Background technology
The final quality grade of determining directly to affect plasma panel of plasma panel process parameters importance was by the EXPERIMENTAL DESIGN of construction period in the past and verify the importance of determining parameter in the plasma panel manufacture process repeatedly.Along with production programming extension and 7*24 hour uninterrupted volume production, if need to continue traditional test design method will waste great amount of manpower and kinetic energy consuming cost, need production line to stop the debugging design that line carries out test parameters importance simultaneously.This method has reduced the production efficiency of volume production on the one hand, and test design method will expend great amount of manpower and kinetic energy consuming cost simultaneously.Therefore need a kind of new method dynamically to carry out determining and optimization of process parameters importance.
In the plasma panel manufacture process volume production process, separate unit plasma panel process of producing product device parameter is included the storage of MES system in and is reached 1.2 ten thousand, more than the every day data volume 10G, the parameter that relates to is above 9000, in quantity, has the big data characteristics of magnanimity on dimension and the data generation speed.
So the plasma panel manufacturing process data of magnanimity has determined that further definite method of traditional parameter importance is inapplicable, particularly " feature extraction " reach " invariant feature extraction " technology and must add plasma panel manufacturing process data characteristic.
1. the Feature Extraction Technology of plasma panel process parameters data: feature extraction is the characteristic distributions according to data, chooses a series of feature in order to make up the process of model from feature database.An important hypothesis of feature extraction is to contain a lot of redundancies or incoherent feature in the data, and the purpose of feature extraction filters out these redundant informations exactly, find out can characterization data key character.In this invention, we have explored multiple basic Feature Extraction Technology according to the plasma panel industrial nature, comprise information gain, information gain ratio, SVMs, relief etc.
2. the invariant feature extractive technique of plasma panel process parameters data: integrated feature extraction is to find out the feature with general character from the result of a plurality of feature extractions, can be from the feature of a plurality of sides characterization data.By each feature extracting methods, we can obtain the feature of a series of characterization datas; But every kind of method can only be removed decryption from a certain side.The purpose that the invariant feature of plasma panel process parameters data extracts is that the result with a plurality of feature extractions merges effectively, therefrom extracts the consistency between several different methods gained feature, with obtain can characterization data invariant feature.
Summary of the invention
In order to overcome the above-mentioned shortcoming of prior art, the invention provides dynamically determining and optimization method of a kind of plasma panel process parameters importance, product hierarchy and bad type according to the plasma panel production technology, adopt Feature Extraction Technology, behind the result of given a plurality of feature extractions, the method that various features is extracted organically is fused in the process of integrated feature extraction, obtain stable characteristics extraction result more from a plurality of angles, thereby overcome drawback and the unsteadiness of using single feature extracting method, finally formed the more healthy and stronger of article on plasma display screen creation data, more stable, have more the displaying of practical significance.
The technical solution adopted for the present invention to solve the technical problems is: dynamically determining and optimization method of a kind of plasma panel process parameters importance comprises the steps:
On the date of step 1, the full operation creation data that provides according to the user, make up the database of the creation data of full operation production technology;
Step 2, the data content of all process steps in the full operation is done series connection handle, and divide according to grade or the bad type of PDP display, the PDP display that will belong to specific grade or specific bad type is divided into one group, other PDP display is divided into one group, then the parameter in the corresponding production technology of PDP display between each group is alignd, be formed into full operation data;
Step 3, the full operation data that step 2 is formed are carried out missing values and are handled, and the measured value on all parameters is carried out normalized;
Step 4, employing feature extracting method carry out feature extraction to the product of different grouping, filter out the parameter list that can maximize three groups of key characters that comprise information gain, the ratio of gains, minimizing redundant degree and the maximization degree of correlation;
The parameter list of step 5, three groups of key characters that step 4 is obtained is integrated, and obtains one group of unified important parameter tabulation, comprising appear at parameter at least two group key characters in three groups of key characters tabulations.
