CN1297146A - Data processing method for evaluating size and quality in making car body - Google Patents

Abstract

A data processing method for evaluating the size and quality in manufacturing car bodies is disclosed. Small specimen sampling techanique for production in batches is used. A prior Bayes method is used to estimate the instantaneous pass rate on current day. The trend items are separated from the wave items by small-wave filtering and index weighing slide average methods. The evaluating indexes for current time and stage are created. The multi-element slide 6-sigma index for evaluating size control power is given out. Its advantage is high accuracy.

Description

The data processing method of evaluating size and quality in making car body

The present invention relates to a kind of data processing method of evaluating size and quality in making car body, be used for the technological ability evaluation of quality assessment of vehicle body fitted position and automobile assembling production line, belong to body of a motor car workmanship control technology field.

Automobile is the trunk industry of Chinese mechanical industry, and Domestic Automotive Industry was finished 2,300 hundred million yuan of the output values altogether in 1999, and the output value 2,500 hundred million is finished in expection in 2000.The internationalization competition of current automobile market is fierce day by day, and quality becomes the life of enterprise, and this improves the competitive power of enterprise self with regard to the workmanship that forces automobile enterprise to improve automobile.

The control of quality with improve at first to have one accurately, quality evaluating method rapidly.For vehicle body, the dimensional accuracy of assembling is main assembling quality requirement, and at present domestic main car load factory does not all have the assay method of a cover system.The vehicle body assembling process is the process of a production in enormous quantities, owing to be difficult to accomplish 100% detection in most of manufacturing enterprises at present, the evaluation of size quality can only be adopted the method for off-line small sample sampling observation.Therefore, by to the sampling standard and the mathematical analysis of measurement data come exactly vehicle body fitted position quality to be estimated, the concurrent Useful Information that excavates is an inexorable trend to carry out Fault Diagnosis, has very strong realistic meaning.While has also proposed challenge to the technique of excavation of sampling theory and data.

The evaluation of vehicle body fitted position quality mainly is objectively the instability of production line to be provided an index.This instability is a kind of data fluctuations at random.The basis of analyzing for the data of this random variation is a statistical method, and the overall of the application requirements sample of statistical method is stationary stochastic process, and often needs sample more than 30 to guarantee the validity of adding up.For the small sample situation of having only 1-2 sample every day, run up to suitable sample size and take nearly one month time.In the so long time; usually have adjustment and other determinacy influence factor of frock clamp; these all can reflect certain trend on data; some disturbing factors in this deterministic trend and the steady production process; the fluctuation that is random state that causes such as loosening as the wearing and tearing of register pin, anchor clamps mixes, and is difficult to distinguish.Directly have influence on the evaluation of size fluctuation variation and determining of fault basic reason.

Traditional method of quality control is with adding up control technology, and the manufacture process of monitoring product is analyzed the cause specific variation and the common cause variation that exist in the production run, helps to eliminate cause specific, makes process be in slave mode, and all process steps all reaches stable state.But this technology is only applicable to single argument control, and for the large-scale assembly structure product of vehicle body one class, its size quality often reflects by the deviation of up to a hundred measuring points arranging.Have the multivariate situation of a plurality of parameters for this class, need more scientific methods and come the dimensional conditions of integral body is estimated effectively, and excavate out wherein deviation rule, the foundation of fault diagnosis is provided.

In the actual production quality assessment, automobile assembling production-line technique ability assessment is another difficult problem, and the rational index of neither one reflects the dimension process level of whole production line with gearing to actual circumstances at present, does not also find the pertinent literature report.

The objective of the invention is at the above-mentioned deficiency of prior art and the needs of practical application, a kind of data processing method of scientific and effective evaluating size and quality in making car body is provided, can provide the evaluation in instant and stage to vehicle body manufacturing dimension quality precision, and provide the integrated artistic merit rating method of production line fitted position quality, for the control and the raising of the quality of production provides foundation.

