Detailed description of the invention
Fig. 1 is the element block diagram of the quality control system 100 of the embodiment of the present invention.As it is shown in figure 1, product
Matter management system 100 comprises data base 10 and processor 11.Data base 10 is in order to store production batch
Quality history data.Quality history data referred herein are criticized after many platform systems, respectively for producing
The data of the various different anomalous events (Issue) that platform systems is recorded.Processor 11 is coupled to number
According to storehouse 10.Here the processor 11 of indication can be the processor on personal computer, in Analysis server
Processor or work table on processor etc..In the present embodiment, processor 11 will utilize number
According to the quality history data of the production batch in storehouse 10, algorithm is utilized to integrate the number in all platform systems
According to, and analyze its statistical property and regression curve.And processor 11 is according to the result after analyzing, will be pre-
Survey the risk factor corresponding to high-reliability interval producing batch quality characteristic.And the work people on production line
Member i.e. can utilize this to produce the risk factor corresponding to high-reliability interval of batch quality characteristic, selects conjunction
Suitable production is criticized sampling and is tested.And how quality control system 100 will predict production batch quality characteristic
The risk factor corresponding to high-reliability interval, its algorithm will be described under.
Fig. 2 describes in Fig. 1 embodiment, and processor 11 is integrated each platform systems and produced showing of batch data
It is intended to.In fig. 2, produce batch data D and criticize after many platform systems for producing, each platform systems
The data acquisition system of the various different anomalous events (Issue) recorded, these data are stored in data base 10.
And processor 11 is by extracting production batch data D in data base 10 after, according to its platform systems, will produce
Batch data D classifies as N number of anomalous event set, such as the anomalous event set O in Fig. 21To anomalous event
Set ON.The corresponding different platform systems of each anomalous event set.In fig. 2, anomalous event collection
Close O1It is the record data for the 1st platform systems, anomalous event set O2It is for the 2nd platform systems
Record data, anomalous event set ONIt is the record data for n-th platform systems.In this enforcement
In example, each anomalous event set has the anomalous event of identical or different quantity.Such as anomalous event
Set O1In have M1Individual anomalous event, anomalous event set O2In have M2Individual anomalous event, anomalous event
Set ONIn have MNIndividual anomalous event.It is the positive integer more than 1 used herein of N, and M1To MNFor
Positive integer.It follows that processor 11 can by the anomalous event correspondence in each anomalous event set extremely
Suitably anomalous event grade (Issue Grade).Such as by the anomalous event set O in Fig. 21In
M1The ascending sequence of individual anomalous event, and respectively the anomalous event after these sequences is corresponded to abnormal thing
Part grade I11To anomalous event grade I1M1, and anomalous event grade I11For less serious anomalous event,
Anomalous event grade I1M1For more serious anomalous event.Processor 11 can be by anomalous event set O2In
M2The ascending sequence of individual anomalous event, and respectively the anomalous event after these sequences is corresponded to abnormal thing
Part grade I21To anomalous event grade I2M2, and anomalous event grade I21For less serious anomalous event,
Anomalous event grade I2M2For more serious anomalous event.Similarly, processor 11 can be by anomalous event collection
Close ONIn MNThe ascending sequence of individual anomalous event, and respectively by corresponding for the anomalous event after these sequences
For anomalous event grade IN1To anomalous event grade INMN, and anomalous event grade IN1For less serious different
Ordinary affair part, anomalous event grade INMNFor more serious anomalous event.Therefore, according to N number of anomalous event
The anomalous event grade that set is corresponding, can define an anomalous event index (Issue Code), for
Under:
In (1) formula, as i-th anomalous event set OiInterior jth anomalous event grade IijIt is 0
Or the when of 1, anomalous event index IC in (1) formula i.e. represents that the anomalous event linked is closed in production batch
Number.For example, production batch has suffered from the 1st anomalous event set O1The 2nd abnormal thing
Part I12, the 2nd anomalous event set O2The 3rd anomalous event I23And the 3rd anomalous event collection
Close O3The 1st anomalous event I31, then the numerical value of anomalous event index IC is I12+I23+I31。
But, observe anomalous event index IC defined in (1) formula and only can know the index of anomalous event
And the number of anomalous event, and cannot to obtain the risk of anomalous event in the way of quantization tight (Index)
Principal characteristic.Therefore, in order to analyze the anomalous event shock extent for production risk, processor 11 further
Can be by i-th anomalous event set OiInterior jth anomalous event grade Iij(imply that each abnormal thing
Part) set corresponding risk factor Bij(Risk Coefficient).As in figure 2 it is shown, the 1st exception
Event sets O1Anomalous event grade I11To anomalous event grade I1M1Corresponding risk factor is B11Extremely
