CN109177101A - A kind of injection molding machine batch process fault detection method - Google Patents
A kind of injection molding machine batch process fault detection method Download PDFInfo
- Publication number
- CN109177101A CN109177101A CN201810691180.0A CN201810691180A CN109177101A CN 109177101 A CN109177101 A CN 109177101A CN 201810691180 A CN201810691180 A CN 201810691180A CN 109177101 A CN109177101 A CN 109177101A
- Authority
- CN
- China
- Prior art keywords
- test
- data
- matrix
- gauss
- train
- 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.)
- Pending
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B29—WORKING OF PLASTICS; WORKING OF SUBSTANCES IN A PLASTIC STATE IN GENERAL
- B29C—SHAPING OR JOINING OF PLASTICS; SHAPING OF MATERIAL IN A PLASTIC STATE, NOT OTHERWISE PROVIDED FOR; AFTER-TREATMENT OF THE SHAPED PRODUCTS, e.g. REPAIRING
- B29C45/00—Injection moulding, i.e. forcing the required volume of moulding material through a nozzle into a closed mould; Apparatus therefor
- B29C45/17—Component parts, details or accessories; Auxiliary operations
- B29C45/76—Measuring, controlling or regulating
Abstract
The present invention discloses a kind of injection molding machine batch process fault detection method, the problem of this method handles data Length discrepancy by data minimum length method, achieve the effect that eliminating data estimates further through the data normalization in both direction, then pivot analysis PCA dimensionality reduction is utilized, to reduce the calculation amount of data modeling, interference is reduced;Then by automatically adjusting gauss component number based on the unsupervised mixed Gaussian algorithm of improvement EM algorithm;Finally calculate the global monitoring index of the new monitoring point of real time data.This method combines multivariate statistical method with mode identification method, it is applied in fault detection, not only avoid the processing to mass data, the influence for reducing operand and outlier improves the accuracy of model built, and calculation amount is concentrated mainly on the off-line modeling stage, it ensure that the real-time of on-line monitoring, global monitoring index used has merged the integrated monitoring index of local message, realizes continuous on-line monitoring, improves the verification and measurement ratio of failure.
Description
Technical field
The invention belongs to injection moulding process detection fields, and in particular to a kind of injection molding machine batch process fault detection method.
Background technique
In modernization industry production process, injection moulding process is very typical multioperation stage batch process, and interval data
With non-linear, dynamic, the multistage, asynchronous sampling, data Length discrepancy, higher-dimension, it is highly coupled, data defect and noise
The features such as.Inaccurate data modeling, bring are not only low qualification rate caused by the discovery not in time of failure, are more likely to bring
Be that injection molding machine repairs a large amount of economic losses of bring enterprise not in time.
Handle interval data, different modeling methods, effect to stress face also somewhat different.In existing method, have
Sub-period division methods based on PCA are had the statistical model established using MPCA, also have the method closed on based on K to be modeled
Deng.The shortcomings that sub-period based on PCA is could not to consider the autocorrelation performance of process variable in time;Multinomial pivot analysis
For the multistage, cannot preferably be handled in high coupling data processing;Based on the method that K closes on, this method is related to heavy
It calculates, this will seriously affect the efficiency of detection.
