TW200534147A - The suitable method for selection of manufacturing factors and the best combination of interactive factors under multiple quality properties - Google Patents

Abstract

This research uses Data Mining to effectively analyze and confirm that the factors affecting discharge for Electrical Discharge Machining (EDM) are, pulse-on time, pulse-off time, open discharge voltage, or interval voltage. The Taguchi Method is used to experiment with electrodes of the same size and different shapes, based on key factors acquired under the consideration of the interaction between factors, and analyze the experimental results respectively with average values and S/N ratio. Eventually, the research will determine and calculate the best factor combination of each quality property (machining and reaming amount, surface roughness, electrode corner loss, and material removal rate (MRR)) by analyzing predicated values and prove it with experimental data. The experiment proves that the interaction between factors really exists and that the best factor combination acquired has favorable effects and credibility.

Description

200534147 0 IX. Description of the invention: [Technical field to which the invention belongs] The present invention relates to the improvement of various industrial processes, and generally refers to the aspects of the process where the industry is at a loss and the parameters are complex and difficult to judge. In particular, considering the interaction between process parameters, how to effectively find the main parameters that affect the process, and at the same time to find the optimization method of process parameter combinations with multiple quality characteristics. [Previous technology] In order to improve product quality, the industry mostly adopts the accumulated experience of past actual operation or the method of trial and error as a reference when processing. ♦ Most processing machine manufacturers only provide judgment systems or reference books. Help the manufacturer choose processing conditions or process planning according to processing requirements. The purpose is to meet the processing requirements and choose a better parameter combination control method. In recent years, the academic community has focused on the establishment of algorithms that meet multiple quality characteristics and the use of various methods to solve the optimization of processing processes. Among the earlier literatures (Osyczka et al., 1982) proposed the use of multi-objective decision-making ( multi-criteria) optimization method improvement | multi-quality characteristics problem. (Lin et al., 2000) proposed a combination of Taguchi method and fuzzy theory to analyze the optimal parameter combination for multi-quality characteristics. In summary, although the above-mentioned literature combines various methods to improve the quality characteristics, it also lacks the basis to influence the selection of the main parameters of the process. At the same time, in the course of the experiment, each paper only considered the quality characteristics and ignored the interaction relationship between the factors, so that the final results were not the same, and could not be effectively applied to industry practice. Therefore, an effective analysis of the interaction between the main parameters of the process and the consideration of factors can really improve the quality problem. 6 200534147 Rough set theory (Data Mining) is mainly used for analysis of uncertain or inaccurate data. It mainly assists decision-makers to use classification and attribute reduction methods for a large database. Minimal decision algorithm that enables decision makers to quickly find the source of the problem. Related literature studies such as (Kusiak and Kurasek, 2001) use rough set theory to identify the factors that cause solder balls in the printed circuit during soldering to find out the problem. (SriniVasan and Moon, 1999) A group of clustering algorithms is proposed based on many quality characteristics. The Xiang average link method is used as the clustering rule. As long as the product characteristics that have an effect on performance are included in the algorithm, the crowds are clustered on the basis. . (Including Han et al., HU to analyze semiconductor process problems, to analyze the strategic tree, find the shadow station and pass through the machine _, thereby narrowing the source of process ㈣. = Star 4, touch side secret and rough set theory In the process of chemical products, find the classification of each attribute and possible value of the system, and use the Bayesian network to find the probability value for each condition. In order to effectively analyze the influence of control parameters from many process parameters, the discriminant matrix method (D1SCernibllity matrix) in rough set theory is used. The main purpose of this method is to perform a set of all attributes (parameters) in an information system. Attribute simplification. The process is to put all the data related to the attributes (parameters) in the database into a matrix. After logical operations, the core in the database is obtained. (reducts) attributes. Sh Taguchi method is mainly aimed at the development of low-cost and robust production port 'at the same time pay attention to the optimization of products and manufacturing methods. In the past; ^ Taguchi method of literature research or practical application Most of them focus on the processing of single-characteristic system 7 200534147 1 # process optimization, but the quality characteristics of the product process is often not unique. If you only seek to optimize the process with a single quality characteristic, except that the true process may not be achieved. Demand, it also loses many opportunities to improve product quality characteristics and reduce costs. For example, the early (Phadke, 1989) applied the Taguchi method to super In the polycrystalline silicon deposition process of the integrated circuit, if the optimization factor level selection conflicts between different quality characteristics, use relevant engineering experience or knowledge to make a judgment to determine the overall optimal factor level. Combination. This method depends on the relatively subjective knowledge and experience of the engineer. Although the analysis is simple, there are no clear judgment rules ^, which also makes engineers with less work experience incomprehensible. In addition, if the same product or process needs to consider the quality at the same time When the characteristics are more complicated, it is bound to rely on human experience alone. • Judgement may not be easy to achieve. (Logothetis and Haigh, 1988) The field method combined with multiple regression analysis and multi-objective decision analysis is applied to plasma etching. For each quality characteristic, this method first calculates the noise performance statistic (NPS) and target performance statistic (TPS) for each quality characteristic value, and determines each quality characteristic separately. The best level combination of the control factors of the value. With reference to multiple regression analysis, each quality characteristic value is a dependent variable, and the control factor is an independent variable. The linear regression formula is obtained. Finally, the multi-objective decision analysis method is used to find the optimal combination of factor levels in the process. Although this method simplifies the problem from the multi-dimensional space through the addition of weights to the one-dimensional space to simplify the problem, the correlation between the quality characteristics has not been included in the decision-making process. Whether the results can be truly the best combination is open to question. Moreover, the calculation process of this method is complicated, and it is doubtful whether it can be applied in practice. (Xiao Gangheng, 1990) Take the Taguchi method applied to the iron sintering process as an example. First, each quality characteristic value measured in each experiment is converted into an S / N ratio, and 8 200534147 is assigned.

