JP2008146621A - Device and method for analyzing quality improvement condition of product, computer program, and computer readable recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To easily analyze the quality of a product by displaying a combination of operating condition where the change of the quality to the change of the operation is relatively large, a combination of operation factors effective for quality improvement is extracted automatically, and a predetermined quality level is established, of all the combination of operation factors. <P>SOLUTION: The operating condition is divided into a plurality of ranges; an operating condition mesh obtained by combining them among a plurality of operation factors is created, and a quality index is calculated based on the probability density of the quality data in each operating condition mesh. The maximum value and minimum value are selected from the quality indices of operating condition meshes in the combination of all operation factors, and the combination of the operation factors with large influence factor calculated from the difference between the maximum and minimum values is selected and presented. A predetermined target quality index is compared with the quality index of the operating condition mesh in the combination of the operation factors having a relatively large effect for quality improvement; the operating condition mesh achieving the target quality and the operating condition corresponding to the mesh are selected, and guidance is performed. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a product quality improvement condition analysis apparatus, an analysis method, a computer program, and a computer-readable recording medium. In particular, in order to obtain the desired quality by specifying the combination of operating factors and operating conditions that are highly related to the occurrence of quality mismatches and operating conditions in the entire manufacturing process where the quality is determined as the operating result, and further predicting the quality from the current operating conditions. The present invention relates to a technique suitable for use in guiding the operating conditions to an operator and an engineer who performs quality control.

  Conventionally, in the process where the quality is determined as the operation result, as the operation analysis method for analyzing the influence of the operation condition on the quality, a correlation analysis method that evaluates using a correlation coefficient between a single operation factor and quality, There is known a method of creating and evaluating a multiple regression model in which the operating factors of the above are input and the quality is output.

  Further, in the technique disclosed in Patent Document 1, in the steel process, the physical property value such as the carbon amount of the slab, the operation value such as the casting width, the temperature value of the cooling zone and the like are the operation factors, and the surface defects of the steel sheet are determined. A quality prediction apparatus using a multi-layer neural network as quality data learns the relationship between the quality data and the operation data and performs quality control diagnosis.

  Further, in the technique disclosed in Patent Document 2, a probability distribution of the frequency of occurrence of quality data in a divided region formed by dividing a range that each parameter of a plurality of operation factors of a manufacturing process can take and consisting of a division of each operation parameter. Then, by calculating the quality data value with a predetermined cumulative probability, it is possible to analyze the relationship between the operation data and the quality data that cannot be captured by the correlation analysis or the multiple regression model. Furthermore, when the operating condition of the product is determined from the relationship between the operating parameter and the quality obtained by the above analysis, the obtained quality can be predicted.

JP-A-6-304723 Japanese Patent No. 3733057

  In the method using the above correlation coefficient and multiple regression model, the correlation coefficient and regression are based on the precondition that the operation and quality data to be analyzed can be expressed by a single linear model in the entire operation range. A model is derived and analyzed. Therefore, when analyzing operation and quality data obtained from a process in which a plurality of quality mismatch factors having different characteristics exist, there is a problem that the relationship between the two cannot be correctly grasped.

  In the method disclosed in Patent Document 1, a model in which the relationship between operation factors and quality is learned using a multilayer neural network is created and applied to quality control diagnosis. However, the multilayer neural network has a problem that it is extremely difficult for humans to read what logic the control diagnosis is made based on, and the operator cannot judge the rationality of the control diagnosis result.

Furthermore, in the actual manufacturing process, it is rare that all the data of operating factors affecting quality can be collected, and the data used for factor analysis and quality control are almost the same operating conditions. In most cases, the data has a large variation including a lot of data having greatly different quality. When the method using the correlation coefficient or the multiple regression model is applied to data having such a large variation, there is a problem in that it is difficult to find operation conditions related to quality. Further, in the method disclosed in Patent Document 1, when learning the relation between the operation factor and the quality of the data with large variation with the multilayer neural network, the learning does not converge due to the variation and has sufficient accuracy. There was a problem that a control model could not be obtained.

  The technique disclosed in Patent Document 2 makes it possible to find the relation between operation and quality that cannot be found by correlation analysis or multiple regression analysis even for the above-mentioned data with large variations. For the combination of operating conditions, the operating conditions are divided into a plurality of areas, the probability density of the occurrence frequency of quality data in each divided area is derived, and the quality index that gives a predetermined cumulative probability is obtained. It is a technique to analyze the relationship between and quality. In other words, it is based on the idea that quality variation under the same operating conditions is regarded as probability density, and the probability density of quality changes due to changes in operating conditions. It is possible to find out.

  However, in an actual manufacturing process, there may be 200 to 300 or more operating factors that may cause quality mismatch, and the number of combinations of operating factors becomes enormous. With the method disclosed in 2, there is a practical problem that it takes a lot of time to analyze the relationship between the combination of each operation condition and the quality individually, and to find the operation condition effective for quality improvement from among them. .

  The present invention has been made in view of the above points, and for all combinations of operation factors, the change in the probability density of the quality with respect to the change in operation is relatively large, and the operation is effective for quality improvement. A combination of factors is automatically extracted, and a combination of operation conditions satisfying a predetermined quality level can be automatically guided. Accordingly, it is an object of the present invention to provide a product quality improvement condition analysis apparatus and analysis method capable of quickly executing a countermeasure action such as a change in operation conditions in response to occurrence of quality mismatch.