Summary of the invention
In view of the deficiencies of the prior art, present invention combination PCA algorithm proposes a kind of injection molding machine batch process fault detection side
Method, the method reduce operands, improve efficiency.Specific technical solution is as follows:
A kind of injection molding machine batch process fault detection method, which is characterized in that this method comprises the following steps:
(1) data Length discrepancy is handled:
The n batch data under injection molding machine works normally is acquired, the shortest lot data of data is selected in n batch data,
Then other lot datas are intercepted, so that all lot datas data length having the same, to obtain equal long numbers
It is used as training data according to X (I × J × K), I is batch number, and J is process variable number, and K is sampling instant number;
(2) data normalization is carried out to training data X (I × J × K), and is converted to X (IK × J), then to X (IK × J)
Progress transposition is X (J × IK), is then denoted as Ttrain(M×N);
(3) to Ttrain(M × N) carries out the training of the gauss hybrid models based on EM algorithm, ttrainFor TtrainThe column of (M × N)
Vector, K are the quantity of gauss component, give mono- suitable initial value of K;
(3.1) desired step: using any initial value α (0), μ (0), and ∑ (0) substitutes into following formula:
In formula, p(s)(Ck|ttrain) indicate in s step iteration, sample ttrainBelong to k-th of gauss component CkPosteriority
Probability, wherein α (0) has K value, the prior probability of respectively K gauss component, and μ (0) is K M dimensional vector, respectively K
The mean value of a gauss component, ∑ (0) are the covariance matrixes of K gauss component, wherein probability density function are as follows:
(3.2) maximization steps
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (3.2.1), acquires s+1 step iteration
In each gauss component mean μk (s+1):
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (3.2.2), acquires s+1 step iteration
In each gauss component prior probability αk (s+1):
The s+1 that (3.2.3) acquires the posterior probability of the corresponding each gauss component of each sample acquired and step (3.2.1)
Walk the mean μ of each gauss component in iterationk (s+1)Following formula is substituted into, the covariance matrix ∑ of each gauss component is acquiredk (s+1)
(3.3) judgement convergence
The obtained gauss component prior probability of the step iteration is counted first less than 10-4Number, and K is subtracted into the number
Value assigns K, then judges the difference of s+1 step iteration and s step iteration prior probability, if | αk (s+1)-αk (s)| < δ, wherein δ be greater than
Zero number, then iteration ends, otherwise return to desired step (3.1) and continue iteration;
(4) global monitoring index P is integrated using following formulaglobal, whether it is failure for monitoring the currently monitored point in new samples
Point
Wherein,For ttestFor the local probability index of k-th of model, p (Ck|ttest) it is test data
Score vector ttestBelong to the posterior probability of k-th of model, detailed process is as follows:
(4.1) test data X is sought with PCAtestThe score vector t of (K × J)testThe matrix T of compositiontest(K × R), K are to adopt
Sample moment number, J are process variable number, and R is number of principal components, seeks XtestThe covariance matrix X of (K × J)T testXtestEigenvalue λj
(j < J), and R maximum characteristic values are used as principal component before choosing, by the corresponding feature of a maximum characteristic value of preceding R to
Amount, the i.e. corresponding load vector p of principal componentj, to form P (J × R) load matrix, test matrix is acquired by following formula,
That is principal component matrix
Ttest(K × R)=Xtest(K×J)P(J×R)
(4.2) it is calculate by the following formula the score vector t of test datatestBelong to the posterior probability p (C of k-th of modelk|
ttest), thus model belonging to judging
Wherein, g (ttest|Ck) be k-th of gauss component distribution parameter, αk(k=1,2 ..., K) is each Gauss model
Prior probability;
(4.3) t is calculatedtestFor the local probability index P of k-th of model(k) local(ttest);
(4.4) by p (Ck|ttest) and P(k) local(ttest) substitute into and acquire general indices PglobalIf Pglobal> 0.95, then table
Show that the currently monitored point in monitoring new samples is fault point.
Further, the step (2) carries out data normalization using following method:
(2.1) training data X (I × J × K) is launched into two-dimensional matrix X (I × JK) according to the direction batch I;
(2.2) data normalization is carried out to matrix X (I × JK) using following formula, to show injection molding machine interval mistake
Change information between journey different batches
(2.3) matrix X (I × J × K) then is rearranged into the matrix X (I × JK) after standardization, then incited somebody to action again
The matrix X (I × J × K) arrived is launched into X (J × IK) according to variable direction J;
(2.4) obtained data X (J × IK) is standardized using following formula, then to the X (J after standardization
× IK) it is denoted as Ttrain(M×N);
Further, in order to reduce off-line calculation amount, interference is reduced, after standardizing using PCA method to step (2.4)
The transposition X (IK × J) of X (J × IK) carries out modeling processing:
(3.1) the covariance matrix X of X (IK × J) is soughtTThe eigenvalue λ of Xj, j < J, and choose preceding R maximum characteristic values and make
For principal component;
(3.2) pass through the corresponding feature vector of the maximum characteristic value of preceding R, the i.e. corresponding load vector p of principal componentj,
To form P (J × R) load matrix;
(3.3) principal component matrix T is acquired by following formulatrain(IK × R), the training matrix as Gaussian Mixture training
Ttrain(IK × R)=X (IK × J) P (J × R)
Then to Ttrain(IK × R) carries out transposition, is denoted as Ttrain(M×N)。
Further, the initial value ∑ (0) of gauss component covariance matrix is arranged as unit square in the step (3)
Battle array.
