CN107562696A - Tire product quality on-line checking and control method - Google Patents
Tire product quality on-line checking and control method Download PDFInfo
- Publication number
- CN107562696A CN107562696A CN201710817985.0A CN201710817985A CN107562696A CN 107562696 A CN107562696 A CN 107562696A CN 201710817985 A CN201710817985 A CN 201710817985A CN 107562696 A CN107562696 A CN 107562696A
- Authority
- CN
- China
- Prior art keywords
- tire
- msub
- data
- mrow
- temperature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000002159 abnormal effect Effects 0.000 claims abstract description 29
- 230000008859 change Effects 0.000 claims abstract description 11
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 238000003062 neural network model Methods 0.000 claims abstract description 6
- 230000007774 longterm Effects 0.000 claims abstract description 5
- 230000000737 periodic effect Effects 0.000 claims abstract description 5
- 238000004458 analytical method Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 14
- 230000007935 neutral effect Effects 0.000 claims description 12
- 210000002569 neuron Anatomy 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 7
- 238000007596 consolidation process Methods 0.000 claims description 6
- 230000014509 gene expression Effects 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000005259 measurement Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 2
- 210000005036 nerve Anatomy 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 description 14
- 230000006835 compression Effects 0.000 description 5
- 238000007906 compression Methods 0.000 description 5
- 241001269238 Data Species 0.000 description 3
- 238000005452 bending Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000003908 quality control method Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012372 quality testing Methods 0.000 description 2
- 238000004073 vulcanization Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000003490 calendering Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000011076 safety test Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
Landscapes
- Tires In General (AREA)
Abstract
A kind of tire product quality on-line checking and control method, comprise the following steps:(1) the whole piece record of useless index and missing data is deleted;For there is periodic index, the tire record data in incomplete cycle is deleted;(2) abnormal data in tire real time data is judged using 3 σ principles in mean variance method and extreme difference;(3) on the basis of sentencing at the beginning of the statistic, according to time series it is basic the characteristics of, the feature of data behind is hidden in by analysis, new time series feature is extracted and is analyzed, carries out secondary detection;The data of sampled data and standard sample database progress trend is compared, if Long-term change trend exceedes given threshold value, it may be abnormal tire to judge the tire;(4) the real-time and on-line study ability based on on-line study machine, the parameter of neural network model is trained using the history tire pressure and temperature data of acquisition, regularly updates standard database.It the method increase the time efficiency and accuracy rate of judgement.
Description
Technical field
The present invention relates to a kind of for tire product quality on-line checking and the method for control, belong to tire quality control skill
Art field.
Background technology
With developing rapidly for China's economy, the car owning amount of resident is continuously increased, and causes auto manufacturing to tire
Demand also continue to increase.Also increasingly paid close attention to accordingly, with respect to the quality security problem of tire production by industry, especially
Quality safety test problems before tire shipment.How accurately and effectively wheel tyre defect to be detected, ensure tire
Outgoing, it is the problem of must currently paying attention to.
At present, the detection both at home and abroad on tire quality is all to carry out afterwards, substandard product is either scrapped or low
Valency processing, brings huge economic losses.Tire quality is usually to carry out corresponding quality testing using the mode of image procossing, profit
With x-ray bombardment finished tire, then tire image is shown by computer, finally passes through tire defect detection equipment again
Detected, but using the above method equipment it is sufficiently expensive, most domestic is detected by manual method, efficiency compared with
Low, artificial disturbance factor is more.
In the quality control of tire, the detection except carrying out finished tire using the above method can also be according to tire
Production process carry out quality testing and control, to improve the qualification rate of product, improve the production efficiency of product.Wheel is viviparous
The technique of production is divided into banburying, calendering, shaping, four master operations of vulcanization, and wherein vulcanization process is related to the inherent quality of tire,
Directly influence the product quality and service life of tire.In order to preferably carry out tire curing procedure, it is necessary to be carried by boiler
The steam of confession realizes the Optimum Matching of curingprocess rate, curing temperature and pressure as thermal source.Most domestic tire plant, mainly
Using the mode manually rechecked, pedestrian is entered to the Periodic Temperature in every tires production process of every production line, pressure curve
The mode of work selective examination, to exclude unqualified tire, take that big, accuracy is not high and artificial disturbance factor is more.
The content of the invention
The present invention is in order to make up the deficiency of existing tire quality detection technique, there is provided a kind of compressed data is fast, reliability
High, practical tire product quality on-line checking and control method, forecast model can adaptively be adjusted according to real time data
Parameter, product abnormal prediction accuracy rate is improved, avoids detecting the economic loss that occurrence product are brought afterwards.
The tire product quality on-line checking and control method of the present invention, including data prediction, statistic are just sentenced, are based on
The secondary judgement that knowledge instructs and the judgement based on neutral net, step specific as follows:
(1) data prediction:Delete the whole piece record of useless index and missing data;For there is periodic index, delete
Unless the tire record data of complete cycle.
(2) statistic is just sentenced:Judged using 3 σ principles in mean variance method and extreme difference different in tire real time data
Regular data.
(3) the secondary judgement that knowledge based instructs:On the basis of sentencing at the beginning of the statistic, according to the basic spy of time series
Point, the feature of data behind is hidden in by analysis, new time series feature is extracted and is analyzed, carry out secondary detection;Will
The data of sampled data and standard sample database (data of standard sample database are the sampled data of normal tire) carry out trend comparison,
If Long-term change trend exceedes given threshold value, it may be abnormal tire to judge the tire.
