CN106054840A - Whole process product quality online control system - Google Patents

Abstract

The invention provides a whole process product quality online control system, which can achieve the on-line dynamic optimization of the product quality of procedures of the whole process, and thus improving the stability of the product quality. The system includes a specification making module, a process control system of each procedure, and a controller, the process control system includes a CPS-based dynamic product quality control module, the specification making module sets the quality indexes of the whole process and technological standards, the dynamic product quality control module acquires the product quality information of a previous procedure and the technological parameter setting value in the previous procedure, performs dynamic quality control and optimization of the obtained product quality information and the technological parameter setting value according to the set quality index and process specification of the whole process, and sends the optimized technological parameters to the controller, and the controller executes the optimized technological parameters. The whole process product quality online control system is applicable to the technical field of automatic control.

Description

Online management and control system for quality of full-flow product

Technical Field

The invention relates to the technical field of automatic control, in particular to a full-process product quality online control system.

Background

For the steel, petrochemical and other flow-type industries, a plurality of continuous and coupled working procedures are involved in the manufacturing process of products, and each working procedure requires that the set values and quality indexes of process parameters are controlled within a determined range to ensure the final quality of finished products. In the prior art, a centralized manufacturing mode can be adopted for products with relatively single variety and large batch, namely, according to production experience, each procedure is accurately controlled through preset process parameters to ensure the final quality of finished products.

However, with the arrival of the industrial 4.0 era, the requirements of customers on the quality of steel products are stricter and more obvious, and the conflict between large-scale production and customized manufacturing becomes more obvious. In the prior art, a centralized manufacturing mode and a quality control method adopted by iron and steel enterprises are difficult to adapt to instantaneously changing market requirements and stability requirements on product quality.

Disclosure of Invention

The invention aims to provide a full-process product quality online control system to solve the problem that the stability requirement of product quality cannot be met in the prior art.

In order to solve the above technical problem, an embodiment of the present invention provides an online control system for quality of a full-process product, including: a standard formulation module, a process control system of each procedure and a controller; wherein the process control system comprises: a CPS-based dynamic product quality control module;

the specification formulation module: the method is used for setting the quality index and the process specification of the whole process;

the dynamic product quality control module is used for acquiring product quality information of a previous procedure and a process parameter set value in the previous procedure, performing dynamic quality control and optimization on the acquired product quality information and the process parameter set value according to a set quality index and a set process specification of a whole flow, and sending an optimized process parameter to the controller;

and the controller is used for executing the optimized process parameters.

Further, the specification formulation module: the CPS-based dynamic product quality control module is specifically used for setting quality indexes and process specifications of the whole process and transmitting the set quality indexes and process specifications to the CPS-based dynamic product quality control module in the process control system of each process.

Further, the dynamic product quality management and control module comprises: the system comprises a quality online judgment submodule based on big data analysis, a quality abnormity diagnosis submodule based on contribution rate analysis and a process parameter optimization submodule based on an evolution direction algorithm of a local low-dimensional main manifold of adjacent points.

Further, the quality online judgment submodule is used for judging whether the product quality of the previous process or the subsequent process exceeds a preset product quality controllable range due to the set values of the process parameters in the previous process, wherein the product quality controllable range is determined by the set quality indexes and process specifications of the whole process, historical product quality information and corresponding historical process parameters;

if the product quality exceeds the preset controllable range of product quality, outputting a quality judgment or waste judgment result;

otherwise, judging whether the distance from the acquired process parameter set value in the previous procedure to the preset spherical center of the hyper-sphere is smaller than the radius R of the hyper-sphere;

and if the radius is not less than the radius R of the hyper-sphere, judging that the quality of the product deviates from a quality controllable area to generate quality deviation, wherein R represents a control limit, and the quality controllable area is an area in the hyper-sphere.

Further, a radius R of the hypersphere is determined by a nonlinear method, and the radius R of the hypersphere is expressed as:

$R^2=\left|\right|\mathrm{\Φ}(x_k-o)|{|}^2=k(x_k,x_k)-2\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\α}}_{i}k(x_k,x_i)+\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}k(x_i,x_j)$

wherein phi (x)_k) Representing points on the hypersphere of a hypersphere, k (.) representing a gaussian kernel function, o representing the sphere center of the hypersphere, n representing the number of sample points in a preset historical training set, α_i、α_jRespectively representing lagrange multipliers; x is the number of_kRepresenting the kth sample point in the historical training set within the controllable range of the product quality; x is the number of_i,x_jRespectively representing the ith sample point and the jth sample point in the historical training set.