Further, t in the step (4.3)testFor the local probability index of k-th of model
It is first to calculate ttestTo the mahalanobis distance of each modelProbability is converted by card cloth distribution table again
Wherein,Indicate ttestTo the mahalanobis distance of k-th of gauss component, ∑-1Indicate k-th of gauss component
Inverse, the μ of covariance matrixkFor the mean value of k-th of gauss component,For chi square distribution, m is freedom degree, that is, vector ttest
The number of variable.
Compared with prior art, beneficial effects of the present invention:
Fault detection method of the invention combines multivariate statistical method with mode identification method, is applied to fault detection
In, the processing to mass data is not only avoided, the influence of operand and outlier is reduced, improves the accurate of model built
Property, and calculation amount is concentrated mainly on the off-line modeling stage, ensure that the real-time of on-line monitoring, global monitoring index used
The integrated monitoring index for having merged local message, realizes continuous on-line monitoring, improves the verification and measurement ratio of failure.
Detailed description of the invention
Fig. 1 is algorithm flow chart of the invention;
Fig. 2 three-dimensional matrice presses the data structure diagram of batch direction expansion;
Fig. 3 three-dimensional matrice presses the data structure diagram of process variable direction expansion.
Specific embodiment
Below according to attached drawing and preferred embodiment the present invention is described in detail, the objects and effects of the present invention will become brighter
White, below in conjunction with drawings and examples, the present invention will be described in further detail.It should be appreciated that described herein specific
Embodiment is only used to explain the present invention, is not intended to limit the present invention.
Injection moulding process is a typical interval industrial process, and the following examples are exactly injection molding process fault detection
One embodiment, most of variable states of injection moulding process can be collected by respective sensor.
As shown in Figure 1, a kind of injection molding machine batch process fault detection method acquires under 25 groups of operating conditions in this example
11 process variables, related data such as table 1, including 22 groups of normal datas and 3 groups of fault datas, each operation batch is with 20ms
Obtain 1000 or so sampled values altogether for the sampling period.
1 injection moulding process variable data of table
Variable serial number | Variable name | Unit |
1 | Nozzle temperature | ℃ |
2 | Nozzle exit pressure | Bar |
3 | Injection speed | mm/sec |
4 | Screw rod rotation speed | RPM |
5 | Screw stroke | mm |
6 | Oil cylinder working-pressure | Bar |
7 | Plasticizing pressure | Bar |
8 | Cavity pressure | Bar |
9 | SV1 valve opening | % |
10 | SV2 valve opening | % |
11 | Mold temperature | ℃ |
Since this method is when actual samples, the interval time of sampling for various reasons will not be stringent, therefore meeting
The sampled point for different batches occur is different as a result, so using " shortest length method " the obtained original historical data of injection molding machine
Carry out equal length treatment.
(2) to 22 groups of normal datas, shortest 996 1 groups of data length is chosen and is used as full-length, obtains data X (22
× 11 × 996), it is denoted as X (I × J × K).
(3) then obtained three-dimensional matrice is carried out array as shown in Figure 2 is unfolded by batch direction, is denoted as X (I × JK),
Then data normalization is carried out using following formula to the matrix:
Then the matrix X (I × JK) after standardization is rearranged into matrix X (I × J × K) again, then will obtained again
Matrix X (I × J × K) is launched into X (J × IK) according to variable direction, as shown in figure 3, then carrying out data standard to the matrix
Change, formula is as follows:
It is above-mentioned that data are done with centralization and normalized square mean processing simultaneously on batch direction, it reduces to a certain extent
The non-linear and dynamic characteristic of data, and the change information of data between different batches is also highlighted in this way;Then by two-dimemsional number
According to it is reinflated be three-dimensional data, then do in variable dimension same standardization processing, such purpose is prominent each variable
Variation on the entire period.