Compared with prior art, good effect of the present invention is: saved by mode spent time and the resources costs of repetition test design with checking, promoted the accuracy of determining concerning the process parameters importance of plasma panel quality.And by the new historical data accumulation after adjusting, carry out this analytical model method of screw type recycling and seek out the process parameters importance of plasma panel quality, manufacturing technology personnel are according to parameter importance, carry out the management and control of emphasis parameter, not only reach the purpose that continues spiral lifting plasma panel yields, but also reduce the percent defective of manufacture process indirectly.Simultaneously, in the use of method model, can not relate to the breaks in production of production line, avoid the unnecessary line loss of stopping to lose.
Embodiment
Dynamically determining and optimization method of a kind of plasma panel process parameters importance comprises the steps:
On the date of step 1, the full operation creation data that provides according to the user, make up the database of the creation data of full operation production technology;
Step 2, the data content of all process steps in the full operation is done series connection handle, and divide according to grade or the bad type of PDP display, the PDP display that will belong to specific grade or specific bad type is divided into one group, other PDP display is divided into one group, then the parameter in the corresponding production technology of PDP display between each group is alignd, be formed into full operation data;
Step 3, the full operation data that step 2 is formed are carried out missing values and are handled and normalized:
Described missing values is handled and is referred to: for missing values, fill according to the mean value of the measured value of the identical parameters of other PDP display, target be the missing values that makes this filling to after analytical framework do not exert an influence;
After missing values is handled, measured value on all parameters is carried out normalized, be about to the measured value unification of this parameter on different PDP display and in this interval of 0 average and unit variance, guarantee simultaneously that also the relative size of this measured value on the initial different screens is constant, illustrate as follows:
If α
t=(α
11, α
12,, α
In) representing a group observations of i parameter of n PDP display, the adjustment mode of new argument measured value is after the normalization:
In the formula, μ
iBe α
iAverage, σ
iBe α
iVariance.
Step 4, employing feature extracting method carry out feature extraction to the product of different grouping, filter out the important parameter tabulation that can maximize the important evaluation index that comprises information gain, the ratio of gains, minimizing redundant degree and the maximization degree of correlation, concrete grammar is as follows:
Definition of T is a creation data set, comprise (x, y)=(x
1, x
2..., x
n, y), x wherein
iThe value of i parameter in the PDP display data is produced in ∈ vals (i) representative, grade or the bad code of current this PDP display of y representative.
(1) computing information gain (Information Gain):
Wherein
Be comentropy, p (x
i) representative observes x
iProbability, can be by occurring x in the historical data
iNumber of times calculate.
(2) calculated gains is than (Gain Ratio):
IGR (T, a)=IG (T, a)/IV (T, a) wherein, IG (T a) is information gain, IV (T, a) account form is as follows:
(3) calculate maximal correlation degree and minimum redundancy (minimum Redundancy and Maximum Relevancy, mRMR)
1) computational minimization redundancy (mR):
minW
I,
Wherein,
, be defined as the interactive information between two variable a and the b;
2) calculate the maximization degree of correlation (MR):
maxV
I,
3) calculating the mode can satisfy mR and MR to the full extent is that method for searching by incremental finds the feature (parameter) that satisfies following condition:
maxΦ(W
I,V
I), Φ=V
I-W
I
Suppose that existing key character (parameter) set is S
M-1, wherein having comprised m-1 key character (parameter), target is in order to find m key character (parameter) to make it can maximize Φ from residue character (parameter) set.
More than the computational methods of each feature selecting can export one group of key character (parameter), namely by IG, three stack features (parameter) that IGR, mRMR export respectively characterize feature (parameter) combination of discrimination is arranged in the current PDP display grouping most.
Step 5, three groups of important parameters tabulations that step 4 is obtained are integrated, and obtain out one group of unified important parameter tabulation, comprising in three groups of key characters tabulations, appearing at parameter at least two groups:
The inventive method can be divided into two classes with the target of calculating at different comparison task:
1) judges at the important parameter of the PDP display of specific grade;
2) judge at the important parameter of the PDP display of specific bad type.