B1M1, anomalous event grade I of the 2nd anomalous event set O221To anomalous event grade I2M2Corresponding
Risk factor is B21To B2M2, n-th anomalous event set ONAnomalous event grade IN1To abnormal thing
Part grade INMNCorresponding risk factor is BN1To BNMN.In the present embodiment, due in same exception
The severity of the anomalous event in event sets is that therefore corresponding risk factor is served as reasons to longer spread by little
Big to minispread, imply that risk factor is the biggest and represent that corresponding anomalous event is the lightest to the impact degree of quality
Micro-, risk factor is the least represents that corresponding anomalous event is the most serious to the impact degree of quality.Therefore at Fig. 2
In, B11>B12…>B1M1, B21>B22…>B2M2, BN1>BN2…>BNMN.Therefore, according to N number of anomalous event
The risk factor of the anomalous event grade that set is corresponding, can define a risk score index (Risk
Value), under:
But, seriousness and the correspondence thereof of present invention anomalous event in same anomalous event set are different
The grade of ordinary affair part is not limited to the ascending sequence described in embodiment, and corresponding to anomalous event grade
Risk factor be also not limited to descending arrangement.For example, in other embodiments, abnormal thing
Coefficient corresponding to part grade can be to longer spread by little.And in the ensuing explanation of the present embodiment, can profit
By the relational expression of a matrix, release the risk system of anomalous event grade corresponding to all anomalous event set
Number, then takes back in (2) formula by the risk factor calculated so that (2) formula becomes one and has
The linear regression equations of individual parameter, whereinRepresent the anomalous event sum of all considerations, and line
Property regression equation be that monotonicity is incremented by (Monotonically Increasing) or monotonicity and successively decreases
(Monotonically Decreasing), and processor 11 will obtain according to linear regression equations
About quality characteristic baseline (Base Line), the upper limit (Upper Bound) and lower limit to risk factor
(Lower Bound).Processor 11 can be according to these statistical results above-mentioned, it was predicted that produce batch quality characteristic
The risk score corresponding to high-reliability interval.
Here illustrate that processor 11 is integrated each platform systems and produced batch data and prediction with an example
Produce the flow process of the risk factor corresponding to high-reliability interval of batch quality characteristic.Here, in order to simplify
Describing, data base 10 considers the data of 2 platform systems records, as shown in the table:
In upper table, two platform systems record the anomalous event set of 2 kinds respectively, and each is different
Often event sets has 3 anomalous events, and the 1st anomalous event of first category is the thing without quality loss
Part, therefore its seriousness is relatively low, and the 2nd anomalous event of first category is that quality loss is below 10%
Event, its seriousness is medium, and the 3rd anomalous event of first category is that quality loss is more than 10%
Event, its seriousness is the highest.1st anomalous event of second category is the event without quality loss,
Therefore its seriousness is relatively low, and the 2nd anomalous event of second category is quality loss thing below 5%
Part, its seriousness is medium, and the 3rd anomalous event of second category is quality loss event more than 5%,
Its seriousness is the highest.Quality characteristic here can be the standard of any production quality, such as yield rate
(Yield) etc..
It follows that each anomalous event in each anomalous event set can be set by processor 11
Corresponding risk factor.Therefore, the 1st anomalous event of first category is set corresponding risk factor and is
B11, it is B that the 2nd anomalous event of first category is set corresponding risk factor12, the 3rd of first category
It is B that individual anomalous event is set corresponding risk factor13, it is right that the 1st anomalous event of second category is set
Answering risk factor is B21, it is B that the 2nd anomalous event of second category is set corresponding risk factor22,
It is B that 3rd anomalous event of second category is set corresponding risk factor23.In order to release all abnormal things
The risk factor of the anomalous event grade that part set is corresponding, first can be according to 2 platforms in data base 10
The data of system record, set up an anomalous event ranking matrix.This anomalous event ranking matrix is one
Individual sparse matrix (Sparse Matrix), and each element of this anomalous event ranking matrix is 0 or 1,
And the definition of anomalous event ranking matrix is that producing each product criticized is classified as first category
What all anomalous events of anomalous event set and the anomalous event set of second category combined presents.Citing
For, if producing the product quantity criticized in embodiment is NLOT.Representing for convenience, embodiment will produce
Product in Pi is expressed as numbering 1 to NLOTThe product of index value, then anomalous event ranking matrix can represent
Under for:
Upper table is in the anomalous event set of anomalous event set and the second category considering first category,
Producing batch middle quantity is NLOTThe combination of anomalous event kind that met with of product.For the present embodiment,
Because the anomalous event of each classification has three kinds, therefore produce batch anomalous event kind suffered from i.e.
