CN103093051A - Product performance conductive knowledge mining method - Google Patents

Abstract

The invention discloses a product performance conductive knowledge mining method. Product performance relevant parameter data changing conditions before or after the conversion of an extension configuration design element are mined, conductive knowledge of performance and special parameters are obtained, and direct conductive relationships among product performances are produced. The method comprises the following steps: extracting the performance information before or after the conversion, obtaining conductive performance features through threshold value judgment, quantificationally describing performance conductive influence degree by conductivity, obtaining extension classification knowledge caused by conductive conversion through performance correlation functions after the conductive conversion, and guiding product design process of performance drive. According to the product performance conductive knowledge mining method, the conductive knowledge of the performance and the special parameters are obtained, and the direct conductive relationships among the product performances are produced, so that the method can help a designer to make a favorable decision in the product design process of demand drive.

Description

A kind of properties of product conduction Knowledge Discovery Method

Technical field

The present invention proposes properties of product conduction Knowledge Discovery Method, in order to obtain the conduction knowledge between properties of product, generate the conduct the relation between such properties of product.

Background technology

Product is constantly updated the replacement in evolution, product example quantity is rapid growth thereupon also.Effective development efficiency that builds the good management product example information of energy and improve product of case library.Yet, be stored in the product-related data of the magnanimity in various data mediums, in the situation that lack strong instrument, oneself understanding through having exceeded from far away the people and the ability of summary.Therefore, in the process of product design, how to go out the useful Knowledge and information of product design by effective method from these extracting data is the problem of enterprises pay attention always.

Client's requirement description is fuzzyyer, and is general just to the recapitulative requirement of properties of product.Therefore need to carry out semantic qualitative, quantitative description to it, the Obtaining Accurate client requirement information carries out after similarity calculates and give respective weights each attribute of performance, searches out like product.Like product can not satisfy the client fully to the requirement of properties of product, therefore need by rule of thumb product structure to be carried out opening up conversion, and then test, make the contradiction problem between the properties of product of customer demand and performance that like product has be cleared up, realize satisfying of a certain or a few performances, thereby obtain corresponding scheme solution.

The performance of product has ambiguity and dynamic, causes the search very complex based on the similar example of complex product of performance, needs integrated multidisciplinary, multi-field interior performance knowledge and practical experience; In addition, can open up conversion and change to satisfy a certain attribute of performance except causing its effective object to produce, also due to conduction, can cause the conduction transformation of relative other objects, make some attribute of performance of originally having met customer need reconstitute the contradiction problem.For these problems, the present invention proposes product example storehouse modeling and the conduction knowledge excavation technology of performance driving, the relation that affects by performance structure before and after the excavation conversion, obtain the conduction knowledge of performance and its special parameter, generate the direct conduct the relation between such properties of product, help the designer to make favourable decision-making in the product design process that demand drives, avoid the generation of new contradictory problems and obtain the extension classification knowledge of conduction transformation.

Summary of the invention

In order to overcome the deficiency that can't excavate properties of product conduction knowledge in the existing product design process, the invention provides a kind of properties of product conduction Knowledge Discovery Method, based on the demand of client to properties of product, the deviser searches out like product in the product example storehouse, by product structure is carried out opening up conversion, clear up the contradiction problem that like product performance not exclusively satisfies client's performance requirement.Can open up in conversion process and can cause conduction transformation, if can excavate the relation that affects of performance structure before and after conversion, obtain the conduction knowledge of performance and its special parameter, generate the direct conduct the relation between such properties of product, can help the designer to make favourable decision-making in the product design process that demand drives.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of properties of product conduction Knowledge Discovery Method, described conduction Knowledge Discovery Method comprises the following steps:

1), build the product example storehouse: develop, build the product example storehouse on drawing analysis software platform;

2), client's performance requirement matter-element is described: the product performance demands matter-element is with this product O _mBe object, c _mBe requirement product performance characteristic, v _mFor about c _mThe orderly tlv triple that consists of of value, if this product O _mA plurality of feature c are arranged ₁, c ₂..., c _nAnd corresponding value v ₁, v ₂..., v _n, represent with multidimensional performance requirement matter-element:

$M=\left[\begin{array}{ccc}{O}_m& c_1& v_1\\ & c_2& v_2\\ & .& .\\ & .& .\\ & .& .\\ & c_n& v_n\end{array}\right];$

3), performance requirement product example retrieval:

Based on opening up apart from the demand similarity algorithm: establish performance requirement data dimensionless and process between the back zone

Case library performance data dimensionless is processed Y=[y between the back zone ₁, y ₂], and

Apart from as follows with distance similarity model formula:

$d(V,Y)=|\frac{y_1+y_2}{2}-\frac{v_{\mathrm{\α}1}^l+v_{\mathrm{\α}1}^r}{2}|-|\frac{y_2-y_1}{2}-\frac{v_{\mathrm{\α}1}^r-v_{\mathrm{\α}1}^l}{2}|---\left(3\right);$

Process concrete numerical value if x is the performance requirement dimensionless, the left side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}{y}_1-x,x\≤\frac{y_1+y_2}{2},\\ x-\frac{y_1+y_2}{2},x>\frac{y_1+y_2}{2}.\end{array}\right.---\left(4\right);$

The right side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}\frac{y_1+y_2}{2}-x,x<\frac{y_1+y_2}{2},\\ x-y_2,x\&GreaterEqual;\frac{y_1+y_2}{2}.\end{array}\right.---\left(5\right);$

Calculate the most similar case similarity sim by similarity _max, obtain the most similar example L;

4), clear up based on the performance structure contradiction that can open up conversion: compare demand example and each performance parameter of the most similar example, do not meet the demands if any performance parameter, performance requirement object instance G exists with performance condition example L and conflicts, the contradiction problem represents with P=G*L, expand by the performance topology requirement and analyze, acquisition methods thing meta-model, then carry out opening up conversion; Comprise the following steps:

(4.1), demand and condition question are described: for given problem P=G*L, for performance requirement object instance and performance condition example correlation function K (P)＜0, the conversion T of target and condition _P=(T _G, T _L) be the solution conversion of P, make K (P)〉0, satisfy the client for the performance requirement of product;

(4.2), product thing meta-model is expanded and is analyzed: by calculating more similar example performance characteristic and demand performance correlation function value, obtain similar example performance feature association degree value less than zero, do not satisfy performance requirement, need to carry out emission analysis to contradictory problems, then analysis is contained in thing unit, obtain to open up conversion scheme, calculate T _GAfter the degree of association, this process makes K (P) repeatedly〉0 realize the performance requirement goal satisfaction;

