Hierarchical clustering method in algorithm is illustrated as follow [11]: Case base {x 1 , x 2 … x N } values {xc1 ,xc2 ,… ,xcN }. Each case is regarded as one class to get the final clustering value c. The steps of this algorithm are: 1) At the beginning, set Rj={xj }, ∀ j∈I ,I ={1 ,2 ,…,N}. 2）Using euclidean distance stands for the class separation distance. For one-dimensional problems, there is D(R i, R j) =| xci- xcj|. Seek a pair of satisfying conditions: clustering R j and R k of Δ(R i, R k ) = I i , jmin∈ ∀{Δ(R j, Ri)}. Thereinto, Δ(R j, R i) =I i , jmin∈ ∀{D(R j, R i) }. 3) Let class Rj merge into class Rk, and delete class Rj. 4) Delete j from index set I. If the cardinal number of I is equal to c, the algorithm stop or turn to step 2). There gives an applied example using hierarchical clustering methods from attribute b of table 1. In figure 2, numbers on the lines stand for the nth clustering and class separation distance, and show the possible classification of this attribute. Figure 2 Example of Classification Clustering Algorithm According to definition 5, KC (D)= card ( posC ( D) )/ card (U) stands for support of C to D. And posC ( D) is D’ positive region C and it is a case set which stands for that the classification information of U/C could accurately belong to decision table D. Let C contain n attributes in total, which includes k quantitative attributes and n-k qualitative attributes. Each quantitative attribute may have dk kinds of classification methods. Encode all kinds of discrete ranges by 1, 2 … Discrete problems transform into solving max KC( D) optimization problems. It could use some optimization methods to solve problems, for example, genetic algorithm. For avoiding too detailed classification in the premise of keeping support, we select the results which have the least number of classifications as the final results [12]. The important degree of all attributes discretized can be judged by formula (1) and (2): rC (D) = card ( posC ( D) ) / card (U) （1） rC-a(D) = card ( posC-a (D)) / card (U) （2） Using rC (D)－rC-a (D) judges the important degree and the basis for selecting weight. We’ll review the changes of classification positive range after throwing off attribute a from condition attributes. If it does not change rC-a (D), it will prove to effect on original classification without attribute a. That is to say, attribute a isn’t the essential one, and it can be deleted in attribute value table. Deleted unnecessary attribute is one reduction of condition attribute, we expresses it as RedD( A). Attribute reductions usually have several ones. We always focus our attention on the reduction which has minimal number of attributes; we call the minimal reduction, represented it as mRedD (A). Minimal reduction represents the characteristic set which includes the complete characteristics but minimal number attributes. This paper establishes index based on the minimal reduction of the characteristic attributes. 4. Similarity degree calculation The design based on case realizes the similarity matching by mainly adopting the nearest neighbor algorithm. The similarity degree is obtained by mainly adopting distance calculation algorithm. But this method is not analyzed from the composing element and correlativity and ignores the comparability of composing element characteristic. It leads to calculate the similarity coefficient that can not commendably reflect the similarity degree of case. We do not only take into account the quantity of two similarity elements, but also calculate the similarity value of the public similarity elements. The calculation formula based on the similarity degree preferably expresses the similarity extent between the similarity cases [13]. Its representation is shown as follow: S= ) u ( qn l knin1 ii ∑ = − +β （3） The letter k,l and n severally expresses the composing elements quantity and the public similarity elements quantity of the similarity cases on the certain layer in formula 3. The q(u i) is the value of the similarity element i. The iβ is the weighting coefficient of q(u i). The case attribute is pided into the qualitative attributes and quantitative attributes. The two attributes need be processed respectively [14]. The calculation formula of quantitative attributes is shown as follow: ) u ( q i=)} b ( u ), ( u { max)} b ( u ), ( u { mini j i ji j i jαα （4） The ) ( u i jα and ) b ( u i jexpress respectively the characteristic values of the similarity element attribute j in two cases, and 0< ) u ( q i≤1. The calculation formula of qualitative attributes is shown as follow: ) u ( q i=⎪ ⎩⎪⎨⎧=≠ ) b ( u ) ( u 1, ) b ( u ) ( u 0,i j i ji j i jαα （5） 5. Application and calculation The example of stamping die design is introduced in this paper. Because stamping die design involves many attributes, we only select several attributes to analyze and discuss. Suppose the case base consists of cases whose serial numbers are from Ⅰ toⅤ: domain U= {Ⅰ,Ⅱ,Ⅲ,Ⅳ, Ⅴ}. Case characteristics are described by 9 attributes: C={a, b, c, d, e }, D ={ f, g, h }.They are shown in the Table 1 as follow: Table 1 Case characteristic attribute table Case number Die accessory（a） Material thickness（mm）(b) Material strength (Mpa) (c) Circular arc radius (mm) (d) Material Type (e) Die type (f) Fixation mode (g) Size tolerance (mm) (h) Ⅰ Punch 4.2 410 0.65t High carbon steel Single working procedure die Orientation plank 0.030 Ⅱ Punch 7.0 340 0.55t Low carbon steel Single working procedure die Orientation plank 0.020 Ⅲ Cavity Die 9.6 400 0.75t Brass Continuous Die Underlay plank 0.030 Ⅳ Punch 7.5 440 0.60t Low carbon steel Complex die Underlay plank 0.040 Ⅴ Cavity Die 6.8 440 0.55t Low carbon steel Complex die Orientation plank 0.040 The first, we discretize the quantitative attributes in order to become qualitative attributes. The result is shown in table 2 as follow. Table 2 Discretization attribute table Case number Die accessory（a） Material thickness（mm）(b) Material strength (Mpa) (c) Circular arc radius (mm) (d) Material Type 案例检索算法冲压模具英文文献和中文翻译(3):http://www.youerw.com/fanyi/lunwen_33959.html