There is 3 × 3=9 kind possible.Therefore, if producing batch quantity N of middle productLOT> 9, then it represents that in production batch
The product having some different index value can be subjected to identical anomalous event (with the life that general yield rate is higher
Producing for criticizing, quality characteristic is higher at the two the most break-even ratios of kind).Such as in upper table, rope
The product drawing (1), index (2) and index (6) is just subjected to identical anomalous event (quality characteristic is lossless
Lose).Therefore, the anomalous event ranking matrix set up, the dimension of its row is NLOT.Furthermore, with
For the present embodiment, a total of 6 kinds of anomalous events are considered, the anomalous event grade hence set up
Matrix, the dimension of its column be 3+3=6 (but in formula, if considering N number of anomalous event set,
M is had in assuming the n-th anomalous event setnThe anomalous event of individual quantity, then anomalous event ranking matrix its
The dimension of column isWherein ∑ is for even putting in marks).And by above-mentioned anomalous event grade
Matrix launches to be represented by down:
Wherein IMRepresenting anomalous event ranking matrix, its dimension is NLOT×6。
By anomalous event ranking matrix IMAfter foundation completes, according to the relation in (2) formula, each risk
Coefficient can be corresponding to each anomalous event.Therefore, by the risk in the anomalous event set of first category
Coefficient B11, risk factor B12, risk factor B13, and by the anomalous event set of second category
Risk factor B21, risk factor B22, risk factor B23Launch, and produce a risk factor vector.
In the present embodiment, i.e. there are 6 elements in risk factor vector (but in formula, if considering N number of
Anomalous event set, it is assumed that have M in the n-th anomalous event setnThe anomalous event of individual quantity, then risk
I.e. have in coefficient vectorIndividual element, wherein ∑ is for even putting in marks), this risk factor vector can
It is expressed as down:
Wherein B is risk factor vector, and its dimension is 6 × 1.Therefore, when by the abnormal thing in (3) formula
Part ranking matrix IMWhen being multiplied with the risk factor vector B in (4) formula, the vectorial Y of output is production
At the risk factor indicator vector corresponding to the combination of each anomalous event in Pi.But, the present embodiment is raw
The product quantity produced in criticizing is NLOT, risk factor index represented in the vectorial Y of this output is NLOT
The dimension of × 1, shown in lower:
In (5) formula, B11To B23The number of 2 platform systems records of data base 10 internal memory can be utilized
Anomalous event ranking matrix I according to this and in (3) formulaMEstimate its numerical value.Due to anomalous event ranking matrix
IMIt is not square formation, therefore can not directly do matrix inversion (Matrix Inverse).In the present embodiment,
Obtain B11To B23Mode may utilize anomalous event ranking matrix IMVirtual reversal process (Pseudo
Inverse) or use linear regression mode, by Least squares approximation method (Least-Squared
Or minimum approach error method (Minimum-Mean-Squared Error Approach) is forced Approach)
Closely obtain and (in (5) formula, now need one group of micro-margin of error ε1ExtremelyAssist and solve).Although this enforcement
Example formula solves B with the framework of (5) formula11To B23, but the present invention is not limited to this, in other embodiments,
Can also be by anomalous event ranking matrix IMIts dimension is expanded with a constant row vector (Column Vector)
Degree, or risk factor vector B is expanded its dimension with a constant, as follows:
In (6) formula, C is a constant, corresponding IMThe row vector of the leftmost side is also constant vector,
B in this case11To B23Still can use Least squares approximation method (Least-Squared Approach)
Or minimum approach error method (Minimum-Mean-Squared Error Approach) solves.And ask
Solution in Xie Yuyou (5) formula gone out only one constant offset of gap (Constant Offset), the most not
The statistical property of the post analysis that can affect.