I A gives a weight to the S / N ratio of each quality characteristic value, and adds up the S / N ratios of various quality characteristics to serve as a performance measure for multiple quality characteristic values, and finally finds the best combination of factor levels. Although this method is simple and easy to understand, it is not easy to determine the weight of each quality characteristic, so the rationality of this result needs to be studied. In this study, the Taguchi method is used as a method to solve the problem of the best factor level of quality characteristics considering the interaction between process parameters. The experimental data is analyzed by the S / N ratio method, and the prediction value analysis is proposed as a solution. Considering the interaction between process parameters, the problem of factor level conflict caused by quality characteristics is the final indicator of product development or process. Although many scholars have proposed many ways to solve the optimization of multiple quality characteristics, the importance of quality characteristics has never been demonstrated. (Derringer and Suich, 1980) The ideal function is applied to the process problem with multiple quality characteristics, and the multiple regression analysis method is used to construct a linear regression formula for each quality characteristic and controllable factor. Although this method can simplify the problem of multiple quality characteristics, the calculation process of this method is complicated, and it is doubtful whether it can be applied in practice. (Nian and Tarng, 1999) Normalize the loss function of each quality characteristic of the lathe process, and then give different weights to the different importance of product quality to calculate the total loss function, and then convert the loss function into S / N ratio Then select the parameters and levels, and finally define the best parameters. This method is not easy to determine the weight of each quality characteristic, so the rationality of this result needs to be studied. (Pignatiel lo etal., 1993) redefines a loss function that takes into account the correlation of quality characteristics as a performance evaluation index for multiple quality characteristics products and processes, and proposes strategies for optimizing multiple quality characteristics. This method is not easy to accurately estimate the quality loss function, and more experiments are needed to estimate the accurate total variation matrix. (Tong and Su, 1997) proposed the use of principal component analysis to solve the problem of optimizing multiple quality products and processes. The example does not explain how to choose the best combination when two or more principal components are disposed. Therefore, it is questionable whether the best combination can be obtained without fully using all the information of each quality characteristic value. Many scholars in the literature have proposed methods to deal with multiple quality characteristics, most of which are mainly regression analysis and grey correlation analysis. However, they have the following disadvantages. Large amounts of data.