(4) in order to reduce off-line calculation amount, reduce interference, using PCA method to after step (2.4) standardization X (J ×
IK transposition X (IK × J)) carries out modeling processing:
(4.1) the covariance matrix X of X (IK × J) is soughtTThe eigenvalue λ of Xj, j < J, and choose preceding R maximum characteristic values and make
For principal component;
(4.2) pass through the corresponding feature vector of the maximum characteristic value of preceding R, the i.e. corresponding load vector p of principal componentj,
To form P (J × R) load matrix;
(4.3) principal component matrix T is acquired by following formulatrain(IK × R), the training matrix as Gaussian Mixture training
Ttrain(IK × R)=X (IK × J) P (J × R)
Then to Ttrain(IK × R) carries out transposition, is denoted as Ttrain(M×N)。
Since history modeling data amount is generally very big, and since data dimension is too many and the effect of some dimensions is extremely micro-
It is small, therefore directly the data of standardization directly should not be taken away modeling.High dimensional data is dropped using PCA (principal component analysis)
The advantages of low dimensional, can not only improve operation by the Data Dimensionality Reduction after standardization to the low-dimensional more sensitive to failure in this way
Efficiency, moreover it is possible to improve fault detection rate.
(5) to Ttrain(M × N) is carried out based on the gauss hybrid models training for improving EM algorithm, ttrainFor Ttrain(M×N)
Column vector, K be gauss component quantity, give mono- suitable initial value of K;
(5.1) desired step (E step):
K=7 is enabled,E is unit battle array,
Following formula is substituted into respectively:
In formula, p(s)(Ck|ttrain) indicate in s step iteration, sample ttrainBelong to k-th of gauss component CkPosteriority
Probability, wherein α (0) has K value, the prior probability of respectively K gauss component, and μ (0) is K M dimensional vector, respectively K
The mean value of a gauss component, ∑ (0) are the covariance matrixes of K gauss component, wherein probability density function are as follows:
(5.2) maximization steps (M step)
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (5.2.1), acquires s+1 step iteration
In each gauss component mean μk (s+1):
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (5.2.2), acquires s+1 step iteration
In each gauss component prior probability αk (s+1):
The s+1 that (5.2.3) acquires the posterior probability of the corresponding each gauss component of each sample acquired and step (3.2.1)
Walk the mean μ of each gauss component in iterationk (s+1)Following formula is substituted into, the covariance matrix ∑ of each gauss component is acquiredk (s+1)
(5.3) judgement convergence
First determine whether the step iteration gauss component prior probability less than 10-4Number, and by (this of K- numerical value) assign K,
Then judge the difference of s+1 step iteration and s step iteration prior probability, if | αk (s+1)-αk (s)| < δ, wherein δ is the number greater than zero,
Otherwise so iteration ends return to desired step (5.1) and continue iteration;
Improved EM algorithm is used in the invention.EM algorithm is more sensitive to initial value, if initial value is arranged
With great difficulty do not reach locally optimal solution, therefore the setting of initial value needs certain priori knowledge.And inventive algorithm is a kind of
Unsupervised clustering method meets local Gaussian point according to the initial data in each stage in the case where no priori knowledge
The condition of cloth is trained since any number of gauss component, gradually removes the item that weight is zero by constantly iteration, thus
Obtain best Gauss model number.
(6) global monitoring index P is integrated using following formulaglobal, whether it is failure for monitoring the currently monitored point in new samples
Point
Wherein,For ttestFor the local probability index of k-th of model, p (Ck|ttest) it is test data
Score vector ttestBelong to the posterior probability of k-th of model, detailed process is as follows:
(6.1) test data X is sought with PCAtestThe score vector t of (K × J)testThe matrix T of compositiontest(K × R), K are to adopt
Sample moment number, J are process variable number, and R is number of principal components, seeks XtestThe covariance matrix X of (K × J)T testXtestEigenvalue λj
(j < J), and R maximum characteristic values are used as principal component before choosing, by the corresponding feature of a maximum characteristic value of preceding R to
Amount, the i.e. corresponding load vector p of principal componentj, to form P (J × R) load matrix, test matrix is acquired by following formula,
That is principal component matrix
Ttest(K × R)=Xtest(K×J)P(J×R)
(6.2) it is calculate by the following formula the score vector t of test datatestBelong to the posterior probability p (C of k-th of modelk|
ttest), thus model belonging to judging
Wherein, g (ttest|Ck) be k-th of gauss component distribution parameter, αk(k=1,2 ..., K) is each Gauss model
Prior probability;
(6.3) t is calculatedtestFor the local probability index P of k-th of model(k) local(ttest) it is first to calculate ttestTo each
The mahalanobis distance of modelProbability is converted by card cloth distribution table again
Wherein, μkCorresponding Gauss model is respectively corresponded to ∑,Indicate ttestTo the geneva of k-th of gauss component
Distance, ∑-1Indicate the inverse of the covariance matrix of k-th of gauss component, μkFor the mean value of k-th of gauss component,For card side point
Cloth, m are freedom degree, that is, vector ttestThe number of variable.