(4) judgement based on neutral net:Real-time and on-line study ability based on on-line study machine, utilizes going through for acquisition
History tire pressure and temperature data (normal sample and exceptional sample) are trained to the parameter of neural network model, are regularly updated
Standard database.
Useless index in the step (1) refers to except temperature, left inside pressure, right internal pressure, left hot plate in left inside temperature, the right side
Remaining index beyond eight temperature, right hot plate temperature, left mould sleeving temperature and right mould sleeving temperature indexs.
Missing data refers to data corresponding to no tyre serial number in the step (1).
The detailed process of the step (2) is:The average and variance of each index are calculated, using 3 σ principles, finds out a columns
According to mean μ and variances sigma, then the data outside (μ -3 σ, μ+3 σ) are abnormal data;The extreme difference per column data is calculated simultaneously
(maximum and minimum value), the extreme difference scope of each index is trained, the tire of indication range is exceeded for data, prompt tire to deposit
In exception.
It is to calculate tire to be tested and the dynamic time of each sample in database is curved that trend in the step (3), which compares,
Bent distance, when dynamic time warping distance exceedes given threshold value, then the tire may be abnormal tire, be otherwise normal rounds
Tire.
Two time serieses are represented with X and Y respectively in the step (3), and length is respectively | X | and | Y |;Consolidation path
(Warp Path) form is W=w1, w2..., wk, wherein wkForm be (i, j).Consolidation path distance be defined as D (| X |, | Y
|), wherein
D (i, j)=Dist (i, j)+min [D (i-1, j), D (i, j-1), D (i-1, j-1)].
Each tire sampled point number is ni, i=1 ..., n, T expression sampling times, the left inside temperature of R expressions, B storage right wheels
Tire is numbered, and K storage slopes, the i-th tire is designated as in the slope of j sampled pointWherein rI, jRepresent that the i-th tire exists
The left inside temperature of j sampled point, tI, jRepresent time of i-th tire in j sampled point.
The detailed process of the step (4) is, by temperature in the left inside temperature of each sampled point of single tire of measurement, the right side
Degree, left inside pressure, right internal pressure, left hot plate temperature, right hot plate temperature, left mould sleeving temperature and 8 values of right mould sleeving temperature are used as a sample
This attributive character value, normal tire output is 1, and underproof tire output is 0, obtains corresponding characteristic results set,
Then sample is divided into training sample and test sample;A given activation primitive G (x) and hidden neuron numberTraining
Sample data, judge real time tire sampled data with the presence or absence of abnormal;
The parameter of neural network model is initialized first, selected parameter initialization training set is designated as:
Wherein N0The number of samples needed for parameter initialization is represented, it is general to requireBoth initialization sample number was big
The hidden neuron number in neutral net;xiRepresent input sample, tiDesired output is represented, herein, the 1 normal tire of expression, 0
Represent abnormal tire;Rn, RmInput sample and desired output sample dimension are represented respectively;
The row vector of initial hidden layer output matrix is designated as:
Wherein w, b represent the connection weight and hidden layer deviation of input layer and hidden layer in neutral net respectively,For nerve net
Hidden neuron number in network, N0Represent the number of samples needed for neural network parameter initialization;
Initially output weights are:
Wherein
Hidden layer output vector is:
Givens QR based on OLS algorithms are decomposed:Λ1/2(N) H (N)=Q (N) R (N), wherein Q (N) be each row just
The matrix of friendship, R (N) are a upper triangular matrixs, q (N)=[QT(N)Q(N)]-1/2QT(N)Λ1/2(N) Y (N), Ω (N)=[QT
(N)Q(N)]1/2R (N), q (N) are a vectors, and Ω (N) is a upper triangle square formation, (relative to Gram-Schmidt conversion and
Householder is converted, and Givens conversion has obvious saving internal memory, computational efficiency high, sequential when solving least square problem
The advantages of adjustment facilitates) to calculate newest output weights be β (k)=R (k)-1p(k)。
The present invention is based on tire actual production line temperature and pressure data, and mean variance method, knowledge based is respectively adopted
The method of guidance and the Forecasting Methodology based on line neural network, the abnormal on-line monitoring of tire pressure and temperature data is realized,
Realize the abnormal alarm of unqualified tire curve;By compressed data, improve the time efficiency of judgement, by with master sample
Storehouse compares and regularly updates database, adaptive adjusting parameter, improves the accuracy rate of judgement;Suitable for had an impact product matter
Amount factor can quantify measurable automatic assembly line, have larger application value.
Brief description of the drawings
Fig. 1 is the schematic diagram of the left inside temperature interval of normal tire obtained by 3 σ principles.
Fig. 2 is tendency chart before the left inside temperature and pressure contracting of part tire.
Fig. 3 is tendency chart after the left inside temperature and pressure contracting of part tire.
Fig. 4 is the tire inspection operation result figure selected at random.
Fig. 5 is three abnormal tires and a left inside temperature trend map of normal tire.
Fig. 6 is online tire quality detecting system operation interface.
