CN116088453A - Production quality prediction model training method and device and production quality monitoring method - Google Patents
Production quality prediction model training method and device and production quality monitoring method Download PDFInfo
- Publication number
- CN116088453A CN116088453A CN202310131807.8A CN202310131807A CN116088453A CN 116088453 A CN116088453 A CN 116088453A CN 202310131807 A CN202310131807 A CN 202310131807A CN 116088453 A CN116088453 A CN 116088453A
- Authority
- CN
- China
- Prior art keywords
- production
- quality index
- quality
- training
- parameter
- 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
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 198
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000012549 training Methods 0.000 title claims abstract description 64
- 238000012544 monitoring process Methods 0.000 title claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims abstract description 42
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 229940088594 vitamin Drugs 0.000 claims description 20
- 229930003231 vitamin Natural products 0.000 claims description 20
- 235000013343 vitamin Nutrition 0.000 claims description 20
- 239000011782 vitamin Substances 0.000 claims description 20
- 150000003722 vitamin derivatives Chemical class 0.000 claims description 20
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 10
- FFBHFFJDDLITSX-UHFFFAOYSA-N benzyl N-[2-hydroxy-4-(3-oxomorpholin-4-yl)phenyl]carbamate Chemical compound OC1=C(NC(=O)OCC2=CC=CC=C2)C=CC(=C1)N1CCOCC1=O FFBHFFJDDLITSX-UHFFFAOYSA-N 0.000 claims description 8
- 238000001125 extrusion Methods 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000009413 insulation Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 230000000694 effects Effects 0.000 abstract description 5
- 210000002569 neuron Anatomy 0.000 description 9
- 238000000354 decomposition reaction Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 3
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 229910052757 nitrogen Inorganic materials 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32339—Object oriented modeling, design, analysis, implementation, simulation language
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a production quality prediction model training method, a training device and a production quality monitoring method based on a production quality prediction model, wherein the training method comprises the following steps: producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,……,x m ) And a multi-dimensional quality index set y= (Y) 1 ,y 2 ,……,y n ) As an initial sample; determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Subjective coefficientFor the production parameter x determined based on analytic hierarchy process i For quality index y j Is a weight of influence of (1); calculating each production parameter x in each initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set YIn a matrix formAs input, quality index y j Building training samples (C, y j ) Training a neural network to obtain a pre-production quality prediction model. The invention selects the correlation coefficientSubjective coefficientAnd the input data is subjected to dimension lifting, the mutual connection between the multidimensional production parameters and the multidimensional quality indexes is comprehensively considered, the training effect is enhanced, and the prediction accuracy of the model is improved.
Description
Technical Field
The invention belongs to the technical field of production quality prediction, and particularly relates to a production quality prediction model training method and device and a production quality monitoring method.
Background
In the manufacturing industry, production quality indicators are a very important part of the production line. Because of the influence of various factors, such as unreasonable design of production parameters, certain deviation exists in the quality of products in each production stage, if the quality deviation is not detected and adjusted in time, along with the accumulation and transmissibility of the quality deviation in the production process, it is difficult to trace the factors causing poor final quality. Therefore, during production, quality indexes of products need to be obtained in time, and defects of the production process are located in time.
The most traditional method is to directly measure the current quality index value of the product through a quality detection machine, but the quality detection machine is generally expensive and complex to operate, and can prolong the whole production period, and is time-consuming and labor-consuming.
With the research of neural network deep reinforcement learning, a great number of methods for predicting the process quality by training a neural network model are developed at present, but in the methods, the neural network model is trained by directly taking the process parameters as input and taking the quality as output, and the prediction accuracy of the finally obtained model is not high.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides a production quality prediction model training method and device and a production quality monitoring method, and aims to enhance the training effect by optimizing input data of neural network training, thereby improving the prediction accuracy of the model and providing more accurate basis for production quality monitoring.
