CN114760969A

CN114760969A - Mass measurement method and mass measurement device

Info

Publication number: CN114760969A
Application number: CN202080082342.6A
Authority: CN
Inventors: 平尚大; 坂本雅基
Original assignee: Kao Corp
Current assignee: Kao Corp
Priority date: 2019-11-28
Filing date: 2020-08-20
Publication date: 2022-07-15
Anticipated expiration: 2040-08-20
Also published as: JP7084907B2; CN114760969B; JP2021086457A; WO2021106288A1

Abstract

The invention provides a quality measuring method and a quality measuring apparatus. The quality measurement method is a quality measurement method in a process of manufacturing a product through a plurality of manufacturing steps, and includes: constructing a prediction model between the quality data and the processing process data; and a step of calculating a quality measurement value of the manufactured product from the process data acquired by the on-line measurement based on the prediction model. According to the present invention, it is possible to measure quality that is difficult to measure by on-line measurement without performing a sampling inspection.

Description

Mass measurement method and mass measurement device

Technical Field

The present invention relates to a mass measurement method and a mass measurement apparatus.

Background

In order to ensure the quality of sheet products such as absorbent articles, it is a conventional technique to perform a total inspection by an on-line measurement, a sampling inspection by an off-line measurement method, or the like in a manufacturing process.

In the general inspection based on the on-line measurement, foreign substances on the surface of a material are detected or the size of the material is measured by an image sensor or the like provided in the manufacturing process, and the presence or absence of an abnormality of the product is confirmed.

On the other hand, the quality (for example, sealing strength and absorption performance) which is difficult to measure by on-line measurement is subjected to a sampling inspection such as a destruction inspection to confirm whether or not there is an abnormality in the product.

Patent document 1 describes a method of calculating the amount of change due to a change in the manufacturing process. In this method, in order to improve productivity in a manufacturing plant, the amount of change due to a change in manufacturing process such as a control function and a work method is calculated with high accuracy.

Further, patent document 2 discloses a system and a process that can be configured to associate a manufacturing parameter and a performance feedback parameter with each absorbent article manufactured by a processing device. The product performance data obtained by the product experiment can be used as a tool for manufacturing a future processing machine and/or process adjustment by a manufacturer.

Further, patent document 3 describes a manufacturing method for manufacturing a product through a plurality of manufacturing steps. The method is also used for manufacturing absorbent articles, and can realize improvement of machine operation rate and improvement of product yield.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2013-105313

Patent document 2: japanese patent laid-open publication No. 2016-538652

Patent document 3: japanese patent laid-open publication No. 2018-129030

Disclosure of Invention

The present invention provides a quality measurement method in a process of manufacturing a product through a plurality of manufacturing steps, the quality measurement method including: a step of constructing a prediction model between the quality data and the processing technique (machining technique) data; and a step of calculating a quality measurement value of the manufactured product from the processing process data acquired by the on-line measurement based on the prediction model.

Further, the present invention provides a quality measuring apparatus used for an apparatus that manufactures a product through a plurality of manufacturing steps, the quality measuring apparatus having an analysis server that constructs a prediction model between quality data and process data, the analysis server calculating a quality measurement value from the process data by online measurement based on the prediction model.

The above and other features and advantages of the present invention will become more apparent from the following description with reference to the accompanying drawings as appropriate.

Drawings

Fig. 1 is a flowchart showing a preferred example of the prediction model construction step in the quality measurement method of the present invention.

Fig. 2 is a schematic configuration diagram showing a preferred example of a manufacturing apparatus using the mass measurement method of the present invention, together with various sensors used in the mass measurement method.

Fig. 3 is a diagram showing a specific example of process data to be subjected to the time-series processing, wherein (a) shows an example of a method for associating process data with each other in the time-series processing step, and (B) shows an example of a storage state of the process data after the time-series processing.

Fig. 4 is a flowchart showing a preferred example of an on-line quality measurement step in the quality measurement method of the present invention.

Fig. 5 is a block diagram showing a preferred embodiment of the mass measuring apparatus of the present invention, wherein (a) shows a mode in which a data collection CPU is separated from an analysis server, and (B) shows a mode in which the data collection CPU and the analysis server are integrated.

Fig. 6 is a graph comparing the measured quality values of the sealing strength at the inlet side calculated in examples 1 to 3 with the actual measured values.

Detailed Description

The present invention relates to a quality measuring method and a quality measuring apparatus capable of measuring a quality which is difficult to measure by on-line measurement without performing a sampling inspection.

In a sampling inspection such as a destruction inspection, since a part of all products is sampled and inspected, all the products cannot be inspected. Therefore, if an abnormal product is found in the sampling inspection, it takes much time to find the cause. In addition, when the same product is produced in different production lines, the ranges of production conditions for obtaining appropriate quality may be greatly different depending on differences in mechanical properties, differences in material properties, and the like. In such a case, it is necessary to verify the production conditions for each product type and production line, and a lot of labor is consumed. Therefore, it is desirable to estimate the quality that is difficult to measure by online measurement without much labor.

According to the quality measuring method and the quality measuring apparatus of the present invention, it is possible to measure the quality that is difficult to measure by online measurement without performing a sampling inspection.

Hereinafter, a preferred embodiment of the mass measurement method of the present invention will be described with reference to fig. 1 to 4, taking a mass measurement method of an absorbent article as an example. First, the kind of data used in the quality measurement method of the present invention will be described.

(processing data)

Process data refers to data acquired through on-line measurements at various steps in the manufacture of a product.

The process data is roughly divided into material data obtained from the material of the product and device data obtained from the device. The material data and the device data are obtained in time series by online measurement in the manufacturing step, and are data that change with time.

Specific examples of the material data include a sheet material roll diameter, a sheet material roll storage humidity, a sheet temperature, a sheet tension, a sheet thickness, and the like. Specific examples of the apparatus data include hydraulic pressure, pattern roll temperature, anvil roll temperature, roll rotation speed, motor shaft rotation frequency, motor shaft load, tension roll pressure, and conveyor conveyance speed.

The processing data may include data (hereinafter, referred to as "operation data") that an operator of the manufacturing line can appropriately operate as a set value. As specific examples of the operation data, the device data may be hydraulic pressure, pattern roll temperature, anvil roll temperature, dancer roll pressure, conveyor transport speed, and the like, and the material data may be web roll storage humidity, and the like.