When above-mentioned (5) formula has been set up and via (5) formula by risk factor B11To B23After obtaining, wind
Danger coefficient B11To B23I.e. take back the risk score index in (2) formula, it is possible to (2) formula is become a tool
Having the linear regression equations of 6 parameters, wherein the quantity of parameter i.e. represents the anomalous event of all considerations
Sum is (in formula, if considering N number of anomalous event set, it is assumed that have in the n-th anomalous event set
The anomalous event of Mn quantity, then (2) formula i.e. becomes one and hasThe linear regression side of individual parameter
Formula).Here, linear regression equations is monotonicity is incremented by (Monotonically Increasing)
Or monotonicity successively decreases (Monotonically Decreasing).For example, the linear regression side obtained
Formula LY is deployable as follows:
LY=91.378+0.0 × I11-3.52×I12-8.61×I13+0.0×I21-5.07×I22
-32.32×I23
(7)
In the present embodiment, by (5) formula, risk factor is solved one by one, and disaggregation is combined into B11=0.0,
B12=-3.52, B13=-8.61, B21=0.0, B22=-5.07, B23=-32.32.But, the wind of the present invention
Danger coefficient disaggregation credit union is along with producing that batch quality characteristic is different and difference, and above-mentioned listed solution set is only
For the expression solving set a kind of in embodiment, and it is not used to limit solution procedure and the result of the present invention.This
Risk factor solution calculated by embodiment also should demonstrate,prove aforementioned condition, because the present embodiment is integrating not homology
During the anomalous event set that system is recorded, processor 11 can be by the exception in each anomalous event set
The ascending sequence of event, therefore corresponding its size of risk factor sequence also should meet B11>B12>B13
And B21>B22>B23Condition.And in (7) formula, 91.378 is a predetermined constant, the most not
The statistical property of the post analysis of impact.
After the linear regression equations LY in (7) formula obtains, examined and determine by risk score model, can
Produce datum line (Base Line) and to should the upper limit (Upper Bound) of datum line and lower limit (Lower
Bound) result, as shown in Figure 3.Fig. 3 is that the present invention produces corresponding quality characteristic and the base of risk score
The schematic diagram of directrix, the upper limit and lower limit.For example, when consider new production criticize add fashionable, corresponding new
Produce batch anomalous event index (Issue Code) and anomalous event grade (Issue Grade) also can
It is brought in (7) formula the output producing different linear regression equations LY.And the qualitative control of the present invention
System i.e. can produce such as Fig. 3 according to this in risk score and the statistics of corresponding datum line, the upper limit and lower limit etc.
Characteristic.In figure 3, X-axis is risk score, and Y-axis is quality characteristic, and upper limit UB is for having circular mark
The line segment of point, datum line BL is the line segment with triangle punctuate, and lower limit LB is to have rectangle punctuate
Line segment.Criticize when engineering and at least require that quality characteristic meets QLStandard time, the risk of corresponding upper limit UB is divided
Number is RVL, and the meaning representated by this risk score RVL is and meets quality characteristic QLStandard time, work
Journey criticizes the lower limit of selected risk score.Therefore, the reliability of sampling, processor are criticized in order to increase engineering
11 meetings calculate with the angle of datum line BL and meet quality characteristic QLStandard time, corresponding risk score.
In that case, owing to datum line BL is corresponding to quality characteristic QLRisk score be RVR, Gu Chu
Reason device 11 will be advised that the engineering staff on production line selects and be met the risk factor of datum line BL at least
The engineering of RVR criticizes sampling.
But, the quality characteristic described in Fig. 3 and the datum line of risk score, the upper limit and lower limit, its slope
It is not limiting as the algorithm of the present invention.For example, if the quality characteristic in embodiment is for considering yield rate
(Yield), then the slope of datum line, the upper limit and lower limit will be same as Fig. 3, and its slope is just.Otherwise,
In other embodiments, if quality characteristic for consider defect rate (Defective Rate), then datum line, on
The slope of limit and lower limit will be contrary with Fig. 3, and its slope is negative.
In sum, the present invention describes a kind of being applied to and produces the quality control system criticized and method thereof,
Its idea is the record data integrated and produce batch different platform system, and by the data pair of these anomalous events
Should be to suitable anomalous event grade and risk factor.By setting up anomalous event ranking matrix and risk
Coefficient vector, can obtain all of risk factor by the way of recurrence, and then try to achieve risk score
The linear regression equations of index, the high-reliability interval institute that one-step prediction of going forward side by side produces batch quality characteristic is right
The risk score answered.Therefore, the engineering staff on production line utilizes the risk score of suggestion easily,
Sampling is criticized in the production picking out reliability higher.
The foregoing is only the preferred embodiments of the present invention, all equivalents done according to the claims in the present invention become
Change and modify, all should belong to the covering scope of the present invention.