Embodiment
The tire product quality on-line checking of the present invention is related to data processing with control method, including statistic is just sentenced, base
The secondary judgement instructed in knowledge and the judgement based on neutral net, and form on-line detecting system.Carried out just using statistic
Sentence, the rule contained again by index for just sentencing result carries out secondary judgement, calculates the slope of adjacent record data, finds and close
Key point, compressed data;Then the change of tire curve to be identified and tire curve in database is compared by the DTW algorithms of modification
Trend, trained by mass data, provide judgment threshold, and regularly update database.Specifically include following steps.
One, data predictions
It is used for judging the index of tire quality at present mainly including temperature, left inside pressure, right internal pressure, a left side in left inside temperature, the right side
Eight hot plate temperature, right hot plate temperature, left mould sleeving temperature and right mould sleeving temperature indexs, delete the measurement data of remaining index;By
In two neighboring tire, existence time is poor in process of production, and the period system still records the number such as the temperature of production line, pressure
Judge according to, these data with tire unrelated, delete data corresponding to no tyre serial number;For there is periodic index, for example, it is left
It is warm in (right side), delete the tire data in incomplete cycle.
(1) for database given evidence, left inside temperature and temperature in the right side are completely the same, left mould sleeving temperature and right mould set temperature
Spend completely the same, temperature and the data of right mould sleeving temperature in the deletion right side.
(2) left tire bar code and right tire bar code correspond to synchronization sampled data, i.e. recognizable with a tire coding
Tire, delete left tire coding (delete right wheel tire coding also can).
(3) tire on a production line, is molded the sampled data that the data between matched moulds are tire, meeting between two tires
There is interval, i.e. a tire sampling is completed, and next tire does not start to sample, and this period, system still recorded some data,
These data do not have tyre serial number, useless for tire checking, therefore delete the data of coding missing.
(4) incomplete cycle tire data is deleted.
(5) time changes:Record time format is hh:mm:Ss, it is single that the time is converted into " second " according to sequencing
Position.
Two, statistics are just sentenced
The abnormal data in tire real time data is judged using 3 σ principles in mean variance method and extreme difference.
Calculate the average and variance of each index, using 3 σ principles, find out the mean μ and variance μ of a column data, then (μ -3 σ,
The σ of μ+3) outside data be abnormal data;The extreme difference (maximum, minimum value) per column data is calculated simultaneously, and training each refers to
Target extreme difference scope, the tire of indication range is exceeded for data, system alarm, prompts tire to there may be exception.Wherein, pole
Difference judge in addition to being controlled by indication range, it is also contemplated that the difference of tire or so temperature or pressure.From sample
Data see that left and right temperature and pressure difference is all very small, once some index of a certain moment tire or so or so observation data are inclined
Difference is larger, then product quality may be caused abnormal.
(1) 3 σ principles:The mean μ and variance μ of a column data are found out, then the data outside (μ -3 σ, μ+3 σ) are abnormal
Data.
(2) symbol description:
Assuming that the upper cycle limit of one tire of test is the n seconds.Divide time into L parts.Adjacent time point is at intervals of ε, institute
The index of test in need has M, and all tire numbers are m.
If the jth column data of i-th of tire is in t ∈ [tk, tk+1) in the range of data be x (i, j, t), wherein 0=t0<
t1=t0+ ε < t2=t1+ ε < ... < tk< tk+1=tk+ ε < ... < tL=tL-1+ ε=n.ε=ls, i=1,2 ..., m, j=1,
2 ..., M.
(3) specific algorithm explanation
1. fixed j, extraction t ∈ [tk, tk+1) in all data for one row, statistics number, calculate its mean μ and
Variances sigma, μ is designated as respectivelykAnd σk。
2. if column data number is 0, mean μ and variances sigma and a upper section [tk-1, tk) mean μ and variances sigma phase
Together.
3. if column data number is 1, mean μ, variances sigma and a upper section [t are only calculatedk-1, tk) variances sigma it is identical.
4. if column data number is more than 1, its mean μ and variances sigma are calculated.
Judge:If variances sigma and a upper section [tk-1, tk) variances sigma change it is bigger, it is necessary to according to less σ for mark
It is accurate.The mean μ calculated according to above-mentioned stepskAnd variances sigmak, calculate as t ∈ [tk, tk+1) when, the reasonable interval of numerical value is
(μk-3σk, μk+3σk)。
Fig. 1 illustrates the left inside temperature interval of normal tire obtained by 3 σ principles by taking left inside temperature as an example.
The secondary judgement that three, knowledge baseds instruct
For with the index curve periodically with visible trend changing rule, by compressed data, calculating wheel to be identified
The trend distance of tire curve and tire curve in database, threshold value is trained from historical data, as the reference for judging abnormal tyre
Data.
On the basis of sentencing at the beginning of the statistic, according to time series it is basic the characteristics of, mass data is hidden in by analysis and carried on the back
Feature afterwards, extract new time series feature and analyzed, carry out secondary detection;By the number of sampled data and standard sample database
Trend comparison is carried out according to (data of standard sample database are the sampled data of normal tire), if Long-term change trend exceedes given threshold value,
It may be abnormal tire then to judge the tire.
Two time serieses represent with X and Y respectively, and length is respectively | X | and | Y |;Consolidation Path form is W=w1,
w2..., wk, wherein wkForm be (i, j);Consolidation path distance be defined as D (| X |, | Y |), wherein:
D (i, j)=Dist (i, j)+min [D (i-1, j), D (i, j-1), D (i-1, j-1)].