2. Require the sample to have a good distribution law, especially linear regression is better. If the sample is non-linear, it will cause a lot of calculations. Therefore, it can be known that the regression analysis method has greater limitations, and the gray theory proposes a new analysis method for the above shortcomings, called the gray correlation analysis method. The degree of similarity or dissimilarity between development trends is used to measure factors. The factor of two factors (GreyRe〗 ationaIAnaIy___ T element analysis features can make up for the shortcomings of general statistical regression. Earlier: two:: ° () using gray correlation analysis method to solve the multiple aspects of the semiconductor process =: _ 陶The gray multi-attribute decision-making method and Tian Wan method are used to solve the problem of multi-quality characteristics of 1C process. Therefore, the present invention proposes the Taguchi gray correlation method, which has the advantages of high experiment and robustness and gray correlation 22: the advantages of fewer methods. Not only The best factor level combination of the negative characteristics of Tian σ square, heart data, and multi-factor analysis is used to find all the quality characteristics at each time. According to 200534147%, a set of process parameter optimization combinations meeting multiple quality characteristics is analyzed. [Summary of the Invention] In order to solve the above-mentioned problems such as process improvement, the present invention proposes a An optimized method for improving the manufacturing process. First of all, for the simplification of many parameters in the manufacturing process, the present invention combines the K-means clustering method and the discernibility matrix method as a method for analyzing the main parameters of the shadow process. First, the K- The clustering method is used to analyze the advantages and disadvantages of the experimental results of various quality characteristics. Then the discernibility matrix method is used to reduce the parameter set of all the process parameters in the experiment, and finally the core parameters or the most in the experiment are obtained. Reducts parameters are the main parameters that affect the manufacturing process. The present invention uses the Taguchi orthogonal table experimental plan method as the experimental design framework for optimizing the combination of process parameters with various quality characteristics. The experiment is performed under the consideration of the interaction between process parameters. At the same time, the experimental results are analyzed by comparison. Finally, the predicted value analysis is used to solve the problem of factor level conflicts that are considered during the interaction between process parameters. The experimental results are obtained. The predicted value of the best combination of quality characteristics and process parameters. In addition, the Taguchi method is used in miscellaneous literature research. The problems to be solved are single quality and characteristic. If the problem of multiple quality characteristics must be solved, other methods and theories must be matched. Therefore, the present invention uses the Taguchi method in combination with the holistic gray correlation analysis as a solution to the Taguchi method. The factor level conflict problem under quality characteristics, and at the same time, the optimal combination of process parameters is obtained by considering the interaction between factors and multiple quality characteristics. [Embodiment] > Simplification of process parameters The experimental method of making a full factor. The full factor includes the parameters of the recording and sales process, and the process is involved in money. In order to distinguish the advantages and disadvantages of the experimental results in 200534147, the present invention uses K-means clustering method. According to the expected value of quality characteristics, the experimental results are divided into several groups. The calculation process of the Li-I, K-means clustering method is shown in Figure i. The K-means clustering method performs step I. Determine n groups, and calculate the initial value of the center point of each group. ZM-', Zn {i) which stands for the nth ethnic group, the first-second center point.

Π · Bingxing 乂 The difference between each point and each center point, and then the points are grouped into a suitable group. _ ZjH \ ThenX 6 sxn) fori &j; j-_1 > 2r .. > n {2) where jst is a # material point with a restrictive value, and coffee is a set representing the first classification scale. Calculate the difference between each point in the two groups and each center point, and use the difference in points as the basis for regrouping. After H. shoulder to shoulder, recalculate the center point of each group according to formula ⑺. The calculation will be stopped when the new and old values of the center point are equal, or the points of the scale have not changed. Otherwise, step II must be performed again.

ZXn + Hj ^ = ^ ⑶ where # is 5 / ^; the number of meanings to which the group belongs. If the calculation result meets the convergence to all points in the ethnic group, the calculation is stopped. Then, the present invention adopts RGUgh Set as a method for simplifying process parameters, so as to effectively obtain the most important parameters that affect the process. The calculation process is mainly to put the data related to the database cap ^ related parameters (attributes) into a matrix, and then through logical operations, know the core (core) attributes or the most simplified (reduct) Magic attribute. This method is defined as 12 200534147

The recognition matrix can be expressed by the following equation: ‘domain and &^; ^^^, ···, '}, d is the value of x on the attribute α, so it can be discerned

1, ΦΟ = φ ^),

Where Z) is the decision attribute, α is the condition attribute, and its differentiating step (Walczak and construct the discernible matrix according to the above equation (4)