5), performance conduction knowledge excavation, process is as follows:

(5.1), the judgement of conductive object and conduction feature: given threshold value δ〉0, when the value of the object that conversion occurs or feature poor

(

Value before and after the expression conversion) absolute value satisfies:

$\left|\mathrm{\&Delta;}{v}_{l_i}\right|=|v_{l_i}^{\&prime;}-v_{l_i}|>\mathrm{\&delta;}---\left(14\right);$

The time, think corresponding object or be characterized as conductive object or conduction feature;

(5.2), conductivity is calculated: for the product example structure feature information l of unit _i=(o _ic _iv _i), obtaining information word after conversion is l ' _i=(o _ic _iv’ _i); If there is initiatively conversion

Tl _i=l’ _i=(o _i c _i v’ _i) (15)；

This is another object of transfer pair or feature m initiatively _j=(q _jd _ju _j) the generation conduction transformation

Make m _jBe transformed to m ' _j=(q _jd _ju′ _j), namely can be expressed as

And conductivity γ namely is expressed as

$\mathrm{\&gamma;}=\left|\frac{u_j^{\&prime;}-u_j}{v_i^{\&prime;}-v_i}\right|---\left(17\right);$

(5.3), many initiatively conversion conduction knowledge excavations:

For the structural information l of unit ₁, l ₂, l ₃..., l _hL satisfies condition:

, i=1,2 ..., h; J=1,2 ..., k is if there are a plurality of active conversion to cause the generation of a plurality of conduction transformation, namely

There is conduction transformation

And the transformation parameter of active more than obtaining collection

V(v _i,v’ _i)={(v ₁,v’ ₁),(v ₂,v′ ₂),…,(v _h,v′ _h)} (19)；

The conduction knowledge that many initiatively conversion cause is designated as

When many actives conversion The time, the first conductivity of each conducts information is

${\mathrm{\&gamma;}}_i=\frac{|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(21\right);$

Wherein: $\mathrm{AHP}\left(\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i\right)=1---\left(22\right);$

If conduction feature m _jWeight in whole conduction features is β _j, j=1,2 ..., k, and

Comprehensive conductivity is

$\mathrm{\&gamma;}=\frac{\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_j|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(23\right).$

Further, described step 5) also comprises following process: (5.4), based on the extension classification knowledge of conduction transformation, and the detailed process step is as follows:

For product demand performance collection, k _j(m _j) represent that demand performance and product have the degree of association of example performance now,

Corresponding configuration primitive structure, the initiatively conversion of material properties of performance characteristic thing meta analysis described;

Expression

The conduction transformation that the attribute of performance that causes changes is for domain M _oAbout the performance characteristic attribute transformation

The config set opened up;

Obtain the conduction classificating knowledge that conduction transformation causes;

Before conduction transformation, domain is divided, and according to the attribute of performance correlation function, domain static state is divided into:

$E_{+}^j=\left\{m_j\right|m_j\&Element;M_O,y=k_j\left(m_j\right)>0\}---\left(25\right);$

$E_{-}^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)<0\}---\left(26\right);$

$E_0^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)=0\}---\left(27\right);$

Respectively will

With

Be called the positive territory of domain, negative territory and zero boundary;

After conduction transformation, domain is divided, and by the active conversion, reclassifies for the attribute of performance that conduction occurs:

With

Be respectively the positive qualitative change of the attribute of performance territory that conduction transformation causes, negative qualitative change territory, positive quantity variable domain, negative quantity variable domain and open up the boundary; Described positive qualitative change, negative qualitative change, positive quantitative change, negative quantity become knowledge and are properties of product conduction transformation extension classification knowledge excavation result.

Further, described step 3) comprises following process:

(3.1), attribute of performance quantitative description: raw data is carried out dimensionless process, namely the value of characteristic parameter all is converted into [0,1] interior value;

The raw data normalization mode of a certain performance characteristic value of product is as follows:

$y_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}-{\min x}_j}{{\max x}_j-{\min x}_j}---\left(1\right);$

Or:

$y_{\mathrm{ij}}=\frac{\max x_j-x_{\mathrm{ij}}}{{\max x}_j-{\min x}_j}---\left(2\right);$

In formula, x _ijProduct example storehouse performance data before expression is processed, y _ijProduct example storehouse performance data after expression is processed, i=1,2 ..., m, j=1,2 ..., n, maxx _jAnd minx _jRepresent respectively the minimum and maximum value of product example storehouse j item performance characteristic;

(3.2), based on opening up apart from the demand similarity algorithm: performance requirement data dimensionless is processed between the back zone

Case library performance data dimensionless is processed Y=[y between the back zone ₁, y ₂], and

Formula is as (3), and the left side distance is suc as formula (4), and the right side distance is as shown in the formula (5);

(3.3), the calculating of similarity: similarity is a kind of tolerance of the similarity relation of requirement product and existing procucts example, and the rank according to similarity in Similarity algorithm is different, and similarity can be divided into:

(3.3.1), local similarity (Part Similarity): interval and interval, point is with interval, and point with some local similarity search formula is:

sim(V,Y)=1-|d(V,Y)| (6)；

A, B can be interval, can be also concrete a certain numerical value;

(3.3.2), overall similarity: customer demand and the heavy calculating formula of similarity of product example storehouse all properties Feature Weighting:

$\mathrm{sim}(R,E_i)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_j\mathrm{sim}(R_i,E_i)---\left(7\right);$

Sim (R, E _i) expression example E _iWith the overall similarity of demand configuration design problem R, sim (R _i, E _i) be example E _iConfigure design problem R at attribute of performance c with demand _iOn similarity, the product example number is n, the attribute number is m, determines performance attribute weight w by analytical hierarchy process _j

Calculate the most similar case similarity sim by similarity _max, obtain the most similar example L.