Each tire sampled point number is ni, i=1 ..., n, in the T expression sampling times, R represents left inside temperature, B storage right wheels
Tire is numbered, and K storage slopes, the i-th tire is designated as in the slope of j sampled pointWherein rI, jRepresent that the i-th tire exists
The left inside temperature of j sampled point, tI, jRepresent time of i-th tire in j sampled point.
Due to tire part index number, such as temperature, left and right internal pressure have significantly periodically in left and right, and Long-term change trend is bright
It is aobvious.Therefore mainly by a complete tire to be tested, some data and curves (can be adopted the step by die sinking and matched moulds
Collect the complete documentation data of a tire) trend contrast is carried out with database Plays tire data curve, when trend generation is bright
Aobvious change, then may cause abnormal tyre.Due to tire sampling time interval, sampled data points number also differs, directly
The difference of two groups of sampled datas can not be contrasted using the Euclidean distance of classics by connecing, therefore the present invention program mainly utilizes the dynamic of modification
State Time alignment (DTW) algorithm, calculate the dynamic time warping distance of tire to be tested and each sample in database.When dynamic
Time Warp distance exceedes given threshold value, and system alarm, it may be abnormal tire to prompt the tire, be otherwise normal tire.
(1) key point, compressed data are found
DTW algorithms solve the Similarity Problem of Length discrepancy sequence well, but in the difference for contrasting two sampled datas
In the different time, what is taken is each point in sequence and the distance that a little calculates one by one of another sequence, less efficient, operation time compared with
It is long.For large-scale data, it can not realize that instantaneous completion judges.Because many sampled points change simultaneously for overall trend in sequence
Without too big influence, therefore key point can be found, both ensure the variation tendency of legacy data, reduce data point number again, improved
Operation efficiency.
1. slope calculations
The change of slope is using as the important indicator for finding key point.Provided with N number of tire, each tire sampled point number is
ni, i=1 ..., N, the sampling time is represented with T respectively, R represents left inside temperature, and B stores right tyre serial number, K storage slopes.I-th wheel
Tire is designated as in the slope of j sampled pointWherein rI, jRepresent left inside temperature of i-th tire in j sampled point, tI, j
Represent time of i-th tire in j sampled point.
2. time threshold determines and key point is chosen
According to temperature variation curve, the slope that existing tire all moment correspond to sampled point is calculated, counts the mode of slope
For 0.1, therefore the value is chosen as threshold epsilon1.If kI, j≤ε1, then i-th of tire is deleted in j instance sample point data;Otherwise
Retain i-th of tire in j instance sample point data, the point is a key point.
Contrasted 3. compression is front and rear
A. data point number compares
Table 1 gives sampled point number before and after the tire compression of part, the wherein corresponding upper odd number row of even numbers row, for its compression
Number afterwards.It can be seen that finding key point by slope, the unconspicuous point of slope variation is deleted, sampled data is greatly reduced,
The 1/3 of generally original sampled point number.
Table 1 compresses front and rear sampled point number contrast
524 | 528 | 527 | 527 | 532 | 527 | 523 | 524 | 523 | 490 | 517 |
183 | 151 | 151 | 149 | 142 | 146 | 162 | 154 | 168 | 151 | 157 |
527 | 530 | 521 | 525 | 524 | 525 | 531 | 525 | 529 | 486 | 517 |
150 | 151 | 146 | 148 | 139 | 154 | 155 | 168 | 149 | 143 | 156 |
524 | 521 | 523 | 525 | 513 | 520 | 519 | 518 | 515 | 491 | 521 |
158 | 143 | 143 | 139 | 166 | 151 | 160 | 152 | 159 | 135 | 156 |
519 | 519 | 521 | 520 | 519 | 518 | 511 | 518 | 517 | 524 | 486 |
148 | 163 | 138 | 148 | 172 | 152 | 139 | 138 | 169 | 181 | 130 |
520 | 494 | 489 | 490 | 495 | 492 | 492 | 489 | 483 | 521 | 487 |
157 | 154 | 162 | 144 | 150 | 139 | 151 | 145 | 150 | 149 | 137 |
525 | 525 | 530 | 527 | 525 | 528 | 527 | 523 | 516 | 523 | 521 |
143 | 156 | 138 | 144 | 145 | 166 | 160 | 136 | 147 | 153 | 169 |
525 | 372 | 508 | 512 | 504 | 487 | 518 | 515 | 518 | 510 | 509 |
140 | 107 | 160 | 152 | 143 | 165 | 137 | 164 | 170 | 166 | 163 |
520 | 516 | 519 | 521 | 519 | 514 | 491 | 501 | 513 | 516 | 516 |
166 | 171 | 162 | 163 | 163 | 151 | 160 | 172 | 155 | 7 | 147 |
B. trend comparison
Present invention selection slope have chosen suitable threshold value as index, while data amount check is greatly reduced,
It ensure that the variation tendency of legacy data.Table 2 is 10 tire randomly selected, the bending before compression between any two tire
Distance, table 3 are the deflection distance after compression, it can be seen that deflection distance varies less.Fig. 2 and Fig. 3 is respectively to be adopted before and after compressing
Sampling point tendency chart.
