JP7249320B2 - How to monitor the production process - Google Patents

Description

The present invention relates to a method for monitoring a process of manufacturing a product (thing) or a process of providing a service (hereinafter collectively referred to as "production process").

In Patent Document 1, principal component analysis and cluster analysis are applied to quality data and state data related to the manufacturing process to divide the lot of the manufacturing process into a plurality of groups, determine the quality of each group, and generate criteria in advance. However, a production process monitoring method is disclosed in which the quality of the manufacturing process in operation is determined based on whether the principal component score derived from the manufacturing process in operation satisfies the criteria (for example, Patent Document 1 reference).

JP 2016-167205 A

However, according to the method described in Patent Document 1, since the criterion is generated based on the manufacturing process that has been operated in the past, if the manufacturing process in operation hits a manufacturing state that has not been experienced in the past, However, there is a problem that the quality of the manufacturing process cannot be judged flexibly.

Therefore, there arises a technical problem to be solved in order to accurately predict the quality of the production process during operation, and the present invention aims to solve this problem.

In order to achieve the above object, a method for monitoring a production process according to the present invention is a method for monitoring a production process for producing a product or service, wherein a plurality of actions in an already-executed process, which is a production process performed in the past, are performed. a step of collecting performance status data, which is status data indicating the status of completed lots, and performance quality data, which is quality data indicating the quality of the product or service; and a virtual process for verification set to simulate the production process. setting virtual state data, which is state data indicating states of a plurality of virtual lots corresponding to inexperienced states different from the plurality of operated lots, and executing the virtual process with the virtual state data a step of collecting virtual quality data that is quality data indicating the quality of the product or service of the operated lot on multi-dimensional coordinates in which a plurality of principal component scores derived from the state data are set on each coordinate axis and the step of dividing the virtual lot into a plurality of groups; and determining whether it does not belong either.

According to this configuration, the virtual state data of the virtual lot is set to be different from the actual state data of the operated lot, and the virtual quality data corresponding to the virtual state data of the virtual lot is obtained by a virtual process such as an experiment or a simple simulation. By collecting, the monitoring process during operation based on the determination of whether it belongs to any group that divides the operated lot and the virtual lot set to complement it. The quality of the process can be predicted with high accuracy.

Also, the production process monitoring method according to the present invention preferably further includes the step of predicting quality data indicating the quality of the product or service in the monitoring process from the quality data of the group to which the monitoring process belongs.

According to this configuration, it is possible to predict the quality data of the products or services produced in the monitoring process based on the quality data of experienced lots or virtual lots belonging to the group determined to belong to the monitoring process.

Further, in order to achieve the above object, a production process monitoring method according to the present invention is a production process monitoring method for producing a product or service, comprising: Setting virtual state data, which is state data indicating states of a plurality of virtual lots corresponding to different states in a process, and quality indicating the quality of the product or service when the virtual process is executed with the virtual state data a step of collecting virtual quality data, which is data; a step of dividing the virtual lot into a plurality of groups on multidimensional coordinates in which a plurality of principal component scores derived from the state data are set on each coordinate axis; determining whether the monitoring process belongs to any of the groups or not, based on status data indicating the status of the monitoring process, which is the target production process.

According to this configuration, different virtual state data are set for a plurality of virtual lots, and virtual quality data corresponding to the virtual state data of the virtual lots are collected by virtual processes such as experiments and simple simulations. Accordingly, the quality of the monitoring process in operation can be accurately determined based on the determination as to whether or not the monitoring process in operation belongs to any group classified based on the virtual condition data and the virtual quality data. can be predicted.

In addition, the production process monitoring method according to the present invention includes performance status data, which is status data indicating the status of a plurality of operated lots in an operated process, which is a production process performed in the past, and the quality of the product or service. further comprising the step of collecting performance quality data, which is quality data indicating, wherein the plurality of virtual lots are set corresponding to inexperienced states different from the plurality of operated lots, respectively, and on the multidimensional coordinates, Preferably, the operated lot and the virtual lot are divided into a plurality of groups.