  The product quality improvement condition analysis device of the present invention analyzes the relationship between the product quality data and the operation data of a plurality of operation factors in the manufacturing process, and identifies the operation factor that is highly related to the quality mismatch of the product. And a product quality improvement condition analysis device for guiding operation conditions for improving quality, a data input means for inputting analysis data composed of product quality data and operation data of a plurality of operation factors in the manufacturing process; From the operation data of a plurality of operation factors input by the data input means, an operation factor combination creation means for creating all combinations of mutually different operation factors for a predetermined number, and the operation factor combination creation means Operation factor selection means for selecting one combination from all created combinations, and one combination selected by the operation factor selection means For each of a plurality of operating factors constituting the combination, the operating conditions are divided into a plurality of ranges based on the value of the operation data, and an operating condition mesh is created by combining the operating conditions between the operating factors. A quality data probability distribution calculating means for extracting quality data belonging to each operating condition mesh created by the operating condition mesh creating means and calculating a probability distribution of quality data in each operating condition mesh; and the quality Based on the probability distribution calculated by the data probability distribution calculating means, the quality index calculating means for calculating the quality index indicating the quality of the product quality in each operation condition mesh, and the quality in each operation condition mesh calculated by the quality index calculation means Based on the index, the quality impact is calculated by calculating the quality impact by quantifying the difference in the quality index due to the difference in operating conditions Calculation unit, operation factor selection unit, operation condition mesh creation unit, quality data probability distribution calculation unit, quality index calculation unit, and operation of the quality influence degree calculation unit created by the operation factor combination creation unit Based on the quality influence level of the operation combination evaluation means that sequentially performs all of the factor combinations and all the operation factor combinations created by the operation factor combination creation means, the combination of the operation factors having the large quality influence degree is improved in quality. Effective operating factor selection means to select as a combination of effective operating conditions, and for the combination of operating factors selected by the effective operating factor selection means, the quality index in each operating condition mesh calculated by the quality index calculation means, An operation condition mesh that achieves the target quality index by comparing with a preset target quality index, and the operation condition Quality improvement operation condition extraction means for extracting a range of operation conditions corresponding to the mesh, and a quality improvement guidance output means for outputting a combination of operation factors calculated by the quality improvement operation condition extraction means and the range of operation conditions It is characterized by that.

  The product quality improvement condition analysis method of the present invention includes a data input step of inputting analysis data composed of product quality data and operation data of a plurality of operation factors in the manufacturing process, and a plurality of data input in the data input step From the operation data of the individual operation factors, an operation factor combination creation step for creating all combinations of the operation factors different from each other in a predetermined number, and one of all combinations created in the operation factor combination creation step. An operation factor selection step for selecting one combination and a plurality of operation conditions for one combination selected in the operation factor selection step based on the value of the operation data for each of the plurality of operation factors constituting the combination. The operation condition mesh creating step of creating an operation condition mesh that is divided into a range of the above and combining them among a plurality of operation factors, and the operation condition The quality data belonging to each operation condition mesh created in the cache creation step is extracted, and the quality data probability distribution calculation step for calculating the probability distribution of the quality data in each operation condition mesh and the quality data probability distribution calculation step are calculated. Based on the probability distribution, the quality index calculation step for calculating the quality index indicating the quality of the product quality in each operation condition mesh, and the difference in the operation condition based on the quality index in each operation condition mesh calculated in the quality index calculation step A quality influence calculation step for calculating a quality influence degree obtained by quantifying the difference in quality index, and the operation factor selection step, the operation condition mesh creation step, the quality data probability distribution calculation step, the quality indicator calculation step, and the The quality impact calculation process is performed for all the combinations of operation factors created in the operation factor combination creation process. Based on the quality influence in the combination of operation factors evaluated step by step and the combination of all the operation factors created in the operation factor combination creation step, the combination of the operation factors having a large quality influence degree is the effective operating condition for quality improvement. The effective operation factor selection step selected as a combination, the quality index in each operation condition mesh calculated in the quality index calculation step for the combination of the operation factors selected in the effective operation factor selection step, and a preset target quality Compared with the index, the operation condition mesh that achieves the target quality index, and the quality improvement operation condition extraction step that extracts the range of the operation condition corresponding to the operation condition mesh, and the quality improvement operation condition extraction step were calculated It has a quality improvement guidance output step for outputting a combination of operation factors and a range of operation conditions.

The computer program of the present invention includes a data input procedure for inputting analysis data comprising product quality data and operation data of a plurality of operation factors in a manufacturing process, and a plurality of operation factors input in the data input procedure. From the operation data, select one combination from the operation factor combination creation procedure for creating all combinations of different operation factors for the number specified in advance and all combinations created in the operation factor combination creation procedure. For the operating factor selection procedure and one combination selected in the operating factor selection procedure, the operating conditions are divided into a plurality of ranges based on the magnitude of the value of the operation data for each of the plurality of operating factors constituting the combination. , An operating condition mesh creating procedure for creating an operating condition mesh combining the operating factors, and the operating condition message. The quality data belonging to each operation condition mesh created in the menu creation procedure is extracted, and the quality data probability distribution calculation procedure for calculating the probability distribution of the quality data in each operation condition mesh and the quality data probability distribution calculation procedure are calculated. Based on the probability distribution, the quality index calculation procedure for calculating the quality index indicating the quality of the product quality in each operation condition mesh, and the difference in the operation condition based on the quality index in each operation condition mesh calculated in the quality index calculation procedure The quality impact calculation procedure for calculating the quality impact level by quantifying the difference in quality index, and the operation factor selection procedure, the operation condition mesh creation procedure, the quality data probability distribution calculation procedure, the quality index calculation procedure, and the The quality impact calculation procedure is performed in order for all the combinations of operating factors created in the operating factor combination creating procedure. The combination of operation conditions that are effective in improving the quality of the combination of operation factors with a large quality influence based on the evaluation of the operation combination to be performed and the quality influence on the combinations of all the operation factors created in the operation factor combination creation procedure. For the combination of the operating factor selected in the effective operating factor selection procedure and the operating factor selected in the effective operating factor selection procedure, the quality index in each operating condition mesh calculated in the quality index calculating procedure, and a preset target quality index And the operation condition mesh that achieves the target quality index, the quality improvement operation condition extraction procedure that extracts the range of the operation condition corresponding to the operation condition mesh, and the operation calculated by the quality improvement operation condition extraction procedure A computer executes a quality improvement guidance output procedure for outputting a combination of factors and a range of operating conditions.

  The computer-readable recording medium of the present invention records the computer program of the present invention.

  According to the present invention, when a product quality mismatch occurs, an operation factor highly related to the quality mismatch is quickly identified from a combination of several hundred operating factors, and a predetermined quality level is achieved. Operating conditions can be obtained. Therefore, it is possible to obtain the effects of improving the quality, improving the yield, and avoiding the delivery delay of the product to the customer by quickly executing the countermeasure action such as changing the operation range using the analysis result. it can.