For realizing such purpose, adopt in the technical scheme of the present invention towards the small sample Sampling techniques of producing in enormous quantities, utilize the priori bayes method to estimate the instant percent of pass on the same day, utilize Daubichies wavelet filtering method and EWMA method that measurement data is carried out separating of trend term and fluctuation, set up the instant evaluation index and the stage evaluation index of size fluctuation quality, and provide the polynary slip 6 σ indexs of production line size Control technological ability, production line size Control technological ability is estimated.

This method is towards whole measurement-analysis-evaluation procedure, the fluctuation deviation separation and extraction technology, the dynamically instant of size quality that solve the small sample sampling frequency and sample size standard, measurement data are estimated and multinomial gordian techniquies such as stage evaluation method, production line integrated artistic merit rating, by to the method for sampling, data error separation, the organic integration of data multidimensional statistical appraisal technology, set up size quality systematic analysis evaluation method.

When measuring sampling, the size of body in white is measured by three coordinate measuring machine, the size of main sub-unit is measured by the detection anchor clamps and the portable surveying instrument of special use, the data of these measurements are sent into central computer and are stored and handle, the result after the processing by print, plotting equipment exports.This method realizes data processing and quality assessment by the analyzing software system of central computer.

Concrete operations step of the present invention is as follows:

1, measures sampling

Under the situation of off-line measurement, the time of sampling and size are very important for the assessment of quality.This method has proposed to be suitable for the method for sampling of body of a motor car factory off-line measurement quality evaluation at measuring condition commonly used, requires to sample according to different evaluations.

Determining of sample size:

For instant evaluation, carry out assay according to the measurement sample on the same day, the body in white sample number is 1-2, the sub-unit sample number is 3-5;

For the stage evaluation, statistical sample should be recommended as 30 more than 17;

For production line dimension process merit rating, the integral body of getting measurement data is as sample, i.e. all data that begin from volume production.

Determining of sampling time:

1) samples at a fixed time every day;

2) the sub-unit sampling can be finished continuously, generally gets 3～5/day, 1～2/day of body in white sample frequency;

3) assembly relation of specific demand is sampled as irregular sampling, and will guarantee assembly relation accurately;

2, the error separating of measurement data

The present invention carries out error separating to measurement data and handles, take dual mode, detect data for sub-unit, adopt EWMA (EWMA) method to carry out the calculating of error separating, for the error separating of the detection data of body in white, adopt the wavelet filteration method of Daubechies discrete form to handle.

(1). sub-unit detects the error separating of data

For the detection data of sub-unit, the present invention adopts EWMA (EWMA) method to carry out the calculating of error separating.This method is introduced estimation to current measurement data with the measurement data in past by the form of weighting, thereby weakens the influence of the random signal in the measurement data, and the main trend and the pure wave that obtain data variation move situation.

Specifically according to the following steps:

1) packeting average: the measurement sample with every day is one group, from measuring for the first time, to one group of new data computation mean value X of every day _i

2) recursion is calculated: the EWMA method has following recursive operation pattern and obtains trend term in the data:

X′ _i= X′ _i-1+w( X _i- X′ _i-1)

Recursive operation is suitable for the increase of data finishing in time the mask work of data, has certain real-time.The acquisition of first data is according to the fundamental formular of EWMA ${\stackrel{\‾}{X}}_i^{\′}=\underset{j=0}{\overset{s-1}{\mathrm{\Σ}}}{w}^j{X}_{i-j}$ (0＜w＜1) obtains, and after this just calculates fast with recursion formula.

3) key parameter of EWMA

When carrying out EWMA calculating, there is the parameter of two keys to provide, these two parameters are:

A) the number S of the historical data of Yin Ruing;

In this method, S is taken as 4;

B) weighting coefficient W

In this method, the selection of weighting coefficient has very big influence for the accuracy of handling, when w → 0, and weight coefficient (1-w) w ^j→ w represents that all data have identical power, are equivalent to simple average.When w → 1, weight coefficient (1-w) w ^jGrowth with j decays rapidly, and expression has only considered that nearest data are to X ' _iInfluence.For the body of a motor car assembling, the present invention gets 0.2-0.4.