Further, the distance D from the acquired process parameter set value in the previous process to the preset spherical center of the hyper-sphere²(x_new) Expressed as:

$D^2\left(x_{new}\right)=\left|\right|\mathrm{\Φ}\left(x_{new}\right)-o|{|}^2=1-2\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\α}}_{i}k(x_{new},x_i)+\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}k(x_i,x_j)$

wherein x is_newRepresenting the acquired set values of the process parameters in the previous process; k (.) represents a Gaussian kernel function; o represents the center of the hyper-sphere; phi (x)_new) Representing a process parameter set point x_newMapping to a point in a high dimensional feature space by non-linear mapping α_i、α_jRespectively representing lagrange multipliers; n represents the number of sample points in a preset historical training set; x is the number of_i,x_jRespectively representing the ith sample point and the jth sample point in the historical training set.

Further, the hyper-sphere center o is represented as:

wherein n represents the number of sample points in the preset historical training set α_iRepresenting a lagrange multiplier;representing points located within the hypersphere, x_iRepresenting the ith sample point in the historical training set.

Further, the quality anomaly diagnosis submodule is configured to, when it is determined that a set value of a process parameter in a previous process may cause a deviation of product quality from a quality-controllable area to cause a quality deviation, obtain a contribution value of each process parameter in the previous process to the quality deviation, and obtain a process parameter corresponding to a larger contribution value as a process parameter that causes the deviation of product quality from the quality-controllable area;

wherein the larger contribution value is a contribution value exceeding a predetermined threshold.

Further, the process parameter optimization submodule is configured to analyze the acquired process parameters causing the product quality to deviate from the quality controllable region based on an evolution direction algorithm of the local low-dimensional main manifold at the neighboring point, and determine the regulation and control directions and actual regulation amounts of the process parameters in the previous process in the current process and the subsequent process.

Further, the product quality acquisition module is used for acquiring online and offline product quality information of a previous process, and transmitting the acquired product quality information to the CPS-based dynamic product quality control module in the process control system of each process;

and the process parameter acquisition module is used for acquiring process parameter set values in the previous working procedures and transmitting the acquired process parameter set values to the CPS-based dynamic product quality control module in the process control system of each working procedure.

The technical scheme of the invention has the following beneficial effects:

in the scheme, the quality index and the process specification of the whole process are set through a specification making module; the method comprises the steps that product quality information of a previous procedure and a process parameter set value in the previous procedure are obtained through a dynamic product quality control module, dynamic quality control and optimization are conducted on the obtained product quality information and the obtained process parameter set value according to a set quality index and a set process specification of a whole flow, and optimized process parameters are sent to a controller; and finally, executing the optimized process parameters through a controller. In this way, the CPS-based dynamic product quality control module is embedded into the process control system of each process, the dynamic product quality control module is used for dynamically monitoring the process parameters and the product quality in the previous process, and once a potential product quality problem occurs, the optimized process parameters can be provided in time so as to execute the optimized process parameters in the process and the subsequent processes, thereby ensuring the stability of the product quality.

Drawings

Fig. 1 is a schematic structural diagram of a full-process product quality online control system according to an embodiment of the present invention;

FIG. 2 is a service logic diagram of a CPS-based full-process product quality online control system according to an embodiment of the present invention;

fig. 3 is a schematic service flow diagram of a CPS-based dynamic product quality management and control module according to an embodiment of the present invention;

fig. 4 is a schematic view of a partial manifold formed by neighboring points according to an embodiment of the present invention.

Detailed Description

In order to make the technical problems, technical solutions and advantages of the present invention more apparent, the following detailed description is given with reference to the accompanying drawings and specific embodiments.

The invention provides a full-process product quality online control system aiming at the problem that the existing product quality stability requirement cannot be met.

Example one

As shown in fig. 1, the system for online control of quality of a full-flow product provided in the embodiment of the present invention includes: a specification setting module 11, a Process Control System 12 (PCS) of each Process, and a controller 13; wherein the process control system 12 comprises: a dynamic product quality control module 121 based on Cyber Physical Systems (CPS);

the specification formulation module 11: the method is used for setting the quality index and the process specification of the whole process;

the dynamic product quality control module 121 is configured to obtain product quality information of a previous process and a process parameter set value in the previous process, perform dynamic quality control and optimization on the obtained product quality information and the obtained process parameter set value according to a set quality index and a set process specification of a full process, and send an optimized process parameter to the controller 13;

and the controller 13 is used for executing the optimized process parameters.