(First embodiment)
A product quality improvement condition analysis apparatus, analysis method, computer program, and computer-readable recording medium according to the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration of a product quality improvement condition analysis apparatus 1 according to the present embodiment.

  100 in the figure is a data input unit as a data input means. This data input unit 100 is an operation data of a manufacturing process and a quality corresponding to the operation from a keyboard, an OCR for reading a data sheet, or a LAN in a factory. Enter the data. The data input unit 100 is composed of a CPU or MPU of a computer, a RAM, a ROM, and the like, and inputs operation data and quality data to a memory such as a RAM, for example.

  For example, in the steel process, the operation data is a fluctuation amount of the molten metal surface in a continuous casting process, a heating furnace temperature in a hot rolling process, and the like, and is given as a continuous value. When a plurality of p operation factors x1, x2,..., Xp are given for N products, the input operation data is a matrix of N rows and p columns.

  The quality data is, for example, the number of surface defects per steel plate coil manufactured by a steel process, and is given as a continuous value. When N cases of quality data are given corresponding to the operation data, the input quality data is an N-dimensional vector.

  The data input unit 100 creates analysis data by associating the input N case quality data and operation data, and stores information such as names of operation factors and the like in computer RAM, ROM, or magnetic storage device. Save it in the form of a file on the device.

  Reference numeral 101 denotes an operation factor combination creation unit as an operation factor combination creation unit, which creates a combination list of all the operation factors corresponding to a preset number r of operation factors to be combined. The operation factor combination creating unit 101 is constituted by a CPU or MPU of a computer, a RAM, a ROM, etc., and is implemented by operating a program recorded in the RAM or ROM. For example, considering the case where the number of operating factors is p = 100 items and the number of combinations of operating factors is r = 2, the combinations to be evaluated are 4950 different combinations in which two different ones are extracted from 100 different ones. It becomes.

  The operation factor combination creation unit 101 reads the name information of the operation factors created by the data input unit 100, creates a list of combinations of these 4950 operation factors, and can use this combination list in subsequent processing. The file is stored in the form of a file in a storage device such as a RAM or ROM of a computer or a magnetic storage device.

  Reference numeral 102 denotes an operation combination evaluation unit as an operation combination evaluation unit. The operation combination evaluation unit 102 includes an operation factor selection unit 103, an operation condition mesh creation unit 104, a quality data probability distribution calculation unit 105, a quality index calculation unit 106, and a quality influence calculation unit 107. The operation combination evaluation unit 102 executes a predetermined process on all the rows of the operation factor combination list, thereby creating a list in which the quality influence level described later is added to each row of the operation factor combination list. It is stored in the form of a file in a storage device such as a RAM or ROM of a computer or a magnetic storage device so that it can be used in processing.

  Reference numeral 103 denotes an operation factor selection unit as operation factor selection means. The operation factor selection unit 103 selects and reads one combination sequentially from the top of the operation factor combination list created by the operation factor combination creation unit 101, and operates the operation data and quality of the operation factor described in each row of the list. Data is extracted from the analysis data created by the data input unit 100.

  Reference numeral 104 denotes an operation condition mesh creation unit as an operation condition mesh creation means. The operation condition mesh creation unit 104 uses a plurality of operation data constituting a combination selected based on the list by the operation factor selection unit 103, and sets a plurality of operation conditions in an operation factor space based on each operation factor. A process of dividing the area into individual ranges is performed.

  As a specific dividing means, for example, there is a method in which the maximum value and the minimum value are calculated from the data of each operating factor, and the dividing point is determined so as to equally divide the interval between them by the preset operating condition dividing number m. However, the actual operation data is mostly biased to a specific range, such as the manufacturing conditions of products with a large production volume. If the operation range is simply divided equally, the data per mesh is significantly biased. In some cases, a large number of meshes whose quality probability distributions cannot be evaluated are generated.

  In order to solve this problem, it is effective to determine the division points so that the number of data existing in each division obtained by dividing the operation condition is substantially equal. Specifically, for example, a value obtained by dividing the total number of data N by the operating condition division number m is obtained, the number of data to be included in one division obtained by dividing each operating condition is calculated, and then the operation data is sorted. And the operation data of the array index corresponding to an integer multiple of the number of data to be included in one division is obtained and used as a division point. FIG. 2 shows a case where the operating conditions of each operating factor are equally divided into a plurality of ranges (regions) (a) and an unequal division so that the number of data is equal (b). The operation condition mesh created for is shown.

  As described above, the operation condition mesh creating unit 104 divides the operation condition into a plurality of ranges based on the value of the operation data, and creates an operation condition mesh in which these are combined among a plurality of operation factors. Further, the operation condition mesh creation unit 104 creates an operation condition mesh based on the determined division point information of each operation factor, assigns a serial number to the plurality of operation condition meshes, and stores the division point information together with the computer's information. The file is stored in the form of a file in a storage device such as a RAM, ROM, or magnetic storage device.

  Reference numeral 105 denotes a quality data probability distribution calculating unit as quality data probability distribution calculating means. The quality data probability distribution calculation unit 105 reads out information on the operation condition mesh created by the operation condition mesh creation unit 104, and extracts and selects corresponding quality data for each operation condition mesh assigned with a serial number. . Next, using the selected quality data, a process of calculating a probability distribution of the quality data based on the frequency distribution is performed.

  Specifically, for example, the frequency of occurrence of quality is analyzed in advance with a frequency distribution, a predetermined probability density function suitable for approximating this is set, and the probability density function is calculated using the selected quality data. There are ways to determine the parameters. As a probability density function suitable for approximating the occurrence frequency of quality, there is an exponential distribution function represented by the equation (1).

  Here, y is quality data, e is the base of natural logarithm, and λ is a parameter of the exponential distribution function. For example, when the number of surface defects of a thin steel sheet coil, which is a steel product, is used as quality data, the frequency of occurrence of a coil having a number of defects of 0 is generally the highest. A distribution function is suitable. Another probability density function suitable for approximating the occurrence frequency of quality is a normal distribution function represented by Expression (2).

  Here, μ is an average, σ is a standard deviation, and these two are parameters determined by quality data. Furthermore, as another probability density function suitable for approximating the frequency of occurrence of quality, there is a lognormal distribution function represented by Expression (3).