As material data, the process data preferably comprises sheet tension or sheet thickness or both.

(quality data)

Quality data is data that is difficult to measure by online measurement. For example, by sampling inspection measurement such as a destructive inspection (off-line measurement). Specific examples of the quality data include the sealing strength of diaper side seals, sanitary napkin embossing seals, eye patch hangers seals, individual package periphery seals, and the like; the absorption performance of the sanitary product evaluated by the remaining amount of liquid, the reverse osmosis amount, the absorption speed, the grammage distribution, and the like; the thermal performance of the heating element evaluated from the duration, maximum temperature, rate of temperature rise, etc.; the quality of the coating material evaluated from the viscosity, the amount of solid components, and the like; sensory evaluation of skin touch, hand, appearance and the like; moisture permeability of the moisture permeable film. The seal strength includes a seal strength of a portion where 2 or more sheets are joined.

The seal strength is generally measured by a fracture inspection performed while peeling off the bonded seal portion by a tensile tester. The sealing strength is therefore a quality data.

The bonding of the sheets can use various methods. Specifically, heat sealing, ultrasonic sealing, laser welding, hot melt adhesion, and the like can be mentioned.

The quality measurement method of the present invention is used in a method of manufacturing a product through a plurality of manufacturing steps. The term "product" as used herein includes not only a finished product (end product) but also an intermediate product and a semi-finished product located in the middle of a manufacturing process.

The product to be manufactured is not particularly limited as long as the effects of the present invention can be exerted. For example, a product produced using a sheet, typically, a composite sheet used for diapers and the like can be cited. The sheet may be made of various materials, for example, a fibrous sheet such as a nonwoven fabric.

The product to be manufactured is not limited to a single product cut into one product, and may be a continuous product such as roll paper. When the product is continuous, the process data and the quality data may be processed using values obtained by moving averaging the individual data.

The method of moving the average is not particularly limited as long as the production method of the present invention can achieve the effects. For example, in fig. 3 (a), when moving average is performed with 10 consecutive values, the data X is correlated with the data X₁₀Corresponding moving average value is included data X₁₀The first 10 consecutive data (i.e., data X) therein₁～X₁₀) The sum of which is divided by 10. When there are not 10 previous continuous data items immediately after the start of manufacturing a product, the value obtained by summing all the continuous data items and dividing by 10 is used as a moving average value. Specifically, with data X₃The corresponding moving average is data X₀～X₃The sum of these 4 values is divided by 10 to obtain the value.

The fiber sheet is easily deformed by an external force and is difficult to stabilize in shape. Thus, the fiber sheet may be deformed in the manufacturing step depending on the conditions. Therefore, in the quality measurement method of the present invention, it is preferable to acquire the processing data on the fiber sheet in time series.

In this case, it is preferable to acquire processing data before and after the processing of the fiber sheet. In particular, processing data of sheet tension and sheet thickness are preferred as materials for better estimation of quality data on the fiber sheets (for example, quality of sealing between fiber sheets) when manufacturing the composite sheet.

The quality measuring method of the present invention comprises: a step of constructing a prediction model between the quality data and the processing process data (hereinafter, referred to as "prediction model construction step"); and a step of calculating a quality measurement value of the manufactured product from the processing process data acquired by the on-line measurement based on the prediction model (hereinafter, referred to as "on-line quality measurement step"). The quality measurement value is data of the same kind as the aforementioned quality data calculated by on-line measurement. Thus, in the online quality measurement step, the quality data of the product is not measured by offline measurement, but the measured value of the quality of the product is estimated by online measurement.

Hereinafter, the prediction model building steps will be described in order.

(prediction model construction step)

As shown in fig. 1, the prediction model building step typically includes a process data acquisition step 61, a time sequence processing step 62, a data combination step 63, an end determination step 64, data analysis steps 71 to 73, generalization ability (generalization ability) confirmation steps 81 to 83, and the like.

However, the prediction model building step is not limited to the embodiment shown in fig. 1 as long as the effects of the present invention can be achieved.

In order to efficiently and inclusively (cyclopaedicaly) acquire the processing process data for constructing the prediction model by repeating the processing process data acquisition step 61 to the data combination step 63, a matrix test, a factor fixation test, an experimental design method, or the like can be employed. From the viewpoint of reducing the combination of values of the acquired process data, it is preferable to use an experimental design method.

The orthogonal table used in the experimental design method is not particularly limited. For example, L can be used₁₂Orthogonal table, L₁₈Orthogonal table, L₃₆Orthogonal tables, etc.

Further, by adjusting the level of assignment to the orthogonal table in the experimental design method, the ratio of the variation of the predicted quality measurement value can be changed. For example, when the level of allocation to the orthogonal table is increased, the ratio of the variation of the predicted quality measurement value becomes small. Conversely, when the level of allocation to the orthogonal table is reduced, the ratio of the variation of the predicted quality measurement value becomes large.

(processing data acquisition step)

In the process data acquisition step 61, process data for prediction model construction is acquired from the manufacturing line by on-line measurement. In this step, material data and device data are taken together as process data.

From the viewpoint of acquiring process data directly related to a product to be manufactured, the acquisition of the process data is preferably performed during operation of the manufacturing line.

Various sensors corresponding to the type of data can be used to acquire the process data. For example, a radiation thermometer can be used to acquire process data of temperature. In order to acquire the processing process data of the sheet thickness, a contact displacement meter, a noncontact displacement meter using a laser, an ultrasonic wave, an electrostatic capacitance, or the like can be used.

In the present invention, it is preferable to acquire the processing data before and after the processing, from the viewpoint of judging whether or not the product is acceptable from the state before and after the processing. Acquiring the processing data before and after the processing can be realized by providing various sensors before and after the processing portion of the manufacturing apparatus.

In the case where the product to be manufactured is an unstable material such as a fiber sheet, which may change in thickness or the like in the manufacturing process, obtaining the processing process data before and after the processing is particularly effective from the viewpoint of estimating the quality data with high accuracy.

(time sequence processing step for processing Process data)

The timing processing step 62 is a step of correcting a time delay between processing process data and storing the time delay in an analysis server or the like.