To achieve the above object, according to one aspect of the present invention, there is provided a production quality prediction model training method comprising:
producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,……,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,……,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j The quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2;
determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Subjective coefficient->For multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,……,x m ) Production parameter x of (1) i For quality index y j Is a weight of influence of (1);
calculating each production parameter x in each initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
In a matrix formAs input, quality index y j Building training samples (C, y j ) Training a neural network to obtain a predicted quality index y j Is a production quality prediction model of (a).
In one embodiment, the multi-dimensional production parameter set x= (X) 1 ,x 2 ,……,x m ) The data of (1) is the actual sampling data which is subjected to denoising, data outlier processing, missing value processing and normalization processingAnd (3) obtaining, wherein the denoising adopts a CEEMDAN denoising algorithm.
In one embodiment, the production parameters are selected production parameters, and the method for selecting the production parameters includes:
training an XGBoost feature importance model, taking the production parameters to be screened as model input, taking the quality index as model output, sorting according to feature importance scores, eliminating the production parameters with lower scores of 1% -2%, and taking the rest parameters as the production parameters after screening.
In one embodiment, the objective function Obj used to train the XGBoost feature importance model is:
Ω(f t )=γT+0.5λw 2
wherein ,ft (x i ) Output function of t-th tree, g i () Represents the first derivative, h i () Representing the second derivative, T representing the number of leaf nodes in the tree, w representing the weight of the leaf, gamma representing the number of leaf nodes that can be controlled, and lambda representing the fraction of leaf nodes that can be controlled.
In one embodiment, subjective coefficients are determined based on analytic hierarchy processThe process of (1) comprises:
establishing a hierarchical structure model of a production process, wherein the quality index y j And multidimensional production parameter set x= (X) 1 ,x 2 ,……,x m ) Associating;
construction a= (a pq ) m×m Wherein element a pq Representing the p-th vitamin production parameter versus the q-th vitamin production parameter versus the quality index y j Importance scale of (a) according to experience setting judgment matrix each element value, a pq *a qp =1;
Based on the judgment matrix, calculating the weight of each dimension production parameter by using a root-finding method, wherein the calculation formula is as follows:
wherein ,Wi Weights for the ith vitamin production parameter;
the consistency check is carried out by calculating a consistency index C.I. to judge whether the judgment matrix set according to experience is reasonable, if not, the judgment matrix is adjusted until the consistency check is reasonable, so that the weight W of each dimension production parameter calculated based on the latest set judgment matrix i As subjective coefficients
wherein ,for production parameter x i A covariance matrix formed by covariance of each element in the multi-dimensional quality index set Y; sigma and method for producing the same YY Covariance matrix formed by covariance among each element of multi-dimensional quality index set Y, < ->For sigma YY An inverse matrix of (a); />Is->Is a transposed matrix of (a); />For all production parameters x i Variance of the data.
According to another aspect of the present invention, there is provided a production quality prediction model training apparatus comprising:
data acquisition unit for producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,……,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,……,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j The quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2;
subjective coefficient determining unit for determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Wherein subjective coefficient->For multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,……,x m ) Production parameter x of (1) i For quality index y j Is a weight of influence of (1);
a complex correlation coefficient calculation unit for calculating each production parameter x in the initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
An input unit for arranging in matrixAs input, quality index y j Building training samples (C, y j ) For training by the neural network.
According to still another aspect of the present invention, there is provided a production quality monitoring method including:
obtaining a current multi-dimensional production parameter set X= (X) 1 ,x 2 ,……,x m ) Wherein x is i Is the ith vitamin production parameter, and m is more than or equal to 2;
production parameter x for each dimension i Expanded into two-dimensional input vectorsAnd constructs an input matrixInputting the production quality prediction model and outputting the quality index y j Is a prediction of the data of the model;
wherein the subjective coefficientFor multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,……,x m ) Production parameter x of (1) i For quality index y j The production quality prediction model is trained according to the production quality prediction model training method.
In one embodiment, the quality index y is determined i Whether the predicted data of (a) is compliant or not, if not, according to subjective coefficientsGradually checking the production parameter x i Whether there is a problem.