Principle of work of the present invention: can open up is the new branch of science that is used for the problem of resolving contradiction that Chinese scholar Cai Wen first is born in the nineteen eighty-three foundation.Explore through for many years theoretical research and practical application, now formed the frame form of the rich theoretical degree of depth, mainly comprise: 1, can open up the engineering theory basis; 2, collection can be opened up and logic can be opened up; 3, extension data mining method;

Traditional product design for the performance requirement driving when performance does not satisfy, by structure corresponding to property, realizes performance requirement.Present not satisfying for performance requirement, all by rule of thumb product structure to be carried out certain remodeling, and then test, to realize satisfying of a certain or a few performances, do not note utilization and excavation for data after conversion, the situation of change that comprises other performances of example is not more noted influencing each other between data.The conduction knowledge excavation is the data digging method with contradictory prerequisite, quantitatively judge conversion front and back performance structure conductive object collection or conduction feature collection, and by the conductivity quantitative description influence degree of transfer pair conductive object or feature initiatively, obtain simultaneously the extension classification knowledge based on conduction transformation, excavate the properties of product structure conduct the relation of conversion front and back.By excavating the relation that affects of performance structure before and after conversion, help the designer to make favourable decision-making in the configuration design process that demand drives, avoid the generation of new contradictory problems.

Beneficial effect of the present invention shows as: 1, the description of the qualitative, quantitative demand of client to properties of product of knowing clearly; 2, improved the quality and reliability of case-based reasoning; 3, simple and be easy to computer system and realize; 4, formalized description conduction transformation knowledge; 5, the result that obtains is scientific and reasonable.

Description of drawings

Fig. 1 is case library structural texture figure.

Fig. 2 is based on the product performance demands that can open up conversion and clears up process flow diagram to similar example.

Fig. 3 is configuration design performance conduction knowledge excavation figure.

Embodiment

Below in conjunction with accompanying drawing, the present invention is further described.

With reference to Fig. 1 ~ Fig. 3, a kind of properties of product conduction Knowledge Discovery Method comprises the following steps:

1), build the product example storehouse: case library is divided into product example storehouse, module instance storehouse, parts case library; Comprise product in case library and have routine storehouse, and formed module library case library and the parts case library with certain knowledge, wherein comprised concrete product component and parts information;

2), the customer demand matter-element is described: the product demand matter-element is with this product O _mBe object, c _mBe requirement product performance characteristic, v _mFor about c _mThe orderly tlv triple that consists of of value.If this product O _mA plurality of feature c are arranged ₁, c ₂..., c _nAnd corresponding value v ₁, v ₂..., v _n, can represent with multidimensional demand matter-element:

$M=\left[\begin{array}{ccc}{O}_m& c_1& v_1\\ & c_2& v_2\\ & .& .\\ & .& .\\ & .& .\\ & c_n& v_n\end{array}\right];$

3), performance requirement retrieval

(3.1), attribute of performance quantitative description: for guaranteeing equivalence and the same sequence of target data, for the ease of similarity search, need to carry out dimensionless to raw data and process, namely the value of characteristic parameter all is converted into [0,1] value in, different situation disposal routes is also different;

It is as much as possible little in a certain interval that the client requires a certain performance characteristic value of product, as the weight of screw air compressor, noise etc.General raw data normalization mode is as follows:

$y_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}-{\min x}_j}{{\max x}_j-{\min x}_j}---\left(1\right);$

It is large as much as possible in a certain interval that the client requires a certain performance characteristic value of product, as the air capacity of screw air compressor, power input etc.General raw data normalization mode is as follows:

$y_{\mathrm{ij}}=\frac{\max x_j-x_{\mathrm{ij}}}{{\max x}_j-{\min x}_j}---\left(2\right);$

In formula, x _ijProduct example storehouse performance data before expression is processed, y _ijProduct example storehouse performance data after expression is processed, i=1,2 ..., m, j=1,2 ..., n.Maxx _jAnd minx _jRepresent respectively the minimum and maximum value of product example storehouse j item performance characteristic;

(3.2), based on opening up apart from the demand similarity algorithm: performance requirement data dimensionless is processed between the back zone

Case library performance data dimensionless is processed Y=[y between the back zone ₁, y ₂], and

After improving, formula is as follows:

$d(V,Y)=\mathrm{\&rho;}(V,Y)=|\frac{y_1+y_2}{2}-\frac{v_{\mathrm{\&alpha;}1}^l+v_{\mathrm{\&alpha;}1}^r}{2}|-|\frac{y_2-y_1}{2}-\frac{v_{\mathrm{\&alpha;}1}^r-v_{\mathrm{\&alpha;}1}^l}{2}|---\left(3\right);$

The client wishes that a little eigenwerts of certain performance less in given range (or lower) are better, and the side distance that can open up in is not suitable for similarity retrieval.For example, the right side above using is apart from formula, and the distance between computation interval [6,10] and numerical value 6 is 1, there is no discrimination, does not meet the characteristics that Search Results is the bigger the better in interval.

Process concrete numerical value if x is the performance requirement dimensionless, after improving, the left side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}{y}_1-x,x\&le;\frac{y_1+y_2}{2},\\ x-\frac{y_1+y_2}{2},x>\frac{y_1+y_2}{2}.\end{array}\right.---\left(4\right);$

After improving, the right side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}\frac{y_1+y_2}{2}-x,x<\frac{y_1+y_2}{2},\\ x-y_2,x\&GreaterEqual;\frac{y_1+y_2}{2}.\end{array}\right.---\left(5\right);$

(3.2), the calculating of similarity: similarity is a kind of tolerance of the similarity relation of requirement product and existing procucts example, and the rank according to similarity in Similarity algorithm is different, and similarity can be divided into:

(3.2.1), local similarity (Part Similarity): refer to different performance characteristic attribute similarity, be the process of each performance requirement attribute and each example performance attributes match, its similarity is calculated different attribute classification and the value scope that depends on the performance characteristic attribute;

Interval and interval, point is with interval, and point with some local similarity search formula is:

sim(V,Y)=1-|d(V,Y)| (6)；

B, B can be interval, can be also concrete a certain numerical value.

(3.2.2), overall similarity (Overall Similarity): refer to customer demand and example similarity degree; After the local similarity value of knowing all properties feature, can obtain the total similarity value of example of customer demand example and case library; By composing weight to different performance characteristics, obtain total similarity value again, case-based reasoning system can adopt multiple overall similarity calculating method;

Customer demand and the heavy calculating formula of similarity of product example storehouse all properties Feature Weighting:

$\mathrm{sim}(R,E_i)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_j\mathrm{sim}(R_i,E_i)---\left(7\right);$

Sim (R, E _i) expression example E _iWith the overall similarity of demand configuration design problem R, sim (R _i, E _i) be example E _iConfigure design problem R at attribute of performance c with demand _iOn similarity, the product example number is n, the attribute number is m, determines performance attribute weight w by analytical hierarchy process (AHP) _j

Calculate the most similar case similarity sim by similarity _max, obtain the most similar example L;

(4), clear up based on the performance structure contradiction that can open up conversion: compare demand example and each performance parameter of the most similar example, do not meet the demands if any performance parameter, performance requirement object instance G exists with performance condition example L and conflicts, and the contradiction problem represents with P=G*L.Expand by the performance topology requirement and analyze, acquisition methods thing meta-model, then carry out opening up conversion.

