JP2010115692A - Device and method for predicting quality in manufacturing process, program and computer-readable recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform quality prediction with high accuracy in a manufacturing process which has characteristic which is non-linear and fluctuated timewisely and in which the ease of understanding of a relational formula is required. <P>SOLUTION: The whole operation range of the manufacturing process is divided into a plurality of local regions, a local relation formula is prepared in each local region and a nonlinear characteristic is predicted with high accuracy in order to construct the whole relational formula by their combination. The relationship between operation variables and the quality variables is readable from the constitution of the whole relation formula and the value of coefficient of each local relational formula by humans. Furthermore, when the predicting accuracy of the local relation formula is deteriorated by the change of the relationship between the operation variable and the quality variable, the coefficient of the local relational formula is renewed so as to reduce the predicting error. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a quality prediction apparatus, a prediction method, a program, and a computer-readable recording medium in a manufacturing process, and more particularly, to a manufacturing process in which quality is determined as a result of a plurality of operating factors, and between operation variables and quality variables. This is a manufacturing process in which the relationship between operating variables and quality variables changes due to changes in the state of equipment over time, etc., and humans can understand the relational expressions for estimating quality variables from operating variables. It relates to the quality prediction of the manufacturing process that is required to exist. In the present invention, the entire operating range of the manufacturing process is divided into a plurality of local regions, a local relational expression is created for each local area, and the entire relational expression is constructed by combining the local relational expressions. Human beings can predict the relationship between the operation variable and the quality variable from the configuration of the whole relational expression and the coefficient value of each local relational expression. Further, when the relationship between the operation variable and the quality variable changes and the prediction accuracy of the local relational expression deteriorates, the coefficient of the local relational expression is updated so as to improve the prediction error. Therefore, the present invention is useful for guidance to operators and quality prediction for quality control in a manufacturing process having non-linear and time-varying characteristics and requiring easy understanding of relational expressions for quality prediction. This is the preferred technique to use.

  Conventionally, in a manufacturing process in which quality is determined based on operating conditions, as a method for predicting quality in the manufacturing process of a product, a physical model created based on knowledge about the mechanism of occurrence of quality defects, or operation data and quality data is used. 2. Description of the Related Art There is a well-known method of calculating product quality prediction values by inputting product operation data into a model using a linear model obtained by applying multiple regression analysis (hereinafter, multiple regression model).

  Further, in the technique disclosed in Patent Document 1, in order to perform quality prediction adapted to the change in the characteristics of the manufacturing process caused by the change in the state of the equipment over time, the edge drop of the rolling mill expressed in a line format For a quantity prediction model, a method for quality prediction adapted to changes in characteristics is realized by updating the parameters of the linear model sequentially by the least square method.

  Furthermore, in the method disclosed in Patent Document 2, for the purpose of predicting the width change amount of the material to be rolled in the finish rolling of hot rolling, a past rolling example similar to the material to be rolled is selected from the database, Production with nonlinearity between operation variables and quality variables by creating a multiple regression model online from the past case operation data and finish width change data, and calculating the predicted width fluctuation of the material to be rolled In the process, a method for predicting with high accuracy is realized. In this method, based on the rolling performance of the material to be rolled, the contents of the database are updated to a new case, so that the characteristics of the operating conditions and the amount of change in the finish width due to changes in rolling conditions and equipment over time, etc. A method of forecasting reflecting changes is also realized.

  Further, in the methods disclosed in Patent Document 3 and Patent Document 4, based on operation data and quality data, an operation factor space having an operation variable as a base vector is divided into several local regions, and each local region The relationship between operational factors and quality is modeled in a linear format that is easy for humans to understand intuitively. Then, an activity function that expresses the contribution rate indicating how much each local line format affects the overall quality as a function of the coordinate of the operation variable space is obtained from the operation data, and the overall operation variable and quality are determined. By deriving a relational expression that expresses the relationship with a variable, a relational expression between a multivariate operation variable having complex nonlinear characteristics and a quality variable is expressed by a mathematical expression that is easy to understand for humans, and an apparatus used for quality prediction Realized.

Japanese Patent Laid-Open No. 2004-90074 JP 2007-50413 A Japanese Patent No. 3875875 JP 2007-140694 A Yoshiharu Naoya “Sequential parameter estimation using vector-type forgetting factor and its application to actual plant” Vol., No. 5, 579/585 (1989) “Information Statistics” by Keiyuki Sakamoto et al. (1983)

  However, the method based on the physical model has a problem that a physical model cannot be created by a manufacturing process in which the mechanism of occurrence of defective quality is not sufficiently elucidated. In addition, in the method using multiple regression model, the operation variable and the quality variable are in a linear relationship, and the model is created based on the premise that both can be expressed in a single line format in the entire operation range. . For this reason, a quality prediction model with the required accuracy is created for manufacturing processes that have a non-linear relationship between operational variables and quality variables, and for manufacturing processes that have multiple quality failure factors with different characteristics. There was a problem that could not be done.

  According to the method disclosed in Patent Document 1, for a manufacturing process capable of expressing operation variables and quality variables in a linear format and realizing sufficient accuracy, even if there is a change in characteristics, sufficient prediction can be made by adapting to the change. It is possible to maintain accuracy. However, there is a problem that it cannot be applied when there is nonlinearity between the operation variable and the quality variable.

  In the method disclosed in Patent Document 2, there is a non-linearity between the operation variable and the quality variable, and even when the relationship between the operation variable and the quality variable is changed due to a change in the manufacturing process equipment over time, etc. It is possible to perform prediction with high accuracy. However, since it is a method of creating a linear model by multiple regression, selecting past case data similar to the product you want to predict online, it is difficult to extract information about the model such as linear coefficient, for example, in a certain operating range There is a problem that it is difficult to perform an analysis such as reading from the model what kind of influence the quality has when the operation variable is changed.

  In the methods disclosed in Patent Document 3 and Patent Document 4, an operation variable space having an operation variable as a basis vector is divided into several local regions, and the relationship between the operation variable and the quality variable in each local region is determined by a human. Modeling in a linear format that is easy to understand intuitively, and calculating the overall relational expression by the sum of the product of the linear form and the activity function of the local region. Therefore, the validity of the model is compared with the conventional experience based on the operation range to which the linear model of each local region is applied and the coefficient value of the linear model in each operation range. It is possible to evaluate together. However, if the relationship between operational variables and quality variables changes, it takes time to accumulate enough data to recreate the model, so it takes time to improve the prediction accuracy. There was a problem.