  Here, ln is a natural logarithm, μ ′ is a logarithmic average, and σ ′ is a logarithmic standard deviation. In this case, μ ′ and σ ′ are parameters to be determined by the quality data. To determine which probability density function should be used in actual analysis, create a frequency distribution of quality occurrence frequency in advance using all quality data, and select the probability density function of the distribution form closest to the frequency distribution. It ’s fine. In addition, the probability density function used in the analysis of the present invention is not limited to the above function, and is a case where a probability density function proposed in the field of probability statistical theory such as Poisson distribution or binomial distribution is used. There may be.

  FIG. 3 is a diagram illustrating a case where the probability density distribution approximating the frequency distribution of quality in each mesh of the operation condition mesh is an exponential distribution, taking a two-dimensional operation factor space as an example. FIG. 3 shows a frequency distribution in which one part of a mesh is extracted from the operating condition mesh, the horizontal axis indicates quality data values such as the number of surface defects, and the vertical axis indicates the frequency. An exponential distribution is selected as a probability density distribution that approximates the frequency distribution, and this exponential distribution function is indicated by a broken line.

  Further, the quality data probability distribution calculation unit 105 calculates the probability density function parameters of each operation condition mesh using the set probability density function and the quality data of each operation condition mesh, and corresponds to the serial number of the operation condition mesh. The file is stored in the form of a file in a storage device such as a RAM or ROM of a computer or a magnetic storage device.

  Reference numeral 106 denotes a quality index calculation unit as quality index calculation means. The quality index calculation unit 106 reads out the serial number of each operation condition mesh and the parameters of the probability density function created by the processing up to the quality data probability distribution calculation unit 105, and the cumulative probability equal to a preset quality evaluation probability value. Quality data value (quality index) is calculated.

  FIG. 4 is a diagram schematically illustrating the principle of the calculation for deriving the quality index performed by the quality index calculation unit 106, taking as an example the case where the exponential distribution is used for the probability density distribution. The parameter λ of the probability density function corresponding to the operation condition mesh to be evaluated is read, and the distribution of the probability density f (y) with respect to the quality y is obtained.

  At this time, the quality y integrated in the range of 0 to Y0 is defined as cumulative probability in the field of probability statistics. This cumulative probability corresponds to the probability that the quality can be in the range from 0 to Y0. When Y0 is infinite, that is, when all values that the quality can take are included, the cumulative probability is 100%. The quality index Y0 obtained so that this cumulative probability becomes equal to a preset quality evaluation probability value α (for example, 80%) means the worst quality index that can be generated with the probability α in this operating condition mesh. . Note that it is proved from the probability statistical theory that, when the exponential distribution function of Expression (1) is assumed, the quality index Y0 with the cumulative probability α is calculated using Expression (4).

  Therefore, the quality index calculation unit 106 calculates, for each operation condition mesh, a quality index Y0 corresponding to the probability α based on the equation (4) and displaying the quality of the product quality, and the serial number of the operation condition mesh. The file is saved in the form of a file in the form of a file in a RAM or ROM of a computer or a storage device such as a magnetic storage device.

  In the case where the normal distribution function, the log normal distribution function, and other probability density functions are used, the calculation formula of the quality index Y0 corresponding to the probability density function used in place of the expression (4) is similarly used. Process. FIG. 5 is a diagram illustrating the distribution of quality indices in the operation condition mesh using a two-dimensional operation factor space as an example. FIG. 5 shows the quality index Y0 calculated for each mesh of the operation condition mesh, and the level of the quality index Y0 is displayed corresponding to the shade of the color.

  Reference numeral 107 denotes a quality influence degree calculation unit as quality influence degree calculation means. The quality influence calculation unit 107 reads the operation condition mesh and the calculation result of the quality index in a certain combination of operation factors obtained by the processing up to the quality index calculation unit 106, and obtains the maximum value and the minimum value of the quality index. Then, the process of adding information to the corresponding row of the operation factor combination list is performed using the difference as the degree of influence in the combination of the operation factors.

  Here, the difference amount of the quality index in the operation condition mesh is an index that reflects how much the quality occurrence frequency changes when the operation condition is variously changed in the combination of the operation factors. That is, a combination of operation factors having a large influence degree is a combination of operation factors effective in quality improvement because the frequency of occurrence of quality is changed in both a good direction and a bad direction by changing operation conditions. On the other hand, in the combination of operation factors having a small influence degree, the occurrence frequency of quality does not change relatively even if the operation conditions are changed. As described above, the quality influence calculation unit 107 calculates the quality influence obtained by quantifying the difference in the quality index due to the difference in the operation condition based on the quality index in each operation condition mesh.

  Reference numeral 108 denotes an effective operation factor selection unit as effective operation factor selection means. The effective operation factor selection unit 108 reads the list to which the quality influence degree calculated by the quality influence degree calculation unit 107 is added, and sorts the list so that the combination of the operation factors having the large quality influence degree is higher. Next, a combination of operation factors is selected from the list in descending order of influence based on a preset operation factor selection criterion index.

  As the selection criteria for specific operation factors, for example, all the combinations of operation factors that are equal to or higher than a preset criterion in which the difference from the target quality index of the quality index under the influence degree, that is, the quality evaluation probability is α, are selected. There is a way. Alternatively, for example, there is a method in which the number q of combinations of operation factors to be selected is set in advance, and q sets of combinations of operation factors are selected in descending order of influence.

  As described above, the effective operation factor selection unit 108 selects a combination of operation factors having a large quality influence degree as a combination of operation conditions effective for quality improvement. Combinations of operation factors that have cleared the operation factor selection criteria, and information on the division position and quality index of the operation condition mesh in each combination, such as computer RAM, ROM, or magnetic storage device, etc., can be used in subsequent processing Save to the storage device in the form of a file.

  Reference numeral 109 denotes a quality improvement condition extraction unit as quality improvement condition extraction means. The quality improvement condition extraction unit 109 sequentially reads out the combination of the operation factors that have cleared the operation factor selection criteria created by the effective operation factor selection unit 108, the operation condition mesh, and the quality index, and sets the preset target quality The index and the quality index in each operation condition mesh are compared, and the operation condition mesh achieving the target quality index is specified. As a method of comparing the target quality index set in advance with the quality index in each operation condition mesh, the difference between the two is calculated and compared.