Dynamic bending distance before table 2 compresses
0 | 0.04806 | 0.078621 | 0.076654 | 0.050763 | 0.079083 | 0.08251 | 0.050888 | 0.077544 | 0.096605 |
0.04806 | 0 | 0.091549 | 0.090281 | 0.0778 | 0.088734 | 0.075019 | 0.06725 | 0.081791 | 0.085117 |
0.078621 | 0.091549 | 0 | 0.072104 | 0.089201 | 0.054579 | 0.054608 | 0.080595 | 0.055682 | 0.105247 |
0.076654 | 0.090281 | 0.072104 | 0 | 0.060511 | 0.057739 | 0.0532 | 0.073845 | 0.081348 | 0.134884 |
0.050763 | 0.0778 | 0.089201 | 0.060511 | 0 | 0.08356 | 0.072871 | 0.067606 | 0.095372 | 0.141965 |
0.079083 | 0.088734 | 0.054579 | 0.057739 | 0.08356 | 0 | 0.040236 | 0.08054 | 0.044813 | 0.108039 |
0.08251 | 0.075019 | 0.054608 | 0.0532 | 0.072871 | 0.040236 | 0 | 0.085784 | 0.043272 | 0.108695 |
0.050888 | 0.06725 | 0.080595 | 0.073845 | 0.067606 | 0.08054 | 0.085784 | 0 | 0.08718 | 0.096095 |
0.077544 | 0.081791 | 0.055682 | 0.081348 | 0.095372 | 0.044813 | 0.043272 | 0.08718 | 0 | 0.100932 |
0.096605 | 0.085117 | 0.105247 | 0.134884 | 0.141965 | 0.108039 | 0.108695 | 0.096095 | 0.100932 | 0 |
Dynamic bending distance after table 3 compresses
C. ONLINE RECOGNITION
Online acquisition tire data, the data between die sinking and matched moulds are a complete sampled data of tire.Treat for i-th
The deflection distance of j-th of tire in the tire and database of identification is D (i, j).D (i, j) is smaller, illustrates the curved of two sequences
Qu Chengdu is closer, because the sample data in database is the sampled data of normal tire, all sequences trend almost one
Cause.CalculateIf M exceedes given threshold value, illustrate that the tire sampled data is bent with sample in database
Degree difference is larger, that is, is possible to have larger fluctuation between temperature change in process and normal variation rule, so as to shadow
Ring tire quality.According to the deflection distance between sequence in database, by regular exercise, obtain as threshold parameter ε2.If M
> ε2, then system alarm, it is abnormal tire to prompt this tire, and record may be abnormal tire right wheel tire coding, if M≤
ε2, then to be normal.Fig. 4 gives the numbering of abnormal tire in tire to be detected.Fig. 5 give the left inside temperature of abnormal tire with
The trend control of the left inside temperature of normal tire.
Judgements of four, based on neutral net
Real-time and on-line study ability based on on-line study machine, utilizes the history tire pressure and temperature data of acquisition
(normal sample and exceptional sample) is trained to the parameter of neural network model, regularly updates standard database.
By temperature, left inside pressure, right internal pressure, left hot plate temperature in the left inside temperature of each sampled point of single tire of measurement, the right side
The attributive character value of degree, right hot plate temperature, left mould sleeving temperature and 8 values of right mould sleeving temperature as a sample, normal tire
Export as 1, underproof tire output is 0, obtains corresponding characteristic results set, sample then is divided into training sample and survey
Sample sheet;A given activation primitive G (x) and hidden neuron number, training sample data, judge that real time tire samples
Data are with the presence or absence of abnormal.
(1) initial phase:Give an initial training collection
Learning algorithm is initialized by following initialization procedure first:
1. give input weight wiWith deviation biAssignment (for additivity hidden neuron), or center wiWith width biIt is (relative
In RBF hidden neurons).
2. calculate initial hidden layer output matrixWherein
3. the initial output weights of estimationWherein
(2) the Sequence Learning stage:For follow-up input (Xi, ti).If k=0.
1. calculate hidden layer output vector hk+1。
2. with output weights β (k)=R (k) that the Givens QR decomposition computations based on OLS algorithms are newest-1p(k)。
3. k=k+1, return to second step.
The iteration of algorithm is mainly two Main Stages:Initial phase is before single hidden layer is trained with original EL M methods
To neutral net, and initial phase, once completing, these training datas are just dropped, and reduce computer memory space.Net
In network after data initialization, algorithm will learning training data one by one, once and the learning process of training data complete, then delete
The training data, so as to be correspondingly improved the pace of learning of algorithm.
Five, on-line detecting systems
Using JAVA by above-mentioned three kinds of models (statistic just sentence (3 σ principles), knowledge based instruct secondary judgement, be based on
The judgement of neutral net) processing method be packaged into a real time tire quality of production on-line monitoring system.The system, including 1.
Import sampled data;2. model selects;3. Input Online samples real time data;4. export prediction result;5. carry out production control;
6. circulate.
Fig. 6 gives online tire quality detecting system operation interface, and production line is carried out using the operation of menu button formula
The real-time monitoring of quality, have it is simple to operate, it is feature-rich, the characteristics of easy left-hand seat.
Claims (7)
1. a kind of tire product quality on-line checking and control method, it is characterized in that, including sentence at the beginning of data prediction, statistic,
The secondary judgement that knowledge based instructs and the judgement based on neutral net, step specific as follows:
(1) data prediction:Delete the whole piece record of useless index and missing data;For there is periodic index, delete non-
The tire record data of complete cycle.