  The present invention has been made in view of the above points, and predicts the quality of a manufacturing process that is easy for humans to understand and has non-linear characteristics with high accuracy. An object of the present invention is to provide a quality prediction device, a prediction method, a program, and a computer-readable recording medium that automatically update a model and maintain high prediction accuracy even when the relationship between an operation variable and a quality variable changes. And

The quality predicting device manufacturing in the manufacturing process of the present invention is the operation in each local region i obtained by dividing the operation variable space in which the relational expression y representing the relationship between the operation variable and the quality variable in the process is the basis vector. a local relation y _i representing the relationship between the variables and said quality variables, the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i A quality predicting apparatus in a manufacturing process that represents a sum of all the local regions i, and inputs operation data, which is an actual value of an operation variable, into the relational expression y to calculate a quality predicted value,
Operation data input means for taking out and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Storage means;
Based on the activity function [Phi _i in the respective local region i and the operation data, and the contribution ratio calculating means for calculating a contribution ratio of each of the local region i to quality prediction value for該操industry data,
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i Relational expression calculation means;
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value A value calculating means;
Quality predicted value output means for outputting the quality predicted value to the outside;
When the quality of the product is measured, quality data input means for extracting and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, And local relational expression updating means for updating the coefficient value of the local relational expression y _i .
Another feature of manufacturing the quality prediction apparatus in the manufacturing process of the present invention is that the local relational expression y _i is a linear polynomial having all or at least one of the plurality of operation variables as independent variables. is there.
Another feature of the quality prediction apparatus in the manufacturing process of the present invention is that the activity function Φ _i is a normal membership function composed of a normal distribution function centered on the center of gravity of the local region i. It is in.
Another feature of the quality predicting apparatus according to the manufacturing process of the present invention is that the local relational expression updating means sets the coefficient of the linear polynomial before the update in the local region i to W _i (T−1) and the operation. The coefficient is calculated using the following equation (1), where u (T) is a vector consisting of data.

Another feature of the quality prediction apparatus in the manufacturing process of the present invention is that the initial value of the coefficient of the local relational expression y _i and the division pattern store the operation data and the quality data of a plurality of products. From the database
(A) data extraction means for extracting the operation data and quality data corresponding to a predetermined selection condition;
(B) division pattern candidate creation means for creating a plurality of p division pattern candidates for dividing the operation variable space into a plurality of M local regions i;
(C) an activity function calculating means for calculating a coefficient of the activity function Φ _i in each local region i based on the divided coordinate information of each of the divided pattern candidates;
(D) For each of the divided pattern candidates, a local relational expression calculating means for calculating a coefficient of the local relational expression _yi based on the operation data and the quality data in each local region i;
(E) For each of the divided pattern candidates, the product of the local relational expression y _i of each local area and the activity function Φ _i is added for all the local areas to calculate the relational expression y, and the operation A relational expression calculating means for deriving a relational expression y representing the relation between the operation variable and the quality variable in the entire variable space;
(F) The extracted operation data is input to the relational expression y derived for each of the division pattern candidates to calculate a quality prediction value, and the quality prediction value and the extracted quality data are A minimum error relational expression selecting means for calculating a prediction error which is a sum of squares of differences and selecting a division pattern candidate having the smallest value of the prediction error;
(G) learning error evaluation means for comparing the smallest prediction error value with an error determination index set in advance by a predetermined error determination method to determine whether the convergence is sufficient or insufficient. And
When the convergence is insufficient, the plurality of M is incremented by 1 based on the division pattern candidates selected by the minimum error relational expression selection unit, and a series of processing by the units (b) to (g) is performed. This is the initial value of the coefficient of the local relational expression y _i calculated by repeating and the division pattern.
Another feature of the quality prediction apparatus in the manufacturing process of the present invention is that the manufacturing process is a steel process, and the quality variables include surface defects of the product, mechanical strength characteristic values, flatness of the shape, product It is the process value that affects the size, internal stress, or quality. For example, the manufacturing process is a hot rolling process of a thin steel plate that is a steel product, and the quality is a change amount of a plate width on a rough rolling exit side and a plate width of a rolled material on a rolling-in side, As the operation variable of the manufacturing process, at least one or more is selected from the target plate width, target plate thickness, equivalent carbon amount, finish rolling target temperature, and scraping target temperature of the material to be rolled. Alternatively, the manufacturing process is a hot rolling process of a thin steel plate that is a steel product, and the quality is a scraping temperature at the outlet side of the hot-rolled runout table, and the operation variable of the manufacturing process is C amount, Si amount, Mn amount, P amount, S amount, Cu amount, Ni amount, Cr amount, Mo amount, Nb amount, V amount, Ti amount, B in molten steel at the time when the material finishes the refining process Amount, Al amount, N amount, O amount, Ca amount, target plate width of rolled material, target plate thickness, equivalent carbon amount, finish rolling target temperature, scraping target temperature, rolling speed, cooling water amount density, Select at least one of the cooling water temperatures.

The quality prediction method in the manufacturing process according to the present invention includes the operation in each local region i obtained by dividing the operation variable space having the operation variable as a basis vector, the relational expression y representing the relationship between the operation variable and the quality variable in the manufacturing process. a local relation y _i representing the relationship between the variables and said quality variables, the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i Is a quality prediction method in a manufacturing process that represents a sum of all the local regions i, and inputs operation data, which is an actual value of an operation variable, into the relational expression y to calculate a quality prediction value,
A step of inputting operation data using operation data input means for taking out and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Storing an activity function using storage means;
Based on the activity function [Phi _i in the respective local region i and the operational data, the contribution ratio with the contribution ratio calculating means for calculating a contribution ratio of each of the local region i to quality prediction value for該操Industry Data A calculating step;
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i A step of calculating a local relational expression using a relational expression calculating means;
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value Calculating a quality predicted value using a value calculating means;
Outputting a quality prediction value using quality prediction value output means for outputting the quality prediction value to the outside;
When the quality of the product is measured, the step of inputting the quality data using quality data input means for taking out and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, And updating the local relational expression using local relational expression updating means for updating the coefficient value of the local relational expression y _i .
The program according to the present invention includes the operation variable and the quality variable in each local region i obtained by dividing the operation variable space having the operation variable as a base vector by using the relational expression y representing the relationship between the operation variable and the quality variable in the manufacturing process. a local relation y _i representing the relationship between, the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i, all of the A program for calculating a quality prediction value by inputting operation data, which is an actual value of an operation variable, into the relational expression y, which is expressed as a sum of local areas i,
Operation data input processing for extracting and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Amnestics,
Based on the activity function [Phi _i in the respective local region i and the operation data, and the contribution ratio calculation process for calculating the contribution ratio of each of the local region i to quality prediction value for該操industry data,
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i Relational expression processing,
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value Value calculation processing,
A quality prediction value output process for outputting the quality prediction value to the outside;
When the quality of the product is measured, a quality data input process for extracting and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, And causing the computer to execute local relational expression update processing for updating the coefficient value of the local relational expression y _i .
The computer-readable recording medium of the present invention is characterized by recording the program of the present invention.

  As described above, the present invention divides an operation variable space composed of multivariate operation variables into a plurality of local regions when performing quality prediction of a manufacturing process having nonlinear characteristics between operation variables and quality variables. The relationship between the operating factors and quality in each local area is expressed by a mathematical formula that can be understood by humans, and the overall relational expression is calculated by the sum of the product of the mathematical formula and the activity function of the local area. Is measured, the coefficient of the mathematical formula is updated based on the actual quality value and the quality predicted value by the model.