(4.1), demand and condition question are described: for given problem P=G*L, for performance requirement object instance and performance condition example correlation function K (P)＜0; The conversion T of target and condition _P=(T _G, T _L) be the solution conversion of P.Make K (P)〉0, satisfy the client for the performance requirement of product;

(4.2), set up the problem correlation function

Definition one: establishing x is arbitrary value, X ₀=＜a, b〉be the arbitrary interval on real number field, some x and interval X ₀=＜a, b〉distance be:

$\mathrm{\&rho;}(x,X_0)=|x-\frac{a+b}{2}|-\frac{b-a}{2}---\left(8\right);$

Definition two: establish X ₀=＜a, b 〉, X=＜c, d 〉, and

Point x is about interval X ₀Nest of intervals place value with X:

Definition three: side is apart from conceptual description

The left side distance $x_0\&Element;(a,\frac{a+b}{2}),$ $\mathrm{\&rho;}(x,{\mathrm{<figure-callout\; id="0"\; label="x"\; filenames="HDA00002746120500021.png"\; state="\{\{state\}\}">x</figure-callout>}}_0,X_0)=\left\{\begin{array}{c}a-x,x\&le;a\\ \frac{(b-x_0)}{a-x_0}(x-a),x\&Element;<a,x_0>\\ x-b,x\&GreaterEqual;x_0\end{array}\right.---\left(10\right);$

The right side distance $x_0\&Element;(\frac{a+b}{2},b),$ $\mathrm{\&rho;}(x,{\mathrm{<figure-callout\; id="0"\; label="x"\; filenames="HDA00002746120500021.png"\; state="\{\{state\}\}">x</figure-callout>}}_0,X_0)=\left\{\begin{array}{c}x-b,x\&GreaterEqual;b\\ \frac{(a-x_0)}{(b-x_0)}(b-x),x\&Element;<x_0,b>\\ a-x,x\&le;x_0\end{array}\right.---\left(11\right);$

Correlation function is set up:

$k\left(x\right)=\frac{\mathrm{\&rho;}(x,{\mathrm{<figure-callout\; id="0"\; label="x"\; filenames="HDA00002746120500021.png"\; state="\{\{state\}\}">x</figure-callout>}}_0,X_0)}{D(x,{\mathrm{<figure-callout\; id="0"\; label="X"\; filenames="HDA00002746120500021.png"\; state="\{\{state\}\}">X</figure-callout>}}_0,X)}---\left(12\right);$

Performance requirement object instance and performance condition example correlation function:

(4.3), product thing meta-model is expanded and is analyzed: by calculating more similar example performance characteristic and demand performance correlation function value, obtain similar example performance feature association degree value less than zero, do not satisfy performance requirement.Need to carry out emission analysis to contradictory problems, then analysis is contained in thing unit, obtains opening up conversion scheme, calculates T _GAfter the degree of association, this process makes K (P) repeatedly〉0 realize the performance requirement goal satisfaction;

(5), performance conduction knowledge excavation

(5.1), conductive object and conduction feature judgement:

: for specific product instance objects and characteristic, tend to occur the variation of the corresponding value in conversion front and back, the difference before and after some characteristics of objects value conversion is very little.Conductive object and conduction feature for what is described more accurately, common given threshold value δ〉0, when the object that conversion occurs or the value poor (conduction effect) of feature (

Value before and after the expression conversion) absolute value satisfies

$\left|\mathrm{\&Delta;}{v}_{l_i}\right|=|v_{l_i}^{\&prime;}-v_{l_i}|>\mathrm{\&delta;}---\left(14\right);$

The time, think corresponding object or be characterized as conductive object or conduction feature;

(5.2), conductivity is calculated: initiatively the conduction of conversion has great impact to conductive object or conduction feature actually, need to be quantitatively described by conductivity.

For the product example structure feature information l of unit _i=(o _ic _iv _i), obtaining information word after conversion is l ' _i=(o _ic _iv’ _i); If there is initiatively conversion

Tl _i=l’ _i=(o _i c _i v’ _i) (15)；

This is another object of transfer pair or feature m initiatively _j=(q _jd _ju _j) the generation conduction transformation

Make m _jBe transformed to m ' _j=(q _jd _ju′ _j), namely can be expressed as

And conductivity γ namely can be expressed as

$\mathrm{\&gamma;}=\left|\frac{u_j^{\&prime;}-u_j}{v_i^{\&prime;}-v_i}\right|---\left(17\right);$

(5.3), many initiatively conversion conduction knowledge excavations: for the similar example that does not satisfy performance requirement, be not often to the initiatively conversion of single structure feature, but need to carry out the active conversion to a plurality of architectural features of configuration primitive; These many initiatively conversion comprise the material characteristics parameter that configures primitive, structural characteristic parameter, interface characteristics parameter etc.And use these the many initiatively conversion of conductivity quantitative description and the conduction transformation of other performance characteristics of example of causing;

For the structural information l of unit ₁, l ₂, l ₃..., l _hL satisfies condition: , i=1,2 ..., h; J=1,2 ..., k.If have a plurality of active conversion to cause the generation of a plurality of conduction transformation, namely

There is conduction transformation

And the transformation parameter of active more than obtaining collection

V(v _i,v’ _i)＝{(v ₁,v’ ₁),(v ₂,v’ ₂),…,(v _h,v’ _h)} (19)；

The conduction knowledge that many initiatively conversion cause is designated as

When many actives conversion

The time, the first conductivity of each conducts information is

${\mathrm{\&gamma;}}_i=\frac{|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(21\right);$

Wherein: $\mathrm{AHP}\left(\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i\right)=1---\left(22\right);$

If conduction feature m _jWeight in whole conduction features is β _j, j=1,2 ..., k, and Comprehensive conductivity is

$\mathrm{\&gamma;}=\frac{\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_j|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(23\right);$