There are cases where a plurality of sensors for acquiring process data are provided in a process of manufacturing a product. In this way, the process data acquired at the same time by the sensor provided upstream and the sensor provided downstream in the machine flow direction are data observed for different products. Therefore, in the present invention, in order to correspond each processing process data to the same product, it is preferable to perform time-series processing on the processing process data. As the timing processing, a delay (time lag) in time is appropriately corrected in accordance with the speed of the mechanical flow. Specifically, the time delay between sensors that acquire process data is corrected based on the velocity of the mechanical flow and the data is stored. Thus, the products can be arranged in correspondence with the processing process data, and data analysis can be realized. In addition, the processing process data and the quality data may be associated with each other in the data combining step 63 described later.

A specific method of the sequence processing will be described with reference to the manufacturing apparatus 11 shown in fig. 2.

Various sensors used in the mass measurement method of the present invention are disposed in the manufacturing apparatus 11. When the shaft rotation speed acquired from the motor 24E is associated with other processing process data, the position of the motor 24E and the distance from the position of another sensor are confirmed in advance. As other sensors, fig. 2 may include a temperature sensor 33B for acquiring process data of the sealing unit temperature, a sheet blank roll sensor 34A for acquiring process data of the upper sheet blank roll diameter, and the like. Here, each distance is made to correspond to the number of pieces (the number of products) to be cut. For example, the temperature sensor 33B is 34 th from the motor 24E. The web roll sensor 34A is the 40 th from the motor 24E. The number of pieces corresponding to the distance may vary depending on the manufacturing apparatus. By corresponding to the number of sheets in this manner, the distance from the motor 24E to the temperature sensor 33B and the sheet blank roll sensor 34A can be determined.

Thereby, as shown in fig. 3 (a), data Z as the shaft rotation speed of the motor 24E₄₀The processing process data of the time point of (1) and the sealing unit temperature data Y₄₀The diameter of the upper shaft sheet blank coil is data X₄₀. However, these treatment process data are not comparable to the observed data Z₄₀The product of (1). The number of products from the motor 24E (see fig. 2) to the temperature sensor 33B (see fig. 2) is 34. Thus, the data Y acquired by the temperature sensor 33B at the time of tracing back the time corresponding to 34 pieces of the sheet is obtained from the speed of the mechanical flow₆Is and observed data Z₄₀Processing process data of the sealing unit temperature corresponding to the product. In addition, with respect to the upper axial slice blank roll diameter (refer to fig. 2), data X of the time corresponding to 40 slices is traced back₀Is and observed data Z₄₀The diameter of the upper shaft sheet blank material roll corresponding to the product.

In this way, the processing process data can be individually associated with each other for each product, and the time delay can be corrected.

The corresponding process data are preferably stored for each product in the same row or column of the table calculation software for the arrangement. In the above example, it is preferable that the data X corresponding to the same product is stored₀Data Y₆And data Z₄₀As shown in fig. 3 (B), stored in the same row of the table calculation software.

(data combining step)

The data combining step 63 is a step of associating and combining the processing process data of each product with the quality data of each product. The method of combining is not particularly limited, but mass data may be stored in the same row or column of the table calculation software for each product and arranged as in the case of processing process data.

The prediction model building step includes a data combining step 63, and thus the quality data can be associated with various process data. For example, material data, device data, and operational data can be associated and combined with quality data.

From the viewpoint of rapid data analysis, it is preferable to use the processing process data corrected for the time delay in the time-series processing step 62 in the data combining step 63.

For 1 product, the corresponding process data and quality data are stored as a set of data sets in an analysis server or the like. In other words, as many data sets as the number of products can be acquired.

The end judgment step 64 is a step of judging whether or not the number of data sets acquired through the processing process data acquisition step 61 to the data combination step 63 is sufficient for data analysis. In the case where the number of data sets is sufficient, a data analysis step 71 described later is performed. On the other hand, when the number of data sets is insufficient, the processing process data acquisition step 61 to the data combining step 63 are repeated while appropriately changing the combinations of the values of the processing process data until a sufficient number of data sets can be acquired. In this case, it is preferable to efficiently acquire the data set by the experimental design method.

(1 st data analysis step)

The data analysis step 71 is a step of performing the 1 st data analysis after the data set is acquired.

In data analysis, the acquired data set is divided into 2 parts. One of the 2 divided parts is used as learning data, and the other part is used as test data. The learning data is data mainly used for constructing a model that can be a candidate of a prediction model (hereinafter, referred to as a "prediction model candidate"). The test data is data mainly used for verifying a prediction model candidate constructed from the learning data and deciding a prediction model.

In data analysis, various analyses are performed on learning data and test data to construct a plurality of prediction model candidates. The scores described later are calculated for the plurality of prediction model candidates thus constructed.

By separating the data set into learning data and test data, overfitting can be avoided and unknown quality data can be easily predicted. The division ratio of the learning data and the test data may be determined arbitrarily, and is generally set as 8: 2. 5: 5 equal ratio division. In addition, the segmentation of the learning data and the test data may also use cross-validation.

The method of judging and selecting the prediction model candidates is not particularly limited. For example, a method of using a prediction model in which the result of analysis using learning data and test data is a certain level or more as a prediction model candidate can be cited.

The method of analyzing the learning data and the test data is not particularly limited as long as the effects of the present invention can be obtained. Examples thereof include various linear regressions and various nonlinear regressions.

Specific examples of the linear regression include multiple regression analysis, PLS (Partial least squares regression) regression analysis, lasso regression analysis, ridge regression analysis, elastic net regression analysis, sgd (Stochastic Gradient Descent) regression analysis, and support vector regression analysis.

Specific examples of the nonlinear regression include nonlinear support vector regression analysis, kernel ridge regression analysis, random forest regression analysis, MLP (multi layer perceptron) regression analysis, and the like.

In each analysis method, adjustment of the hyper-parameter is required. By adjusting the hyper-parameters, the accuracy of the prediction model can be improved. The method of adjusting the hyper-parameter is not particularly limited as long as the effect of the present invention can be achieved.

The generalization ability confirmation step 81 is a step of confirming the generalization ability of the analysis result of the data analysis step 71. Generalization capability in this specification refers to the performance of a constructed predictive model that can be used in an online measurement for a measure of quality.

As a specific method of determining the generalization ability in the generalization ability confirmation step 81, it is preferable to determine a prediction model candidate having the highest generalization ability among the plurality of prediction model candidates constructed as the prediction model. By performing the determination in two stages, i.e., the analysis of the learning data and the analysis of the test data, it is possible to efficiently select a prediction model (optimal prediction model).