According to the full-process product quality online control system disclosed by the embodiment of the invention, the quality index and the process specification of the full process are set through the specification setting module; the method comprises the steps that product quality information of a previous procedure and a process parameter set value in the previous procedure are obtained through a dynamic product quality control module, dynamic quality control and optimization are conducted on the obtained product quality information and the obtained process parameter set value according to a set quality index and a set process specification of a whole flow, and optimized process parameters are sent to a controller; and finally, executing the optimized process parameters through a controller. In this way, the CPS-based dynamic product quality control module is embedded into the process control system of each process, the dynamic product quality control module is used for dynamically monitoring the process parameters and the product quality in the previous process, and once a potential product quality problem occurs, the optimized process parameters can be provided in time so as to execute the optimized process parameters in the process and the subsequent processes, thereby ensuring the stability of the product quality.

In the embodiment of the present invention, the preceding steps refer to all the steps before the present step.

In an embodiment of the present invention, for example, in the steel manufacturing industry, the full-process product quality online control system may include: a specification formulation module 11 of quality indexes and process specifications of the whole process, a process control system 12 of each procedure of a dynamic product quality control module 121 embedded with a CPS, and a controller 13 for executing optimized process parameters; wherein, the specification formulation module 11 may include: the ERP and MES system is used for formulating the quality index and the process specification of the whole process; the controller 13 is a PLC controller.

In a specific embodiment of the above full-process product quality online control system, the system further includes:

a product quality acquisition module, configured to acquire online and offline product quality information of a previous process, and transmit the acquired product quality information to the CPS-based dynamic product quality control module 121 in the process control system 12 of each process;

and the process parameter acquisition module is used for acquiring process parameter set values in the previous working procedures and transmitting the acquired process parameter set values to the CPS-based dynamic product quality control module 121 in the process control system 12 of each working procedure.

In the embodiment of the present invention, the collected product quality information and the process parameter setting value may be transmitted to a CPS-based dynamic product quality control module 121 embedded in the process control system 12 of each process through a real-time communication network, and the dynamic product quality control module 121 of each process dynamically controls and optimizes the process parameters and the product quality information of each process, and sends the optimized control parameters to the PLC controller. Thus, the CPS-based dynamic product quality control module 121 can enhance the dynamic monitoring, diagnosis and optimization capabilities of each process parameter and product quality information.

In the embodiment of the invention, the product quality information is collected for the online optimization of the process parameters, and once the process parameters of the previous process have deviation or part of the product quality deviates from the set quality index, the process parameters need to be dynamically adjusted in the process and the subsequent process so as to correct the quality deviation caused by the previous process.

As shown in fig. 2, in the embodiment of the present invention, the product quality information may include, but is not limited to: the method comprises the following steps of (1) quality information such as component data, casting blank quality, dimensional accuracy, material performance and the like, wherein process parameters of each process can include but are not limited to: technological parameters in the steel-making process, technological parameters in the continuous casting process, technological parameters in the steel rolling process, technological parameters in the heat treatment process and the like.

In a specific embodiment of the above full-process product quality online control system, further, the specification formulation module 11: specifically, the method is used for setting the quality index and the process specification of the whole process, and transmitting the set quality index and the set process specification to the CPS-based dynamic product quality control module 121 in the process control system 12 of each process.

In a specific embodiment of the foregoing full-process product quality online control system, further, the dynamic product quality control module 121 includes: the system comprises a quality online judgment submodule based on big data analysis, a quality abnormity diagnosis submodule based on contribution rate analysis and a process parameter optimization submodule based on an evolution direction algorithm of a local low-dimensional main manifold of adjacent points.