(2) statistic is just sentenced:The abnormal number in tire real time data is judged using 3 σ principles in mean variance method and extreme difference
According to.
(3) the secondary judgement that knowledge based instructs:On the basis of sentencing at the beginning of the statistic, according to time series it is basic the characteristics of, lead to
The feature that analysis is hidden in data behind is crossed, new time series feature is extracted and is analyzed, carries out secondary detection;By hits
Compared according to trend is carried out with the data of standard sample database, if Long-term change trend exceedes given threshold value, judge that the tire may be to be different
Normal tire.
(4) judgement based on neutral net:Real-time and on-line study ability based on on-line study machine, utilizes the history wheel of acquisition
Tire pressure and temperature data are trained to the parameter of neural network model, regularly update standard database.
2. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (1)
In useless index refer to except left inside temperature, the right side in temperature, left inside pressure, right internal pressure, left hot plate temperature, right hot plate temperature, a left side
Remaining index beyond eight indexs of die sleeve temperature and right mould sleeving temperature.
3. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (1)
Middle missing data refers to data corresponding to no tyre serial number.
4. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (2)
Detailed process be:The average and variance of each index are calculated, using 3 σ principles, finds out the mean μ and variances sigma of a column data, then
Data outside (μ -3 σ, μ+3 σ) are abnormal data;The extreme difference per column data is calculated simultaneously, trains the extreme difference model of each index
Enclose, the tire of indication range is exceeded for data, prompt tire to exist abnormal.
5. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (3)
In trend to compare be the dynamic time warping distance for calculating tire to be tested and each sample in database, when dynamic time is curved
Bent distance exceedes given threshold value, then the tire may be abnormal tire, be otherwise normal tire.
6. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (3)
In two time serieses represent that length is respectively with X and Y respectively | X | and | Y |;Consolidation Path form is W=w1,w2,…,wk,
Wherein wkForm be (i, j);Consolidation path distance be defined as D (| X |, | Y |), wherein:
D (i, j)=Dist (i, j)+min [D (i-1, j), D (i, j-1), D (i-1, j-1)].
Each tire sampled point number is ni, i=1 ..., n, in the T expression sampling times, R represents left inside temperature, B storage right wheel tire volumes
Number, K storage slopes, the i-th tire is designated as in the slope of j sampled pointWherein ri,jRepresent the i-th tire at j
The left inside temperature of sampled point, ti,jRepresent time of i-th tire in j sampled point.
7. tire product quality on-line checking according to claim 1 and control method, it is characterized in that, the step (4)
Detailed process be:By temperature, left inside pressure, right internal pressure, left heat in the left inside temperature of each sampled point of single tire of measurement, the right side
The attributive character value of plate temperature, 8 right hot plate temperature, left mould sleeving temperature and right mould sleeving temperature values as a sample, normally
Tire output is 1, and underproof tire output is 0, obtains corresponding characteristic results set, sample then is divided into training sample
And test sample;A given activation primitive G (x) and hidden neuron numberTraining sample data, judge that real time tire is adopted
Sample data are with the presence or absence of abnormal;
The parameter of neural network model is initialized first, selected parameter initialization training set is designated as:
<mrow>
<msub>
<mi>N</mi>
<mn>0</mn>
</msub>
<mo>:</mo>
<mo>{</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>|</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>R</mi>
<mi>n</mi>
</msup>
<mo>,</mo>
<msub>
<mi>t</mi>
<mi>i</mi>
</msub>
<mo>&Element;</mo>
<msup>
<mi>R</mi>
<mi>m</mi>
</msup>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>N</mi>
<mn>0</mn>
</msub>
<mo>,</mo>
<msub>
<mi>N</mi>
<mn>0</mn>
</msub>
<mo>&GreaterEqual;</mo>
<mover>
<mi>N</mi>
<mo>~</mo>
</mover>
<mo>}</mo>
<mo>,</mo>
</mrow>
Wherein N0The number of samples needed for parameter initialization is represented, it is general to requireBoth initialization sample number was more than nerve
Hidden neuron number in network, xiRepresent input sample, tiDesired output is represented, herein, 1 represents normal tire, and 0 represents different
Normal tire;Rn, RmInput sample and desired output sample dimension are represented respectively;
The row vector of initial hidden layer output matrix is designated as:
<mrow>
<msub>
<mi>h</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<msup>
<mrow>
<mo>&lsqb;</mo>
<mi>G</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>w</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>G</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>w</mi>
<mover>
<mi>N</mi>
<mo>~</mo>
</mover>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mover>
<mi>N</mi>
<mo>~</mo>
</mover>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mi>T</mi>
</msup>
<mo>,</mo>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<msub>
<mi>N</mi>
<mn>0</mn>
</msub>
<mo>;</mo>
</mrow>
Wherein w, b represent the connection weight and hidden layer deviation of input layer and hidden layer in neutral net respectively,To be hidden in neutral net
Layer neuron number, N0Represent the number of samples needed for neural network parameter initialization.