According to this configuration, in addition to the virtual state data and virtual quality data of the virtual lot, by considering the actual state data and actual quality data of the operated lot, it is possible to accurately predict the quality of the monitoring process during operation. can be done.

The present invention can accurately predict the success or failure of the monitoring process during operation.

1 is a schematic diagram showing the configuration of a distillation column to which a production process monitoring method according to one embodiment of the present invention is applied; FIG. Table showing status and quality data for operated lots and virtual lots. A table showing the amount of information and groups of principal components for each operated lot. A graph plotting principal component scores for each operated lot. A table showing the amount of information and quality data of principal components for each virtual lot. A graph plotting principal component scores for each virtual lot. A graph plotting principal component scores for operated lots and virtual lots.

An embodiment of the present invention will be described based on the drawings. In addition, hereinafter, when referring to the number, numerical value, amount, range, etc. of the constituent elements, unless otherwise specified or clearly limited to a specific number in principle, it is limited to the specific number It does not matter if the number is greater than or less than a certain number.

The analysis method according to this embodiment is applied to the production process. The production process includes a process consisting only of mechanical equipment and all processes are automated, a process including manual work processes by workers, and a manufacturing process automated by mechanical equipment and manual work by workers. Processes including steps are included. In addition, the production process is not limited to the process of producing things, and includes services such as cleaning of waste liquids and analysis of clinical trial results in drug development.

A case where this analysis method is applied to the purification of NMP (N-methyl-2-pyrrolidone), which is an example of the production process, will be described below. Needless to say, the production process to which the present invention is applied also includes other manufacturing lines and processes for providing services.

FIG. 1 is a schematic diagram showing a distillation column 1 that purifies NMP. The distillation column 1 purifies high-purity NMP from a raw material (NMP before purification) recovered from, for example, a Li-ion secondary battery manufacturing process. Pre-purified NMP mainly contains NMP, H2O, NMP precursors, safety-related substances and impurities. Note that the proportions of various products contained in the pre-purification NMP differ for each Li-ion secondary battery manufacturing process and are not necessarily uniform.

The distillation column 1 is divided into a stripping section 2 provided at the bottom and an enrichment section 3 provided at the top, and the raw material is supplied from the border between the stripping section 2 and the enrichment section 3 .

The raw material introduced into the distillation column 1 is heated by steam supplied from the bottom of the recovery unit 2 by a heating device (reboiler) 4, and the low boiling point component evaporates first. rise towards

A shelf-like structure is installed inside the distillation column 1, and vapor rising from the bottom and liquid falling from the top come into contact with each other. Vapor-liquid contact (gas-liquid contact) transfers the heat necessary for the liquid to evaporate (latent heat of vaporization) between the vapor and the liquid. At this time, the generated vapor contains more components with a boiling point lower than that of the liquid, while the liquid contains more components with a boiling point higher than that of the vapor.

Further, in the distillation column 1, gas-liquid contact is performed, so vapor is returned to liquid again (reflux) using a cooling device (condenser) 5 at the top of the column.

In this way, in the distillation column 1, since gas-liquid contact is performed in the entire column, cooling at the top of the column and heating at the bottom are performed at the same time to reflux, and at the top of the distillation column 1, components with a low boiling point are removed. , at the bottom of the distillation column 1, components with high boiling points are separated and recovered.

The distillation column 1 is provided with sensors (not shown) that measure various values. Objects to be measured by the sensor include the composition of the raw material, the amount of raw material input, the temperature in the tower, the set temperature of the heating device 4, the reflux amount of the liquid that flows back into the tower via the cooling device 5, and the like.

Various components constituting the distillation column 1 are controlled in operation by a control device 10 . The control device 10 includes, for example, a device control section 11 having a CPU, memory, etc., an input/output section 12 for controlling data input/output, a display section 13 for displaying data, and a storage section 14 for storing data. I have. Note that the functions of the control device 10 may be realized by controlling using software, or may be realized by operating using hardware.

The control device 10 performs the processing described later based on the state data indicating the operating state of the distillation column 1 measured by the sensor and the quality data indicating the quality information of the purified NMP. The state data includes manufacture data indicating the purification conditions of the production process (operating conditions of various devices constituting the distillation column 1, etc.) and material data indicating the composition of NMP before purification.