(6.4) by p (Ck|ttest) and P(k) local(ttest) substitute into and acquire general indices PglobalIf Pglobal> 0.95, then table
Show that the currently monitored point in monitoring new samples is fault point.
Judge which Gauss model current sampling point belongs to by integrated monitoring index, if break down.Integrated prison
Monitor control index T2 and the SPE control figure that index is different from traditional single model is controlled, but has merged the information of whole models, energy
Enough more accurate response data features, to improve fault detection rate.
It will appreciated by the skilled person that being not used to limit the foregoing is merely the preferred embodiment of invention
System invention, although invention is described in detail referring to previous examples, for those skilled in the art, still
It can modify to the technical solution of aforementioned each case history or equivalent replacement of some of the technical features.It is all
Within the spirit and principle of invention, modification, equivalent replacement for being made etc. be should be included within the protection scope of invention.
Claims (5)
1. a kind of injection molding machine batch process fault detection method, which is characterized in that this method comprises the following steps:
(1) data Length discrepancy is handled:
The n batch data under injection molding machine works normally is acquired, the shortest lot data of data is selected in n batch data, then
Other lot datas are intercepted, so that all lot datas data length having the same, to obtain isometric data X (I
× J × K) it is used as training data, I is batch number, and J is process variable number, and K is sampling instant number;
(2) data normalization is carried out to training data X (I × J × K), and is converted to X (IK × J), then X (IK × J) is carried out
Transposition is X (J × IK), is then denoted as Ttrain(M×N);
(3) to Ttrain(M × N) carries out the training of the gauss hybrid models based on EM algorithm, ttrainFor TtrainThe column of (M × N) to
Amount, K are the quantity of gauss component, give mono- suitable initial value of K;
(3.1) desired step: using any initial value α (0), μ (0), and ∑ (0) substitutes into following formula:
In formula, p(s)(Ck|ttrain) indicate in s step iteration, sample ttrainBelong to k-th of gauss component CkPosterior probability,
Wherein, α (0) has K value, the prior probability of respectively K gauss component, and μ (0) is K M dimensional vector, respectively K Gauss
The mean value of ingredient, ∑ (0) are the covariance matrixes of K gauss component, wherein probability density function are as follows:
(3.2) maximization steps
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (3.2.1), is acquired every in s+1 step iteration
The mean μ of a gauss componentk (s+1):
The posterior probability of the corresponding each gauss component of each sample acquired is substituted into following formula by (3.2.2), is acquired every in s+1 step iteration
The prior probability α of a gauss componentk (s+1):
(3.2.3) changes the s+1 step that the posterior probability of the corresponding each gauss component of each sample acquired and step (3.2.1) acquire
The mean μ of each gauss component in generationk (s+1)Following formula is substituted into, the covariance matrix ∑ of each gauss component is acquiredk (s+1)
(3.3) judgement convergence
The obtained gauss component prior probability of the step iteration is counted first less than 10-4Number, and by K subtract this numerical value assign
K is given, then judges the difference of s+1 step iteration and s step iteration prior probability, if | αk (s+1)-αk (s)| < δ, wherein δ is greater than zero
Number, then iteration ends, otherwise return to desired step (3.1) and continue iteration;
(4) global monitoring index P is integrated using following formulaglobal, whether it is fault point for monitoring the currently monitored point in new samples
Wherein,For ttestFor the local probability index of k-th of model, p (Ck|ttest) be test data score
Vector ttestBelong to the posterior probability of k-th of model, detailed process is as follows:
(4.1) test data X is sought with PCAtestThe score vector t of (K × J)testThe matrix T of compositiontest(K × R), when K is sampling
Number is carved, J is process variable number, and R is number of principal components, seeks XtestThe covariance matrix X of (K × J)T testXtestEigenvalue λj(j<
J), and before choosing R maximum characteristic values are used as principal component, by the corresponding feature vector of a maximum characteristic value of preceding R, i.e.,
The corresponding load vector p of principal componentj, to form P (J × R) load matrix, test matrix is acquired by following formula, i.e., it is main
Component matrix
Ttest(K × R)=Xtest(K×J)P(J×R)
(4.2) it is calculate by the following formula the score vector t of test datatestBelong to the posterior probability p (C of k-th of modelk|ttest),
To judge affiliated model
Wherein, g (ttest|Ck) be k-th of gauss component distribution parameter, αk(k=1,2 ..., K) is the elder generation of each Gauss model
Test probability;
(4.3) t is calculatedtestFor the local probability index P of k-th of model(k) local(ttest);
(4.4) by p (Ck|ttest) and P(k) local(ttest) substitute into and acquire general indices PglobalIf Pglobal> 0.95, then it represents that prison
Controlling the currently monitored point in new samples is fault point.