In one embodiment, the predicted cable quality index data is produced by a three-layer cable coextrusion process, wherein the multi-dimensional production parameters comprise temperature, pressure, linear speed and quality index y i Is any one of the thickness of the inner screen, the thickness of the outer screen and the insulation eccentricity.
In general, the above technical solutions conceived by the present invention, compared with the prior art, enable the following beneficial effects to be obtained:
the invention considers that in actual production, various production parameters are involved at the same time and various quality indexes are also required to be considered, thus the invention collects the multi-dimensional production parameter set at the same time and corresponds to the production parametersThe multi-dimensional quality index set of the set. Corresponding to each production parameter, calculating a production parameter x i Complex correlation coefficient with multi-dimensional quality index set YProduction parameter x i Relative to the quality index y j Subjective coefficient of>Wherein the complex correlation coefficient->Reaction of production parameter x i Linear correlation with the multidimensional quality index set Y, subjective coefficient +.>The reaction is that the quality index y is influenced j Production parameter x among all production parameters of (2) i For quality index y j Is used to influence the weight. Production parameter x for each dimension i Expansion into three-dimensional input vector +.>And reconstruct the input matrix +.>To train the sample (C, y j ) Training the neural network. The invention selects the correlation coefficient->Subjective coefficient->And the input data is subjected to dimension lifting, the mutual connection between the multidimensional production parameters and the multidimensional quality indexes is comprehensively considered, and the training effect is enhanced, so that the prediction accuracy of the model is improved.
Drawings
FIG. 1 is a flow chart of steps of a method for training a production quality prediction model in one embodiment.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
A flowchart of the steps of the method for training a production quality prediction model is shown in fig. 1, and each of the steps is described in detail below.
Step S100: producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,……,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,……,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j And the quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2.
The invention needs to acquire a large amount of initial sample data to form an initial sample set.
The present invention contemplates a number of production parameters involved in the production and a number of quality indicators for the aspects under consideration. For example, for producing a high-voltage cable by using a three-layer co-extrusion process, the temperature of each extruder, the temperature of a vulcanizing tube, the nitrogen pressure, the linear speed and the like are important production parameters in the three-layer co-extrusion process, and the thickness of an inner screen, the thickness of an outer screen and the insulation eccentricity are important quality indexes to be considered, so that a plurality of production parameters and values corresponding to a plurality of quality indexes need to be acquired first during model training.
In one embodiment, after the original sampling value is obtained, preprocessing is performed first, which specifically includes denoising, data outlier processing, missing value processing, and normalization processing.
Specifically, a CEEMDAN denoising algorithm may be selected to denoise. The method comprises the following steps:
1) Setting the iteration times N, adding positive and negative Gaussian white noise of N groups into an original signal x (t), and obtaining N new signals in total. The new signal after the j-th Gaussian white noise addition is:
wherein x (t) is the original signal, ε 0 Is the standard deviation of noise, w j And (5) the jth Gaussian white noise meeting the standard normal distribution.
2) Let E (-) be the EMD decomposition of the new signal after the j th addition of Gaussian white noise, then the j th IMF can be obtained by the j th decomposition 1 The components are as follows:
in the formula ,IMFj 1 (t) is the j-th IMF 1 Component, r j (t) is the j-th residual component.
3) N IMFs decomposed by 1 The components are summed and averaged to obtain the final IMF 1 (t):
4) Calculation of the 1 st residual component r 1 (t)::
r 1 (t)=x(t)-IMF 1 (t)
5) At r 1 Adding N groups of auxiliary noise which adopts EMD decomposition to the paired positive and negative Gaussian white noise in the step 1 into (t), and setting E i (. Cndot.) is the ith modal component after EMD decomposition, then the new signal after the jth addition of auxiliary noise is: :
by aligning x j 1 (t) performing j-th decomposition to obtain the j-th IMF 2 The components, the decomposition results are:
then for the decomposed N IMFs 2 Adding and averaging to obtain final IMF 2 (t):
Calculation of the 2 nd residual component r 2 (t):
r 2 (t)=r 1 (t)-IMF 2 (t)
6) Repeating the step 5) until the obtained residual component can not be subjected to EMD decomposition, and ending the algorithm. Assuming that the number of IMF components obtained at this time is k, the original signal x (t) is decomposed into:
specifically, the data exceeding the upper limit and the lower limit can be removed by using the standard threshold value, so that the abnormal value of the data is processed. For a small amount of discontinuously missing data, the interpolation method is adopted to fill the data, the average value of the data before and after the interpolation method is adopted, for a large amount of continuously missing data, the interpolation method is difficult, even if interpolation is carried out, the change rule of the data is possibly not met, and then the deletion processing is directly carried out.