In a specific embodiment of the above full-process product quality online control system, further, the quality online determination submodule is configured to determine whether a set value of a process parameter in a previous process may cause product quality of the current process or a subsequent process to exceed a preset product quality controllable range, where the product quality controllable range is determined by a set quality index and a set process specification of the full process, historical product quality information, and a historical process parameter corresponding to the historical product quality information;

if the product quality exceeds the preset controllable range of product quality, outputting a quality judgment or waste judgment result;

otherwise, judging whether the distance from the acquired process parameter set value in the previous procedure to the preset spherical center of the hyper-sphere is smaller than the radius R of the hyper-sphere;

and if the radius is not less than the radius R of the hyper-sphere, judging that the quality of the product deviates from a quality controllable area to generate quality deviation, wherein R represents a control limit, and the quality controllable area is an area in the hyper-sphere.

In the embodiment of the invention, the quality online judgment submodule firstly simply judges the set value of the process parameter in the previous process according to the set quality index and the set process specification of the whole process, and if the set value of the process parameter in the previous process can cause the product quality of the process or the subsequent process to exceed the preset controllable range of the product quality, a quality judgment or waste judgment result is given, so that the waste of energy, cost and time caused by the continuous manufacturing of the process or the subsequent process is avoided. If the process parameter set value of the previous process does not cause the product quality of the current process or the subsequent process to exceed the preset controllable range of the product quality, the quality on-line judgment sub-module further analyzes the process parameter set value to ensure that the final quality of the product of the current process and the subsequent process is in the controllable range;

in the embodiment of the invention, the product quality within a certain range of each process parameter can be determined to be controllable by performing big data analysis on a large amount of historical product quality information and corresponding historical process parameters according to the set quality index and process specification of the whole process. In the embodiment of the present invention, the step of further analyzing the set values of the process parameters by the quality online determination submodule includes:

obtaining a preset historical training set { x_i}，x_i∈R^dI is 1,2 … n, where x_iRepresents the ith sample point in the historical training set, each x_iVarious process parameters may be included, such as, for example, continuous annealing heating temperature,continuous annealing soaking temperature, continuous annealing fast cooling outlet temperature, continuous annealing aging outlet temperature, continuous annealing slow cooling outlet temperature, C content, Mn content, P content, S content, hot rolling heating furnace temperature, finish rolling inlet temperature, finish rolling outlet temperature, coiling temperature and other technological parameters; n represents the number of sample points in the historical training set, and is mapped by non-linearityMapping sample points in a historical training set into a high-dimensional feature space to expect in a high-dimensional F^sSeeking a hypersphere, using as small a hypersphere F (R, o, ξ) as possible_i) To contain as many sample points as possible, and therefore, this problem can be described as:

$\min F(R,o,{\mathrm{\ξ}}_i)=R^2+\frac{1}{n\mathrm{\υ}}\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\ξ}}_i$

wherein R is^dRepresenting a d-dimensional space before mapping; f^sRepresenting the mapped s-dimensional space;indicating a point located within the hyper-sphere ξ_iMore than or equal to 0, i is 1,2 …, n; r is a hyper-sphereRadius, o being the centre of the hyper-sphere, which method allows a small number of sample points outside the hyper-sphere, for which a relaxation variable ξ is introduced_iPunishment is carried out on the error of the ith sample point, the parameter upsilon is a constraint condition on the size of the hyper-sphere and the error rate of all samples, and a Lagrangian multiplier α (α)_i、α_j) Converting formula (1) into a dual problem:

$max\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\α}}_{i}k(x_i,x_i)+\underset{i=1}{\overset{n}{\Σ}}\underset{j=1}{\overset{n}{\Σ}}{\mathrm{\α}}_i{\mathrm{\α}}_{j}k(x_i,x_j)$

$s.t.\underset{i=1}{\overset{n}{\Σ}}{\mathrm{\α}}_i=1,0\≤{\mathrm{\α}}_i\≤\frac{1}{nv}---\left(2\right)$

wherein, k (x)_i,x_j) Usually, a gaussian kernel function is selected, but other functions are also possible, and are not limited herein; x is the number of_i,x_jRespectively representing the ith sample point and the jth sample point in the historical training set; wherein, k (x)_i,x_j) Can be expressed as:

$k(x_i,x_j)=e^{\frac{-\left|\right|x_i-x_j|{|}^2}{{\mathrm{\σ}}^2}}---\left(3\right)$

solving equation (2) can obtain lagrange multiplier α_iα is mainly used as the value of (A)_i0 indicating that the sample spot is located within the hypersphere α_i1/n.v., meansThe spotting sample point is located outside the hypersphere; 0<α_i<1/nv, then represents the sample point on the hypersphere, typically α_iThe sample points corresponding to No. 0 are collectively referred to as support vectors, i.e., points located outside the hypersphere and on the hypersphere surface are support vectors.