2. injection molding machine batch process fault detection method according to claim 1, which is characterized in that the step (2)
Data normalization is carried out using following method:
(2.1) training data X (I × J × K) is launched into two-dimensional matrix X (I × JK) according to the direction batch I;
(2.2) data normalization is carried out to matrix X (I × JK) using following formula, to show injection molding machine batch process not
With the change information between batch
(2.3) matrix X (I × J × K) then is rearranged into the matrix X (I × JK) after standardization, then will obtained again
Matrix X (I × J × K) is launched into X (J × IK) according to variable direction J;
(2.4) obtained data X (J × IK) is standardized using following formula, then to after standardization X (J ×
IK) it is denoted as Ttrain(M×N);
3. injection molding machine batch process fault detection method according to claim 2, which is characterized in that in order to reduce offline meter
Calculation amount reduces interference, is modeled using transposition X (IK × J) of the PCA method to the X (J × IK) after step (2.4) standardization
Processing:
(3.1) the covariance matrix X of X (IK × J) is soughtTThe eigenvalue λ of Xj, j < J, and preceding R maximum characteristic values are chosen as master
Ingredient;
(3.2) pass through the corresponding feature vector of the maximum characteristic value of preceding R, the i.e. corresponding load vector p of principal componentj, thus group
At P (J × R) load matrix;
(3.3) principal component matrix T is acquired by following formulatrain(IK × R), the training matrix as Gaussian Mixture training
Ttrain(IK × R)=X (IK × J) P (J × R)
Then to Ttrain(IK × R) carries out transposition, is denoted as Ttrain(M×N)。
4. injection molding machine batch process fault detection method according to claim 1, which is characterized in that the step (3)
The initial value ∑ (0) of middle gauss component covariance matrix is set as unit matrix.
5. injection molding machine batch process fault detection method according to claim 1, which is characterized in that the step
(4.3) t intestFor the local probability index of k-th of modelIt is first to calculate ttestTo the geneva of each model
DistanceProbability is converted by card cloth distribution table again
Wherein,Indicate ttestTo the mahalanobis distance of k-th of gauss component, ∑-1Indicate the association side of k-th of gauss component
Poor inverse of a matrix, μkFor the mean value of k-th of gauss component,For chi square distribution, m is freedom degree, that is, vector ttestVariable
Number.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810691180.0A CN109177101A (en) | 2018-06-28 | 2018-06-28 | A kind of injection molding machine batch process fault detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810691180.0A CN109177101A (en) | 2018-06-28 | 2018-06-28 | A kind of injection molding machine batch process fault detection method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109177101A true CN109177101A (en) | 2019-01-11 |
Family
ID=64948712
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810691180.0A Pending CN109177101A (en) | 2018-06-28 | 2018-06-28 | A kind of injection molding machine batch process fault detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109177101A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114734604A (en) * | 2022-03-28 | 2022-07-12 | 浙江凯华模具有限公司 | Mold temperature online control method in injection molding process |
CN115157601A (en) * | 2022-09-06 | 2022-10-11 | 南通飞旋智能科技有限公司 | Injection molding machine and detection method thereof |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009057113A2 (en) * | 2007-10-31 | 2009-05-07 | Rosetta Genomics Ltd. | Diagnosis and prognosis of specific cancers by means of differential detection of micro-rnas/mirnas |
US20150161232A1 (en) * | 2013-12-10 | 2015-06-11 | University Of Southern California | Noise-enhanced clustering and competitive learning |
CN105574587A (en) * | 2016-01-21 | 2016-05-11 | 华中科技大学 | On-line condition process monitoring method for plastic injection moulding process |
CN107065843A (en) * | 2017-06-09 | 2017-08-18 | 东北大学 | Multi-direction KICA batch processes fault monitoring method based on Independent subspace |
CN107272655A (en) * | 2017-07-21 | 2017-10-20 | 江南大学 | Batch process fault monitoring method based on multistage ICA SVDD |