  According to the present invention, it is possible to build a prediction model for predicting the quality of a manufacturing process having a non-linear characteristic between an operation variable and a quality variable in a format that is highly accurate and easy for humans to understand the contents. It is possible to perform quality prediction with high accuracy while the human being judges the rationality regarding the prediction result. Further, even when the relationship between the operation variable and the quality variable changes due to a change in equipment over time, the model can be automatically updated to maintain high prediction accuracy.

Preferred embodiments of the present invention will be described below with reference to the accompanying drawings.
(First embodiment)
FIG. 1 is a block diagram showing a configuration of a quality prediction apparatus 100 in the manufacturing process of the present embodiment. As will be described in detail below, the quality prediction apparatus 100 uses a relational expression y representing the relationship between an operation variable and a quality variable in a manufacturing process, and each local region i obtained by dividing an operation variable space having the operation variable as a basis vector. a local relation y _i representing the relationship between該操industry variables and said quality variables, the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i The operation data, which is the actual value of the operation variable, is input to the relational expression y to calculate the quality prediction value.

  Reference numeral 101 in FIG. 1 denotes an operation data input unit that functions as an operation data input unit. The operation data of the prediction target product collected from the manufacturing process 200 is input to the quality prediction apparatus 100. Specific operation data input unit 101 includes, for example, a keyboard, an OCR for reading a data sheet, and measurement signals from sensors installed in the factory, collected and stored sequentially, and data is transmitted via a LAN or the like at a preset timing. Think of a computer that captures.

  For example, in the steel process, the operation data includes the amounts of various elements of the molten steel measured in the refining process, the amount of molten steel surface fluctuation and casting speed in the continuous casting process, and the rolling load and scraping temperature in the hot rolling process. It is. In addition, the quality to be predicted based on these operation data includes, for example, in the case of the steel process, the number of surface defects and internal defects, tensile strength, yield stress, elongation rate, etc. of various products such as thin plates and thick plates. These are various quality indicators of the final product directly requested by customers, such as mechanical strength characteristic values, shape flatness such as wave height, product size such as plate thickness, plate width and plate length, and internal stress. In addition, the process values that affect these final qualities are also the quality to be predicted.

  Here, when the quality is predicted based on a plurality of p operation variables u1, u2,..., Up, the quality can be regarded as a function in a p-dimensional operation variable space having u1 to up as basis vectors. Therefore, the relational expression y representing the relation between the operation variable and the quality variable can be generally expressed by the expression (3) via the mapping function f (·).

The mapping function f (•) is a non-linear and multivariate complex function in the case of a normal manufacturing process, and it is difficult to find an appropriate functional expression over the entire operation variable space. Therefore, in the present invention, the operation variable space is divided into a plurality of local regions, and a local relational expression y _i ^ (u_) representing the relationship between the operation variable and the quality variable in each local region i (in this embodiment, ^ is attached). Note that y _i ^ is assumed that ^ is attached to y _i , u_ is attached to _ below u, and so on.) And local relational expression y at each point on the entire operation variable space _The whole relational expression y is expressed by Expression (4) which is the sum of products of the activity functions Φ _i (u_) representing the contribution rate of _i ^. Here, Σ represents the sum of terms, and M represents the number of local regions.

  The operation data input by the operation data input unit 101 is data in which numerical values corresponding to the operation variables u1 to up are described, and by performing arithmetic processing by each unit described below, An operation equivalent to equation (4) is executed to calculate a quality prediction value.

Reference numeral 102 in FIG. 1 denotes an activity function storage unit that functions as an activity function storage unit, which calculates and stores the coefficient of the activity function Φ _i in equation (4). The coefficient of the activity function Φ _i is calculated based on the division coordinate information of the division pattern in which the operation variable space is divided into M local regions. FIG. 2 shows a schematic diagram of an example of a division pattern when an operation variable space composed of two operation variables u1 and u2 is divided into three local regions. The operation data can be regarded as a point existing on the operation variable space. In the example of FIG. 2, the operation data is located in the boundary region between the local region 2 and the local region 3. As specific division coordinate information, for example, there is a method of using coordinate values of vertices of each local region. Alternatively, as shown in FIG. 2, in the case where the division of the local region is limited to only the division parallel to the axis of the operation variable, each local region that is a super rectangular parallelepiped in the p-dimensional space obtained for each local region i. The coordinates c ⁱ _j of the center of gravity point and the length Δ ⁱ _j of the hypercube in the axial direction of each operation variable may be used. Here, j = 1... P is a subscript indicating an operation variable. In this embodiment, from the viewpoint of easy understanding of division by humans, the division pattern is limited to only division parallel to the operation variable, and the division coordinate information also uses the center of gravity point and the length of the super rectangular parallelepiped.

As the activity function Φ _i , an arbitrary function that satisfies the normal condition of Expression (5) in which the sum of the function values is 1 at an arbitrary position in the operation variable space can be used.

In this embodiment, a normal membership function obtained by substituting the p-dimensional normal distribution function μ _i shown in Equation (6) into Equation (7) as a function for smoothly connecting the boundaries of adjacent local regions is activated. A degree function Φ _{i is} assumed. Here, c ⁱ _j represents the barycentric point of the local region, and σ ⁱ _j represents the standard deviation of the normal distribution function, and the barycentric point and the standard deviation of the normal distribution function are coefficients in the normal membership function.

The coordinates c ⁱ _j of the centroid point of each local area in the divided coordinate information may be used as the centroid point of the local area in the above equation (6), and the standard deviation σ ⁱ _j is the axial direction of each operation variable. There is a method of calculating, for example, by the following equation (8) using the length Δ ⁱ _j of the super rectangular parallelepiped.

Here, α is a non-negative real value, and the smaller the α value, the smaller the standard deviation of the normal distribution function in equation (6). Therefore, the activity function Φ _i calculated in equation (7) This corresponds to setting the activity function so that the portion where the regions overlap each other at the boundary portion of the local region becomes small. FIG. 3 is a diagram schematically illustrating the difference in the activity function Φ _i when the α value is changed when the one-dimensional operation variable space is divided into two local regions. The α value may be set empirically from physical knowledge about the operation range of the target process corresponding to the local region, or the value obtained by changing the value while checking the prediction accuracy. May be used. FIG. 4 shows an activity function when the operation factor space composed of u1 and u2 is divided into three regions as an example of the activity function in the two-dimensional operation variable space.

In the activity function storage unit 102, the coefficient of the activity function Φ _i is calculated by the above-described procedure, the coefficient is stored, and can be used in the subsequent units. Note that the division coordinate information of the division pattern used for the calculation of the coefficient may be determined by the apparatus described in the second embodiment, or division may be set based on experience with the manufacturing process.