The hypersphere has two key parameters: the center o and radius R are as in formula (4)

$\begin{array}{c}{R}^2=\left|\right|\mathrm{\&phi;}(x_k-o)|{|}^2\\ =(x_k,x_k)-2{\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&alpha;}}_{1}k(x_k,x_i)+{\&Sigma;}_{i=1}^n{\&Sigma;}_{j=1}^n{\mathrm{\&alpha;}}_i{\mathrm{\&alpha;}}_{j}k(x_i,x_j)\end{array}---\left(4\right)$

Wherein phi (x)_i) Representing an arbitrary point, x, within the hypersphere_kRepresents the kth sample point, phi (x), in the historical training set within the controllable range of the product quality_k) Representing points lying on a hypersphere.

Typically, the radius R of the hyper-sphere will be determined during monitoring²The sample point can be used as a control limit to carry out real-time monitoring, the quality deviation of the product quality can not be caused when the sample point is positioned in the hyper-sphere, and the sample point is also usedThat is, the region within the hypersphere is the region of controllable quality.

When testing a new sample point x to be tested_newThen, a new sample point x to be measured can be calculated_newDistance to the center of the hyper-sphere:

$\begin{array}{c}{D}^2\left(x_{new}\right)=\left|\right|\mathrm{\&Phi;}\left(x_{new}\right)-o|{|}^2\\ =1-2\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_{i}k(x_{new},x_i)+\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{\mathrm{\&alpha;}}_{j}k(x_i,x_j)\end{array}---\left(5\right)$

wherein phi (x)_new) Representing a process parameter set point x_newA point in the high dimensional feature space is mapped by a non-linear mapping.

As shown in FIG. 3, if D²(x_new)<R²The new sample point to be tested is in the quality controllable area, namely, the process parameters in the new sample point to be tested are in a normal state; conversely, the process parameters may cause quality deviations.

Examples of the invention，x_newVarious process parameters can be included, for example, continuous annealing heating temperature, continuous annealing soaking temperature, continuous annealing fast cooling outlet temperature, continuous annealing aging outlet temperature, continuous annealing slow cooling outlet temperature, C content, Mn content, P content, S content, hot rolling heating furnace temperature, finish rolling inlet temperature, finish rolling outlet temperature, coiling temperature and the like.

In the embodiment of the invention, the boundary determining method of the hypersphere support vector can be used for processing the single complex boundary problem under the conditions of high dimension, strong coupling and nonlinearity. The method converts the boundary problem of the hyper-sphere into a radius R from the sphere center to the boundary through nonlinear Gaussian kernel mapping. And if the distance from the sample point to be detected to the sphere center is greater than or equal to the radius R, judging that the quality of the product is possible to have quality deviation.

In the embodiment of the invention, in practical industrial application, the R value can be properly adjusted according to the requirement of a client on the product quality. For products with more strict quality requirements, the reliability of the product quality can be improved by reducing the R value; the R value can be properly increased for the steel grade with loose requirements on the product quality so as to reduce the production cost.

In the embodiment of the present invention, the quality online determination sub-module mainly adopts a big data analysis technology, and also can embed a mechanism module or an analysis module into the dynamic product quality control module 121 to implement dynamic optimization of product quality. In the quality online judgment submodule based on big data analysis, the establishment of a historical training set plays a decisive role, and the perfection of historical sample points is a continuous verification and continuous optimization process, as shown in fig. 3.

In a specific embodiment of the above full-process product quality online control system, further, the quality anomaly diagnosis sub-module is configured to, when it is determined that a set value of a process parameter in a previous process may cause a deviation of product quality from a quality-controllable area and a quality deviation occurs, obtain a contribution value of each process parameter in the previous process to the deviation of product quality, and obtain a process parameter corresponding to a larger contribution value as a process parameter that causes the deviation of product quality from the quality-controllable area;

wherein the larger contribution value is a contribution value exceeding a predetermined threshold.