(5.3), based on the extension classification knowledge of conduction transformation, detailed process is as follows:

The conduction transformation that causes of conversion initiatively can be with the quantitative description of the conductivity relation of conversion value and conduction effect initiatively; Based on the extension classification knowledge representation of conduction transformation the degree of the conduction transformation performance characteristic magnitude variations that causes of conversion initiatively, namely because quantitative change (positive and negative quantitative change) or qualitative change (positive and negative qualitative change) have occured the conduction transformation performance characteristic;

For product demand performance collection, k _j(m _j) represent that demand performance and product have the degree of association of example performance now,

The initiatively conversion of attribute such as configuration primitive structure corresponding to performance characteristic thing meta analysis, material have been described; Expression

The conduction transformation that the attribute of performance that causes changes is for domain M _OAbout the performance characteristic attribute transformation

The config set opened up;

The characteristics of extension classification maximum are its dynamics, and existing sorting technique is almost static, do not consider the classification situation after performance characteristic changes; This paper mainly obtains the conduction classificating knowledge that conduction transformation causes;

Before conduction transformation, domain is divided, and according to the attribute of performance correlation function, can be divided into domain static state:

$E_{+}^j=\left\{m_j\right|m_j\&Element;M_O,y=k_j\left(m_j\right)>0\}---\left(25\right);$

$E_{-}^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)<0\}---\left(26\right);$

$E_0^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)=0\}---\left(27\right);$

Respectively will

With

Be called the positive territory of domain, negative territory and zero boundary;

After conduction transformation, domain is divided, and by the active conversion, reclassifies for the attribute of performance that conduction occurs:

With Be respectively the positive qualitative change of the attribute of performance territory that conduction transformation causes, negative qualitative change territory, positive quantity variable domain, negative quantity variable domain and open up the boundary.

Example: a kind of properties of product conduction Knowledge Discovery Method comprises the following steps:

The first step, use api interface and Access2003, VC++6.0 carry out the exploitation of screw air compressor on the three-dimensional graphics software of Solidworks, and set up corresponding screw air compressor product example storehouse on development platform; Air compressor machine is the second largest power source that is only second to electric power, be again the technique source of the gas that serves many purposes, its range of application spreads all over industry and the departments such as oil, chemical industry, metallurgy, electric power, machinery, light industry, weaving, automobile making, electronics, food, medicine, biochemistry, national defence, scientific research.A large amount of screw air compressor product type convergences are serious.In order to satisfy diversified customer demand, be necessary to set up corresponding screw air compressor product example storehouse in existing a large amount of similarly screw air compressor products.As space is limited, now provide the partial data in the existing screw air compressor case library of a certain enterprise, as shown in table 1.

Partial data in table 1 screw air compressor case library

Second step, obtain the client performance requirement of screw air compressor is described: air capacity is medium; Discharge pressure 1Mpa; Motor rated power 20kw left and right; Noise is moderate, and the people is had no significant effect; Weight 600kg left and right, convenient in carrying; 22 liters of left and right of lubricants capacity; Unit operation stability will be got well.By above-mentioned Fuzzy Demand performance specification, client's Fuzzy Demand is converted into the first parameter model of configuration performance base:

The 3rd step, performance requirement retrieval obtain like product example: the screw air compressor attribute of performance is carried out qualitative, quantitative description; Utilize formula (1 and the 1 data normalization pre-service of (2) his-and-hers watches, as table 2

Table 2 screw air compressor part case library performance normalization data

Describe according to the customer demand matter-element, found the solution by formula (3), (4), (5) and (6).Similarity for air capacity is calculated, and the client always wishes to be the bigger the better in certain range of attributes, calculates local similarity by formula (5); For the weight attribute, the client wishes as much as possible little, calculates local similarity by formula (4); Calculate the similarity matrix of demand and case library by local similarity:

$s=\left[\begin{array}{ccccccc}0.6553& 1& 0.9182& 0.975& 0.9262& 0.9184& 0.95\\ 0.7369& 0.8& 0.3636& 0.725& 0.1144& 0.1224& 0.8833\\ 0.6421& 0.8& 0.9182& 0.975& 0.9262& 0.9184& 0.7166\\ 0.8421& 0.8& 0.5& 0.725& 0.1144& 0.1224& 0.8833\\ .& .& .& .& .& .& .\\ .& .& .& .& .& .& .\\ .& .& .& .& .& .& .\\ 0.8947& 1& 0.6818& 0.875& 0.4465& 0.3265& 0.8833\\ 0.7369& 1& 0.9818& 0.935& 0.9631& 0.9592& 0.95\\ 0.579& 0.8& 0.1818& 0.725& 0.1144& 0.1224& 0.8833\end{array}\right]$

Obtain each performance characteristic weights according to analytical hierarchy process (AHP) as follows:

w=[0.3691,0.2417,0.0675,0.1334,0.0406,0.0267,0.1212]

By formula (7), can computing client to performance requirement and the heavy overall similarity of compressor case library all properties Feature Weighting of screw compressor:

sim(R,E _i)=[0.7850 0.5406 0.8183 0.5992 0.6499 0.8476 0.6676 0.8467

0.8448 0.8613 0.6820 0.7319 0.8001 0.8544 0.6828 0.8281 0.8153

0.8527 0.7087 0.6336 0.8673 0.7245 0.8346 0.8327 0.8249 0.4542] ^T

Calculate the most similar case similarity sim by similarity _max, obtaining the retrieval product example is LG-6.3/10.

The 4th step, screw air compressor contradictory problems conflict resolution:

The screw air compressor contradictory problems is described: describe according to performance requirement object instance G and performance condition example L contradictory problems, foundation can be opened up the contradictory problems model:

Set up each demand performance correlation function of screw compressor:

By screw air compressor performance requirement evaluating characteristic, set up each evaluating characteristic correlation function k _i(x _i), and by judgement

Extract the condition example aspects that needs to improve conversion.With reference to factors such as GB and properties of product experience, measurement index, demands, can quantitatively determine its degree of association.And using formula (8) ~ (13), carry out correlation function and calculate.

For air capacity performance requirement correlation function k ₁(x ₁), require between desired region and the qualitative change interval is X=＜6.23m ³/ min, 10m ³/ min 〉, and at M=10m ³Reach maximum during/min, set up its simple correlation function:

$k_1\left(x_1\right)=\left\{\begin{array}{c}\frac{x-6.23}{4},x<10,\\ \frac{10-x}{4},x>10,\\ 1,x=10.\end{array}\right.$

Obtaining similar example air capacity performance parameter is 6.3m ³/ min, degree of association k ₁=0.075〉0.