Reference numeral 103 in FIG. 1 denotes a contribution rate calculation unit that functions as a contribution rate calculation unit, and calculates the rate at which each local region contributes to the predicted value by substituting the value of the operation data into the activity function Φ _i. . When the activity function Φ _i is a normal membership function, numerical values corresponding to u1 to up input from the operation data input unit 101 are set in u1 to up in the expression (6), and the expression (6) And the contribution rate is calculated by performing mathematical formula calculation of Formula (7). In addition, when the trapezoidal membership function and the triangular membership function proposed as research results in the field of fuzzy inference are used for the activity function Φ _i , the operation data is stored in the variables corresponding to the operation variables of these functional expressions. What is necessary is just to calculate by setting a numerical value.

Reference numeral 104 in FIG. 1 denotes a local relational expression calculation unit that functions as a local relational expression calculation means, and calculates the value of the local relational expression y _i ^ for each local region based on the coefficient value of the local relational expression and the operation data. Process. Here, as the local relational expression y _i ^, an arbitrary mathematical expression that can express the relation between the quality variable and the operation variable that are continuous values can be adopted. More specifically, a method of using a linear polynomial of equation (9) having sufficient accuracy if the change behavior of the quality variable with respect to the increase or decrease of the operation variable is easier for humans to read than the coefficient value and the region division is appropriate. There is. Here, w _ij is a coefficient of a linear polynomial.

  The local relational expression calculation unit 104 stores a mathematical expression of a local relational expression set in advance as in Expression (9) and its coefficient value, and corresponds to u1 to up input from the operation data input unit 101. Set a numeric value and execute local relational expression processing. In the local relational expression, it is not necessary to limit to using the operation variables corresponding to all of the input operation data u1 to up, and the local relational expression using only a part may be used. This corresponds to a case where some of the operation variables are variables used only for dividing the operation variable space, and there are variables that are not used in the local relational expression for predicting quality.

  In addition, the coefficient value of the local relational expression is updated by the local relational expression updating unit 104 described later. However, the calculation of the initial value may be performed in the second embodiment, or may be appropriately performed based on experience. An initial value may be set. When a mathematical expression other than a linear polynomial is used as the local relational expression, the local relational expression may be calculated by setting the numerical value of the operation data in a variable corresponding to the operation variable of the mathematical expression.

Reference numeral 105 in FIG. 1 denotes a quality prediction value calculation unit that functions as a quality prediction value calculation unit. By calculating the product of the calculation result of the local relational expression y _i ^ calculated for each local region i and the calculation result of the contribution rate, and calculating the sum of all the local regions with the value of this product, A quality prediction value based on the relational expression between the operation variable and the quality variable in the equation (4) is calculated.

  Reference numeral 106 in FIG. 1 denotes a quality prediction value output unit that functions as a quality prediction value output unit. The quality prediction value calculated by the quality prediction value calculation unit 105 is displayed as guidance information or used for quality control. Output to the control system. Based on the displayed quality prediction value, the operation operator can determine the operation condition for obtaining the desired quality, and the inspector conducts the priority inspection of the product with the concern of the quality defect, It is possible to prevent defective products from being shipped to customers. Also, when used for quality control, the control system performs the process of calculating the target value of the operation variable at the operation end so that the quality falls within the desired range based on the predicted quality value and reflects it in the operation of the manufacturing process. As a result, the occurrence of poor quality can be suppressed.

  Reference numeral 107 in FIG. 1 denotes a quality data input unit that functions as a quality data input unit. When quality of a product is measured, the quality data is extracted from the manufacturing process and input. For example, in the case of a steel process, the number of surface defects and internal defects by an automatic defect inspection device, mechanical strength measurement values in a sample test process, wave height measurement values by a shape meter, dimensional information by a plate thickness / width measurement meter, internal Information such as the amount of elemental components, temperature, and size in the intermediate process stage that affects the stress and the final quality is quality data. These quality data include, for example, a keyboard, an OCR that reads a data sheet, and a measurement signal from a sensor. It is input by a computer that sequentially collects and transmits via a LAN or the like.

Reference numeral 108 in FIG. 1 denotes a local relational expression update unit that functions as a local relational expression update unit. The local relational expression update unit calculates a prediction error that is a difference between the quality data and the quality predicted value in each local region, Based on the contribution ratio of the local region, the value of the coefficient of the local relational expression y _i ^ is updated. When the linear polynomial of Expression (9) is used for the local relational expression y _i ^, the coefficient is updated as follows, for example. Now, a vector composed of operation data of a product used for updating is represented as u (T), a vector composed of coefficients before updating of the linear polynomial in the local region i is represented as W _i (T−1), and quality data The quality data of the product used for the update input from the input unit 107 is y (T). At this time, the coefficient W _i (T) of the local relational expression y _i ^ in the local region i is updated by the following expression (1). Here, K _i (T) is an update rate matrix in the local region i, and is calculated by Expression (2).

Expressions (1) and (2) achieve the same effect as the update of the successive least squares method described on page 73 of Non-Patent Document 1, for example, with the local relational expression y _i ^ of the linear polynomial in the present invention. Therefore, it is a renewal formula that was newly devised this time. Formula (1) is obtained by multiplying the error of the local relational expression calculated in parentheses in the second term on the right-hand side by the update matrix K _i (T) as a correction amount and using the coefficient W _i (T− The updated coefficient is calculated by adding to 1). Here, according to Equation (2), in the case of a local region i where the contribution rate of the activity function is Φ _i (u_ (T)) ≈0 with respect to the operation data u_ (T), K _i It can be seen that (T) ≈0. Therefore, when the contribution ratio of the local region is Φ _i (u_ (T)) ≈0, even when the error in the local relational expression is large, the correction amount is almost zero, and the coefficient W _i (T) is almost updated. Not. That is, a local relational expression having almost no contribution rate to the operation data means that it is not updated.

Note that the adaptive gain matrix P _i (T−1) of the local region i in Expression (2) is updated by the following Expression (10) in preparation for the next update of W _i (T + 1).

Here, I is a unit matrix, and Λ _i is a pair of reciprocals of forgetting coefficients λ _k (k = 0, 1,..., P) corresponding to each of p + 1 coefficients of the local relational expression y _i ^. It is a matrix of Expression (11) as a corner component.

  The forgetting factor is a factor for setting to what extent the correction based on the latest operation and quality data is reflected in the correction amount when calculating the correction amount when the coefficient is updated. ).

The forgetting factor λ takes a range of 0 to 1, and the smaller the λ, the greater the forgetting effect, and the update matrix K _i (T) in Equation (1) is greatly evaluated. Conversely, when λ = 1, the forgetting effect is zero, and the coefficient is not updated regardless of the magnitude of the prediction error. According to Equation (12), when the second term in the root sign on the right side is viewed, λ → 0 and the forgetting effect increase when the local line format error is large. _In the case of a local region where _i (u_ (T)) ≈0, it can be seen that the second term is almost zero and the forgetting effect does not act. That is, it means that the forgetting effect does not act on the relational expression y _i ^ of the local region having almost no contribution rate to the operation data. Note that the coefficient g _k in the equation (12) is an adjustment coefficient for individually setting the forgetting effect for each operation variable, and may be set empirically from physical knowledge of each operation variable. You may adjust so that quality prediction accuracy may improve, performing operation.