Specifically, a min-max normalization method can be adopted, and normalized production parameter data and quality index data can be obtained according to the following formula.
wherein xi For any of the ith vitamin production parameters in the initial sample set, max is the maximum value of the ith vitamin production parameters in the initial sample set, and min is the minimum value of the ith vitamin production parameters in the initial sample set.
In an embodiment, when the production parameters are more, in order to reduce the calculation degree on the premise of ensuring the training effect, the production parameters can be screened, and the production parameters with higher value can be selected to participate in the training of the model.
Specifically, training an XGBoost feature importance model, taking production parameters to be screened as model input, taking quality indexes as model output, sorting according to feature importance scores, eliminating 1% -2% of production parameters with lower scores, and taking the rest parameters as screened production parameters. The objective function Obj used by the XGBoost feature importance model is as follows:
Ω(f t )=γT+0.5λw 2
wherein ,ft (x i ) Output function of t-th tree, g i () Represents the first derivative, h i () Representing the second derivative, T representing the number of leaf nodes in the tree, w representing the weight of the leaf, gamma representing the number of leaf nodes that can be controlled, and lambda representing the fraction of leaf nodes that can be controlled.
Step S200: determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Wherein the subjective coefficientFor multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,……,x m ) Production parameter x of (1) i For quality index y j Is used to influence the weight.
Although the initial sample is constructed and various quality indexes are acquired, the model trained by the invention is only used for predicting one quality index, and the invention considers that the various quality indexes of the same product are mutually influenced by mutual constraint, so that the training data is firstly subjected to specific processing by utilizing the various quality indexes.
In step S200, one of the quality indexes y is used j As output of the predictive model, i.e. the trained predictive model is the predictive quality index y j Predictive model of data. For the quality index y j Each parameter in the multi-dimensional production parameter set is possible to be used for the quality index y j Has influence and different influence degrees. Thus, step S200 is to determine a multi-dimensional production parameter set x= (X) based on Analytic Hierarchy Process (AHP) 1 ,x 2 ,……,x m ) Each production parameter x of (a) i For quality index y j I.e. the production parameter x i Relative to the quality index y j Subjective coefficient of (2)It should be noted that each production parameter correspondingly obtains an influence quality index y j For example, in the case of a cable three-layer coextrusion process, after the process is fixed, the subjective coefficient of the temperature affecting the thickness of the inner screen, that is, the influence weight of the temperature on the thickness of the inner screen is determined, and the subjective coefficient is not changed due to the change of the temperature sampling value, which reflects the influence of the production parameter of the dimension on the quality index macroscopically.
Among them, the Analytic Hierarchy Process (AHP) is a qualitative and quantitative combined decision analysis method for solving the complex problem of multiple targets. The method combines quantitative analysis and qualitative analysis, judges the relative importance degree between the standards which can be realized among the measurement targets by using the experience of a decision maker, and reasonably gives the weight of each standard of each decision scheme. The invention relates to a method for systematically analyzing social, economic and management problems by using an Analytic Hierarchy Process (AHP), which is used for determining the influence weight of each production parameter on a certain quality index so as to guide the training of a follow-up neural network.
Specifically, the production parameter x is determined based on analytic hierarchy process i Influence of quality index y j Subjective coefficient of (2)The process of (1) comprises:
step S210: establishing a hierarchical structure model of a production process, wherein the quality index y j And multidimensional production parameter set x= (X) 1 ,x 2 ,……,x m ) And (5) associating.