-
2018
- 2018-06-28 CN CN201810691180.0A patent/CN109177101A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2009057113A2 (en) * | 2007-10-31 | 2009-05-07 | Rosetta Genomics Ltd. | Diagnosis and prognosis of specific cancers by means of differential detection of micro-rnas/mirnas |
US20150161232A1 (en) * | 2013-12-10 | 2015-06-11 | University Of Southern California | Noise-enhanced clustering and competitive learning |
CN105574587A (en) * | 2016-01-21 | 2016-05-11 | 华中科技大学 | On-line condition process monitoring method for plastic injection moulding process |
CN107065843A (en) * | 2017-06-09 | 2017-08-18 | 东北大学 | Multi-direction KICA batch processes fault monitoring method based on Independent subspace |
CN107272655A (en) * | 2017-07-21 | 2017-10-20 | 江南大学 | Batch process fault monitoring method based on multistage ICA SVDD |
Non-Patent Citations (1)
Title |
---|
王静 等: "基于GMM的间歇过程故障检测", 《自动化学报》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114734604A (en) * | 2022-03-28 | 2022-07-12 | 浙江凯华模具有限公司 | Mold temperature online control method in injection molding process |
CN114734604B (en) * | 2022-03-28 | 2023-08-29 | 浙江凯华模具有限公司 | On-line control method for mold temperature in injection molding process |
CN115157601A (en) * | 2022-09-06 | 2022-10-11 | 南通飞旋智能科技有限公司 | Injection molding machine and detection method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110262450B (en) | Fault prediction method for cooperative analysis of multiple fault characteristics of steam turbine | |
CN103927412B (en) | Instant learning debutanizing tower soft-measuring modeling method based on gauss hybrid models | |
CN104914723B (en) | Industrial process soft-measuring modeling method based on coorinated training partial least square model | |
CN108508865B (en) | A kind of fault detection method based on distributing OSC-PLS regression model | |
CN105574587B (en) | A kind of online operating mode course monitoring method of plastic injection molding process | |
CN108803520A (en) | A kind of dynamic process monitoring method rejected based on the non-linear autocorrelation of variable | |
CN112613186B (en) | Aero-engine gas circuit fault fusion diagnosis method based on statistical distribution characteristics | |
CN109538311B (en) | Real-time monitoring method for control performance of steam turbine in high-end power generation equipment | |
CN109298633A (en) | Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization | |
CN110009020A (en) | A kind of non-linear process monitoring method based on multicore Principal Component Analysis Model | |
CN110516394A (en) | Aero-engine steady-state model modeling method based on deep neural network | |
CN116340796B (en) | Time sequence data analysis method, device, equipment and storage medium | |
CN106896219A (en) | The identification of transformer sub-health state and average remaining lifetime method of estimation based on Gases Dissolved in Transformer Oil data | |
CN106647650A (en) | Distributed industrial process monitoring method based variable weighting PCA (Principal Component Analysis) model | |
CN109177101A (en) | A kind of injection molding machine batch process fault detection method | |
CN108445867A (en) | A kind of nongausian process monitoring method based on distributing ICR models | |
CN108469805A (en) | A kind of distributing dynamic process monitoring method based on dynamic optimal selection | |
CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
Liu et al. | Deep & attention: A self-attention based neural network for remaining useful lifetime predictions | |
CN108572639A (en) | A kind of dynamic process monitoring method rejected based on principal component autocorrelation | |
Ruan et al. | Remaining useful life prediction for aero-engine based on lstm and cnn | |
CN106649202B (en) | Industrial process flexible measurement method based on diversity variable weighting PLSR model | |
CN108427398A (en) | A kind of dynamic process monitoring method based on distributing AR-PLS models | |
Wenqiang et al. | Remaining useful life prediction for mechanical equipment based on temporal convolutional network | |
CN115618733A (en) | Multi-scale hybrid attention mechanism modeling method for predicting remaining service life of aircraft engine |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190111 |