The device control unit 11 includes a control unit 11a that controls each device, an analysis unit 11b that performs processing described later on state data, a verification unit 11c that performs verification of a virtual lot described later, and a monitoring process that will be described later. The function is divided into the determination unit 11d and the determination unit 11d.

The input/output unit 12 includes, for example, a keyboard, a mouse, a communication control device, a printer, and the like. The display unit 13 has, for example, a display. Various data used in various processes are stored in the storage unit 14 .

Next, the procedure of the production process monitoring method according to the present embodiment will be described.

[Operated lot analysis]
First, the step of analyzing the status and quality of a plurality of operated lots in the production process will be specifically described.

The analysis unit 11b collects sensor-measured state data (actual state data) and post-refining NMP quality data (actual quality data) for each operational lot (step S1).

The state data and quality data collected in step S1 are stored in the storage unit 14 . FIG. 2 is a table containing status data and quality data of operated lots. Lot belonging to the type "experienced" in the table of FIG. 2 corresponds to the operated lot, and " The "raw material composition" column is state data indicating the composition of the NMP before purification that was put into the distillation column 1 in each operated lot. It is quality data indicating quality, and "O" in the table means that the refined NMP has cleared the shipping standards and no safety problems have occurred.

Next, the condition data and quality data collected in step S1 are standardized and converted into an intermediate function (step S2).

The state data standardization process performed in step S2 is a known process. Specifically, the analysis unit 11b performs the state data standardization process based on Equation 1.

Next, a principal component loading amount and a principal component score are obtained based on the intermediate variables obtained in step S2 (step S3).

Specifically, first, a correlation coefficient matrix is created for the intermediate variables, and eigenvalues and eigenvectors of the correlation coefficient matrix are derived. In the correlation coefficient matrix, the first principal component PC1 is expressed by Equation 2 when the intermediate variables are x1, x2, x3, . . . Also, the N-th principal component PCn is expressed as shown in Equation 3. A correlation coefficient matrix is formed by using the coefficients a11, a12, a13, .

Next, a principal component score is obtained from the eigenvectors of the correlation coefficient matrix. Also, the contribution rate of each principal component is obtained from the eigenvalues of the correlation coefficient matrix. The contribution rate of the principal component is obtained by dividing the eigenvalue by the sum of the eigenvalues. Here, the first principal component, second principal component, . . . Nth principal component are determined in descending order of eigenvalue.

Specifically, based on the intermediate variables x1, x2, and x3 of each lot and each coefficient of the correlation coefficient matrix, the analysis unit 11b calculates the values of the first principal component PC1, the second principal component PC2, . , to calculate the principal component score. FIG. 3 is a table showing the amount of information of the principal components for each operated lot.

Next, the analysis unit 11b applies cluster analysis to the principal component scores shown in FIG. 3 to classify each lot into a plurality of groups (step S4).

“Cluster analysis” is a method of classifying data to be analyzed (clusters) into a plurality of groups by focusing on similarity, and hierarchical clustering, classification optimization clustering, and the like are known. The “similarity” focused on by the cluster analysis in this embodiment refers to the distance between the principal component scores of each lot.

As a method of cluster analysis, for example, agglomerative hierarchical clustering, which is one of hierarchical clustering methods, and the like are known. As a method for calculating the distance between clusters, Ward's method or the like, which can stably obtain a solution, is used. "Ward's method" selects a cluster that minimizes the increase in the sum of squared deviations when two clusters are merged. For example, when clusters A and B are merged to generate cluster C, sums of squared deviations Sa, Sb, and Sc within clusters A, B, and C are represented by Equations 4 to 6, respectively.

From Equations 4 to 6, the deviation sum of squares Sc within the cluster C is as follows.

ΔSab in Equation 7 means an increment of the sum of squared deviations when cluster C is generated by merging clusters A and B. Therefore, clustering proceeds by selecting and merging clusters such that ΔSab is minimized at each merging step.