Initially output weights are:
<mrow>
<mi>&beta;</mi>
<mrow>
<mo>(</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<msubsup>
<mi>H</mi>
<mn>0</mn>
<mi>T</mi>
</msubsup>
<msub>
<mi>H</mi>
<mn>0</mn>
</msub>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<msubsup>
<mi>H</mi>
<mn>0</mn>
<mi>T</mi>
</msubsup>
<msub>
<mi>T</mi>
<mn>0</mn>
</msub>
<mo>;</mo>
</mrow>
Wherein
Hidden layer output vector is:
<mrow>
<msub>
<mi>h</mi>
<mrow>
<mi>k</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>=</mo>
<msup>
<mrow>
<mo>&lsqb;</mo>
<mi>G</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>w</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mn>1</mn>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
<mo>...</mo>
<mo>,</mo>
<mi>G</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>w</mi>
<mover>
<mi>N</mi>
<mo>~</mo>
</mover>
</msub>
<mo>,</mo>
<msub>
<mi>b</mi>
<mover>
<mi>N</mi>
<mo>~</mo>
</mover>
</msub>
<mo>,</mo>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>&rsqb;</mo>
</mrow>
<mi>T</mi>
</msup>
<mo>;</mo>
</mrow>
Givens QR based on OLS algorithms are decomposed:Λ1/2(N) H (N)=Q (N) R (N), wherein Q (N) be one respectively arrange it is orthogonal
Matrix, R (N) are a upper triangular matrixs, q (N)=[QT(N)Q(N)]-1/2QT(N)Λ1/2(N) Y (N), Ω (N)=[QT(N)Q
(N)]1/2R (N), q (N) are a vectors, and Ω (N) is a upper triangle square formation;It is β (k)=R to calculate newest output weights
(k)-1p(k)。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710817985.0A CN107562696B (en) | 2017-09-12 | 2017-09-12 | On-line detection and control method for tire product quality |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710817985.0A CN107562696B (en) | 2017-09-12 | 2017-09-12 | On-line detection and control method for tire product quality |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107562696A true CN107562696A (en) | 2018-01-09 |
CN107562696B CN107562696B (en) | 2020-06-05 |
Family
ID=60980599
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710817985.0A Expired - Fee Related CN107562696B (en) | 2017-09-12 | 2017-09-12 | On-line detection and control method for tire product quality |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107562696B (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108268901A (en) * | 2018-01-25 | 2018-07-10 | 中国环境监测总站 | A kind of algorithm that environmental monitoring abnormal data is found based on dynamic time warping distance |
CN108415393A (en) * | 2018-04-19 | 2018-08-17 | 中江联合(北京)科技有限公司 | A kind of GaAs product quality consistency control method and system |
CN108536777A (en) * | 2018-03-28 | 2018-09-14 | 联想(北京)有限公司 | A kind of data processing method, server cluster and data processing equipment |
CN108921335A (en) * | 2018-06-15 | 2018-11-30 | 山东大学 | Intelligent box substation based on short-term load forecasting runs combined optimization method |
CN110321366A (en) * | 2019-06-27 | 2019-10-11 | 中国科学院自动化研究所 | Method and system are determined based on the statistic of on-line study |
CN111123837A (en) * | 2019-12-31 | 2020-05-08 | 江苏南高智能装备创新中心有限公司 | Forging press fault prediction device and method thereof |
CN111221810A (en) * | 2020-01-13 | 2020-06-02 | 苏宁云计算有限公司 | Commodity main data abnormity identification method, system, computer equipment and storage medium |
CN112470446A (en) * | 2018-05-31 | 2021-03-09 | 法国大陆汽车公司 | Method for reconfiguring a motor vehicle tyre monitoring device |
CN114877760A (en) * | 2022-07-12 | 2022-08-09 | 东方空间技术(北京)有限公司 | Polarity testing method and system for spacecraft |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103302777A (en) * | 2013-06-26 | 2013-09-18 | 广东工业大学 | Neural-network-based energy consumption anomaly detection method of tire vulcanizing machine |
CN103745279A (en) * | 2014-01-24 | 2014-04-23 | 广东工业大学 | Method and device for monitoring energy consumption abnormity |
CN103886372A (en) * | 2014-04-12 | 2014-06-25 | 青岛科技大学 | Tire body temperature soft measurement method based on wavelet neural network |
CN105444856A (en) * | 2015-11-12 | 2016-03-30 | 云南电网有限责任公司物流服务中心 | Tire pressure dynamic weighing device based on neural network |
CN106228002A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | A kind of high efficiency exception time series data extracting method based on postsearch screening |
US20170231519A1 (en) * | 2014-08-19 | 2017-08-17 | The General Hospital Corporation | System and method for annotating and analyzing eeg waveforms |
-
2017
- 2017-09-12 CN CN201710817985.0A patent/CN107562696B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103302777A (en) * | 2013-06-26 | 2013-09-18 | 广东工业大学 | Neural-network-based energy consumption anomaly detection method of tire vulcanizing machine |
CN103745279A (en) * | 2014-01-24 | 2014-04-23 | 广东工业大学 | Method and device for monitoring energy consumption abnormity |
CN103886372A (en) * | 2014-04-12 | 2014-06-25 | 青岛科技大学 | Tire body temperature soft measurement method based on wavelet neural network |
US20170231519A1 (en) * | 2014-08-19 | 2017-08-17 | The General Hospital Corporation | System and method for annotating and analyzing eeg waveforms |
CN105444856A (en) * | 2015-11-12 | 2016-03-30 | 云南电网有限责任公司物流服务中心 | Tire pressure dynamic weighing device based on neural network |