Regarding the level of generalization ability, R can be calculated²Scores such as (coefficient of determination), R (correlation), MSE (Mean Square Error), RMSE (Root Mean Square Error), MAE (Mean Absolute Error), AIC (Akaike's Information Criterion), BIC (Bayesian Information Criterion), and the like are determined by the level of the score. In addition, when R is used²The score of (2) is dependent on the object to be predicted, but if it exceeds 0.6, it can be judged that the accuracy of the prediction model is high.

A score calculated from the learning data (hereinafter referred to as "learning data score") and a score for the test data (hereinafter referred to as "test data score") are calculated in the data analysis step 71. Therefore, in the generalization ability confirmation step 81, the highest score among the test data scores in the prediction model candidates having the learning data score and the test data score at a certain level or higher is determined as the prediction model.

And if the learning data score and the test data score reach a certain number of prediction model candidates, determining the highest score of the test data as a prediction model, and ending the prediction model construction step. Otherwise, a data analysis step 72 described later is performed.

(2 nd data analysis step)

The data analysis step 72 is a step of 2 nd data analysis performed after the data analysis step 71.

When data analysis is performed in the data analysis step 72, the data analysis step 72 is performed so as to delete a part of the data set used in the data analysis step 71 or delete a part of the processing data constituting the data set (reduce the types of the processing data) by using data different from the data analysis step 71. That is, the number of data used in the data analysis step 72 is smaller than the number of data used in the data analysis step 71. By reducing the amount of data used for analysis, the calculation time may be shorter.

The specific method of the data analysis step 72 is the same as that of the data analysis step 71. That is, the data set to be analyzed is divided into 2 parts, one part being learning data and the other part being test data. The learning data and the test data are subjected to various analyses in the same manner as in the data analysis step 71, and a plurality of prediction model candidates are constructed. Learning data scores and test data scores are calculated for the plurality of prediction model candidates constructed.

The generalization ability confirmation step 82 is a step of confirming the generalization ability of the analysis result of the data analysis step 72. The generalization ability confirmation step 82 is similar to the generalization ability confirmation step 81 in specific manner.

When there is a prediction model candidate in which the learning data score and the test data score reach a certain value, the highest score of the test data is determined as a prediction model, and the prediction model building step is ended. Otherwise, a data analysis step 73 described later is performed.

(3 rd and subsequent data analysis steps)

The data analysis step 73 is a step of performing the 3 rd and subsequent data analyses after the data analysis step 72.

When data analysis is performed in the data analysis step 73, data analysis is performed after appropriate data addition and deletion is performed on the data sets used in the data analysis steps 71 and 72 in order to use a data set different from the previous data analysis. That is, the number of data used in the data analysis step 73 may be larger or smaller than the number of data used in the data analysis steps 71 and 72.

When data is added and data analysis is performed in the data analysis step 73, data is acquired by adding data as appropriate. Specifically, a series of operations performed in the process data acquisition step 61, the sequence processing step 62, and the data combining step 63 are performed in the same manner as described above.

The specific method of the data analysis step 73 is the same as the data analysis steps 71 and 72. That is, the data set to be analyzed is divided into 2 parts, one part being learning data and the other part being test data. The learning data and the test data are subjected to various analyses in the same manner as in the data analysis steps 71 and 72 to construct prediction model candidates. Learning data scores and test data scores are calculated for the plurality of prediction model candidates constructed.

The generalization ability confirmation step 83 is a step of confirming the generalization ability of the analysis result of the data analysis step 73. The specific determination method in the generalization capability confirmation step 83 is the same as the generalization capability confirmation steps 81 and 82.

And when the prediction model candidate with the learning data score and the test data score reaching a certain value exists, determining the highest score of the test data as the prediction model, and finishing the construction step of the prediction model. If not, the data analysis step 73 and the generalization ability determination step 83 are repeated until a prediction model in which the learning data score and the test data score reach a certain value can be constructed.

And after the step of constructing the prediction model is finished, the online quality measurement step is carried out. The online quality measurement procedure is explained below.

(Online quality measuring step)

The online quality measurement step is a step of inferring a quality measurement value based on a predictive model in a process of manufacturing a product through a plurality of manufacturing steps. The quality measurement value is a value calculated by applying process data to the prediction model, and corresponds to an estimated value of the quality data. By calculating the quality measurement value in the online quality measurement step, a value of a target quality such as a sealing strength, which is difficult to achieve in online measurement, can be estimated.

As shown in fig. 4, the online quality measurement step typically includes a process data acquisition step 91, a timing processing step 92, a quality measurement value estimation step 93, a quality measurement value transmission step 94, and an acceptance determination step 95. However, the online quality measuring step is not limited to the mode shown in fig. 4 as long as the effects of the present invention can be obtained.

(processing data acquisition step and timing processing step)

In the processing process data acquisition step 91 and the sequence processing step 92 in the online quality measurement step, the acquisition of the processing process data and the correction of the time delay are performed in the same manner as in the processing process data acquisition step 61 and the sequence processing step 62 in the prediction model construction step. In this case, it is preferable to manufacture the product by the same manufacturing apparatus as that used in the prediction model building step. By using the same apparatus, the predictive model can be adapted to the online quality measurement step.

(Mass value estimation step)

The quality measurement value estimation step 93 is a step of estimating a quality measurement value by combining the process data with the prediction model. Specifically, the process data acquired in the process data acquisition step 91 is applied to the prediction model determined in the prediction model construction step, and the measured quality value is calculated. In this way, a value that is difficult to obtain by online measurement can be estimated.

(quality measurement value transmitting step)

The measured-quality-value transmitting step 94 is a step of transmitting the measured quality value calculated in the measured-quality-value estimating step 93 to a CPU (Central Processing Unit, for example, a manufacturing-apparatus-side CPU13, which will be described later) directly connected to the manufacturing apparatus 11. By transmitting the measured quality value to the CPU directly connected to the manufacturing apparatus 11, the CPU directly connected to the manufacturing apparatus 11 can determine whether or not the CPU is qualified in a qualification judgment step 95 described later.

When the CPU directly connected to the manufacturing apparatus 11 calculates the quality measurement value, the quality measurement value transmission step 94 may not be necessary depending on the method of data exchange. The manner of data exchange will be described later.