Massart, 1999) is as follows: 1. First determine the decision attributes (D): 2. If the decision attributes (D) are equal, the condition attributes are substituted with 0. 3. If the decision attributes (D) are not equal, then judge the condition attributes sequentially. 4. If the condition attributes are equal, the condition attributes are substituted with 丨. 5. If the condition attributes are not equal, substitute the condition attribute. After the construction of the discernible matrix is completed, an OR operation is performed on each attribute value in each block, and then an AND operation is performed with the remaining blocks. The obtained calculation formula is finally obtained by logical operation to obtain its core attribute or the most Where the reducts attribute is. Among them, if the calculated result is an individual independent factor ', this factor is called the core attribute of the experimental table, otherwise it is called the most simplified attribute of the pass table. > Optimization of process parameter combination of each quality characteristic The present invention uses Taguchi orthogonal experiment plan method as the experimental design framework for analyzing the optimization of each quality characteristic process parameter combination, and considers the main parameters affecting the process. The experiment was performed under the interaction between process parameters, and the experimental results were analyzed by S / N ratio. 13 200534147 Finally, predictive value analysis is used to solve the problem of level conflict. And find the factor of heart & / interaction. Benfa_ Records her He's predictions based on the actual production process. Centering === r depends on a. Determine the quality characteristic α. Quality characteristics are the ultimate issue in product development or manufacturing processes. According to _ teacher lose Xiaoling ^ miscellaneous eyes = determine b. Determine the level of the factor The parameters of the ancient Sr-free process will be called factors. Each of its factors has a different value, "Fan_ 卩 is a revolution." The setting of the transition value must be set at a different level. The effect on the quality characteristics is very obvious. Due to the setting of C. Orthogonal form, the parameter factor is calculated according to the degree of freedom. The post is the most suitable for delivery form configuration. The nose is shown in formula (5).

DOF = DOF = ^ + 7 + 7 (5 where Z is the number of hours multiplied by (the number of revolutions _1), γ is the interaction axis (level, 1 ··· is the mean freedom degree of error. When choosing the orthogonal table Please pay attention when ... a. The number of experiments 2 degrees of freedom b. In the case of interaction, select the most suitable orthogonal table according to its point and line graph. Obtain the experimental degrees of freedom from formula (5). If the interaction between parameters is considered, When selecting an orthogonal table, you need to match the dot-line diagram. 14 200534147 d · Experiment e. Analysis of experimental results The present invention uses S / N (signal / noise) ratio analysis to obtain the best quality characteristics for the obtained experimental data. The combination method. Then, the predicted value analysis of each combination is used to calculate the best predicted value for each combination. Finally, the obtained predicted value is used as the basis for solving the problem of the optimal combination factor level conflict of the parameters. You Guangyi + (M-her-ton-her- PHASE_YI Guang 岣 ⑹ Among them is the total average value of the quality characteristics, which is the level setting value, ~ < is the best factor average value of the independent factor 'Μ is the best factor level average value of the interactive side. The S / N ratio analysis is in Taguchi experimental design riding towel 'Dr. Tian Pi will be more than Signal_t〇 _N〇ise (S / N) 'Unit: decibel, dedba (defined here) is defined as a criterion for evaluating robustness, which is converted logarithmically, multiplied by H), and shown in reverse. The hidden value is the smaller the value and the smaller the value The better is called the small-magnitude property, and the better the value is called the large-magnitude property, if the quality characteristic has a set target value, it is called the long-range characteristic. The present invention first The measured values of each quality characteristic are converted into _ ratio from formulas ⑺ and 依 according to different characteristics, and then the overall average m and factors (including interactions _ each level average =, U.) are obtained from the formula ⑼ side. Calculate the results of the data, and select the best sub-rule according to the thief value of each quality characteristic. 1, type S / N = -Wlogifx ητ ^ ι 丨 1 domain S / N = —1〇di nTti X 2 15 (9) 200534147 The meaning represents the test result of M, and 〃 is the total number of experiments. M =

L

(10) Which represents the recorded value of the experimental results, 〃 is the total number of experiments. ~ Represents the 7th factor extraction value, and is the total number of experiments at the ζ level. In the present invention, the main parameter factors and the expected values of the quality characteristics are used as the basis for selecting the optimal combination selection of the interaction parameters without including the interaction parameters at different levels according to formula (10). In the present invention, the main parameter factors and the thief value of each quality characteristic are used as the selection.