Taking the three-layer co-extrusion procedure of the cable as an example, a hierarchical structure model of the three-layer co-extrusion procedure of the cable is established, wherein the thickness of the inner screen is related to materials, specific production flow and the like, and the materials are related to temperature, pressure and linear speed, and the upper layer factors and the lower layer factors are combed out as an example.
Step S220: construction a= (a pq ) m×m Wherein element a pq Representing the p-th vitamin production parameter versus the q-th vitamin production parameter versus the quality index y j Importance scale of (a) according to experience setting judgment matrix each element value, a pq *a qp =1。
The values (scales) of the elements in the judgment matrix are determined empirically and with reference to a judgment scale reference table. For example, see table 1 below for a reference table for a judgment scale, which is also a currently common reference table.
Table 1 judgment scale references
For example, the judgment matrix obtained for the three-layer coextrusion process of the cable is shown in table 2 below, wherein the quality index y i Is the thickness of the inner screen.
Table 2 criterion layer judgment matrix
Wherein, the influence scale of temperature to the quality index is 1/3, and the influence scale of pressure to the quality index is 3.
Step S230: based on the judgment matrix, calculating the weight of each dimension production parameter by using a root-finding method, wherein the calculation formula is as follows:
wherein ,Wi Is the weight of the ith vitamin production parameter.
Step S240: the consistency check is carried out by calculating a consistency index C.I. to judge whether the judgment matrix set according to experience is reasonable, if not, the judgment matrix is adjusted until the consistency check is reasonable, so that the weight W of each dimension production parameter calculated based on the latest set judgment matrix i As subjective coefficients
The calculation formula of the consistency index C.I. is as follows:
wherein ,λmax To determine the largest feature root of the matrix.
Step S300: calculating each production parameter x in each initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
Taking (X, Y) as an initial sample, acquiring a large amount of sample data, and calculating a production parameter X corresponding to each sample based on each specific sample data i Complex correlation coefficient with multi-dimensional quality index set Y in corresponding initial sample/>
wherein ,for production parameter x i A covariance matrix formed by covariance of each element in the multi-dimensional quality index set Y; sigma and method for producing the same YY Covariance matrix formed by covariance among each element of multi-dimensional quality index set Y, < ->For sigma YY An inverse matrix of (a); />Is->Is a transposed matrix of (a); />For all production parameters x i Variance of the data.
It should be noted that, the steps S200 and S300 are not performed in a specific order, and in other embodiments, the complex correlation coefficient may be calculated first and then the subjective coefficient may be determined, or both may be performed simultaneously.
Step S400: in a matrix formAs input, quality index y j Building training samples (C, y j ) Training a neural network to obtain a predicted quality index y j Is a production quality prediction model of (a).
Each production parameter x in any initial sample data i Expanded into three-dimensional vectorsm production parameters form a matrix of m.times.3>To (/ ->y j ) Training the neural network as a training sample to obtain a predicted quality index y j Is a production quality prediction model of (a). Wherein data of different dimensions is input to different channels.
It will be appreciated that if a different quality index needs to be predicted, the model needs to be retrained as per the above steps.
Specifically, 75% of the data can be randomly taken as training and the remaining 25% can be tested in a large number of training samples.
Specifically, the neural network input layer includes N neurons, the hidden layer includes L neurons, the output layer includes M neurons, and the definition: the input vector is X, the actual output vector is Y, and the model output vector is Y * F (x) is an activation function, the calculated error of the network training is epsilon, and the network learning rate is eta. The step of training the neural network model includes:
1) If the number of neurons in the input layer is N, the input vector set x= (X) 1 ,x 2 ,…,x N ),W i If the initial weight value is the node, the input formula of the hidden layer neuron is as follows:
2) If the number of neurons in the hidden layer is L, W ij For the connection weight value of the node j and the node i, theta j For the threshold of node i, the output of the hidden layer neuron is
3) If the number of output layer neurons is M, the output vector set y= (Y) 1 ,y 2 ,…,y M ),W jk For the connection weight value of the node j and the node k, theta k Being the threshold of node k, the output of the hidden layer neuron is
4) The activation functions of each layer are as follows:
5) The error epsilon of each node of the hidden layer and the output layer is
6) The weight of each layer of error reverse propagation is corrected by the following correction formula:
and repeatedly adjusting and training the weight and the threshold value of each layer by calculating an error function by utilizing the difference between the expected output and the actual output of the network, finishing training when the error reaches the preset precision, and otherwise, selecting the next training sample and the corresponding expected output for learning.
