive principle, in model performance evaluation study, considers all kinds of indexs of model comprehensively, comprehensive degree of accuracy, benefit, The various aspects such as operational efficiency, evaluate mining model service behaviour the most all sidedly.

(2) balance principle, in the investigation of mining model performance evaluation factor, needs to investigate and can reflect the correct of model capability Sexual factor, it is also desirable to consider the resource consumption factor of model.I.e. need to pay close attention to the professional ability of mining model, can not ignore The cost of model various aspects.

(3) practical principle, the final target that designs of mining model is for application service, and not only which compares in correctness The modeling method of model is abstruse or complicated, and the evaluation of mining model needs also exist for investigating the benefit can brought for excavation business, Therefore, the factor in terms of appraisement system needs comprehensive value.

The present invention achieves the data of DMME-OAF on the basis of setting up mining model performance indications evaluation system structure and digs Pick evaluation model design.2 explanations are had during application:

(1) index of level can extend, and when needs consider more many factors, can extend assessment according to actual needs and refer to Mark.

(2) being ultimately used to as the index set of decision-making is a subset of this index system, can select different according to actual condition Index set.Such as:

U={ degree of lifting benefit coverage compatible degree hit rate can be run time complexity } as decision index system Collection, it is also possible to index " compatible degree " changes into its subset, and { terseness definitiveness practicality Interest Measure novelty can Explanatory visualization }, i.e. U '={ lifting degree benefit coverage terseness definitiveness practicality Interest Measure novelty can Explanatory visualization hit rate runs time complexity }.

When test case allows, index can be got and more refine, such result is more accurate, when test condition is insufficient Words, can evaluate roughly by more general index.Use U ' as evaluation indice herein.

Step 2 designs mining model performance Comment gathers

In all of evaluation index, not every index can quantitative Analysis, part index number is qualitative description, the most logical Cross expert opinion and obtain comment, as fuzzy language value.Comment gathers is that the evaluation result possible made evaluation object is formed Set, be expressed as V={v₁, v₂, v₃..., v_n}.Consider the rational density of opinion rating, in mining model appraisement system Use Pyatyi Comment gathers, i.e.

V={ is fine, preferably, in, poor, very poor.

Step 3 solves the comment of each factor of evaluation of mining model

For each factor of evaluation of mining model, system provides concrete quantization.When all of qualitative index provides qualitative evaluation After, need to quantify qualitative evaluation, and set up membership function.Qualitative evaluation quantifies assignment and carries out by table 2.

Table 2 qualitative evaluation quantifies assignment

Corresponding membership function formula is as follows:

$r_{ij}=\left\{\begin{array}{cccc}1& x_{ij}\&le;a& & \\ \frac{b-x_{ij}}{b-a}& a<x_{ij}<b& i=1,2,...,m& j=1,2,...,n\\ 0& x_{ij}\&GreaterEqual;b& & \end{array}\right.$

Wherein a representing matrix element x_ijUpper limit threshold value；B matrix element x_ijLower limit threshold value.

By quantization and the membership function of comment, evaluation index matrix (subordinated-degree matrix) R can be drawn:

$R=\left[\begin{array}{cccc}{r}_{11}& r_{12}& \mathrm{...}& r_{1n}\\ r_{21}& r_{22}& \mathrm{...}& r_{2n}\\ .& .& & .\\ .& .& & .\\ .& .& \mathrm{...}& .\\ r_{m1}& r_{m2}& \mathrm{...}& r_{mn}\end{array}\right]$

Order

$a_i=\frac{k_i}{\underset{i=1}{\overset{m}{\&Sigma;}}{k}_i}$

Each factor is normalized, obtains R⁰For:

$R^0=\left[\begin{array}{cccc}{r_{11}}^0& {r_{12}}^0& \mathrm{...}& {r_{1n}}^0\\ {r_{21}}^0& {r_{22}}^0& \mathrm{...}& {r_{2n}}^0\\ .& .& & .\\ .& .& & .\\ .& .& \mathrm{...}& .\\ {r_{m1}}^0& {r_{m2}}^0& \mathrm{...}& {r_{mn}}^0\end{array}\right]$

Step 4 utilizes the method seeking weight above to obtain weight sets W

Step 5 calculates evaluation result matrix B

By the Judgement Matrix R after weight matrix W and normalized⁰Be multiplied, evaluation result matrix B B=W R⁰.According to maximum membership grade principle, select optimal case.