  Further, the quality improvement condition extraction unit 109 extracts the division position information of the operation condition mesh and stores a storage device such as a RAM or ROM of a computer or a magnetic storage device as a range of operation conditions for realizing the target quality index. Save it in the form of a file.

  Reference numeral 110 denotes a quality guidance output unit as quality guidance output means. The quality guidance output unit 110 performs a process of presenting a combination of operation factors and a range of operation conditions for realizing the target quality index created by the quality improvement condition extraction unit 109 to an operator or a person in charge of quality control work. Do. Specifically, output processing such as displaying on a display device (such as a CRT display) for displaying information or transmitting information by e-mail is performed. Based on the presented information, the operator and the person in charge can select the one judged to be effective from the presented operational conditions for quality improvement, and can quickly take a countermeasure action. Further, the quality guidance output unit 110 may display the distribution of the quality index in the two-dimensional operation factor space as shown in FIG. 5 on the display device.

  Next, a product quality improvement condition analysis method according to the present embodiment will be described with reference to a process flowchart shown in FIG.

  First, the process proceeds to the data input process of S201. This data input process is performed by the data input unit 100 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Here, the operation data of the manufacturing process and the quality data corresponding to the operation are input, and both are linked to create analysis data.

  Next, it progresses to the operation factor combination preparation process of S202. This operation factor combination creation step is performed by the operation factor combination creation unit 101 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Using the analysis data input in the data input step, a process of creating a combination list of all operation factors corresponding to a preset number r of operation factors to be combined is performed.

  Next, it progresses to the operation factor selection process of S203. This operation factor selection step is performed by the operation factor selection unit 103 in FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Select one combination from the list of operating factor combinations.

  Next, it progresses to the operation condition mesh preparation process of S204. This operation condition mesh creation step is performed by the operation condition mesh creation unit 104 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Here, a plurality of operation data selected based on the list in the operation factor selection step S203 is used to perform a process of dividing the operation condition into a plurality of ranges in the operation factor space based on each operation factor. .

  Next, the process proceeds to the quality data probability distribution calculation step of S205. This quality data probability distribution calculation step is performed by the quality data probability distribution calculation unit 105 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Information on the operation condition mesh created in the operation condition mesh creation step S204 is read out, and corresponding quality data is selected for each of the operation condition meshes.

  Further, a process for calculating a probability distribution is performed using the selected quality data. Specifically, for example, a predetermined probability density function suitable for approximating the occurrence frequency of quality is set in advance, and parameters of the probability density function are determined using the selected quality data.

  Next, the process proceeds to a quality index calculation step in S206. This quality index calculation step is performed by the quality index calculation unit 106 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. The serial number of each operation condition mesh and the parameters of the probability density function created in the processing up to the quality data probability distribution calculation step S205 are read out, and the quality data value (quality index) that becomes the cumulative probability equal to the preset quality evaluation probability value ) Is calculated.

  Next, the process proceeds to the quality influence degree calculation step of S207. This quality influence degree calculation step is performed by the quality influence degree calculation unit 107 in FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Using the operation condition mesh and the calculation result of the quality index obtained in the process up to the quality index calculation step S206, the maximum value and the minimum value of the quality index are obtained, and the difference between these values is calculated. A process of calculating the influence level in the combination is performed.

  Next, processing for determining whether or not the processing from S203 to S207 has been completed for all combinations of operating factors is performed. If not completed yet, the processing after the operating factor selection step S203 is executed again. Here, the process of executing the processing after S203 again is referred to as an operation combination evaluation process. If the process has been completed for all combinations of operation factors, the process proceeds to the effective operation factor selection step S208.

  Next, the process proceeds to the effective operation factor selection step of S208. This effective operation factor selection step is performed by the effective operation factor selection unit 108 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. For all the operating factor combinations, sorting is performed so that the combination of operating factors having a large quality influence level is higher. Next, processing for selecting combinations of operation factors from the list in descending order of influence is performed based on a preset operation factor selection criterion index.

  Next, it progresses to the quality improvement condition extraction process of S209. This quality improvement condition extraction step is performed by the quality improvement condition extraction unit 109 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment. Using the combination of operation factors that cleared the operation factor selection criteria, information on the operation condition mesh, and the quality index, the target quality index is achieved by comparing the preset target quality index with the quality index in each operation condition mesh. Specify the operating condition mesh. Next, a process for extracting the division position information of the operation condition of the operation factor corresponding to the operation condition mesh and calculating as a range of the operation condition for realizing the target quality index is performed.

  Next, it progresses to the quality improvement guidance output process of S210. This quality improvement guidance output step is performed by the quality improvement guidance output unit 110 of FIG. 1 in the product quality improvement condition analysis apparatus of the present embodiment.

  In the embodiment of the present invention described above, the operation condition is divided into a plurality of ranges, and the quality index is calculated based on the probability density of the quality data in each operation condition mesh created by combining the operation conditions among a plurality of operation factors. By calculating, it is possible to evaluate the relationship between quality and operation even from data with large variations.

  In addition, for all combinations of operating factors, select the maximum and minimum values from the quality index of each operating condition mesh, and select and present the combination of operating factors that have a large influence calculated from the difference. Thus, it is possible to obtain a combination of operation factors that are relatively effective for quality improvement.

  In addition, the operating condition mesh that achieves the target quality by comparing the quality index of the operating condition mesh in the combination of the predetermined target quality index and the operating factor that is relatively effective for quality improvement, and the operation corresponding to that mesh It is possible to automatically search for conditions and provide guidance.

(Second Embodiment)
In the second embodiment, an example in which the result of probability distribution correlation analysis in the quality data probability distribution calculation unit 105 is used as a quality prediction model as a specific example and application example of the first embodiment described above. To do. As shown in FIG. 7, the product quality improvement condition analysis apparatus 1 according to the second embodiment includes an online data input unit 120 that inputs operation data (online data) to be predicted, and a quality data probability distribution calculation unit 105. A prediction for calculating a quality index that is predicted to occur at a predetermined probability for the operation data to be predicted input by the online data input unit 120 using the probability distribution of the quality data in each operation condition mesh calculated in And a unit 121. In addition, the same code | symbol is attached | subjected to the component already demonstrated by 1st Embodiment mentioned above, and the detailed description is abbreviate | omitted.