(step of judging acceptability)

The pass-or-fail judging step 95 is a step of judging whether or not the manufactured product has a quality of a certain level or more, using the measured value of quality as an index. Therefore, in the present invention, it is preferable to perform the determination of the quality measurement value as to whether it is acceptable by providing the acceptance determination step 95 or the like. It is possible to confirm whether or not a product has a quality above a certain level by confirming whether or not a measured value of the quality is within a certain range of values without performing a sampling inspection of the product.

If the measured value of the quality is outside the predetermined numerical range, the measured value of the quality can be brought within the predetermined numerical range by appropriately changing the set values (operation data and the like) of the manufacturing apparatus. By the method, the online measurement result can be received and appropriately fed back, so that the quality of the product is stabilized above a certain level, and the productivity of the product is improved. Further, automatic operation of the manufacturing line can be facilitated.

The manufacturing apparatus preferably issues control commands based on the quality measurements.

For example, when the quality measurement value is greatly deviated from a certain numerical range, the manufacturing line can be automatically stopped quickly when a defective product is generated by instructing the manufacturing apparatus to stop the manufacturing line. In addition, it is possible to notify an abnormality to the operator as soon as possible when the manufacturing line is stopped. By taking countermeasures as early as possible when an abnormality occurs, the quality of the product can be maintained at a certain level.

Conversely, during the calculation of a normal quality measurement, a command can be issued to continue operation of the manufacturing line, which may be automated.

As a manufacturing method including the quality measurement method of the present invention, various methods can be cited, and the method can be used for various apparatuses that manufacture products through a plurality of manufacturing steps. For example, the present invention can be applied to an apparatus for manufacturing a composite sheet as shown in fig. 2.

The manufacturing apparatus 11 shown in fig. 2 is an apparatus for manufacturing a composite sheet 41 by bonding two

sheets

42 and 43 by sealing. The manufacturing apparatus 11 includes a sealing unit 21, a pattern roll 22, an anvil roll 23, motors 24A to 24E, conveyors 25A to 25H, and the like.

The manufacturing apparatus 11 manufactures a product, i.e., the composite sheet 41, through a plurality of manufacturing steps. As a specific example of the manufacturing method using the manufacturing apparatus 11, there can be mentioned a method including a step of feeding the

sheets

42 and 43 from the sheet stock rolls 44 and 45, a step of conveying the fed

sheets

42 and 43 by the conveyors 25A to 25H, a step of forming the composite sheet 41 by bonding the conveyed

sheets

42 and 43 inside the sealing unit 21, a step of cutting the composite sheet 41 into individual products by a cutter, and the like.

In the manufacturing apparatus 11, when the quality measuring method of the present invention is performed, various sensors are provided to acquire process data. In the manufacturing apparatus 11, as sensors for acquiring process data by online measurement, thickness sensors 31A to 31C, tension sensors 32A and 32B, temperature sensors 33A to 33C, sheet blank roll sensors 34A and 34B, a camera 35, and the like are arranged. The number and type of the sensors are not particularly limited as long as the effects of the present invention can be obtained.

In the composite sheet manufacturing apparatus 11, the fiber sheet is likely to be unstable in shape. Therefore, it is preferable that a pair of sensors (for example, tension sensors 32A and 32B) be disposed on the upstream side and the downstream side of the sealing unit 21 that performs the sealing process. Although not shown, it is preferable that another thickness sensor is disposed downstream of the sealing unit 21 in addition to the thickness sensors 31A to 31C.

The process data acquired by the various sensors is transmitted to the mass measuring device, and the above-described sequence processing, data combination, data analysis, and the like are performed.

The quality measuring method of the present invention can be used for manufacturing apparatuses for various products in addition to the manufacturing apparatus 11 for a composite sheet. For example, an apparatus for producing a coating material by mixing iron powder, activated carbon, and water as raw materials of the coating material (hereinafter, also referred to as "coating material production apparatus") can be mentioned.

The paint manufacturing apparatus is composed of a mixing tank, a stirring blade, a motor and the like, and manufactures the paint through a plurality of manufacturing steps. Specific examples of the production method using the apparatus for producing a coating material include a method including a step of adding iron powder to a preparation tank, a step of adding activated carbon to the preparation tank, a step of adding water to the preparation tank, a step of stirring raw materials with a stirring blade, a step of applying the produced coating material to a substrate sheet, and the like. Through these steps, a sheet product having, for example, appropriate heat generation characteristics can be manufactured.

In a paint manufacturing apparatus, when the mass measurement method of the present invention is performed, the mass measurement values such as the viscosity and the solid content of the paint can be calculated from the processing data such as the amount of iron powder added, the amount of activated carbon added, the amount of water added, the particle size of iron powder, the particle size of activated carbon, the rotational speed of the stirring blade, and the stop time of the stirring blade.

Next, a preferred example of the mass measuring apparatus used in the mass measuring method of the present invention will be described with reference to fig. 5. However, the mass measuring device is not limited to the embodiment shown in fig. 5 as long as the effects of the present invention can be obtained.

The quality measuring apparatus 12 shown in fig. 5 (a) has a manufacturing apparatus side CPU13, a data collection CPU14, and an analysis server 15. In the mass measuring device 12, various sensors (not shown) are disposed as a mechanism for performing online measurement in the manufacturing apparatus 11. In the step of manufacturing a product, the manufacturing apparatus CPU13 acquires the processing data 1 from various sensors disposed in the manufacturing apparatus 11 as needed. Further, the manufacturing apparatus side CPU13 issues a manufacturing line control command 4 to the manufacturing apparatus 11 based on the quality measurement value 3. For example, a threshold value of the quality measurement value 3 is set in advance in the manufacturing apparatus side CPU13, and when the quality measurement value 3 falls outside the threshold value range, the manufacturing apparatus side CPU13 transmits a product discharge command, a device stop command, and the like to the manufacturing apparatus 11. In this way, the manufacturing apparatus 11 and the manufacturing apparatus-side CPU13 exchange data.

The manufacturing apparatus side CPU13 also exchanges data with the data collection CPU 14. Specifically, the data collection CPU14 acquires the processing data 1 from the manufacturing apparatus-side CPU 13. The data collection CPU14 exchanges data with the analysis server 15 and transfers the process data 1 and the like to the analysis server 15.