4) fluctuation is extracted:

The value of the every bit in the trends calculated sequence has been represented the trend of data variation, and this method obtains the high-frequency fluctuation data of a point by the method for the difference of calculating raw data and trend data.This data have reflected that under not adjustment situation the deviation of frock and the deviation of part are to the influence of measurement data.

(2). body in white detects the error separating of data

Because body in white adopts three coordinate measuring machine to measure, and measures sample still less, reflect that more significantly tendency changes in the feasible data of measuring.For body in white, the trend of measuring point data changes the size function that the drift of being introduced does not influence final products.Therefore, focus on providing the index of fluctuating level for the size evaluation of body in white.

Estimate accurately for the fluctuation of dialogue body dimensions provides one, need from a large amount of measurement data, fluctuation information be extracted, as the foundation of estimating.Because vehicle body is measured the non-stationary of original signal, make traditional convert the signal into the method that frequency domain carries out Filtering Processing and do not prove effective.For this reason, introduced the wavelet filteration method that becomes the yardstick windowing process.

Adopted the wavelet filtering method of Daubechies discrete form to carry out Filtering Processing in this method.

3, stages 6 σ estimates

For the fluctuation of a certain size, the possibility in the scope of positive and negative three times of standard deviations is 99.73%, can think that 6 σ are exactly the scope of change in size, and all sizes all change in this scope.Stages 6 σ reflection be the intensity of removing resulting high-frequency fluctuation after the size sudden change that factor such as artificial adjustments causes in a period of time.High 6 σ explanation production status instability need take measures to control.Application data is separated later fluctuation item when carrying out stages 6 σ calculating.

In carrying out stages 6 σ evaluation, this method is carried out according to following step:

1), and, determines in conjunction with the concrete time (as statistics monthly) according to the regulation of the sample size of front

The statistics capacity of sample.

2) calculate 6 σ values of this measuring point;

4, assembly integral 6 σ estimate

Assembly integral 6 σ estimate be to assembly over a period to come the total data fluctuating level carry out whole evaluation.For the factor that some are accidental is got rid of, this method is got 6 σ values of 95 hundredths as assembly integral 6 σ evaluation indexes.

5, instant percent of pass evaluation

The employed data of the evaluation of percent of pass are without the raw data of separating.What instant percent of pass evaluation reflected is the percent of pass level of certain all measured value of measuring point on one day a certain assembly.For the size percent of pass to small sample situation following every day is estimated immediately, this method has adopted experience Bayes Bayes method, be configured to the prior distribution of percent of pass as empirical data by passing by the information of known percent of pass, revise with the sample situation on the same day again, obtain the estimated value that the same day, percent of pass distributed.Owing to introduced the empirical data of steady production process, made the percent of pass under the small sample estimate more accurate.

Gordian technique in carrying out the EB processing procedure is:

1) determining of prior distribution: getting prior distribution in this method is that β distributes, and gets one section stable historical measurement data, and per 10 one batch totals are calculated its percent of pass, tries to achieve the parameter a of prior distribution, b with moments method.

2) dynamically the instant percent of pass of recursion is estimated: distribute as the priori of estimation next time with each posteriority

Distribute.

6, the stage percent of pass is estimated

What stage percent of pass evaluation reflected is the average level of each measuring point percent of pass in a period of time.The time span of estimating is decided according to the concrete condition of enterprise.Generally can monthly estimate by week.Specifically be calculated as the ratio that the measuring point of this section period in tolerance range accounts for whole measuring points.

7, production-line technique merit rating

The technological ability evaluation of production line is that the whole dimension control ability of an assembly line is assessed.This method is a foundation with the measurement data of body in white, and whole measurement data in inlet postpartum are carried out statistical study, and obtaining with 6 σ is the technological ability evaluation index of the production line of index.Specific as follows:

1) rounds individual measurement sequence as analytic target;

2), obtain the dynamic change sequence of 6 σ to the 6 σ calculating of sliding;

3) this sequence is carried out statistical study, determine 95% fiducial interval;

4) this fiducial interval has promptly reflected the size Control level that production line can reach;

For understanding technical scheme of the present invention better, below describe in further detail by drawings and Examples.