  The quality data probability distribution calculation unit 105 calculates the probability distribution of quality data in each operation condition mesh. Therefore, as shown in Table 1 below, for each operation condition mesh serial number (mesh No.), the analysis result is saved as a text file or the like in the range of each operation condition and the corresponding probability density function parameter format. And can be used as a prediction model.

  As a calculation procedure using the prediction model in the prediction unit 121, first, the operation condition (for example, current operation condition) input from the online data input unit 120 is compared with the analysis result of Table 1, and the corresponding mesh No. Search for. Next, the corresponding mesh No. The parameter of the probability density function corresponding to is taken out. Here, it is assumed that an exponential distribution function is used, and the parameter is λ. In this operation condition mesh, since the quality index follows the exponential distribution function expressed by the equation (1), the maximum value of the quality index generated at a certain quality evaluation probability value (for example, 80%) is calculated. Then, the calculation result is output as a quality index (predicted score value) with a probability α with respect to the input operation condition.

  An example of calculation is shown. FIG. 8 shows a prediction result when three input variables (operation factors) in the prediction model are changed. Here, the operation factor u1 is changed from 65 to 80 to 125 (see FIG. 8A), the operation factor u2 is changed from 2000 to 2500 to 3000 (see FIG. 8B), and the operation factor u3 is changed. Is constant (see FIG. 8C) (each operating factor includes a noise component). Here, the steel process is assumed, and the predicted score value is the number of predetermined defects. The horizontal axis of each figure is a steel plate coil number. As shown in FIG. 8D, it can be seen that the corresponding operation condition mesh changes with the change of the operation condition, and the predicted score value at the probability of 95% estimated from the probability density function changes.

  Next, an example will be described in which an operation condition for realizing a quality index desired to be a target is searched by applying a result (Table 1) of probability distribution correlation analysis in the quality data probability distribution calculation unit 105. As a calculation procedure, first, a quality index Y0 and a probability α to be targeted are determined (for example, an operation condition in which a predicted score value is “1” or less with a probability of 95% is searched). Next, mesh Nos. On the other hand, a quality index Yi having a probability α is calculated using a probability density function. Then, the quality index Y0 desired to be targeted is compared with the quality index Yi at the probability α of each operation condition mesh, and an operation condition mesh that is equal to or less than the target value is searched. The operation condition corresponding to the operation condition mesh obtained as a result is presented as an improved operation plan. At this time, the operation condition meshes that can be gathered together may be presented together. Table 2 shows the results of searching for operating conditions with a probability score of 95% and a predicted score value of 1 or less.

  Next, an example of processing in the quality guidance output unit 110 will be described. When it is determined that the relationship between two operational factors and quality is worth presenting to the operator as guidance, when a button (not shown) set as a prediction model is pressed, a guidance display is displayed as shown in FIG. Done.

  As shown in FIG. 9, in the guidance display, the distribution of the quality index in the two-dimensional operation factor space is displayed, and the quality index is color gradation.

  In addition, the prediction unit 121 predicts a quality index for the current operation condition, and a mark is displayed on the guidance display.

  Here, the guidance display shown in FIG. 9A is quality prediction and recovery operation guidance by a combination of the own process operations (own process operations 1 and 2). The quality is predicted from the combination of the current operating conditions in the own process. The operator can know whether the current operating conditions are good from the viewpoint of quality or which direction (the improved operating direction indicated by the arrow in the figure) should be changed to improve the quality.

  The guidance display shown in FIG. 9B is quality prediction and recovery operation guidance by a combination of the upper process operation and the own process operation. The quality is predicted from a combination of the operating conditions in the upper process determined as the actual results and the current operating conditions in the own process. Although the performance in the upper process cannot be changed, the operator can find the operation change direction (recovery operation direction indicated by an arrow in the figure) of the own process for recovering the quality.

  The guidance display shown in FIG. 9C is quality prediction guidance based on a combination of upper process operations (upper process operations 1 and 2). The quality is predicted from the combination of the operating conditions in the upper process determined as the actual results. In other words, the quality predicted based on the operation results in the upper process is displayed as guidance as feedforward information. Since the recovery action is difficult, an alert for prevention of outflow of the priority inspection target or the like may be output. In addition, this information will be presented to the upper process, and measures will be taken to improve quality through feedback, such as prompting measures such as changing the operating standard.

  In any of the guidance displays, an alert is issued to take an outflow prevention action, such as a priority inspection target, for coils that are predicted to be of poor quality due to circumstances such as a change in operation not in time. May be equal.

  Next, the quality of the number of surface flaws on thin steel sheets for automobiles, which are steel products, and the operation data on 200 operating factors in the continuous casting process, hot rolling process, cold rolling process, and hot dip galvanizing process. The example applied to the product quality improvement condition analysis apparatus will be described. The quality improvement condition analysis method according to the present embodiment is realized by a program on a computer.

  The data used for the analysis uses the operation data stored in the database 112 of the database server and the number data of the surface defects measured by the automatic defect inspection device in the hot dip galvanizing process. This database 112 is obtained by processing operation and quality data measured in each manufacturing process into a coil unit or a length unit, and further, a coil number for use in a search, a manufacturing date and time in each process, manufacturing Steel type codes for distinguishing specifications are stored in a given format.

  When surface flaws that cause quality problems occur in the hot dip galvanizing process, the operator or the person in charge of quality control operations can specify the steel type code to be analyzed and the range of production date and time for the quality improvement conditions for this product. By inputting the data into the analysis device, the data input unit 100 obtains operation and quality data matching the specified conditions from the database server via the LAN, or the process computer 111 receives an instruction from the data input unit 100 and specifies it. Send operational and quality data that meets your requirements.