The analysis server 15 executes a process data acquisition step 61, a sequence processing step 62, a data combination step 63, an end judgment step 64, data analysis steps 71 to 73, generalization capability confirmation steps 81 to 83, a process data acquisition step 91, a sequence processing step 92, and a quality measurement value estimation step 93. The quality data 2 used in combination with the process data is measured off-line by a sampling inspection device 16 such as a tensile tester and stored in the analysis server 15.

In the prediction model construction step, the analysis server 15 constructs a prediction model between the quality data 2 and the process data 1. Further, in the online quality measurement step, the analysis server 15 calculates the quality measurement value 3 from the process data 1 by online measurement based on the prediction model. The measured quality value 3 is transferred from the analysis server 15 to the data collection CPU14, and further transferred to the manufacturing apparatus-side CPU 13.

The system shown in fig. 5 (B) is different from the system shown in fig. 5 (a), and does not include the data collection CPU 14. That is, in the mode shown in fig. 5 (B), the analysis server 15 also functions as the data collection CPU 14. Otherwise, the same manner as shown in fig. 5 (a) is adopted.

In the case of the method shown in fig. 5 (a), since data is stored not only in the analysis server 15 but also in the data collection CPU14, the load on the analysis server 15 can be reduced.

On the other hand, when the method shown in fig. 5 (B) is adopted, the manufacturing apparatus-side CPU13 and the analysis server 15 can exchange data directly without going through the data collection CPU 14.

The present invention further discloses the following mass measurement method, product manufacturing method, and mass measurement device in relation to the above embodiment.

<1>

A quality measurement method in a process of manufacturing a product through a plurality of manufacturing steps, the quality measurement method comprising:

constructing a prediction model between the quality data and the processing process data; and

a step of calculating a quality measurement value of the manufactured product from the process data acquired by the on-line measurement based on the prediction model.

<2>

The mass measurement method according to <1>, wherein the construction of the prediction model is performed using an experimental design method.

<3>

The quality measurement method according to <1> or <2>, wherein a plurality of prediction model candidates are constructed in the construction of the prediction model, and the prediction model candidate having the highest generalization ability is determined as the prediction model.

<4>

The quality measurement method according to any one of <1> to <3>, wherein the process data is subjected to time series processing in constructing the prediction model.

<5>

The quality measurement method according to any one of <1> to <4>, wherein the quality data is measured by a sampling inspection.

<6>

The quality measurement method according to any one of <1> to <5>, wherein a pass or fail determination of the quality measurement value is performed.

<7>

The quality measurement method according to any one of <1> to <6>, wherein a control instruction is issued based on the quality measurement value.

<8>

The mass measurement method according to any one of <1> to <7>, wherein the product is a composite sheet.

<9>

The mass measurement method of any one of <1> to <8>, wherein the process data includes sheet tension or sheet thickness, or both of them.

<10>

The quality measurement method according to any one of <1> to <9>, wherein the quality measurement value is a sealing strength of a portion where 2 or more pieces are joined.

<11>

The quality measurement method according to <10>, wherein the bonding is performed by heat sealing, ultrasonic sealing, laser welding, or hot melt adhesion.

<12>

The mass spectrometry method according to any one of <1> to <11>, wherein the manufacturing apparatus-side CPU acquires the process data from a sensor disposed in the manufacturing apparatus.

<13>

The quality measurement method according to <12>, wherein the manufacturing apparatus-side CPU issues a control command for a manufacturing line to a manufacturing apparatus based on the quality measurement value.

<14>

The quality measurement method according to any one of <1> to <13>, wherein the step of constructing a prediction model includes a process data acquisition step, a time series processing step, a data combination step, an end judgment step, a data analysis step, and a generalization ability confirmation step.

<15>

The method for quality measurement according to <14>, wherein in the process data acquisition step, material data and device data are acquired together as the process data.

<16>

The quality measurement method according to <14> or <15>, wherein the time-series processing step is a step of correcting a time delay between the process data and storing the data in an analysis server.

<17>

The quality measurement method according to any one of <14> to <16>, wherein the data combining step is a step of combining the treatment process data of each product in correspondence with quality data of each product.

<18>

The quality measurement method according to any one of <14> to <17>, wherein the end judgment step is a step of judging whether or not the number of data acquired through the process data acquisition step, the time-series processing step, and the data combining step is sufficient for data analysis.

<19>

The quality measurement method according to any one of <14> to <18>, wherein in the generalization capability confirmation step, the judgment of the prediction model is performed at 2 stages of analysis of learning data and analysis of test data.

<20>

The quality measurement method according to any one of <1> to <19>, wherein the step of calculating the quality measurement value is performed at a time when the step of constructing the prediction model is completed.

<21>

The quality measurement method according to any one of <3> to <20>, wherein in the step of calculating the quality measurement value, the quality measurement value is estimated based on the prediction model.

<22>

The quality measurement method according to any one of <1> to <21>, wherein the step of calculating the quality measurement value includes a process data acquisition step, a time-series processing step, a quality measurement value estimation step, a quality measurement value transmission step, and a pass/fail determination step.

<23>

The quality measurement method according to <22>, wherein in the process data acquisition step and the timing processing step, acquisition of the process data and correction of a time delay are performed.

<24>

The quality measurement method according to <22> or <23>, wherein the quality measurement value presumption step is a step of presuming the quality measurement value by combining the processing process data with the prediction model.

<25>

The quality measurement method according to any one of <22> to <24>, wherein the quality measurement value transmission step is a step of transmitting the quality measurement value to a CPU directly connected to a manufacturing apparatus.

<26>

The quality measurement method according to any one of <22> to <25>, wherein the non-defective determination step is a step of determining whether or not the manufactured product has a quality of a certain level or more, using the measured quality value as an index.

<27>

A method for manufacturing a product, comprising the method for measuring mass according to any one of <1> to <26 >.

<28>

A mass measuring apparatus for use with an apparatus for manufacturing a product through a plurality of manufacturing steps,

the quality measurement device has an analysis server,

the analytical server constructs a predictive model between the quality data and the process data,

the analysis server calculates a quality measurement from the process data by an online measurement based on the predictive model.

<29>

The mass measurement device according to <28>, wherein there are a control device side CPU and a data collection CPU.