The selection method is based on the best combination of parent interaction parameters. At the same level, the basis is calculated according to the result calculated by formula (10). Optimization of multi-quality characteristic process parameter combination Gray correlation analysis is mainly used to analyze the degree of discrete order and correlation. It has the characteristics of less data and rural factor analysis, which can just make up for many limitations of traditional statistical regression. In addition, the scary _b Gangca__, this time, the relative degree has also increased, so if you rank these gray correlation degrees, you can form a matrix ′ called the gray correlation matrix. Then through this relationship between the seams of the gray-gray matrix, 200534147 t *, 1998) is analyzed among various factors. This financial method is called the title M analysis (Jiang Jinshan et al {when performing regular gray correlation analysis' The sequence must conform to the following rules: (1) Factor space false ° is printed as a theme and the factor obtained is 2nd order) 4 ^ = 口, Yu Jikou '^ is the relationship of influence 1 口 has the following characteristics, at this time Then it is said that the household (X) must be a factor space. • Brother and Jian's existence of miscellaneous factors. • Towel contains the existence of difficult factors. 2. Countability of connotation 0: the number of factors is limited and countable. Factor expansion: In addition to key factors, other factors can be added. 4. Independence of factors: each factor is independent a, and can be regarded as independent.

Each factor of this study is related to the whole and (b), the comparability of the sequence. Let an original sequence be: where VO represents the sequence χ, the experimental record value of the middle moxibustion item; moxibustion = / 2

For the full collection. ... 〇'1 ..., n'X If the sequence satisfies the following three conditions, then the sequence is said to be comparable. 1. Dimensionless (, dimen_: Regardless of the type of the unit of measure of the factor, it must be processed into a dimensionless form. 2. Scaling: The values in each sequence \ belong to The difference between the same grade or grade is greater than 2. 3. Same polarity (p〇larizati〇n): The factor description status in the sequence must be in the same direction. If the difference between the grades of the measured data values of the quality characteristics is greater than 2, it will not meet the same grade. Graded conditions. Therefore, in order to effectively perform gray correlation analysis, the present invention must modify the obtained data according to the expected value of each quality characteristic, and modify the gray correlation generation process (as shown in formulas 12, 13 17 200534147) to meet the experimental data comparable I hope that the large-scale: when the goal is bigger the better ___all i _wax # ⑻—wz >? E) ㈤ all i all i (12) I hope the small-scale: when the goal is smaller the better X: ⑻ = 顾 X '⑻W-⑻ all i _ (13) maxx (f \ k) ^ min ^ ik) all i all i 4% ¼ The original experimental record value of the moxibustion item in the table sequence; wi? XX / ~ all sequences The largest original experimental record value of the moxibustion item; the smallest original experimental record value of the second moxibustion item in the sequence meaning; 々 = 7 , 2, ..., "; ζ · = 0,7, · ", π. (B). The four axioms of the gray correlation measure (Axiom) satisfy the space formed by factor space and comparability called gray correlation space. Suppose it is a gray-related measurement space, and the factory is the size of the measurement. The furnace (1 > factory) must satisfy four theorems to make it more complete. The following 1, normative: 〇 < y {xlixJ) <l; V /, Vy / (χΆ) = 7 when it is said to be completely related, when it is said to be unrelated 2. Even symmetry: when there are only two groups of sequences Ηά) = / (ά) 3. Holisticity · When the sequence is greater than three groups (including (3 groups) (14) 18 (15) (16) 200534147 > 1 4. Proximity: The main control item of, that is, the gray relation satisfies the above theorem, then the size of 丨 X / ⑷ 一 Χ / ㈣ is the whole , The size of h (4'W) degree must be related to this item. If a function factory is found in the gray correlation space, it is the gray correlation degree in the gray correlation space.

(4) Gray correlation coefficient In the gray correlation space {house (you), · Γ}, there is a series of experimental record values of the moxibustion item in the 'Ya representative series'; moxibustion = person two ... Collection (17)

The grayness of the X body is related to the degree of correlation between the two sequences, and is further divided into the degree of locality and the overall 1 degree of correlation. The gray correlation coefficient is defined as follows:

Local grey correlation coefficient: when there is only one sequence ~ (moxibustion) as the reference sequence. (18) ΛχΜΦ)) = ^-ζ ^

Aoi [k) + ζΔΜαχ where is the reference sequence and 'is the comparison sequence. ζ · = 7 乂, again, w; moxibustion = 厶 2, again ... :: #, (identification coefficient yield [~?], the value of general identification coefficient is taken as 0. = h⑻- 调 丨 4 丫 ㈠ For the difference between the sequence% and the sequence of the moxibustion term experiment and take the absolute value min. Min.