  In the analysis, the number of combinations of operation factors can be selected within a range of 1 to 10, and the number of divisions of operation conditions can be set within a range of 3 to 20. Furthermore, the frequency of the number of occurrences was analyzed in advance for various surface defects of hot-dip galvanized steel sheets for automobiles, and the probability density function was the exponential distribution of equation (1), normal distribution of equation (2), You can select and analyze one of the lognormal distributions. As for the quality evaluation probability value α, a probability of 80% is set as a standard value based on various analysis results, but the analyst can arbitrarily change it within a range of 10% to 99%.

  Regarding the selection of operation factors, either the case of presenting all combinations of operation factors that have a certain degree of influence exceeding a certain threshold value or the case of designating the number of combinations of operation factors to be selected can be selected. Furthermore, for the target quality index, select either the total number of surface defects per coil or the number density of defects per unit area, and the specific target quality index can be set to 0 or a positive numerical value. It was supposed to be possible.

  The above-mentioned product quality improvement condition analysis device enables the operation factors highly related to the occurrence of surface flaws to the operators of continuous casting process, hot rolling process, cold rolling process, hot dip galvanizing process, and the person in charge of quality control work. As a result of implementing and operating a system that presents operating conditions that improve quality, the effects of reducing surface defects, improving product yield, saving product maintenance, avoiding delays in delivery due to quality problems, etc. Can be obtained.

  In the present embodiment, the present invention is realized as a program on a computer. However, the present invention may be configured by hardware combining an arithmetic device, a memory, and the like. In addition, the analysis device of the present invention may be composed of a plurality of devices or a single device.

  Further, the above-described embodiment is configured by a CPU or MPU of a computer, a RAM, a ROM, and the like, and is implemented by operating a program recorded in the RAM or ROM. Therefore, means for supplying software program codes for realizing the functions of the above-described embodiments to the computer, for example, storage media storing such program codes are included in the scope of the present invention.

It is a figure which shows the structure of the product quality improvement condition analysis apparatus which concerns on the 1st Embodiment of this invention. Created for a case where the operating conditions of each operating factor are equally divided into a plurality of ranges (regions) (a) and a case where the number of data is equally divided (b) taking a two-dimensional operating factor space as an example It is a figure which shows the operation condition mesh. It is a figure for demonstrating the probability density distribution of the quality in each mesh of an operation condition mesh taking a two-dimensional operation factor space as an example. It is a schematic diagram which calculates the quality data value used as a predetermined quality evaluation probability value in an exponential distribution function. It is a figure for demonstrating the quality parameter | index in an operation condition mesh taking a two-dimensional operation factor space as an example. It is a process flowchart of the quality improvement condition analysis method of the product which concerns on embodiment of this invention. It is a figure which shows the structure of the product quality improvement condition analysis apparatus which concerns on the 2nd Embodiment of this invention. It is a figure for demonstrating the example of the prediction result using a prediction model. It is a figure for demonstrating the example of a guidance display.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Data input part 101 Operation factor combination preparation part 102 Operation combination evaluation part 103 Operation factor selection part 104 Operation condition mesh preparation part 105 Quality data probability distribution calculation part 106 Quality index calculation part 107 Quality influence calculation part 108 Effective operation factor selection part 109 Quality improvement condition extraction unit 110 Quality improvement guidance output unit 111 Process computer 112 Database 120 Online data input unit 121 Prediction unit

Claims (14)