For discharge pressure performance requirement correlation function k ₂(x ₂), set up the discrete type correlation function:

$k_2\left(x_2\right)=\left\{\begin{array}{c}1,x=1,\\ -1,x\&NotEqual;1.\end{array}\right.$

Obtaining similar example discharge pressure performance parameter is 1Mpa, degree of association k ₂=1〉0.

For motor rated power performance requirement correlation function k ₃(x ₃), require between desired region and the qualitative change interval is X=＜35kw, 46kw, and reach maximum when M=46kw, set up its simple correlation function:

$k_3\left(x_3\right)=\left\{\begin{array}{c}\frac{x-35}{11},x<46,\\ \frac{46-x}{11},x>46,\\ 1,x=46.\end{array}\right.$

Obtaining similar example air capacity performance parameter is 45kw, degree of association k ₃=0.91〉0.

Noiseproof feature demand correlation function k ₄(x ₄) set up X=＜50dB, 107dB, X ₀=＜60dB, 70dB 〉, optimal value x ₀=62dB, set up its simple correlation function:

$k_4\left(x_4\right)\left\{\begin{array}{c}\frac{60-x}{|x-78.5|-|x-65|-23.5},x\&le;60,\\ \frac{4(60-x)}{|x-78.5|-|x-65|-33.5},60<x<62,\\ \frac{x-70}{|x-78.5|-|x-65|-33.5},62\&le;x\&le;70,\\ \frac{x-70}{|x-78.5|-|x-65|-23.5},x>70.\end{array}\right.$

Obtaining similar example noiseproof feature parameter is 80dB, degree of association k ₄(80)=-0.27＜0.

For complete machine weight performance requirement correlation function k ₅(x ₅), require between desired region and the qualitative change interval is X=＜900kg, 1000kg, and reach minimum when M=900kg, set up its simple correlation function:

$k_5\left(x_5\right)=\left\{\begin{array}{c}\frac{x-900}{100},x<900,\\ \frac{1000-x}{100},x>900,\\ 1,x=900.\end{array}\right.$

Obtaining similar example weight performance parameter is 1100kg, degree of association k ₅=-1＜0.

Lubricants capacity performance requirement correlation function k ₆(x ₆) set up, require between desired region and the qualitative change interval is X=＜30 liter, 35 liters 〉, and reach minimum when M=30 rises, set up its simple correlation function:

$k_6\left(x_6\right)=\left\{\begin{array}{c}\frac{x-30}{4},x<30,\\ \frac{34-x}{4},x>30,\\ 1,x=30.\end{array}\right.$

Obtaining similar example weight performance parameter is 35 liters, degree of association k ₆=-0.25＜0.

For stability demand correlation function k ₇(x ₇) set up, set up discrete correlation function

Obtain similar example stability parameter for well, degree of association k ₇=1〉0.

According to above like product analysis result, obtain performance requirement object instance and performance condition example correlation function K (P)=k ₁∧ k ₂∧ k ₃∧ k ₄∧ k ₅∧ k ₆∧ k ₇=0.075 ∧ 1 ∧ 0.91 ∧ (0.27) ∧ (1) ∧ (0.25) ∧ 1＜0 need to carry out first expansion of thing to above-mentioned inconsistent problem and analyzes and carry out opening up conversion.By calculating more similar example performance characteristic and demand performance correlation function value, obtain similar example noise, weight and lubricants capacity performance characteristic degree of association value less than zero, do not satisfy performance requirement.

The 5th step, large-scale screw air compressor conduction knowledge excavation: the like product LG-6.3/10 that obtains, in case library, its performance characteristic parameter is as follows:

Table 3LG-6.3/10 performance characteristic parameter

By the comparative analysis to like product and performance requirement product, clear up model by the contradiction of setting up performance requirement and similar example that can open up the Matter-element Method qualitative, quantitative to screw air compressor, build simultaneously the contradictory problems correlation function; And realize the performance structure mapping in conjunction with the first analytical approach of expanding of thing, obtain multiple contradictory problems solution; Utilization can be opened up transform method (as table 4), by property demand characteristic and object instance feature, obtains following several conversion scheme (as table 5).

Table 4 basic transformation type and priority

Table 5 screw air compressor Local Property Structural Transformation initial scheme

Calculate and the mode such as l-G simulation test is obtained performance characteristic data after conversion by theory, and these data are added in performance characteristic database after conversion.

The pressure loss as silencer internal wall and gas current friction generation:

Δ P _RubFrictional resistance (Pa), ξ _RubCoefficient of frictional resistance, a sound suppressor effective length (m), d _eNoise damping channel equivalent diameter (m) rectangular tube is 2ab/ (a+b), and ρ is atmospheric density 1.29kg/m ³, v is flow velocity m/s, g is acceleration of gravity

Local pressure loss

Δ L=Δ L _Office+ Δ P _Rub

Calculate by theory, obtain each performance parameter data of screw air compressor of many initiatively conversion front and back:

Example performance characteristic parameter after the conversion of table 6 screw air compressor

First according to data message before and after the most similar screw air compressor sound suppressor being implemented many initiatively conversion, the difference (as table 7) of front and back data message unit is obtained in calculating, can find to occur the performance characteristic of conduction transformation from the data difference of conversion front and back, which namely searches out is conduction feature.

The difference of table 7 screw air compressor conversion front and back data message unit

By given threshold value δ _i={ δ ₁=0.5, δ ₂=10, δ ₃=1, δ ₄=1, δ ₅=0.5, δ ₆=0.005, δ ₇=0.5}〉0, the conduction feature that causes after judgement active conversion.The screw air compressor conduction feature that many initiatively conversion cause comprises noise, admission pressure, weight and air capacity (as table 8).

Table 8 screw air compressor conduction feature

By the judgement of performance conduction feature, obtain single order conduction transformation collection (noise, admission pressure, weight), simultaneously according to the correlativity between the performance characteristic of screw air compressor, obtain n conduction transformation feature (admission pressure, air capacity).