Fig. 1 is that hardware device of the present invention connects block diagram.

Among the figure, body in white and main sub-unit cubing 1 are connected to three coordinate measuring machine 2, the size of body in white is measured by three coordinate measuring machine 2, sub-unit detects anchor clamps 3 and is connected to portable surveying instrument 4, the size of main sub-unit is measured by the detection anchor clamps 3 and the portable surveying instrument 4 of special use, the output of three coordinate measuring machine 2 and portable surveying instrument 4 is connected to data analysis computing machine 5, body in white and main sub-unit three-dimensional coordinates measurement data are sent into computing machine 5 by dedicated line and are stored and handle, the sub-unit measurement data manually is downloaded to computing machine 5 and stores and handle, the output of data analysis computing machine 5 connects intranet 7 and output device 6, and the result after the Computer Processing can print by output device 6, drawing is exported.

Fig. 2 is a data analysis schematic flow sheet of the present invention.

Among the figure, the method for sampling at first according to the present invention is sampled to measurement data, the drafting of the evaluation of the quality that fluctuates then, trend constitutional diagram, and the evaluation of percent of pass.Data after the sampling are carried out separating of high-frequency signal and low frequency signal, obtain trend data and fluctuation data.Trend data is depicted as trend map, adjusts the monitoring foundation of situation as frock.Fluctuation signal is for further carrying out the foundation of quality assessment.Mainly provide the stage fluctuation index of single measuring point, the stage fluctuation index of assembly integral and the technological ability index of production line size fluctuation control.What the evaluation of percent of pass was adopted is not separated data, mainly provides the instant percent of pass of reflection percent of pass level on the same day and the stage percent of pass of interior percent of pass level of reflection a period of time.

Fig. 3 is a workflow block diagram of the present invention.

Fig. 3 provides the specific implementation flow process in as shown in Figure 2 the data flow.The method of sampling according to the present invention is measured sub-unit and body in white.In the calculating of percent of pass, the dynamic percent of pass method of estimation of Bayes that adopts the present invention to provide is calculated, and obtains the percent of pass index of every day.After after a while, the data of all measurements are handled with stage percent of pass computing method, obtain stage percent of pass index.After this, the measurement data of sub-unit and the measurement data of body in white are carried out data separating with the method for EWMA method and wavelet filtering respectively, data and trend data obtain fluctuating.Wherein, every of the method for wavelet filtering is calculated at regular intervals, calculates once or secondary as every month, decides on the actual requirement of enterprise.Because the algorithm of EWMA is a recursive algorithm, then carries out single treatment every day.When reaching official hour at interval the time, just intercept the fluctuation data after separating during this, each measuring point is carried out the calculating of 6 σ, obtain reflecting the stages 6 σ value of single measuring point fluctuating level.With descending ordering of 6 σ of all measuring points on this sub-unit or the body in white, 6 σ that are positioned at the measuring point of 95 hundredths are exactly the assembly stages 6 σ value of reflection assembly size fluctuation quality.With analysis time section expand as from volume production and begin to end to when statistics, and then body in white measuring point fluctuation data 6 σ that slide are calculated, the value of getting 95 hundredths of every day is an analytic target, all analytic targets are carried out statistical study, and 95% fiducial interval of its statistical distribution is the technological ability index of production line size fluctuation control.The trend data that sub-unit that obtains after the data separating and body in white are measured is adjusted the foundation of condition monitoring as carrying out frock.

Fig. 4 carries out the synoptic diagram of multi-layer analysis for adopting wavelet transform to signal.

Among the figure, the raw data that obtains of sampling is carried out successively data separating by wavelet filter.Ground floor obtains a high-frequency signal and a low frequency signal after separating; The low frequency signal that obtains carries out the second layer and separates as partial input, obtains low frequency signal and high-frequency signal of time one deck; So analogize, low frequency signal after each layer separation all as the input signal of following one deck, passes through the processing of multilayer at last, is able to the low frequency signal of desirable reflection data variation trend, again this signal and original signal are asked poor, the result is the high-frequency fluctuation signal of reflection data fluctuations situation.