Amin = Vy e 〇νψ0 (Α:) ^ x (19) (20) 200534147 > Λ max = V / e 〇V ^ || x0 (k)-xj (A: | (21) where zlmh is a sequence Between the sequence and the sequence \, the minimum value of the recorded phase difference between all experimental items is taken. Zlmox is the sequence X, and the sequence ', the maximum value of the recorded phase difference between all experimental items (moxibustion). Integrity gray Correlation coefficient: when any sequence '0) is a reference sequence (22) Fine dirty Δ ^) + ζΔιηαχ

Among them, 'is a reference sequence,' is a comparison sequence, / = /, 2,3, · ίoxia = ren two 3, · ί yd, ((identification coefficient) eM,-the values of general contact are all taken as 0.5. 4 丫 则 冰)-training | (23)

Aj (k) ^. Χ is the absolute value of the difference between the experimental items of the moxibustion term between the sequence 'and the sequence'.

(24) max. Max ^ ax = \ fjEm \\ Xi (k) ~ Xj (q (25) The ordering difference between the postal and postal phases is known as postal and postal space), and the postal postal distance is the largest one The identification coefficients are taken from the recorded phase difference between all the experimental items and sequences (the value of 0 is to increase the difference of the results. The values are generally taken. The identification coefficients are changed from m and the identification coefficients (()) Head 0.5 'but can be adjusted according to actual needs 20 200534147

J I size does not affect the ordering of results. (V) Global Grey Correlation Degree In the global grey correlation degree, since each sequence can become a standard sequence, after obtaining all the grey correlation degrees (formula 26), it can be added in the form of the characteristic value A. Sort. The calculation steps are as follows: I η r (^ i9xj) z: z-yLr (^ i (k) fxj (k)) (26)

η μ, J

The meaning is the reference sequence, 'is the comparison sequence; · = person two 3,…, / 2, moxibustion = 7 乂 3, ···, ", r {xXk \ x (k)) / is the global gray correlation coefficient. Assume that there are multiple sequences of ~ = W / K (4…, \ W): · = ·: ⑵,…, χ ») The sequence of each sequence of human 乂 is a standard sequence, the others are comparison sequences, and the sequence is based on the obtained values. Sorting, that is to establish a transfer / ^, this-matrix is called "gray correlation matrix."

R r η γ r2I γ 13 22 lrmI γ m2

Ylmr2m rt (27) Each line in each matrix represents the effect of the same-reference sequence on different comparison sequences, the size of the sequence, and the effect of eight identical standard sequences on the same comparison sequence. Because of the factors that can be identified according to the grey correlation, which are the advantages and disadvantages, they are called holistic 21 200534147 grey correlation analysis. When the matrix type is generated, since each element is the value of the gray correlation degree, then the eigenvalues of the a matrix are calculated. He = the eigenvector p of the shape correction matrix is formed: ρ, = ~ α, υη} (28) where ' If you send from… 乂 to represent a diagonal matrix, which has eigenvalues quot " ν · .Λ on its main diagonal (the rest are zero), the order of the eigenvectors used to form the matrix will determine the eigenvalues. Now ~. ·. · Λ} 's order on the main diagonal. The household matrix is composed of η linearly independent eigenvectors in gray-associated moments. Widely inverse matrix of ρ. Finally, the feature vector corresponding to the maximum feature value L is taken, and the numerical value of each corresponding element in the feature vector is the weight Id (take an absolute value). This weight% can be expressed as a gray correlation matrix and 'can be expressed as a ratio of the importance of the main diagonal elements in the gray matrix to the system towel, and its size arrangement can represent the best sequence in the system. Criterion (Wu Hanxiong et al., 1996) 〇 If the overall gray correlation analysis step does not meet the comparability of the measured data, the experimental data must first be substituted into (Equations 12, 13) for gray generation processing to make each sequence data Meet comparability. Then perform a global gray correlation analysis. The calculation steps are as follows: Step name-: First select a certain quality quality generation value as the fresh order #, and the remaining quality characteristic generation values are comparison sequences. Value, and then calculate the size of the difference sequence by formula (23). Then, the maximum difference "like and minimum difference" between the two poles is obtained by formulas (24) and (25). Step 2 • Calculate the overall gray correlation coefficient according to formula (22). Take the identification coefficient 22 200534147 ξ = 0.5000 Find the generated value of the selected quality characteristic as the overall gray of the standard sequence ^