Obtain the many initiatively conversion of sound suppressor about the conductivity of each performance conduction feature of screw air compressor.By chapter 4 many actives transformation parameter collection that many initiatively conversion obtain to sound suppressor be

V _i(v _j,v′ _j)={V ₁,V ₂,…,V ₆}={{(v ₁,v’ ₁),…,(v ₅,v′ ₅)};…;{(v ₁,v’ ₁),…,(v ₈,v′ ₈)}}

={{(9500,5000),(60,85),(60,85),(180,340),(140,161.5),(0,80.75),(0,40.35)};

{(9500,6000),(60,85),(60,85),(180,340),(140,161.5),(0,80.75),(0,40.35)};

{(9500,5000),(60,85),(60,85),(180,340),(140,161.5),(0,80.75),(0,40.35),(161.5,58.5)};

{(9500,5000),(60,85),(60,85),(180,340),(140,161.5),(0,80.75),(0,40.35),(161.5,58.5)};

{(9500,5000),(60,85),(60,85),(180,340),(140,161.5)};

{(9500,5000),(60,85),(60,85),(180,340),(140,161.5)}}

Obtain initiatively conversion weight by analytical hierarchy process

$\mathrm{AHP}\left(\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i\right)=\{\left(\mathrm{0.024,0.109,0.090,0.427,0.242,0.242,0.059,0.049}\right);$

(0.024,0.109,0.090,0.427,0.242,0.059,0.049);

(0.018,0.132,0.114,0.353,0.195,0.041,0.036,0.111);

(0.018,0.132,0.114,0.353,0.195,0.041,0.036,0.111);

(0.030,0.096,0.074,0.520,0.280);

(0.030,0.096,0.074,0.520,0.280);}

Utilize publicity (15), (16) to calculate the conductivity value of each conductive performance feature;

Table 9 screw air compressor conduction feature conductivity

Influence degree by conductivity value judgement active transfer pair conduction feature; Due to the correlativity of air capacity and admission pressure, cause initiatively conversion to cause repeatedly conduction, the conductivity value is also to weigh repeatedly to conduct the influence degree index.Have a plurality of single order conduction features and secondary-conduction feature in table 5-6, single conductivity evaluation method can not effectively be determined the quality of scheme, introduces based on extension classification conduction Knowledge Discovery Method and obtains performance conduction rule change, provides optimal case.

To like product LG-6.3/10, to analyze and obtain final six kinds of conversion scheme, and obtain noise, admission pressure, weight, air capacity are conduction feature, and each Autocorrelation Function of determining by calculating chapter 4, and six kinds of conversion scheme are carried out extension classification.

The noiseproof feature extension classification that causes due to conduction transformation for LG-6.3/10:

Negative quantity becomes

Positive qualitative change

The weight performance extension classification that causes due to conduction transformation for LG-6.3/10:

Positive quantitative change

Negative qualitative change

The variation of the air capacity performance parameter that causes due to conduction transformation for LG-6.3/10 is the variation of the admission pressure that produces due to Structural Transformation, thereby causes the conversion of air capacity, belongs to secondary-conduction, the extension classification that causes due to conduction repeatedly:

Positive quantitative change

Negative qualitative change

Obtain the transformation results scheme by analyzing many initiatively conversion c

The performance conduction feature noise that causes is positive qualitative change, and weight is positive quantitative change, and air capacity is positive quantitative change.Extracting the conduction knowledge that this active conversion causes is:

Claims (3)

1. properties of product are conducted Knowledge Discovery Method, and it is characterized in that: described conduction Knowledge Discovery Method comprises the following steps:

1), build the product example storehouse: develop, build the product example storehouse on drawing analysis software platform;

2), client's performance requirement matter-element is described: the product performance demands matter-element is with this product O _mBe object, c _mBe requirement product performance characteristic, v _mFor about c _mThe orderly tlv triple that consists of of value, if this product O _mA plurality of feature c are arranged ₁, c ₂..., c _nAnd corresponding value v ₁, v ₂..., v _n, represent with multidimensional performance requirement matter-element:

$M=\left[\begin{array}{ccc}{O}_m& c_1& v_1\\ & c_2& v_2\\ & .& .\\ & .& .\\ & .& .\\ & c_n& v_n\end{array}\right];$

3), performance requirement product example retrieval:

Based on opening up apart from the demand similarity algorithm: performance requirement data dimensionless is processed between the back zone

Case library performance data dimensionless is processed Y=[y between the back zone ₁, y ₂], and

Apart from as follows with distance similarity model formula:

$d(V,Y)=\mathrm{\&rho;}(V,Y)=|\frac{y_1+y_2}{2}-\frac{v_{\mathrm{\&alpha;}1}^l+v_{\mathrm{\&alpha;}1}^r}{2}|-|\frac{y_2-y_1}{2}-\frac{v_{\mathrm{\&alpha;}1}^r-v_{\mathrm{\&alpha;}1}^l}{2}|---\left(3\right);$

Process concrete numerical value if x is the performance requirement dimensionless, the left side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}{y}_1-x,x\&le;\frac{y_1+y_2}{2},\\ x-\frac{y_1+y_2}{2},x>\frac{y_1+y_2}{2}.\end{array}\right.---\left(4\right);$

The right side is apart from as follows:

$d(x,Y)\left\{\begin{array}{c}\frac{y_1+y_2}{2}-x,x<\frac{y_1+y_2}{2},\\ x-y_2,x\&GreaterEqual;\frac{y_1+y_2}{2}.\end{array}\right.---\left(5\right);$

Calculate the most similar case similarity sim by similarity _max, obtain the most similar example L;

4), clear up based on the performance structure contradiction that can open up conversion: compare demand example and each performance parameter of the most similar example, do not meet the demands if any performance parameter, performance requirement object instance G exists with performance condition example L and conflicts, the contradiction problem represents with P=G*L, expand by the performance topology requirement and analyze, acquisition methods thing meta-model, then carry out opening up conversion; Comprise the following steps:

(4.1), demand and condition question are described: for given problem P=G*L, for performance requirement object instance and performance condition example correlation function K (P)＜0, the conversion T of target and condition _P=(T _G, T _L) be the solution conversion of P, make K (P)〉0, satisfy the client for the performance requirement of product;

(4.2), product thing meta-model is expanded and is analyzed: by calculating more similar example performance characteristic and demand performance correlation function value, obtain similar example performance feature association degree value less than zero, do not satisfy performance requirement, need to carry out emission analysis to contradictory problems, then analysis is contained in thing unit, obtain to open up conversion scheme, calculate T _GAfter the degree of association, this process makes K (P) repeatedly〉0 realize the performance requirement goal satisfaction;