In the embodiment of the present invention, as shown in fig. 3, when it is found that the set values of the process parameters in the previous process may cause the quality deviation, the quality abnormality diagnosis sub-module can be used to timely and accurately analyze which processes and which process parameters are the causes of the quality deviation, and find out the causes of the quality deviation, which is beneficial to timely adjust the process parameters of the present process and the subsequent processes, and correct the quality deviation, thereby improving the stability of the product quality. Specifically, the main process parameters causing the deviation of the product quality from the quality controllable region can be searched from the process parameters in the set previous process, i.e. each process parameter in the previous process is calculated, and D in the formula (5)²The process parameters with large contribution values are the main reasons for causing the product quality to deviate from the quality controllable area. As can be seen from equation (5), the third part in the equation is:

$\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{\mathrm{\&alpha;}}_{j}k(x_i,x_j)$

the term is a constant, determined by the samples in the historical training set, once the training samples are determined. The influencing formula (5) is mainly the second part

$\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_{i}k(x_{new},x_j)=\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{e}^{\frac{-\left|\right|x_{new}-x_i|{|}^2}{{\mathrm{\&sigma;}}^2}}---\left(6\right)$

From the formula (6), if | | | x_new-x_i||²The larger the value of the formula (6), the smaller D²(x_new) The larger. From this, the sample point x to be measured in the equation (5)^newThe amount of change in distance to the center of the sphere depends primarily on

$\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i\left|\right|x_{new}-x_i|{|}^2=\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i|\left|\widehat{x_i}\right|{|}^2---\left(7\right)$

Wherein,denotes x_new-x_iThe formula (7) can be further decomposed into

$\begin{array}{c}\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i\underset{j=1}{\overset{p}{\&Sigma;}}{(x_j^{new}-x_{ij})}^2\\ =\underset{j=1}{\overset{p}{\&Sigma;}}\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{(x_j^{new}-x_{ij})}^2\end{array}---\left(8\right)$

Wherein p represents x^newThe number of medium variables, each variable may represent a process parameter;

sample point x to be measured^newThe contribution of the jth variable to the quality deviation may be defined as:

${\mathrm{contrx}}_j^{new}=\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{(x_j^{new}-x_{ij})}^2---\left(9\right)$

in order to eliminate the influence of the variable dimension on the contribution value, the above formula needs to be standardized; due to sample point correspondences with only a few support vectorsOnly the sample points of these support vectors need to be calculatedWherein,representing the jth variable in the ith sample point of the support vector in the preset historical training set, and the dimensionless parameter x to be measured^newThe contribution of the j-th variable to the deviation is expressed as:

${\mathrm{contrx}}_j^{new}=\underset{i=1}{\overset{1}{\&Sigma;}}{\mathrm{\&alpha;}}_i^{*}{(x_j^{new}-x_{ij}^{*})}^2/S_j---\left(10\right)$

wherein S is_jExpressing the variance of the jth variable in the preset historical training set, wherein l is the number of support vectors in the preset historical training set; from p to pThe largest of the several variables were selected, indicating that these variables are the main process parameters responsible for the quality deviation.

In the embodiment of the invention, the contribution value corresponding to the selected variable needs to exceed the preset threshold value.

In a specific implementation manner of the above-mentioned full-process product quality online control system, further, the process parameter optimization submodule is configured to analyze the acquired process parameters that cause the product quality to deviate from the quality controllable region based on an evolution direction algorithm of the local low-dimensional main manifold at the neighboring point, and determine a control direction and an actual adjustment amount of the process parameters in the previous process in the current process and the subsequent process.

In the embodiment of the invention, after finding out which variables (process parameters) cause quality deviation, the process parameters need to be optimized on line, and in addition, the process parameters of the process and the subsequent processes need to be dynamically adjusted, so that the production process can return to a controlled state in time. However, because of the multiple strong coupling and non-linear factors existing between the product quality and the process parameters, the uncertainty of the product quality is caused by simply adjusting the process parameters causing the quality deviation, and therefore, the process parameter optimization algorithm under the conditions of multivariable, strong coupling and non-linearity needs to be researched. Specifically, the method comprises the following steps:

let x_i＝(x₁,x₂…x_m) The subset of the process parameter x variables that have been set for the preceding process is also the main process parameter that causes the quality of the product to deviate from the quality controlled zone, where m<p is the same as the formula (I). According to x_iM variable subsets in the defined previous process, and k variables and x variables are searched from the historical training set_iThe m variable subsets of (a) neighboring normal sample points { y₁,y₂,…y_kIn which y_j＝(y_j1,y_j2,…y_jm,…y_jp) J ═ 1,2 … k, as shown in fig. 4. In FIG. 4, the neighboring points in the ellipsoid region are based on the point x to be optimized_iAt x₁,x₂A set of sample points that are adjacent to the two variables and selected from the historical training set.