The whole data mining model evaluation procedure false code modeled based on body and CBR is as shown in Figure 8.

(5) intelligent integration of local process

This process comprises two aspects: one is initial designs process collection based on Intelligent data mining technology and association knowledge model Becoming, two is that simulation feedback design process based on Intelligent data mining technology and association knowledge model is integrated, and beneficially system is Whole is integrated and perfect.

(6) overall assembly of system, test, perfect

The general frame of present system is as shown in Figure 9, it can be seen that system divides four layers: user interface layer, functional module Layer, technology platform layer, data Layer.

User interface layer: user interface layer is also transaction layer, is user's window of carrying out with computer exchanging, defeated including information Enter and export with information.The mode of information input mainly has following three kinds: products C AD 3-D view is loaded directly into；Man-machine interactively (interface wizard prompting)；Based on UDF characteristic information identification and extraction.

Technology platform layer: this layer primarily illustrates the various major techniques used by the 26S Proteasome Structure and Function realizing native system and platform The such as OWL language used by body realization, storage data base Access used by data, storage interface skill used by data Art ADO.Net etc..

Data Layer: preserve various information with data and document form, includes five data bases: parts information storehouse, manufacture provide Storehouse, source, process planning knowledge base, data mining case searching.

Functional module layer: 6 basic function module defined in present system, respectively: knowledge retrieval module, data Excavate Case retrieval course module, data mining case retrieving module, data mining and Knowledge Discovery module, process knowledge study mould Block and data mining event selection module (i.e. knowledge store module).

(1) knowledge retrieval module

Knowledge retrieval module i.e. process planning knowledge query module, its interface is as shown in Figure 10.Process Planning is carried out when utilizing system When drawing, first carrying out the inquiry of process knowledge according to design objective, if corresponding knowledge can be directly obtained, then system is run Hereto, if satisfied knowledge can not be inquired, then enter data-mining module, wanted by data mining acquisition Knowledge.

(2) data mining Case retrieval course module

When satisfied knowledge can not be directly obtained in knowledge retrieval module, would have to be obtained by data mining technology Required knowledge, at this moment can be imported data mining Case retrieval course module by interface wizard, as shown in figure 11.

In this module, according to interface wizard prompting input relevant information, carry out the retrieval of similar example.Data mining example Retrieval and knowledge retrieval differ.The retrieval of knowledge is the coupling of instantiation information, and precise requirements is higher (to be cut off from Value α >=0.9500)；The retrieval of data mining example is the abstract coupling on concept hierarchy, precise requirements wanting than knowledge retrieval Low (threshold value α ＜ 0.9500).Being returned by data mining Case retrieval course and be typically one group of example, this just requires to be dug by data Pick evaluation mechanism is evaluated obtaining optimal example.If the example obtained by retrieval evaluation does not reaches minimum threshold value (threshold value α=0.9000), at this moment, it has to enter case retrieving module, by case retrieving, to obtain satisfied data mining thing Example.

(3) data mining case retrieving module

Data mining case retrieving module is as shown in figure 12.At present, in data mining case retrieving module, main by amendment The weight of related notion, and add and subtract what concept characteristic regulated.Amendment is complete every time, will be assessed by the example of module Function is estimated, if not reaching more than threshold value (α=0.9000), then continues to adjust according to assessment prompting, until inspection Till the value of rope similarity reaches more than 0.9000, subsequently into next module i.e. data mining and Knowledge Discovery module.

(4) data mining and Knowledge Discovery module

When having obtained the data mining example of a satisfaction, namely mean to establish the data mining model of a satisfaction, Uniform data acess can be carried out.Data mining with Knowledge Discovery module as shown in figure 13, clicks directly on and " performs Data mining " button, i.e. can obtain the knowledge met required for design objective.For the sake of more accurately, set in this module Put knowledge evaluation function, the knowledge obtained can be evaluated, then reapply.Also may be used if evaluation result is " Yes " To enter knowledge learning module and event selection module from this interface.If evaluation result is " No ", then to return to number According to excavating Case retrieval course module, a new cyclic process restarts, until evaluation result is " Yes ".