Embodiment

1. sampling: in this example, the sample frequency of body in white is 2/day, gets 45 day data, and totally 90 perform an analysis.

2. data separating

Present embodiment separates measurement data with the method for wavelet filtering, gets 3 layers of Daubichies wavelet filtering.Be mixed with the data sudden change that the frock adjustment causes in the original signal, if carry out 6 σ when calculating with sort signal, numerical value that obtains and actual deviation are very big.Through behind the wavelet filtering, 6 σ of fluctuation item reflect the true random fluctuation deviation of data.

Fig. 5 carries out the curve map of data separating for the wavelet filtering method, and wherein, Fig. 5 a is the raw measurement data curve, and Fig. 5 b is the trend data curve after separating, and Fig. 5 c is the fluctuation data and curves after separating.

3. stages 6 σ estimates

Data are got 30-60 totally 30 measurement data, calculate 6 σ that separate back fluctuation data, obtain stages 6 σ

Evaluation of estimate is 2.08.The size fluctuation that shows this point is in ± 1.04mm.

4. assembly fluctuation quality index

For this example, assembly promptly is a body in white.After calculating the stages 6 σ value of all measuring points, carry out ordering from big to small, remove 5% of maximum, the fluctuation quality evaluation index as body in white of maximum in all the other 6 σ values.This index is 3.65 in this example, and the fluctuation of main measuring point that shows whole body in white is all in ± 1.8254mm.

5. instant percent of pass:

Choose certain point existing 140 measurement data in the recent period, it is as follows that per 10 one batch totals are calculated percent of pass:

θ=[0.4,0.4,0.7,0.5,0.2,0.6,0.4,0.2,0.1,0.2,0.4,0.2,0.4,0.4]

A. calculate the average and the variance of empirical data: $\stackrel{\‾}{\mathrm{\θ}}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\θ}}_1{---S}_{\mathrm{\θ}}^2=\frac{1}{n-1}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{({\mathrm{\θ}}_1-\stackrel{\‾}{\mathrm{\θ}})}^2$

B. getting prior distribution is that β distributes, and determines super parameter a by moments method, and b is defined as a=2.8855, and b=5.035 obtains prior distribution β (2.8855,5.035).

C. posteriority distributes

The frequency of passing through in the sample is obeyed binomial distribution.Sample size is 2, have 1 qualified, posteriority distribution β (a+1, b+2-1).

D. according to instant percent of pass formula $E\left(\mathrm{\θ}\right|x)=\frac{a+1}{a+b+2}$ Obtaining percent of pass is 41.29%

6. stage percent of pass

After instant percent of pass having been carried out the record of a period of time, be as the criterion at a certain time interval, the mean value that calculates percent of pass is as the stage percent of pass.A plurality of estimated values to Empirical Bayes are averaged, and the stage percent of pass of obtaining is 66.95%

7. production-line technique ability index

This example is carried out the example of production line dimension process merit rating with 6 σ that slide for certain body workshop.To the slide calculating of 6 σ of all measuring point datas since volume production, the value of getting its 95 hundredths has 487 days as objects of statistics, and therefore 487 objects of statistics are arranged.These 487 values are carried out statistical computation, obtain statistical distribution, its 95% fiducial interval is [3.1341,5.3723]; Show that this production line can be effectively be controlled at 3.1341-5.3723 with the fluctuation of size.

Fig. 6 is the whole measurement data of the body in white 6 σ result of calculations of sliding.Wherein Fig. 6 a is the objects of statistics curve that obtains, and Fig. 6 b is the objects of statistics distribution plan.

The present invention can extract the fitted position quality information timely and effectively under the situation that body dimensions off-line small sample is measured, accurately estimate.The evaluation index that provides can be estimated the technological ability of whole body in white size quality and production line.By estimating, for the evaluation of producing stage assembly line frock status level as a trial provides foundation.Because feedback that can be promptly and accurately can shorten the frock adjustment cycle effectively, accelerate the listing speed of new model.And at normal production period, can carry out effective monitoring, in time pinpoint the problems and deal with problems the production line state.