From samples of adjacent points in the region y₁,y₂,…y_kAnd, forming a matrix B, and calculating a covariance matrix C:

C＝B^TB (11)

decomposing the eigenvalue of the covariance matrix C, and respectively obtaining the eigenvalue lambda₁，λ₂，…λ_pAnd corresponding feature vector U₁，U₂，…U_p. U consisting of eigenvectors corresponding to a number of largest eigenvalues (eigenvalues larger than a predetermined value) of the covariance matrix C represents the principal component of the sample set consisting of neighboring points in the local region. Due to the proximity matrix y₁,y₂,…y_kIs at x_iThe m variables in the previous process set in (1) are used as the search basis of the neighboring points, and the neighboring points are all in the quality controllable area, so that the following can be deduced: the principal component, which is formed by the neighborhood of the local area, represents the orientation of the manifold (i.e., the main direction of the process parameter to be adjusted) within the local area.

Since the eigenvectors are orthogonal to each other, the principal component constitutes a tangent space of the local region, which means that a low-dimensional subspace is embedded in the main manifold. X is to be_iThe centroid to matrix B is defined as vector V, whose projection onto tangent space is a local tangent vector, represented as:

T＝VU (12)

the vector T represents the direction of evolution of the local low-dimensional master manifold. Since the vector T is only linearized in a local region, the evolution matrix T can still describe the main direction of evolution of the local region of the non-linear manifold.

In the actual industrial production, besides the process parameter adjusting and controlling direction needs to be mastered, the actual adjusting quantity of each process parameter should be determined,

the adjustment of the process parameter x can be expressed as:

Δx＝TU^T(13)

wherein, Δ x represents the adjustment amount of the process and the subsequent process parameters when x causes the product quality to deviate from the quality controllable region.

In the embodiment of the invention, the abnormal point diagnosis algorithm based on the contribution rate and the evolution direction algorithm based on the local low-dimensional main manifold of the adjacent point can realize online monitoring, diagnosis and optimization of the production process, and the process parameters of the process and the subsequent processes are adjusted in time according to the product quality deviation condition of each process in the production process, thereby improving the stability of the product quality and avoiding the occurrence of batch quality judgment waste.

In summary, the CPS-based dynamic product quality control module 121 is embedded into the process control system 12 of each process, so that the system is changed from the basic mode of centralized control to the basic mode of distributed enhanced control, thereby meeting the requirements of customized manufacturing of products and improvement of product quality stability; specifically, the dynamic product quality control module 121 dynamically monitors the process parameters and product quality information in the previous process, and once a potential product quality problem occurs, the reason causing the quality problem can be diagnosed in time, and optimized process parameters are provided for quality accurate control, so that the optimized process parameters are executed in the current process and the subsequent processes, and the stability of the product quality is ensured.

While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims (10)

1. The utility model provides an online management and control system of full flow product quality which characterized in that includes: a standard formulation module, a process control system of each procedure and a controller; wherein the process control system comprises: a CPS-based dynamic product quality control module;

the specification formulation module: the method is used for setting the quality index and the process specification of the whole process;

the dynamic product quality control module is used for acquiring product quality information of a previous procedure and a process parameter set value in the previous procedure, performing dynamic quality control and optimization on the acquired product quality information and the process parameter set value according to a set quality index and a set process specification of a whole flow, and sending an optimized process parameter to the controller;

and the controller is used for executing the optimized process parameters.

2. The on-line management and control system for the quality of the full-flow product according to claim 1, wherein the specification formulation module: the CPS-based dynamic product quality control module is specifically used for setting quality indexes and process specifications of the whole process and transmitting the set quality indexes and process specifications to the CPS-based dynamic product quality control module in the process control system of each process.

3. The on-line management and control system for the quality of the whole flow product according to claim 1, wherein the dynamic product quality management and control module comprises: the system comprises a quality online judgment submodule based on big data analysis, a quality abnormity diagnosis submodule based on contribution rate analysis and a process parameter optimization submodule based on an evolution direction algorithm of a local low-dimensional main manifold of adjacent points.