  When a function other than the linear polynomial of equation (9) is used for the local relational expression, the update formula corresponding to the function of the adopted local relational expression is used, and the calculation is performed in the same procedure as described above. Update should be done.

Above, the newly devised local equation y _i ^ update equations in the present invention, if the relationship between the operating variables and quality variables has changed, since the coefficient of the topical relation y _i ^ is rapidly updated, It has become possible to greatly reduce the time required for improving the prediction accuracy.

  Next, each process of the quality prediction method performed using the quality prediction apparatus 100 of this embodiment is demonstrated using the processing flowchart shown in FIG.

  Step S101 in FIG. 8 is an operation data input step, which is a step of inputting operation data from the manufacturing process using the operation data input unit 101.

Step S102 in FIG. 8 is an activity function storage process, which uses the activity function storage unit 102 to calculate the coefficient of the activity function Φ _i based on the divided coordinate information and store it in the memory or storage device. I do. When a normal membership function is used as the activity function Φ _i , the center of gravity of the local region is set as the center of gravity, and the standard deviation of the normal distribution function is calculated based on the equation (8). When the operation data is input, the coefficient of the activity function Φ _i is extracted from the memory or the storage device and sent to the next process.

Step S103 in FIG. 8 is a contribution rate calculation step, and using the contribution rate calculation unit 103, the value of the operation data is set to the expression of the activity function Φ _i of Expressions (6) to (8), Processing for calculating the contribution ratio of each local region is performed.

Step S104 in FIG. 8 is a local relational expression calculation step, and the local relational expression calculation unit 104 is used to store the value of the coefficient of the local relational expression y _i ^ in a memory or a storage device. When the is input, the coefficient is extracted and the local relational expression y _i ^ is calculated.

  Step S105 in FIG. 8 is a quality predicted value calculation step, and uses the quality predicted value calculator 105 to perform a process of calculating a quality predicted value based on the relational expression between the operation variable and the quality variable in Expression (4). .

  Step S106 in FIG. 8 is a quality predicted value output process, which uses the quality predicted value output unit 106 to output the predicted value as guidance information on a display screen for an operator or inspector, In order to use it for the control, it is transmitted as a signal to the control system of the manufacturing process.

  Step S107 in FIG. 8 is a quality data input process. When the quality of the product is measured using the quality data input unit 107, the quality data is extracted from the manufacturing process and input.

Step S108 in FIG. 8 is a local relational expression update step, and the local relational expression update unit 108 is used to calculate a prediction error that is a difference between the quality data and the quality predicted value for each local region. Based on the error and the contribution ratio of each local region to the operation data, a process of updating the coefficient value of the local relational expression y _i ^ is performed.

(Second Embodiment)
As a second embodiment, an initial value of a coefficient of the local relational expression y _i ^ and a procedure for calculating a division pattern will be described. FIG. 5 is a diagram showing a configuration for calculating the initial value of the coefficient of the local relational expression y _i ^ and the division pattern.

  Reference numeral 400 in FIG. 5 denotes a database, a set of past operation and quality data in the manufacturing process, code information indicating the type of product, manufacturing time in each manufacturing process, order information such as a destination, This is stored together with the product number for identifying the product.

  Reference numeral 301 in FIG. 5 denotes a data extraction unit that functions as a data extraction unit. Based on the selection conditions relating to the target of quality prediction input from the outside, such as code information indicating the type of product, a predetermined selection condition is obtained from the database 400. A plurality of operation data and quality data corresponding to the above are extracted.

  Reference numeral 302 in FIG. 5 denotes a division pattern candidate creation unit that functions as a division pattern candidate creation unit, and a plurality of p division pattern candidates obtained by dividing an operation variable space composed of p operation variables corresponding to the extracted operation data. Create FIG. 6 schematically shows a procedure for creating candidate division patterns, taking a two-dimensional operation variable space (p = 2) as an example. First, a case where an operation variable space that is not divided at all is divided into two local regions (M = 2) will be described. In the division, since division points are set so as to divide the region into two equal axes on the basis of each operation variable, p = 2 types of division pattern candidates are created.

Further, when the operation variable space is already divided into several local regions, the error of each local relational expression y _i ^ is expressed by the following equation (17) which is a weighted error evaluation function by the activity function Φ _i. The local area with the largest error is divided into two. FIG. 6 shows a procedure for dividing the operation variable space that has already been divided into three into M = 4. If the region with the largest error is the region 3-2, for example, the region 3-2 includes On the other hand, division points are set so as to divide the region into two equal parts with axes parallel to the operation variables of u1 and u2, and in this case as well, p = 2 division pattern candidates are created.

  The subsequent processing from the activity function calculation unit 303 to the relational expression calculation unit 305 is performed for all of the divided pattern candidates created here.

Reference numeral 303 in FIG. 5 denotes an activity function calculation unit that functions as an activity function calculation unit. The activity function Φ _i is calculated based on the division coordinate information for the division pattern obtained by the division pattern candidate creation unit 302. . The activity function Φ _i needs to use the same function expression as that of the activity function setting unit 102 in FIG. 1. In this embodiment, the regular membership function defined by the expressions (6) and (7) is used. Is used.

Reference numeral 304 in FIG. 5 denotes a local relational expression calculation unit that functions as a local relational expression calculation unit, and calculates a coefficient of the local relational expression y _i ^ based on a function form assumed in the local region. When the local relational expression y _i ^ is the linear polynomial of Expression (9), the coefficient is calculated as follows. When N pieces of data are extracted from the data extraction unit 301, the operation data is p-row N-column matrix data, the quality data is N-dimensional vector data, and each local relational expression y _i ^ is expressed by the equation (13). It is represented by the matrix representation of

The taking into account the activity function [Phi _i determine the coefficients w _ij, weighted error evaluation function by activity function [Phi _i, may be determined coefficients as equation (17) is minimized.

  The coefficient that minimizes Expression (17) is equal to the coefficient that satisfies Expression (18), and specifically, is calculated by matrix calculation of Expression (19).

By performing the above matrix operation, the coefficient w _ij of the local relational expression y _i ^ using a linear polynomial can be calculated.

Reference numeral 305 in FIG. 5 denotes a relational expression calculation unit that functions as a relational expression calculation unit. For each divided pattern candidate, the product of the local relational expression y _i ^ and the activity function Φ _i of each local region is calculated for all the local regions. Is added to calculate the relational expression y, and the relational expression y representing the relation between the operation variable and the quality variable in the entire operation variable space is derived. That is, using the activity function Φ _i calculated by the activity function calculating unit 303 and the local relational expression y _i ^ obtained by the local relational expression calculating unit 304, a linear combination formula of Formula (4) is created. Create a relational expression between the quality variable and the operation variable.