CN106228002A (en) * | 2016-07-19 | 2016-12-14 | 北京工业大学 | A kind of high efficiency exception time series data extracting method based on postsearch screening |
Non-Patent Citations (1)
Title |
---|
杨静芳: "轮胎动态平衡在线检测试验机研制及其关键技术研究", 《中国博士学位论文全文数据库 工程科技II辑》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108268901A (en) * | 2018-01-25 | 2018-07-10 | 中国环境监测总站 | A kind of algorithm that environmental monitoring abnormal data is found based on dynamic time warping distance |
CN108268901B (en) * | 2018-01-25 | 2021-05-18 | 中国环境监测总站 | Method for discovering environmental monitoring abnormal data based on dynamic time bending distance |
CN108536777A (en) * | 2018-03-28 | 2018-09-14 | 联想(北京)有限公司 | A kind of data processing method, server cluster and data processing equipment |
CN108415393A (en) * | 2018-04-19 | 2018-08-17 | 中江联合(北京)科技有限公司 | A kind of GaAs product quality consistency control method and system |
CN112470446A (en) * | 2018-05-31 | 2021-03-09 | 法国大陆汽车公司 | Method for reconfiguring a motor vehicle tyre monitoring device |
CN108921335A (en) * | 2018-06-15 | 2018-11-30 | 山东大学 | Intelligent box substation based on short-term load forecasting runs combined optimization method |
CN110321366B (en) * | 2019-06-27 | 2021-06-25 | 中国科学院自动化研究所 | Statistical quantity determining method and system based on online learning |
CN110321366A (en) * | 2019-06-27 | 2019-10-11 | 中国科学院自动化研究所 | Method and system are determined based on the statistic of on-line study |
CN111123837A (en) * | 2019-12-31 | 2020-05-08 | 江苏南高智能装备创新中心有限公司 | Forging press fault prediction device and method thereof |
CN111221810A (en) * | 2020-01-13 | 2020-06-02 | 苏宁云计算有限公司 | Commodity main data abnormity identification method, system, computer equipment and storage medium |
CN111221810B (en) * | 2020-01-13 | 2023-01-06 | 苏宁云计算有限公司 | Commodity main data abnormity identification method, system, computer equipment and storage medium |
CN114877760A (en) * | 2022-07-12 | 2022-08-09 | 东方空间技术(北京)有限公司 | Polarity testing method and system for spacecraft |
CN114877760B (en) * | 2022-07-12 | 2022-09-13 | 东方空间技术(北京)有限公司 | Polarity testing method and system for spacecraft |
Also Published As
Publication number | Publication date |
---|---|
CN107562696B (en) | 2020-06-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107562696A (en) | Tire product quality on-line checking and control method | |
US20240184956A1 (en) | Prediction method of crown of steel plates and strips based on data driving and mechanism model fusion | |
CN115018021B (en) | Machine room abnormity detection method and device based on graph structure and abnormity attention mechanism | |
CN111222549B (en) | Unmanned aerial vehicle fault prediction method based on deep neural network | |
CN106971240A (en) | The short-term load forecasting method that a kind of variables choice is returned with Gaussian process | |
CN112200433A (en) | Power plant thermal performance analysis and optimization system | |
CN106951695A (en) | Plant equipment remaining life computational methods and system under multi-state | |
CN106503035A (en) | A kind of data processing method of knowledge mapping and device | |
AU2020103881A4 (en) | A Diagnosis Method of Train Body Assembly Deviation in the Manufacturing Process of High-Speed Train | |
CN107377634A (en) | A kind of hot-strip exports Crown Prediction of Media method | |
CN112170501B (en) | Prediction method for wear crown and thermal crown of roller | |
CN113723010A (en) | Bridge damage early warning method based on LSTM temperature-displacement correlation model | |
CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
CN105975701A (en) | Parallel scheduling disassembly path forming method based on mixing fuzzy model | |
CN116362376A (en) | Comprehensive energy station construction carbon emission prediction method based on machine learning | |
CN116796275A (en) | Multi-mode time sequence anomaly detection method for industrial equipment | |
CN114548494B (en) | Visual cost data prediction intelligent analysis system | |
CN113858566B (en) | Injection molding machine energy consumption prediction method and system based on machine learning | |
CN117195747B (en) | Uniform heat distribution optimization method for magnetic material drying | |
CN112241832B (en) | Product quality grading evaluation standard design method and system | |
CN110738403B (en) | Data processing method, device and computer storage medium | |
CN110175166A (en) | Method for confirming quality abnormity of tobacco wrapping workshop | |
CN116304849A (en) | Two-dimensional piston pump fault diagnosis method based on local cut space arrangement and gating circulation network | |
CN112628132B (en) | Water pump key index prediction method based on machine learning | |
CN113780405A (en) | Air conditioner parameter regression optimization method based on deep neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right | ||
TR01 | Transfer of patent right |
Effective date of registration: 20201029 Address after: No.3, Shangzhi Road, Shibei District, Qingdao, Shandong Province Patentee after: Qingdao whirlpool Intelligent Technology Co.,Ltd. Address before: 250353 Ji'nan University, Changqing, Shandong Road, No. 3501 Patentee before: Qilu University of Technology |
|
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200605 |