Correspondingly, the invention also relates to a production quality prediction model training device, which comprises:
data acquisition unit for producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,……,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,……,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j The quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2;
subjective coefficient determining unit for determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Wherein subjective coefficient->For multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,……,x m ) Production parameter x of (1) i For quality index y j Is a weight of influence of (1);
a complex correlation coefficient calculation unit for calculating each production parameter x in the initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
An input unit for arranging in matrixAs input, quality index y j Building training samples (C, y j ) For training by the neural network.
The production quality prediction model training device may be used to implement the production quality prediction model training method described above, it being understood that the production quality prediction model training device may further comprise a neural network, the input unit inputting training samples (C, y j ) After the neural network is input, the neural network is trained to update the parameters until the output converges.
Correspondingly, the invention also relates to a production quality monitoring method, which comprises the following steps:
obtaining a current multi-dimensional production parameter set X= (X) 1 ,x 2 ,……,x m ) Wherein x is i Is the ith vitamin production parameter, and m is more than or equal to 2;
production parameter x for each dimension i Expanded into two-dimensional input vectorsAnd constructs an input matrixInputting the production quality prediction model and outputting the quality index y j Is a prediction of the data of the model;
wherein the production quality prediction model is trained according to the above production quality prediction model training method, and subjective coefficientsSubjective coefficient +.>The same applies.
In the invention, when the production quality is monitored, whether the corresponding quality index of the current product has problems or not can be rapidly judged according to the current production parameters by the production quality prediction model, and the problems can be timely found and adjusted.
Further, the monitoring method further comprises the following steps:
judging the quality index y j Whether the predicted data of (a) is compliant or not, if not, according to subjective coefficientsGradually checking the production parameter x i Whether there is a problem.
When the quality index is identified to be problematic, the current production parameters are described to be problematic, and further positioning is needed, so that subjective coefficients are obtainedThe method can reflect the influence weight of each production parameter on the quality index, and firstly check the parameter with larger weight, so that the problem positioning efficiency can be improved.
In summary, the invention selects the correlation coefficient when training the modelSubjective coefficient->Up-scaling the input data, wherein the complex correlation coefficient +.>Reaction of production parameter x i Linear correlation with the multidimensional quality index set Y, subjective coefficient +.>The reaction is that the quality index y is influenced j Production parameter x among all production parameters of (2) i For quality index y j The influence weights of the model are comprehensively considered, the mutual connection between the multidimensional production parameters and the multidimensional quality indexes is comprehensively considered, the training effect is enhanced, and therefore the prediction accuracy of the model is improved. The quality of production is monitored by using the prediction model, so that whether the quality has problems or not can be rapidly and accurately identified.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. A method for training a production quality prediction model, comprising:
producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,......,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,......,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j The quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2;
determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Subjective coefficient->For multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,......,x m ) Production parameter x of (1) i For quality index y j Is a weight of influence of (1);
calculating each production parameter x in each initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
2. The method of claim 1, wherein the multi-dimensional production parameter set x= (X) 1 ,x 2 ,......,x m ) The data of the (C) is obtained by denoising, data outlier processing, missing value processing and normalization processing of actual sampling data, wherein the denoising adopts CEEMDAN denoising algorithm.