Step 2: Calculate the overall gray correlation degree using the equal weight as the standard according to formula 可, and then obtain the selected quality hybrid button as the county's overall gray correlation degree. Step 4: Substitute the gray-degree values of the sequences into the following matrix L in order, and find ^ characteristic values ^. 1 rI2 rI3 y 2! 7 22 ··· m2 rmJ r • · · ύ t Im • · · y, 2m • l • · > · · y • mm Calculate the eigenvalue 乂 size from the matrix, using the largest eigenvalue The corresponding value of the corresponding eigenvector is the weight 1 ^, which is the importance of the # process. I ca n’t make it into everything. Sex is for the entire degree and replaces Pinna A, but Ayazaki ’s characteristic paste

The successor took advantage of the minimum and the big cock to watch the winter (secret, 1965). Firstly, the attribution degree of each quality characteristic is taken as small i A = V Jia (representing Nazaki tear butterfly, that is to obtain the following set (D). Finally, the maximum center in D is taken out (represents the level of important factors for the obtained multiple quality characteristics The combination of the combination and the associated set.) 23 200534147 The important factor level combination corresponding to Dmax is the combination that optimizes the process parameter obtained by using the overall gray correlation analysis while considering the interaction between factors and multiple quality characteristics. [Schematic Brief description] Figure 1 is the flow chart of K-means clustering method. Figure 2 is the experimental flow chart of Taguchi method.

twenty four

Claims (1)

200534147 j »X. Scope of patent application: A method of optimizing parameter combination considering the interaction between process parameters. Contains the following steps: Manufacture of oyster shell characteristics overnight > Simplification of process parameters > 1. Use the K-means clustering method to group periods based on quality characteristics. The expected value is divided into the experimental results. 2. The use of rough set theory (Rough Set) as the process parameters simply affects the most important parameters of the process. Optimization of process parameter combinations for each quality characteristic l = Use the field a cross-tab experiment meter. As an analysis of the optimization of the process parameter combinations for each of the f characteristics of the process, we will conduct experiments under the consideration of the interaction between process parameters. The method is to obtain it effectively. 4. S / N ratio analysis is used as the analysis method of experimental results. Deletion analysis is used to detail the _ process _ the factor level generated by the interactive translation. _4 Seek the bribe to get the predicted value of the best group of each ship's silk number & _ financial verification data for Lin μ hybrid process Selection of the best combination. > With regard to the selection of the optimal combination without interaction parameters, the main parameters of this issue _ are calculated based on formula ⑽ expectation value as the basis for selection under different rotations. Then, in each of the 4 characteristic periods, in terms of selecting the best combination of parent interaction parameters, at the same level, butterfly __ fruit and _ == are used as selection series. Expected value 25 5. 200534147 > Optimization of multiple quality characteristic process parameter combination optimization The whole ship gray-M analysis is used as the experimental method for reading the quality miscellaneous parameter set. 2. The experimental conclusion _ 嶋 means cocoa _ _ _ Lin as a gray generation processing, so that each sequence of data satisfies comparability. 3. First select a certain quality characteristic generation value as the standard phase χ, and the rest of the quality characteristics generate The f value is the comparison order ..., the value of the disparity correlation degree is calculated under the quantified gray correlation degree, and then the size of the sequence is calculated. Then calculate the maximum and minimum difference dmin and the minimum difference dmin. 〇4. Calculate the overall gray correlation coefficient. Take the identification coefficient 'find the selected gray-quality age value as a fresh sequence', the overall gray correlation coefficient soil w, 'w). 5. Calculate the overall gray correlation degree ((Ά)) with the equal weight as the standard. Finally, the selected certain-quality characteristic generated value can be obtained as the overall gray correlation of the standard sequence X /. 6. Sequentially obtain the gray correlation degree values of each sequence into the following moments in order, and find their characteristic values. ’, Yu y 丨 2 ϊ 13 • ·. Γ Ϊ2! Y 22 · · · ... y rmj y m2 · · ·… γ Calculates the eigenvalues from the matrix; the size of L is the maximum eigenvalue; the feature corresponding to 1_ The corresponding value to 1 is the weight ', and this weight can be expressed as the importance of each quality characteristic to the entire process. 26 200534147 7. The butterfly sounds, the attribution of Kawei Kasheng (Zadeh'l965) ° First, the attribution degree of each quality characteristic is taken as Γ value ^ / x: j (k); D = (Dk) ((for multiple Take the intersection of quality characteristics), to get the following set (D). Finally, take the maximum value in D ... (representing the combination of the important combinations of the important factors of the multiple quality characteristics). 8. The important factor corresponding to the obtained l The level combination is the combination of the optimal parameters when recording as the thief μ product threat. ~ T 私 27