The present invention is directed to the offline inspection in the body of a motor car production run, set up small sample sampling-analysis-evaluation flow process, utilize the measurement data error separating technology, vehicle body manufacturing dimension quality precision is provided the evaluation in instant and stage, and provide the integrated artistic merit rating method of production line fitted position quality, for the control and the raising of the quality of production provides foundation, be significantly improved than traditional method precision.

The present invention also is applicable to the detection and the evaluation of size quality in the engineering goods assembling process that other are produced in enormous quantities, small sample detects.

Claims (4)

1, a kind of data processing method of evaluating size and quality in making car body, it is characterized in that adopting small sample Sampling techniques towards producing in enormous quantities, utilize the priori bayes method to estimate the instant percent of pass on the same day, utilize wavelet filtering method and EWMA method that measurement data is carried out separating of trend term and fluctuation, set up the instant evaluation index and the stage evaluation index of size fluctuation quality, and provide the polynary slip 6 σ indexs of production line size Control technological ability, production line size Control technological ability is estimated, and the concrete operations step is as follows:

1) measures sampling: the determining of sample size for instant evaluation, the body in white sample number is 1-2, the sub-unit sample number is 3-5, estimate for the stage, statistical sample is at the 17-30 platform, for production line dimension process merit rating, get all data that begin from volume production, determining of sampling time can be for sampling every day at a fixed time, the sub-unit sampling can be finished continuously, generally get 3～5/day, 1～2/day of body in white sample frequency, the assembly relation of specific demand is sampled as irregular sampling.

2) error separating of measurement data: detect The data EWMA method for sub-unit and handle, handle for the detection The data wavelet filtering method of body in white.

3) stages 6 σ estimates: determining of statistical sample is synchronously rapid 1, and application data is separated later fluctuation item during calculating.

4) assembly integral 6 σ estimate: application data is separated later fluctuation item, and the 6 σ values of getting 95 hundredths are as assembly integral 6 σ evaluation indexes;

5) instant percent of pass evaluation: adopt empirical Bayes method, use is without the raw data of separating, known percent of pass of past is distributed as the priori β that empirical data is configured to percent of pass, revise with the sample situation on the same day again, obtain the estimated value that the same day, percent of pass distributed;

6) the stage percent of pass is estimated: can monthly carry out by week, specifically be calculated as the ratio that the measuring point of this section period in tolerance range accounts for whole measuring points;

7) production-line technique merit rating: a measurement sequence that rounds is as analytic target, to the 6 σ calculating of sliding, obtains the dynamic change sequence of 6 σ, and this sequence is carried out statistical study, and the fiducial interval with 95% is as evaluation index.

2, as the data processing method of the said evaluating size and quality in making car body of claim 1, when it is characterized in that adopting the EWMA method that sub-unit detection data are carried out the error separating processing, measurement sample with every day is one group, from measuring for the first time, to one group of new data computation mean value of every day, with the trend term in the recursive operation acquisition data, the number S of the historical data of introducing is taken as 4, weighting coefficient W gets 0.2-0.4, asks difference to obtain the high-frequency fluctuation data by raw data and trend data.

3, as the data processing method of the said evaluating size and quality in making car body of claim 1, it is characterized in that said wavelet filtering method adopts the discrete form of Daubechies small echo, when body in white detection data are carried out the error separating processing, get three layers of conversion.

4, a kind of DATA REASONING equipment of evaluating size and quality in making car body, it is characterized in that body in white and main sub-unit cubing 1 are connected to three coordinate measuring machine 2, sub-unit detects anchor clamps 3 and is connected to portable surveying instrument 4, the output of three coordinate measuring machine 2 and portable surveying instrument 4 is connected to data analysis computing machine 5, and the output of data analysis computing machine 5 connects intranet 7 and output device 6.

Images