  Reference numeral 306 in FIG. 5 denotes a minimum error relational expression selection unit that functions as a minimum error relational expression selection unit, and inputs the extracted operation data into the relational expression y derived for each divided pattern candidate to predict quality. A value is calculated, a prediction error that is a sum of squares of differences between the quality prediction value and the extracted quality data is calculated, and a division pattern candidate having the smallest value of the prediction error is selected. That is, the processing from the activity function calculation unit 303 to the relational expression calculation unit 305 is performed on all of the plurality of division pattern candidates created by the division pattern candidate creation unit 302, and the relational expression between the quality variable and the operation variable is obtained. Since it is created, a process of selecting a relational expression having the smallest error defined by Expression (21) among these relational expressions is performed.

  Reference numeral 307 in FIG. 5 denotes a learning error evaluation unit that functions as a learning error evaluation unit, which compares the smallest prediction error value with an error determination index (evaluation reference value) set in advance by a predetermined error determination method, and converges. Is sufficient or insufficient. That is, the error of the minimum error relational expression obtained by the minimum error relational expression selecting unit 306 is compared with a preset evaluation reference value, and if the error is equal to or less than the evaluation reference value, the data is sufficiently accurate. It is determined that a relational expression that can explain is constructed. Convergence judgment methods include, for example, a method of comparing the error of the relational expression with the convergence judgment factor, a method of comparing the change amount of the relational expression error with respect to the increment of the local region division with the convergence judgment factor, and considering the number of divisions and error When an index, for example, Akaike's information amount index described on page 43 of Non-Patent Document 2 is used in addition to the learning error as well as the number of local regions, and the index increases when the number of divisions increases The method of aborting the division is used. In any method, when it is evaluated that the convergence is insufficient, the division number M is increased by 1, and a series of processes after the division pattern candidate creation unit 302 is repeatedly executed.

  FIG. 7 schematically shows a procedure for creating a division pattern while increasing the number M of divisions, taking a two-dimensional operation variable space as an example. For the initial division, two division pattern candidates are created, and when the division pattern on the axis parallel to the variable u1 has a small error, the division pattern on the axis parallel to u1 is selected. . Next, the error of each local region is evaluated, and the local region 2-1 having the largest error is selected. Two division pattern candidates obtained by dividing the local area 2-1 are created, and a division pattern on an axis parallel to u2 with a small error is selected. Similarly, the error evaluation of the local region and the creation and selection of the division pattern candidates are repeated.

When it is determined that the learning error evaluation unit 307 has converged, a process of extracting the division coordinate information of the division pattern of the relational expression between the operation variable and the quality variable and the coefficient of the local relational expression is performed. The coordinate information of the division pattern and the coefficient of the local relational expression thus obtained are calculated so as to approximate the relation between the operation variable and the quality variable within a predetermined error based on the past operation data and quality data. Therefore, good prediction accuracy is realized by using the division coordinate information of the division pattern and the initial value of the coefficient of the local relational expression y _i ^ in the first embodiment.

In addition to the second embodiment, the initial value of the division coordinate information of the division pattern and the coefficient of the local relational expression y _i ^ may be set to an appropriate initial value based on, for example, experience with the manufacturing process. Alternatively, it may be set by applying a non-linear optimization method proposed in the research field of mathematical optimization methods.

  The quality prediction apparatus and prediction method in the manufacturing process of the present invention can be realized by a computer. FIG. 9 shows a configuration example of a computer system 1200 that can function as a product material value prediction apparatus of the present invention. The computer system 1200 includes a CPU 1201, a ROM 1202, a RAM 1203, a keyboard controller (KBC) 1205 of a keyboard (KB) 1209, a CRT display (CRT) 1210 as a display unit, a CRT controller (CRTC) 1206, and a hard disk (HD). ) 1211 and a flexible disk (FD) 1212 disk controller (DKC) 1207 and a network interface controller (NIC) 1208 for connection to the LAN 1220 are connected to each other via a system bus 1204 so that they can communicate with each other. Yes.

  The CPU 1201 comprehensively controls each component connected to the system bus 1204 by executing software stored in the ROM 1202 or the HD 1211 or software supplied from the FD 1212. That is, the CPU 1201 reads out and executes a processing program according to a predetermined processing sequence from the ROM 1202, the HD 1211, or the FD 1212, thereby performing control for realizing the operation in the present embodiment.

  A RAM 1203 functions as a main memory or work area of the CPU 1201. The KBC 1205 controls instruction input from the KB 1209, a pointing device (not shown), or the like. A CRTC 1206 controls display on the CRT 1210. The DKC 1207 controls access to the HD 1211 and the FD 1212 that store a boot program, various applications, an edit file, a user file, a network management program, a predetermined processing program in the present embodiment, and the like. The NIC 1208 exchanges data bidirectionally with a device or system on the LAN 1220.

  A software program that describes the means of the quality prediction apparatus according to the embodiment of the present invention and the process for realizing the function of each step of the quality prediction apparatus is supplied to the computer, and the program stored in the computer What was implemented by operating various devices according to a program is included in the category of the present invention.

  In this case, the software program itself realizes the functions of the above-described embodiments, and the program itself is included in the scope of the present invention. As a transmission medium for the program code, a communication medium such as a computer network system that transmits the program as an electric signal can be used.

  Furthermore, means for supplying the program to a computer, for example, a storage medium storing the program is also included in the scope of the present invention. As a storage medium for storing the program code, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a CD-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

  According to the quality prediction device, the prediction method, the program, and the computer-readable recording medium in the manufacturing process according to the present embodiment described above, the quality prediction of the manufacturing process having a non-linear characteristic between the operation variable and the quality variable. When creating the relational expression to perform the operation, the operation variable space consisting of multivariate operation variables is divided into multiple local regions, and the relationship between the operation variables and quality variables in each local region is a mathematical formula that can be understood by humans. And the overall relational expression is calculated by the sum of the product of the mathematical formula and the local area activity function, so that the quality of the manufacturing process having nonlinear characteristics can be predicted with high accuracy, and the contents of the relational expression Can be constructed in a format that is easy to understand. Furthermore, when the quality of the product is measured, the operating variables and quality variables are attributed to changes in equipment over time, etc. by updating the coefficients of the local relational expression based on the actual quality value and the quality predicted value based on the relational expression. Even when the relationship changes, the relational expression can be automatically updated to maintain high prediction accuracy.

(First embodiment)
In the following, for the hot rolling process of thin steel sheets, which are steel products, the amount of change in the strip width of the rolled material on the rough rolling exit side and the strip width of the rolled material on the inlet side of the take-up machine is defined as quality. In the variable, an embodiment is described in which quality prediction is performed using the target plate width, target plate thickness, equivalent carbon amount, finish rolling target temperature, and scraping target temperature of the material to be rolled.