A data input means for inputting analysis data comprising product quality data and operation data of a plurality of operation factors in the manufacturing process;
From the operation data of a plurality of operation factors input by the data input means, operation factor combination creation means for creating all combinations of mutually different operation factors for a predetermined number,
An operation factor selection means for selecting one combination from all the combinations created by the operation factor combination creation means;
For one combination selected by the operation factor selection means, for each of a plurality of operation factors constituting the combination, the operation condition is divided into a plurality of ranges based on the magnitude of the value of the operation data, and this is divided into a plurality of operations. An operating condition mesh creating means for creating an operating condition mesh combined between factors;
Quality data probability distribution calculating means for extracting quality data belonging to each operating condition mesh created by the operating condition mesh creating means, and calculating a probability distribution of quality data in each operating condition mesh;
Based on the probability distribution calculated by the quality data probability distribution calculating means, quality index calculating means for calculating a quality index indicating the quality of the product quality in each operation condition mesh;
Based on the quality index in each operating condition mesh calculated by the quality index calculating means, a quality influence degree calculating means for calculating a quality influence degree by quantifying the difference in the quality index due to the difference in the operating condition;
The operation factor selection means, the operation condition mesh creation means, the quality data probability distribution calculation means, the quality index calculation means, and the quality influence calculation means process of the combination of the operation factors created by the operation factor combination creation means. Operation combination evaluation means to be performed sequentially for all,
Effective operating factor selection means for selecting a combination of operating factors having a large quality influence level as a combination of operating conditions effective for quality improvement based on the quality influence level in all the operating factor combinations created by the operating factor combination creating means When,
For the combination of operation factors selected by the effective operation factor selection means, the quality index in each operation condition mesh calculated by the quality index calculation means is compared with a preset target quality index, and the target quality index is determined. Quality improvement operation condition extraction means for extracting the operation condition mesh to be achieved and the range of operation conditions corresponding to the operation condition mesh;
A quality improvement condition analysis apparatus for a product, comprising quality improvement guidance output means for outputting a combination of operation factors calculated by the quality improvement operation condition extraction means and a range of operation conditions. The quality data probability distribution calculating means approximates the frequency distribution of the quality data in the operation condition mesh using a predetermined probability density function,
2. The quality index calculating means calculates a quality data value having a predetermined cumulative probability based on a probability density function obtained by the approximation process, and uses the quality data value as the quality index in the operation condition mesh. Product quality improvement condition analysis device described in 1.   The quality influence calculation means selects the maximum value and the minimum value from the quality indices in all the operation condition meshes calculated by the quality index calculation means, calculates the difference between the quality influence degree in the combination of the operation factors and The product quality improvement condition analysis apparatus according to claim 1 or 2, wherein   The operating condition mesh creating means determines a dividing range so that the number of data existing in each division of the operating conditions divided into a plurality of ranges is equal, and creates an operating condition mesh combined between a plurality of operating factors. The product quality improvement condition analysis apparatus according to any one of claims 1 to 3. A data input means for inputting operation data to be predicted;
The probability distribution of the quality data in each operation condition mesh calculated by the quality data probability distribution calculation means is used as a prediction model, and the prediction target operation data input by the data input means is predicted to occur with a predetermined probability. 5. The product quality improvement condition analysis apparatus according to claim 1, further comprising: a predicting unit that calculates a quality index.   The product quality improvement condition analysis apparatus according to any one of claims 1 to 5, wherein the quality improvement guidance output means displays a distribution of quality indices in a two-dimensional operation factor space on a display device. . A data input process for inputting analysis data comprising product quality data and operation data of a plurality of operation factors in the manufacturing process;
From the operation data of a plurality of operation factors input in the data input step, an operation factor combination creation step of creating all combinations of mutually different operation factors for a predetermined number,
An operation factor selection step of selecting one combination from all the combinations created in the operation factor combination creation step;
For one combination selected in the operation factor selection step, for each of a plurality of operation factors constituting the combination, the operation condition is divided into a plurality of ranges based on the value of the operation data, and this is divided into a plurality of operations. An operation condition mesh creation process for creating an operation condition mesh combined between factors,
Quality data probability distribution calculation step for extracting quality data belonging to each operation condition mesh created in the operation condition mesh creation step, and calculating a probability distribution of quality data in each operation condition mesh,
Based on the probability distribution calculated in the quality data probability distribution calculating step, a quality index calculating step for calculating a quality index indicating the quality of product quality in each operation condition mesh;
Based on the quality index in each operation condition mesh calculated in the quality index calculation step, a quality influence degree calculation step for calculating a quality influence degree by quantifying the difference in the quality index due to the difference in the operation condition,
The operation factor selection step, the operation condition mesh creation step, the quality data probability distribution calculation step, the quality index calculation step, and the quality influence calculation step for all the combinations of operation factors created in the operation factor combination creation step Operation combination evaluation process to be performed sequentially,
Effective operation factor selection step of selecting a combination of operation factors having a large quality influence degree as a combination of operation conditions effective for quality improvement based on the quality influence degree in all the operation factor combinations created in the operation factor combination creation step. When,
For the combination of operation factors selected in the effective operation factor selection step, the quality index in each operation condition mesh calculated in the quality indicator calculation step is compared with a preset target quality indicator, and the target quality indicator is determined. A quality improvement operation condition extraction step for extracting an operation condition mesh to be achieved and a range of operation conditions corresponding to the operation condition mesh;
A quality improvement condition analysis method for a product, comprising: a quality improvement guidance output step for outputting a combination of operation factors calculated in the quality improvement operation condition extraction step and a range of operation conditions. The quality data probability distribution calculating step approximates the frequency distribution of the quality data in the operation condition mesh using a predetermined probability density function,
8. The quality index calculating step calculates a quality data value having a predetermined cumulative probability based on a probability density function obtained by the approximation process, and sets the quality data value as the quality index in the operation condition mesh. The quality improvement condition analysis method for products described in 1.   The quality influence degree calculation step selects the maximum value and the minimum value from the quality indices in all the operation condition meshes calculated in the quality index calculation step, calculates the difference, and the quality influence degree in the combination of the operation factors and The product quality improvement condition analysis method according to claim 7 or 8, wherein:   The operation condition mesh creation step determines the division range so that the number of data existing in each division of the operation condition divided into a plurality of ranges is equal, and creates an operation condition mesh that is combined among a plurality of operation factors. The product quality improvement condition analysis method according to any one of claims 7 to 9, wherein: A data input process for inputting operation data to be predicted;
The probability distribution of the quality data in each operation condition mesh calculated in the quality data probability distribution calculation step is assumed to be a prediction model, and the prediction target operation data input in the data input step is predicted to occur with a predetermined probability. 11. The product quality improvement condition analysis method according to claim 7, further comprising: a prediction step of calculating a quality index.   The quality improvement condition analysis method for a product according to any one of claims 7 to 11, wherein in the quality improvement guidance output step, a distribution of quality indices in a two-dimensional operation factor space is displayed on a display device. . A data input procedure for inputting analysis data comprising product quality data and operation data of a plurality of operation factors in the manufacturing process;
From the operation data of a plurality of operation factors input in the data input procedure, an operation factor combination creation procedure for creating all combinations of mutually different operation factors for the number specified in advance,
An operation factor selection procedure for selecting one combination from all combinations created in the operation factor combination creation procedure;
For one combination selected in the operation factor selection procedure, for each of a plurality of operation factors constituting the combination, the operation condition is divided into a plurality of ranges based on the value of the operation data, and this is divided into a plurality of operations. An operation condition mesh creation procedure for creating an operation condition mesh combined between factors,
Quality data probability distribution calculation procedure for extracting quality data belonging to each operation condition mesh created in the operation condition mesh creation procedure and calculating a probability distribution of quality data in each operation condition mesh;
Based on the probability distribution calculated in the quality data probability distribution calculation procedure, a quality index calculation procedure for calculating a quality index indicating the quality of product quality in each operation condition mesh;
Based on the quality index in each operation condition mesh calculated in the quality index calculation procedure, a quality impact calculation procedure for calculating a quality impact degree by quantifying a difference in quality index due to a difference in operation conditions;
The operation factor selection procedure, the operation condition mesh creation procedure, the quality data probability distribution calculation procedure, the quality index calculation procedure, and the quality influence calculation procedure for all combinations of operation factors created in the operation factor combination creation procedure The operation combination evaluation procedure to be performed sequentially,
Effective operation factor selection procedure for selecting a combination of operation factors having a large quality influence level as a combination of operation conditions effective for quality improvement based on the quality influence level in all the operation factor combinations created in the operation factor combination creation procedure. When,
For the combination of operation factors selected in the effective operation factor selection procedure, the quality index in each operation condition mesh calculated in the quality index calculation procedure is compared with a preset target quality index, and the target quality index is determined. A quality improvement operation condition extraction procedure for extracting an operation condition mesh to be achieved and a range of operation conditions corresponding to the operation condition mesh;
A computer program for causing a computer to execute a quality improvement guidance output procedure for outputting a combination of operation factors calculated in the quality improvement operation condition extraction procedure and a range of operation conditions.   A computer-readable recording medium on which the computer program according to claim 13 is recorded.

Images