5), performance conduction knowledge excavation, process is as follows:

(5.1), the judgement of conductive object and conduction feature: given threshold value δ〉0, when the value of the object that conversion occurs or feature poor

(

Value before and after the expression conversion) absolute value satisfies:

$\left|\mathrm{\&Delta;}{v}_{l_i}\right|=|v_{l_i}^{\&prime;}-v_{l_i}|>\mathrm{\&delta;}---\left(14\right);$

The time, think corresponding object or be characterized as conductive object or conduction feature;

(5.2), conductivity is calculated: for the product example structure feature information l of unit _i=(o _ic _iv _i), obtaining information word after conversion is l ' _i=(o _ic _iv’ _i); If there is initiatively conversion

Tl _i=l’ _i=(o _i c _i v’ _i) (15)；

This is another object of transfer pair or feature m initiatively _j=(q _jd _ju _j) the generation conduction transformation

Make m _jBe transformed to m ' _j=(q _jd _ju’ _j), namely can be expressed as

And conductivity γ namely is expressed as

$\mathrm{\&gamma;}=\left|\frac{u_j^{\&prime;}-u_j}{v_i^{\&prime;}-v_i}\right|---\left(17\right);$

(5.3), many initiatively conversion conduction knowledge excavations:

For the structural information l of unit ₁, l ₂, l ₃..., l _hL satisfies condition:

I=1,2 ..., h; J=1,2 ..., k is if there are a plurality of active conversion to cause the generation of a plurality of conduction transformation, namely

There is conduction transformation

And the transformation parameter of active more than obtaining collection

V(v _i,v’ _i)={(v ₁,v’ ₁),(v ₂,v’ ₂),…,(v _h,v’ _h)} (19)；

The conduction knowledge that many initiatively conversion cause is designated as

When many actives conversion

The time, the first conductivity of each conducts information is

${\mathrm{\&gamma;}}_i=\frac{|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(21\right);$

Wherein: $\mathrm{AHP}\left(\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i\right)=1---\left(22\right);$

If conduction feature m _jWeight in whole conduction features is β _j, j=1,2 ..., k, and $\mathrm{AHP}\left(\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_j\right)=1,$ Comprehensive conductivity is

$\mathrm{\&gamma;}=\frac{\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_j|u_j^{\&prime;}-u_j|}{\underset{i=1}{\overset{h}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_i|v_i^{\&prime;}-v_i|}---\left(23\right).$

2. a kind of properties of product as claimed in claim 1 are conducted Knowledge Discovery Method, and it is characterized in that: described step 5) also comprises following process:

(5.4), based on the extension classification knowledge of conduction transformation, the detailed process step is as follows:

For product demand performance collection, k _j(m _j) represent that demand performance and product have the degree of association of example performance now, Corresponding configuration primitive structure, the initiatively conversion of material properties of performance characteristic thing meta analysis described;

Expression

The conduction transformation that the attribute of performance that causes changes is for domain M _OAbout the performance characteristic attribute transformation

The config set opened up;

Obtain the conduction classificating knowledge that conduction transformation causes;

Before conduction transformation, domain is divided, and according to the attribute of performance correlation function, domain static state is divided into:

$E_{+}^j=\left\{m_j\right|m_j\&Element;M_O,y=k_j\left(m_j\right)>0\}---\left(25\right);$

$E_{-}^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)<0\}---\left(26\right);$

$E_0^j={\{m}_j|m_j\&Element;M_O,y=k_j\left(m_j\right)=0\}---\left(27\right);$

Respectively will

With

Be called the positive territory of domain, negative territory and zero boundary;

After conduction transformation, domain is divided, and by the active conversion, reclassifies for the attribute of performance that conduction occurs:

With

Be respectively the positive qualitative change of the attribute of performance territory that conduction transformation causes, negative qualitative change territory, positive quantity variable domain, negative quantity variable domain and open up the boundary; Described positive qualitative change, negative qualitative change, positive quantitative change, negative quantity become knowledge and are properties of product conduction transformation extension classification knowledge excavation result.

3. a kind of properties of product as claimed in claim 1 or 2 are conducted Knowledge Discovery Method, and it is characterized in that: described step 3) comprises following process:

(3.1), attribute of performance quantitative description: raw data is carried out dimensionless process, namely the value of characteristic parameter all is converted into [0,1] interior value;

The raw data normalization mode of a certain performance characteristic value of product is as follows:

$y_{\mathrm{ij}}=\frac{x_{\mathrm{ij}}-{\min x}_j}{{\max x}_j-{\min x}_j}---\left(1\right);$

Or:

$y_{\mathrm{ij}}=\frac{\max x_j-x_{\mathrm{ij}}}{{\max x}_j-{\min x}_j}---\left(2\right);$

In formula, x _ijProduct example storehouse performance data before expression is processed, y _ijProduct example storehouse performance data after expression is processed, i=1,2 ..., m, j=1,2 ..., n, maxx _jAnd minx _jRepresent respectively the minimum and maximum value of product example storehouse j item performance characteristic;

(3.2), based on opening up apart from the demand similarity algorithm: performance requirement data dimensionless is processed between the back zone

Case library performance data dimensionless is processed Y=[y between the back zone ₁, y ₂], and

Formula is as (3), and the left side distance is suc as formula (4), and the right side distance is as shown in the formula (5);

(3.3), the calculating of similarity: similarity is a kind of tolerance of the similarity relation of requirement product and existing procucts example, and the rank according to similarity in Similarity algorithm is different, and similarity can be divided into:

(3.3.1), local similarity (Part Similarity): interval and interval, point is with interval, and point with some local similarity search formula is:

sim(V,Y)=1-|d(V,Y)| (6)；

A, B can be interval, can be also concrete a certain numerical value;

(3.3.2), overall similarity: customer demand and the heavy calculating formula of similarity of product example storehouse all properties Feature Weighting:

$\mathrm{sim}(R,E_i)=\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{w}_j\mathrm{sim}(R_i,E_i)---\left(7\right);$

Sim (R, E _i) expression example E _iWith the overall similarity of demand configuration design problem R, sim (R _i, E _i) be example E _iConfigure design problem R at attribute of performance c with demand _iOn similarity, the product example number is n, the attribute number is m, determines performance attribute weight w by analytical hierarchy process _j

Calculate the most similar case similarity sim by similarity _max, obtain the most similar example L.

Images