3. The method for training a production quality prediction model according to claim 1, wherein the production parameters are selected production parameters, and the method for selecting the production parameters comprises:
training an XGBoost feature importance model, taking the production parameters to be screened as model input, taking the quality index as model output, sorting according to feature importance scores, eliminating the production parameters with lower scores of 1% -2%, and taking the rest parameters as the production parameters after screening.
4. A method of training a production quality prediction model as claimed in claim 3, wherein the objective function Obj used to train the XGBoost feature importance model is:
Ω(f t )=γT+0.5λw 2
wherein ,ft (x i ) Output function of t-th tree, g i () Represents the first derivative, h i () Representing the second derivative, T representing the number of leaf nodes in the tree, w representing the weight of the leaf, gamma representing the number of leaf nodes that can be controlled, and lambda representing the fraction of leaf nodes that can be controlled.
5. The production quality prediction model training method of claim 1, wherein subjective coefficients are determined based on an analytic hierarchy processThe process of (1) comprises:
establishing a hierarchical structure model of a production process, wherein the quality index y j And multidimensional production parameter set x= (X) 1 ,x 2 ,......,x m ) Associating;
construction a= (a pq ) m×m Wherein element a pq Representing the p-th vitamin production parameter versus the q-th vitamin production parameter versus the quality index y j Importance scale of (a) according to experience setting judgment matrix each element value, a pq *a qp =1;
Based on the judgment matrix, calculating the weight of each dimension production parameter by using a root-finding method, wherein the calculation formula is as follows:
wherein ,Wi Weights for the ith vitamin production parameter;
the consistency check is carried out by calculating the consistency index C.I. to judge whether the judgment matrix set according to experience is reasonable or not, if not, the judgment matrix is adjusted until the consistency check is reasonable so as to be based on the latestWeights W of production parameters of each dimension calculated by set judgment matrix i As subjective coefficients
6. The method for training a production quality prediction model of claim 1, wherein the complex correlation coefficientsThe calculation formula of (2) is as follows:
wherein ,for production parameter x i A covariance matrix formed by covariance of each element in the multi-dimensional quality index set Y; sigma (sigma) YY Covariance matrix formed by covariance among each element of multi-dimensional quality index set Y, < ->Is the inverse of Σyy; />Is->Is a transposed matrix of (a); />For all production parameters x i Variance of the data.
7. A production quality prediction model training device, characterized by comprising:
data acquisition unit for producing parameter set x= (X) in multiple dimensions 1 ,x 2 ,......,x m ) And the corresponding multi-dimensional quality index set y= (Y) 1 ,y 2 ,......,y n ) As initial samples (X, Y), a plurality of sets of initial samples are obtained, where X i For the ith vitamin production parameter, y j The quality index of the j-th dimension is that m is more than or equal to 2, and n is more than or equal to 2;
subjective coefficient determining unit for determining production parameter x i Relative to the quality index y j Subjective coefficient of (2)Wherein subjective coefficient->For multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,......,x m ) Production parameter x of (1) i For quality index y j Is a weight of influence of (1);
a complex correlation coefficient calculation unit for calculating each production parameter x in the initial sample data i Complex correlation coefficient with corresponding multi-dimensional quality index set Y
8. A method of manufacturing quality monitoring comprising:
obtaining a current multi-dimensional production parameter set X= (X) 1 ,x 2 ,......,x m ) Wherein x is i Production of ginseng for the ith dimensionThe number, m is more than or equal to 2;
production parameter x for each dimension i Expanded into two-dimensional input vectorsAnd constructs an input matrixInputting the production quality prediction model and outputting the quality index y j Is a prediction of the data of the model; />
Wherein the subjective coefficientFor multi-dimensional production parameter set x= (X) determined based on analytic hierarchy process 1 ,x 2 ,......,x m ) Production parameter x of (1) i For quality index y j The production quality prediction model is trained according to the production quality prediction model training method of any one of claims 1 to 6.