<30>

The quality measurement apparatus according to <28> or <29>, wherein the analysis server executes a process data acquisition step, a sequence processing step, a data combination step, an end judgment step, a data analysis step, a generalization ability confirmation step, and a quality measurement value estimation step.

Examples

The present invention will be described in detail below with reference to examples, but the present invention is not limited thereto.

The composite sheet 41 was manufactured using the manufacturing apparatus 11 and various sensors shown in fig. 2 and the mass measuring apparatus 12 shown in fig. 5 (a), and the sealing strength was estimated as in examples 1 to 3 below.

The operating data used hydraulic pressure, pattern roll temperature, anvil roll temperature, dancer roll pressure, web roll moisture retention, and conveyor transport speed.

The processing data other than the operation data are motor shaft rotation speed at 5, motor shaft load at 5, side inlet sheet temperature, side outlet sheet temperature, sealing unit temperature, upper shaft sheet blank roll diameter, lower shaft sheet blank roll diameter, side inlet sheet tension, side outlet sheet tension and sheet thickness at 3. That is, a total of 26 process data including operation data were used.

The quality data adopts the sealing strength of the inlet side and the sealing strength of the outlet side.

The "inlet side" refers to the downstream side in the machine flow direction of the manufacturing apparatus 11, and the "outlet side" refers to the upstream side in the machine flow direction of the manufacturing apparatus 11.

(acquisition of data set)

The construction of the prediction model candidates and the determination of the prediction model are performed according to the flowchart shown in fig. 1.

After the operation data is appropriately set, the sheets for the absorbent article are bonded to each other while acquiring the process data by using the manufacturing apparatus 11 shown in fig. 2, thereby manufacturing a composite sheet. The acquired processing process data is subjected to time sequence processing to correct the time delay between the sensors. On the other hand, all the composite sheets were subjected to sampling inspection, and quality data was measured. The measurement of the inlet side seal strength and the outlet side seal strength was performed as mass data by using a tensile testing machine.

The processing process data and quality data after time series processing are combined to obtain 40 groups of data sets at one time.

The above operations were repeated 18 times, and a total of 720 sets of data sets were obtained. Assigning process data to L using a design of experiment approach when iteratively performed₁₈In the orthogonal table.

(1 st data analysis)

And (3) dividing 720 groups of data sets into 2 parts of 360 groups of study data and 360 groups of test data, and respectively carrying out linear analysis and nonlinear analysis on the data to construct a plurality of prediction model candidates. Validation of generalization ability of prediction model candidates Using R of learning data score and test data score². The linear analysis and the nonlinear analysis use the following method.

■ Linear analysis

Multiple regression

PLS regression

Lasso regression

Ridge regression

Elastic net return

sgd regression

Support vector linear regression

■ nonlinear analysis

Nonlinear support vector regression

Kenling regression

Random forest regression

MLP regression

R of both learning data and test data in any constructed prediction model candidate²Since all of the values of (a) and (b) are 0.6 or less, it is determined that the generalization ability of the prediction model candidate is insufficient, and the 2 nd data analysis is performed.

(2 nd time data analysis)

Part of the processing data constituting the data set used in the 1 st data analysis was deleted, and the number of types was changed from 26 to 14. For a data set consisting of 14 types of processing data, linear analysis and nonlinear analysis were performed in the same manner as in the first data analysis, and a plurality of prediction model candidates were constructed.

Learning R of both data and test data in arbitrary prediction model candidates constructed²All of the values of (a) and (b) are 0.6 or less, and therefore, it is judged that the generalization ability of the prediction model candidate is insufficient, and the 3 rd data analysis is performed.

(analysis of the 3 rd data)

The operation of acquiring the data set was repeated 3 more times, and 120 sets of data sets were newly acquired. And dividing all 840 data sets together with 720 data sets used in the 1 st data analysis into 2 parts of 420 group learning data and 420 group testing data, and respectively performing linear analysis and nonlinear analysis on the data sets in the same way as the 1 st data analysis to construct a plurality of prediction model candidates.

R of both learning data and testing data of lasso regression in constructed prediction model candidates²The value exceeds 0.6, and therefore, the model based on lasso regression is judged to have sufficient generalization ability. Furthermore, R of test data score in a lasso regression based model compared to other prediction model candidates²The highest, therefore, models based on lasso regression are decided as predictive models.

(example 1)

According to the flowchart shown in fig. 4, in the manufacturing apparatus used for acquiring the data set, the sheets for the absorbent article were bonded under the conditions of the pattern roll temperature of 125 ℃ and the anvil roll temperature of 170 ℃, and a total of 100 composite sheets were manufactured. The operational data is unchanged during manufacture.

The processing process data obtained during the bonding process and the operation data are subjected to time sequence processing together, and the time delay between the sensors is corrected.

The lasso regression determined as a prediction model was applied to the processing process data after the time series processing, and values of the entrance-side sealing strength and the exit-side sealing strength were estimated as quality measurement values for 100 composite sheets, respectively. The average of 100 mass measurements was determined for the entry-side seal strength and the exit-side seal strength, respectively.

(example 2)

100 mass measurement values were estimated for the entrance-side seal strength and the exit-side seal strength, respectively, and the average value was determined in the same manner as in example 1, except that the pattern roll temperature was set to 120 ℃ and the anvil roll temperature was set to 160 ℃.

(example 3)

100 mass measurements were estimated for the entrance side seal strength and the exit side seal strength, respectively, and the average value was determined in the same manner as in example 1, except that the pattern roll temperature was set to 115 ℃ and the anvil roll temperature was set to 150 ℃.

(confirmation of quality data)

The entry side seal strength and exit side seal strength were measured for all of the composite sheets manufactured in examples 1 to 3 above by a failure check, which is one of off-line measurements. The seal strength was measured using a tensile tester.

In examples 1 to 3, the average value of the measured values of 100 entrance-side seal strengths and the average value of the measured values of 100 exit-side seal strengths were obtained.

The results of the entry side seal strength are shown in fig. 6.

As shown in fig. 6, in any of examples 1 to 3, it was found that the difference between the average value of the quality measurement values of the entrance seal strength and the average value of the actual measurement values of the entrance seal strength was controlled to be within 0.3N/15mm, and the entrance seal strength was estimated with high accuracy by using the prediction model.