  FIG. 10 is a diagram showing an outline of a hot rolling process of a thin steel plate. A base material of a material to be rolled, called a slab, is extracted from a heating furnace and then rolled into a bar shape having a thickness of 40 to 50 mm by rough rolling. A sheet width meter is installed on the exit side of the rough rolling mill, and the sheet width of the material to be rolled at this point is measured. Next, the material to be rolled is continuously rolled by a finish rolling mill arranged in series, and then wound into a coil by a winder. Here, immediately before the tapping machine, a tapping plate width meter is installed, and processing for determining whether the material to be rolled is within a predetermined width range is performed. If the strip width is not within the specified range, the coil will be demoted to other destinations, or the entire amount will be scrapped, resulting in a large loss. The amount of fluctuation in the sheet width between the rolling-in sides is estimated with a quality prediction model, and the occurrence rate of sheet width defects is reduced by using a control system that reflects this prediction result in the target width setting value for rough rolling. .

  Conventionally, in order to predict this variation in sheet width, a sheet width variation prediction model based on a linear multiple regression model with the target sheet width, target sheet thickness, and equivalent carbon content of steel as operational variables was created and The prediction was performed while maintaining accuracy by updating the coefficient by the square method. Instead of this conventional method, a plate width variation prediction model according to the present invention was applied. In addition to the variables in the conventional linear multiple regression model, the model is created by adding the finish rolling target temperature and the scraping target temperature as operation variables. FIG. 11 shows a frequency distribution of prediction errors (= predicted value−actual value) in order to compare the accuracy of predicting the fluctuation amount of the plate width according to the conventional method with the predictive accuracy according to the present invention. The standard deviation indicating the variation in the prediction error was reduced from 1.82 to 1.25, and the improvement of the prediction accuracy according to the present invention could be confirmed.

  Based on the above results, the width fluctuation prediction model according to the present invention is continuously operated, and as a result, the occurrence rate of plate width mismatch is reduced, and further, the yield is improved, the labor for product maintenance is reduced, and the delivery time delay is avoided. I was able to.

(Second embodiment)
Next, for the hot rolling process of thin steel sheets, which are steel products, the steel sheet temperature in scraping is defined as the quality (scraping temperature at the hot-rolling runout table exit side), and the material to be rolled is refined as the operating variable. C amount, Si amount, Mn amount, P amount, S amount, Cu amount, Ni amount, Cr amount, Mo amount, Nb amount, V amount, Ti amount, B amount, Al in molten steel at the time of completing the process Amount, N amount, O amount, Ca amount, target plate width, target plate thickness, equivalent carbon amount, finish rolling target temperature, scraping target temperature, rolling speed, cooling water density, and cooling water temperature. An embodiment in which quality prediction is performed using the method will be described.

  A non-contact type thermometer with a radiation thermometer is installed at the position of the scraper in FIG. 10, and is an important management index that affects the material properties such as the mechanical property value of the thin steel plate. The steel plate temperature is measured. In order to control the steel sheet temperature within a predetermined range, a cooling facility by spraying water on the steel sheet is provided on the runout table from the finish rolling mill exit side to the scraping. In the hot rolling process, a prediction model is created to estimate the steel sheet scraping temperature, using the steel sheet composition and size, the target temperature at each stage such as finish rolling and scraping, and cooling water density as inputs. The operation of calculating the cooling water amount density in which the scraping temperature falls within a predetermined range using the model and changing the set value of the water amount density is performed.

  Conventionally, the cooling phenomenon when water is sprayed on a steel plate on a runout table has been estimated using a physical model based on a heat transfer equation. Instead of this physical model, the present invention was applied and a scraping temperature estimation model was created and implemented on an actual line. As a result, the standard deviation of the scraping temperature prediction error is improved, and as a result of continuous operation of this model, the occurrence rate of scraping temperature mismatch is reduced, and the quality due to the reduction in variation in mechanical property values is further reduced. Effects such as improvement and yield improvement could be obtained.

  In both the first embodiment and the second embodiment, the quality prediction device is realized as a program on a computer. However, the quality prediction device may be configured by hardware combining an arithmetic device, a memory, and the like. . Moreover, the quality prediction apparatus of this invention may be comprised from a some apparatus, or may be comprised from one apparatus.

It is a block diagram which shows the structure of the quality prediction apparatus in the manufacturing process of this embodiment. It is a figure explaining the example of a division | segmentation pattern which divided | segmented the two-dimensional operation variable space into three local regions. When dividing a one-dimensional operation variable space into two, it is a figure explaining the difference of the activity function at the time of changing the division | segmentation parameter (alpha) value. It is a figure explaining the example of the activity function which divided | segmented the two-dimensional operation factor space into three area | regions. It is a figure which shows the structure in the case of calculating the initial value of the coefficient of a local relational expression, and a division | segmentation pattern in the quality prediction apparatus in the manufacturing process of 2nd Embodiment. It is a schematic diagram of the procedure which produces the candidate of a division | segmentation pattern in a two-dimensional operation variable space, (a) is a figure which shows the case where the operation variable space is not divided at all (when dividing to M = 2), (b) ) Is a diagram showing a case where the operation variable space has already been divided into three (when M = 4). It is a figure which illustrates typically the creation process of a division | segmentation pattern taking a two-dimensional operation factor space as an example. It is a figure which shows the processing flow of the quality prediction method in the manufacturing process of embodiment. It is a block diagram which shows an example of the computer system which can function as a quality prediction apparatus in the manufacturing process of embodiment. It is a figure which shows the outline of the hot rolling process of a thin steel plate. It is a figure which shows the precision of the fluctuation amount prediction model of the board width which is an Example of this invention by the comparison with a conventional system, (a) is a figure which shows the prediction precision by the conventional width fluctuation prediction model, (b) is this figure It is a figure which shows the prediction precision by invention.

Explanation of symbols

100: Quality prediction device 101 in the manufacturing process 101: Operation data input unit 102: Activity function storage unit 103: Contribution rate calculation unit 104: Local relational expression calculation unit 105: Quality prediction value calculation unit 106: Quality prediction value output unit 107: Quality data input unit 108: local relational expression update unit 200: manufacturing process 301: data extraction unit 302: division pattern candidate creation unit 303: activity function calculation unit 304: local relational expression calculation unit 305: relational expression calculation unit 306: minimum Error relational expression selection unit 307: learning error evaluation unit 400: database

Claims (11)

The relational expression y representing the relationship between the operation variable and the quality variable in the manufacturing process is expressed as a local expression representing the relationship between the operation variable and the quality variable in each local region i obtained by dividing the operation variable space with the operation variable as a basis vector. a relational expression y _i, of the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i, the sum of all of said local area i A quality prediction device in a manufacturing process for calculating a quality prediction value by inputting operation data that is a result value of an operation variable into the relational expression y,
Operation data input means for taking out and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Storage means;
Based on the activity function [Phi _i in the respective local region i and the operation data, and the contribution ratio calculating means for calculating a contribution ratio of each of the local region i to quality prediction value for該操industry data,
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i Relational expression calculation means;
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value A value calculating means;
Quality predicted value output means for outputting the quality predicted value to the outside;
When the quality of the product is measured, quality data input means for extracting and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, And a local relational expression updating means for updating a coefficient value of the local relational expression _yi . The quality prediction apparatus in the manufacturing process according to claim 1, wherein the local relational expression y _i is a linear polynomial having all or at least one of the plurality of operation variables as independent variables. The quality predicting device in the manufacturing process according to claim 1, wherein the activity function Φ _i is a normal membership function composed of a normal distribution function centered on the center of gravity of the local region i. . The local relational expression update means uses the following expression (1), where W _i (T−1) is a coefficient of the linear polynomial before update in the local region i, and u (T) is a vector composed of the operation data. The quality prediction apparatus in the manufacturing process according to claim 2, wherein a coefficient is calculated.