10. The method for monitoring the production quality according to claim 8, wherein the predicted quality index data of the cable is produced by a three-layer co-extrusion process of the cable, wherein the multi-dimensional production parameters comprise temperature, pressure and linear velocity, and the quality index is any one of thickness of an inner screen, thickness of an outer screen and insulation eccentricity.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310131807.8A CN116088453A (en) | 2023-02-17 | 2023-02-17 | Production quality prediction model training method and device and production quality monitoring method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310131807.8A CN116088453A (en) | 2023-02-17 | 2023-02-17 | Production quality prediction model training method and device and production quality monitoring method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116088453A true CN116088453A (en) | 2023-05-09 |
Family
ID=86210178
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310131807.8A Pending CN116088453A (en) | 2023-02-17 | 2023-02-17 | Production quality prediction model training method and device and production quality monitoring method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116088453A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116976917A (en) * | 2023-07-31 | 2023-10-31 | 金景(海南)科技发展有限公司 | Agricultural brand certificate system construction method based on blockchain technology |
CN117350517A (en) * | 2023-12-04 | 2024-01-05 | 山东德瑞高分子材料股份有限公司 | Control method, system, equipment and storage medium for chemical production flow |
-
2023
- 2023-02-17 CN CN202310131807.8A patent/CN116088453A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116976917A (en) * | 2023-07-31 | 2023-10-31 | 金景(海南)科技发展有限公司 | Agricultural brand certificate system construction method based on blockchain technology |
CN116976917B (en) * | 2023-07-31 | 2024-05-24 | 金景(海南)科技发展有限公司 | Agricultural brand certificate system construction method based on blockchain technology |
CN117350517A (en) * | 2023-12-04 | 2024-01-05 | 山东德瑞高分子材料股份有限公司 | Control method, system, equipment and storage medium for chemical production flow |
CN117350517B (en) * | 2023-12-04 | 2024-03-29 | 山东德瑞高分子材料股份有限公司 | Control method, system, equipment and storage medium for chemical production flow |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109002686B (en) | Multi-grade chemical process soft measurement modeling method capable of automatically generating samples | |
CN116088453A (en) | Production quality prediction model training method and device and production quality monitoring method | |
CN108764520B (en) | Water quality parameter prediction method based on multilayer cyclic neural network and D-S evidence theory | |
CN110555230B (en) | Rotary machine residual life prediction method based on integrated GMDH framework | |
CN110598299B (en) | Surface integrity evaluation method based on sensor fusion and deep learning | |
CN110561191B (en) | Numerical control machine tool cutter abrasion data processing method based on PCA and self-encoder | |
CN110309609B (en) | Building indoor air quality evaluation method based on rough set and WNN | |
CN111813084A (en) | Mechanical equipment fault diagnosis method based on deep learning | |
CN113822499B (en) | Train spare part loss prediction method based on model fusion | |
CN110569566A (en) | Method for predicting mechanical property of plate strip | |
CN115982141A (en) | Characteristic optimization method for time series data prediction | |
CN115221973A (en) | Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning | |
CN115096357A (en) | Indoor environment quality prediction method based on CEEMDAN-PCA-LSTM | |
CN116502455A (en) | Process parameter determination method and system for laser selective melting technology | |
CN114692507A (en) | Counting data soft measurement modeling method based on stacking Poisson self-encoder network | |
CN117076887A (en) | Pump station unit running state prediction and health assessment method and system | |
CN112966449B (en) | Manufacturing process quality defect root cause diagnosis method based on variable selection and probabilistic neural network | |
CN108446506B (en) | Uncertain system modeling method based on interval feedback neural network | |
CN115828744A (en) | White light LED fault on-line diagnosis and service life prediction method | |
CN115018162A (en) | Method and system for predicting machining quality in industrial finish machining process in real time | |
CN111062118B (en) | Multilayer soft measurement modeling system and method based on neural network prediction layering | |
CN114462873A (en) | Power distribution network reliability assessment and influence factor analysis method thereof | |
CN113361209A (en) | Quantitative analysis method for magnetic anomaly of surface defects of high-temperature alloy | |
CN116595883B (en) | Real-time online system state correction method for numerical reactor | |
CN115062431B (en) | CS-Elman neural network model-based hot rolled plate convexity prediction method |
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 |