4. The full-process product quality online control system according to claim 3, wherein the quality online determination submodule is configured to determine whether a set value of a process parameter in a previous process may cause product quality of the current process or a subsequent process to exceed a preset product quality controllable range, where the product quality controllable range is determined by a set quality index and a set process specification of the full process, historical product quality information, and historical process parameters corresponding to the historical product quality information;

if the product quality exceeds the preset controllable range of product quality, outputting a quality judgment or waste judgment result;

otherwise, judging whether the distance from the acquired process parameter set value in the previous procedure to the preset spherical center of the hyper-sphere is smaller than the radius R of the hyper-sphere;

and if the radius is not less than the radius R of the hyper-sphere, judging that the quality of the product deviates from a quality controllable area to generate quality deviation, wherein R represents a control limit, and the quality controllable area is an area in the hyper-sphere.

5. The full-flow online product quality control system according to claim 4, wherein a radius R of the hyper-sphere is determined by a non-linear method, and the radius R of the hyper-sphere is expressed as:

$R^2=\left|\right|\mathrm{\&phi;}(x_k-o)|{|}^2=k(x_k,x_k)-2\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_{i}k(x_k,x_i)+\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{\mathrm{\&alpha;}}_{j}k(x_i,x_j)$

wherein phi (x)_k) Representing points on the hypersphere of a hypersphere, k (.) representing a gaussian kernel function, o representing the sphere center of the hypersphere, n representing the number of sample points in a preset historical training set, α_i、α_jRespectively representing lagrange multipliers; x is the number of_kRepresenting the kth sample point in the historical training set within the controllable range of the product quality; x is the number of_i,x_jRespectively representing the ith sample point and the jth sample point in the historical training set.

6. The system according to claim 4, wherein the distance D from the process parameter setting value in the previous step to the center of the predetermined hyper-sphere is obtained²(x_new) Expressed as:

$D^2\left(x_{new}\right)=\left|\right|\mathrm{\&phi;}\left(x_{new}\right)-o|{|}^2=1-2\underset{i=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_{i}k(x_{new},x_i)+\underset{i=1}{\overset{n}{\&Sigma;}}\underset{j=1}{\overset{n}{\&Sigma;}}{\mathrm{\&alpha;}}_i{\mathrm{\&alpha;}}_{j}k(x_i,x_j)$

wherein x is_newRepresenting the acquired set values of the process parameters in the previous process; k (.) represents a Gaussian kernel function; o represents the center of the hyper-sphere; phi (x)_new) Representing a process parameter set point x_newMapping to a point in a high dimensional feature space by non-linear mapping α_i、α_jRespectively representing lagrange multipliers; n represents the number of sample points in a preset historical training set; x is the number of_i,x_jRespectively representing the ith sample point and the jth sample point in the historical training set.

7. The on-line whole-process product quality control system according to claim 6, wherein the spherical center o of the hyper-sphere is represented as:

wherein n represents the number of sample points in the preset historical training set α_iRepresenting a lagrange multiplier;representing points located within the hypersphere, x_iRepresenting the ith sample point in the historical training set.

8. The full-flow product quality online control system according to claim 4, wherein the quality anomaly diagnosis sub-module is configured to, when it is determined that a set value of a process parameter in a previous process may cause a deviation of product quality from a quality-controllable area to cause a quality deviation, obtain a contribution value of each process parameter in the previous process to the quality deviation, and obtain a process parameter corresponding to a larger contribution value as a process parameter that causes the deviation of product quality from the quality-controllable area;

wherein the larger contribution value is a contribution value exceeding a predetermined threshold.

9. The full-flow product quality online control system according to claim 8, wherein the process parameter optimization submodule is configured to analyze the acquired process parameters that cause the product quality to deviate from the quality controllable region based on an evolution direction algorithm of a local low-dimensional main manifold at a neighboring point, and determine a control direction and an actual adjustment amount of the process parameters in a previous process in the current process and a subsequent process.

10. The full-flow online product quality control system according to claim 1, wherein the product quality acquisition module is configured to acquire online and offline product quality information of a previous process, and transmit the acquired product quality information to the CPS-based dynamic product quality control module in the process control system of each process;

and the process parameter acquisition module is used for acquiring process parameter set values in the previous working procedures and transmitting the acquired process parameter set values to the CPS-based dynamic product quality control module in the process control system of each working procedure.