The initial value of the coefficient of the local relational expression y _i and the division pattern are obtained from a database storing the operation data and quality data of a plurality of products.
(A) data extraction means for extracting the operation data and the quality data corresponding to a predetermined selection condition;
(B) division pattern candidate creation means for creating a plurality of p division pattern candidates for dividing the operation variable space into a plurality of M local regions i;
(C) an activity function calculating means for calculating a coefficient of the activity function Φ _i in each local region i based on the divided coordinate information of each of the divided pattern candidates;
(D) For each of the divided pattern candidates, a local relational expression calculating means for calculating a coefficient of the local relational expression _yi based on the operation data and the quality data in each local region i;
(E) For each of the divided pattern candidates, the product of the local relational expression y _i of each local area and the activity function Φ _i is added for all the local areas to calculate the relational expression y, and the operation A relational expression calculating means for deriving a relational expression y representing the relation between the operation variable and the quality variable in the entire variable space;
(F) The extracted operation data is input to the relational expression y derived for each of the division pattern candidates to calculate a quality prediction value, and the quality prediction value and the extracted quality data are A minimum error relational expression selecting means for calculating a prediction error which is a sum of squares of differences and selecting a division pattern candidate having the smallest value of the prediction error;
(G) learning error evaluation means for comparing the smallest prediction error value with an error determination index set in advance by a predetermined error determination method to determine whether the convergence is sufficient or insufficient. And
When the convergence is insufficient, the plurality of M is incremented by 1 based on the division pattern candidates selected by the minimum error relational expression selection unit, and a series of processing by the units (b) to (g) is performed. 5. The quality prediction apparatus in a manufacturing process according to claim 1, wherein the quality prediction device is an initial value of a coefficient of a local relational expression y _i calculated by repetition and a division pattern.   The manufacturing process is a steel process, and the quality variable is a product surface defect, mechanical strength characteristic value, shape flatness, product size, internal stress, or a process value that affects the quality. The quality prediction apparatus in the manufacturing process of any one of Claims 1-5 characterized by the above-mentioned.   The manufacturing process is a hot rolling process of a thin steel plate that is a steel product, and the quality is a change amount of a strip width on a rough rolling exit side and a strip width of a rolled material on a rolling-in side, and the manufacturing The operation variable of the process is selected from at least one of a target plate width, a target plate thickness, an equivalent carbon amount, a finish rolling target temperature, and a scraping target temperature of the material to be rolled. Quality prediction device in the manufacturing process.   The manufacturing process is a hot rolling process of a thin steel plate that is a steel product, the quality is a scraping temperature at the outlet side of the hot rolled runout table, and an operation variable of the manufacturing process is a material to be rolled. C amount, Si amount, Mn amount, P amount, S amount, Cu amount, Ni amount, Cr amount, Mo amount, Nb amount, V amount, Ti amount, B amount in molten steel at the time of finishing the refining process, Al amount, N amount, O amount, Ca amount, target strip width, target plate thickness, equivalent carbon amount, finish rolling target temperature, scraping target temperature, rolling speed, cooling water amount density, cooling water temperature The quality prediction apparatus in the manufacturing process according to claim 6, wherein at least one or more is selected. The relational expression y representing the relationship between the operation variable and the quality variable in the manufacturing process is expressed as a local expression representing the relationship between the operation variable and the quality variable in each local region i obtained by dividing the operation variable space with the operation variable as a basis vector. a relational expression y _i, of the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i, the sum of all of said local area i A quality prediction method in a manufacturing process for calculating a quality prediction value by inputting operation data, which is an actual value of an operation variable, into the relational expression y,
A step of inputting operation data using operation data input means for taking out and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Storing an activity function using storage means;
Based on the activity function [Phi _i in the respective local region i and the operational data, the contribution ratio with the contribution ratio calculating means for calculating a contribution ratio of each of the local region i to quality prediction value for該操Industry Data A calculating step;
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i A step of calculating a local relational expression using a relational expression calculating means;
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value Calculating a quality predicted value using a value calculating means;
Outputting a quality prediction value using quality prediction value output means for outputting the quality prediction value to the outside;
When the quality of the product is measured, the step of inputting the quality data using quality data input means for taking out and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, And a step of updating the local relational expression using local relational expression updating means for updating the value of the coefficient of the local relational expression y _i . The relational expression y representing the relationship between the operation variable and the quality variable in the manufacturing process is expressed as a local expression representing the relationship between the operation variable and the quality variable in each local region i obtained by dividing the operation variable space with the operation variable as a basis vector. a relational expression y _i, of the product of the activity function [Phi _i representing the contribution to equation y at each point on the entire operating variable space of the topical relation y _i, the sum of all of said local area i A program for calculating a quality prediction value by inputting operation data, which is an actual value of an operation variable, into the relational expression y,
Operation data input processing for extracting and inputting predetermined plural p operation data of the prediction target product from the manufacturing process;
A predetermined first function form representing a relationship between an operation variable and a quality variable in each local area i based on a division pattern composed of division coordinate information in each local area i obtained by dividing the operation variable space into a plurality of M pieces. An activity function that calculates and stores a coefficient of an activity function Φ _i of a predetermined second function form that represents the contribution ratio of the local equation y _{i to} the equation y at each point on the entire operation variable space. Amnestics,
Based on the activity function [Phi _i in the respective local region i and the operation data, and the contribution ratio calculation process for calculating the contribution ratio of each of the local region i to quality prediction value for該操industry data,
The stored to calculate the initial values of the coefficients of the local equation y _i in each local region i, based on the coefficients of the operational data and said local equation y _i, topical performing the calculation of the topical relation y _i Relational expression processing,
Quality prediction that calculates the sum of all the local areas i of the product of the calculation result of the local relational expression y _i and the contribution rate of each local area i to the quality prediction value for the operation data to obtain a quality prediction value Value calculation processing,
A quality prediction value output process for outputting the quality prediction value to the outside;
When the quality of the product is measured, a quality data input process for extracting and inputting the quality data from the manufacturing process;
A prediction error, which is a difference between the quality data and the quality prediction value, is calculated for each local region i, and based on the prediction error and the contribution rate of each local region i to the quality prediction value for the operation data, A program for causing a computer to execute a local relational expression update process for updating a coefficient value of the local relational expression y _i .   A computer-readable recording medium on which the program according to claim 10 is recorded.

Images