In example 1 of the composite sheet manufactured by setting the pattern roll temperature and the anvil roll temperature to high temperatures, the entrance side seal strength can be maintained at a large value. On the other hand, as shown in examples 2 and 3, it is understood that the entrance side seal strength can be gradually decreased by gradually lowering the pattern roll temperature and the anvil roll temperature. By operating the pattern roll temperature and the anvil roll temperature in this manner, the entry side sealing strength of the composite sheet to be produced can be adjusted.

The present invention has been described in connection with the embodiments and examples thereof, but unless otherwise specified, it is not intended to be limited to any of the details of the description, but rather should be construed broadly within its spirit as set forth in the appended claims.

The present application claims priority based on japanese patent application 2019-215774 filed in the japanese country on day 11/28 in 2019, to which reference is made and the contents of which are incorporated as part of the description of the present specification.

Description of the reference numerals

1 processing of Process data

2 quality data

3 measurement of quality

4 control instruction

11 manufacturing device

12 mass device

13 manufacturing apparatus side CPU

14 data collection CPU

15 analysis server

16 sampling inspection device

21 sealing unit

22 Pattern roll

23 anvil roll

24A-24E motor

25A-25H conveyer

31A-31C thickness sensor

32A, 32B tension sensor

33A-33C temperature sensor

34A, 34B sheet blank roll sensor

35 video camera

41 composite sheet

42. 43 pieces

44. 45 piece blank material roll

61. 91 Process data acquisition step

62. 92 sequence processing step

63 data combining step

64 end judgment step

71-73 data analysis step

81-83 generalization ability confirmation step

93 quality measurement value estimation step

94 quality measurement value transmitting step

And 95, judging whether the product is qualified or not.

Claims

1. A quality measurement method in a process of manufacturing a product through a plurality of manufacturing steps, the quality measurement method characterized by comprising:

2. The quality measurement method according to claim 1, wherein:

the construction of the prediction model is performed using an experimental design method.

3. The quality measurement method according to claim 1 or 2, characterized in that:

in the construction of the prediction model, a plurality of prediction model candidates are constructed, and the prediction model candidate with the highest generalization ability is determined as the prediction model.

4. A mass measurement method according to any one of claims 1 to 3, wherein:

and in the construction of the prediction model, carrying out time sequence processing on the processing process data.

5. The mass measurement method according to any one of claims 1 to 4, wherein:

the quality data is measured by a sampling inspection.

6. The mass measurement method according to any one of claims 1 to 5, characterized by:

and judging whether the quality measurement value is qualified or not.

7. The mass measurement method according to any one of claims 1 to 6, wherein:

and issuing a control command based on the quality measurement value.

8. The mass measurement method according to any one of claims 1 to 7, characterized by:

the product is a composite sheet.

9. The mass measurement method according to any one of claims 1 to 8, characterized by:

the process data includes sheet tension or sheet thickness, or both.

10. The mass measurement method according to any one of claims 1 to 9, wherein:

the quality measurement is the seal strength of the portion where more than 2 sheets are bonded.

11. The quality measurement method according to claim 10, wherein:

the bonding is performed by heat sealing, ultrasonic sealing, laser welding, or hot melt bonding.

12. The mass measurement method according to any one of claims 1 to 11, wherein:

the manufacturing apparatus-side CPU acquires the processing process data from a sensor disposed in the manufacturing apparatus.

13. The quality measurement method according to claim 12, wherein:

the manufacturing apparatus side CPU issues a control command of a manufacturing line to the manufacturing apparatus based on the quality measurement value.

14. The mass measurement method according to any one of claims 1 to 13, wherein:

the step of constructing the prediction model comprises a processing process data acquisition step, a time sequence processing step, a data combination step, an ending judgment step, a data analysis step and a generalization capability confirmation step.

15. The quality measurement method according to claim 14, wherein:

in the processing data acquiring step, material data and device data are acquired together as the processing data.

16. A quality measurement method according to claim 14 or 15, wherein:

the time sequence processing step is a step of correcting time delay among the processing process data and storing the data in the analysis server.

17. The mass measurement method according to any one of claims 14 to 16, wherein:

the data combining step is a step of combining the treatment process data of each product in correspondence with the quality data of each product.

18. The mass measurement method according to any one of claims 14 to 17, wherein:

the end judgment step is a step of judging whether or not the number of data acquired by the processing process data acquisition step, the time-series processing step, and the data combination step is sufficient for data analysis.

19. A mass measurement method as claimed in any one of claims 14 to 18, wherein:

in the generalization ability determination step, the prediction model is judged by 2 stages of analysis of learning data and analysis of test data.

20. The mass measurement method according to any one of claims 1 to 19, wherein:

at the end of the step of building a prediction model, the step of calculating the quality measure is performed.

21. A mass measurement method according to any one of claims 3 to 20, wherein:

in the step of calculating the quality measure, the quality measure is extrapolated based on the predictive model.

22. A mass measurement method as claimed in any one of claims 1 to 21, wherein:

the step of calculating the quality measurement value includes a process data acquisition step, a time sequence processing step, a quality measurement value presumption step, a quality measurement value transmission step, and a pass or fail determination step.

23. The quality measurement method according to claim 22, wherein:

in the processing process data acquisition step and the time-series processing step, the acquisition of the processing process data and the correction of the time delay are performed.

24. A quality measurement method according to claim 22 or 23, wherein:

the quality measurement value estimation step is a step of estimating the quality measurement value by combining the process data with the prediction model.

25. A mass measurement method according to any one of claims 22 to 24, wherein:

the quality measurement value transmitting step is a step of transmitting the quality measurement value to a CPU directly connected to the manufacturing apparatus.

26. The mass measurement method of any one of claims 22 to 25, wherein:

the step of determining whether the manufactured product has a quality of a certain level or more using the measured quality value as an index.

27. A method of manufacturing a product, comprising:

a mass measurement method comprising any one of claims 1 to 26.

28. A mass measurement device used for a device that manufactures a product through a plurality of manufacturing steps, the mass measurement device characterized by:

the quality measurement device has an analysis server,

29. The mass measurement device of claim 28, wherein:

has a control device side CPU and a data collection CPU.

30. A mass measurement apparatus according to claim 28 or 29, wherein:

the analysis server executes a processing process data acquisition step, a time sequence processing step, a data combination step, an end judgment step, a data analysis step, a generalization ability confirmation step, and a quality measurement value presumption step.