(5) process knowledge study module

In data mining and Knowledge Discovery module, when the process knowledge finally excavated is satisfied i.e. display " Yes " word through evaluating During sample, illustrate that the process knowledge that data mining goes out is accurately, then this knowledge just can be stored in knowledge base, under Secondary retrieval uses.Then, by process knowledge study interface as shown in figure 14, knowledge is stored in knowledge base.

(6) data mining event selection module

In data mining and Knowledge Discovery module, when knowledge evaluation is " Yes ", not only demonstrating the knowledge that excavation obtained is Accurately, it is appropriate for also demonstrating Modeling of Data Mining simultaneously, and therefore, this successful example should be stored into data and dig Pick case searching, completes the study of example.Then, by data mining event selection interface as shown in figure 15, example is stored in thing Example storehouse.

Claims (4)

1. new based on data mining technology an intelligent punching press process design method, it is characterised in that include procedure below:

Step one: set up the unified feature information model supporting that native system builds；Research is currently based on the punching press of data mining On the basis of Expert System characteristic information expression model, the spy that research incorporates KBE technology, ontology is increased newly Reference ceases, and sets up the Feature concept definition that whole Process Planning is complete, unified；

Step 2: set up relatedness knowledge representation model；CAD process based on data mining in analyzing stamping process design With the function involved by CAE process based on data mining, data, characteristic parameter setting and shape facility and process knowledge Cause effect relation on the basis of, use knowledge representation based on Ontology to set up the relatedness knowledge representation mould of two processes Type；

Step 3: build the inference system for Modeling of Data Mining；

Step 4: data mining model assessment technique；The evaluation index of data mining model, builds accurate, complete Standby assessment indicator system；The computational methods of agriculture products weight and the computational methods of quantification of targets value；Set up comprehensive utilization The method of these index quantification metric data mining model performances；

Step 5: the intelligent integration of local process；This process comprises two aspects: one is based on Intelligent data mining technology With the initial designs process integration of association knowledge model, two is based on Intelligent data mining technology and the emulation of association knowledge model Feedback Design process integration；

Step 6: the overall assembly of system, test, perfect；This system is divided into four layers: user interface layer, functional module layer, Technology platform layer, data Layer.

A kind of new based on data mining technology intelligent punching press process design method the most according to claim 1, it is special Levy and be: the feature information model in described step one includes: part feature information model, manufacture resource feature information mould Type, process planning feature information model.

A kind of new based on data mining technology intelligent punching press process design method the most according to claim 1, it is special Levy and be: described step 3 comprises the following steps:

Step 3.1: the decomposition technique of design objective and determine the associated data set for excavating；Utilize the RBR of KBE technology Way of realization completes complex task resolves into some enforceability subtasks, and determines the data set that this subtask is relevant；

Step 3.2: build the data mining case searching based on Ontology towards stamping process design: first, from field Collect abundant vocabulary, term, and using the vocabulary being widely recognized as in the industry, term as the concept of body, simultaneously that body is general Thought is categorized as background concepts and result concept；Secondly, connecting each other between concept in analysis field, accurately obtain field originally The various relations of body；Finally, all concepts of body are attached by the relation between them, form a tree network Network structure, is i.e. the data structure of body example knowledge base；

Step 3.3: determine the knowledge reasoning scheme of Modeling of Data Mining subsystem: first, determines the overall retrieval scheme of system, This project uses the 2-level search scheme that knowledge guides and nearest neighbor algorithm combines；Secondly, the similarity algorithm of example is determined； Finally, it is verified that the effect of similarity algorithm and the effect of retrieval scheme.

A kind of new based on data mining technology intelligent punching press process design method the most according to claim 1, it is special Levy and be: the functional module layer in described step 6 includes 6 basic functional modules: knowledge retrieval module, data mining Case retrieval course module, data mining case retrieving module, data mining and Knowledge Discovery module, process knowledge study module and Data mining event selection module.