In this embodiment, as shown in FIG. 4, operated lots are divided into three groups 1 to 3 in a three-dimensional space of first to third eigenvectors (EV1 to EV3 in FIG. 4). FIG. 4 is a graph obtained by projecting the composite vector of EV1 to EV3 in the operated lot on the EV1-EV3 plane (a plane with EV1 on the horizontal axis and EV3 on the vertical axis) plotted on the above-described three-dimensional space. is. Also, in FIG. 4, operated lots constituting the same group are plotted with the same symbols. The right column "group" in the table in FIG. 3 indicates the group to which each operated lot belongs. The number of eigenvectors is not limited to three, and may be two or less or four or more. Also, the number of groups is not limited to three, and may be two or less or four or more.

Next, based on the quality data of the operated lot belonging to each group, the superiority or inferiority of the quality data between groups is determined (step S5). Specifically, the control device 10 calls intermediate variables obtained from quality data (purity of NMP after purification, content of impurities, etc.) for each lot, and determines superiority or inferiority of quality data between groups. The quality of each group was ranked in the order of group 3, group 2, and group 1, with group 3 having the highest quality.

The superiority of quality data is preferably determined based on the average value in a plurality of lots (lot group) forming the same group. As a result, variations in the quality data of a plurality of lots within the group are leveled, and it is possible to comprehensively grasp the tendency of quality data between groups.

[Virtual lot verification]
Next, the step of verifying the quality of a lot (virtual lot) assuming an inexperienced state of the production process will be specifically described.

The verification unit 11c, in the same manner as the above-described analysis of the production process that has been operated (steps S1 to S5), based on the state data (virtual state data) and quality data (virtual quality data) for each of a plurality of virtual lots, Divide virtual lots into multiple groups.

Specifically, first, virtual state data set for each of a plurality of virtual lots in a virtual process imitating a production process is stored in the storage unit 14 (step S6).

Here, the “virtual process” is not the production process that actually operates the distillation column 1, but what kind of virtual quality when it is assumed that the distillation column 1 is operated with arbitrarily set virtual state data. It is set up to mimic an operational production process to verify that data can be obtained.

In addition, the "virtual state data" refers to the operation conditions of the distillation column 1 and the composition of raw materials that are arbitrarily set so as to assume an inexperienced state of the production process, that is, to be in a state different from the above-described operated lot. It means state data that is changed and set.

In this embodiment, only the composition of the raw material to be put into the distillation column 1 is changed at random so that the "virtual state data" does not match the actual state data of the operated lot, and the other refining conditions are set as described above. The status data of the virtual lot matched with the production process was used. The lots belonging to the type "inexperienced" in the table of FIG. 2 correspond to the virtual lots, and the "raw material composition" column related to the virtual lot in the table of FIG. 2 is virtual state data indicating the composition of the raw material in each virtual lot.

Next, virtual quality data for each virtual lot obtained based on the virtual state data is stored in the storage unit 14 (step S7).

The “virtual quality data” means the quality data of purified NMP when the raw material is purified by the distillation column 1 using the virtual state data, and the presence or absence of safety problems. In the present embodiment, "virtual quality data" refers to quality data (presence or absence of safety problems and quality of NMP after purification) obtained by experiments or simple simulations that reproduce the production process based on virtual state data. bottom.

The column of "judgment" for the virtual lot in the table in FIG. 2 is the virtual quality data for each virtual lot. In addition, "Weeping" in the table of FIG. 2 indicates that the raw material in the upper stage of the distillation column 1 excessively dripped and the degree of purification deteriorated. "Flooding" indicates that the steam increased excessively in the distillation column 1, the raw material did not flow down, and the purification degree deteriorated. A "safety problem" indicates that a safety-related substance that exhibits explosive properties in the raw material concentrated at the bottom of the tower has been concentrated to a predetermined concentration or higher.

Next, similarly to steps S2 and S3, the analysis unit 11b standardizes the virtual state data and virtual quality data collected in step S6 and converts them into intermediate functions, and then, based on the intermediate variables, the principal component load amount and A principal component score is obtained (step S8). FIG. 5 is a table showing the amount of information of the principal components and virtual quality data for each virtual lot.

Next, the verification unit 11c divides each virtual lot into four groups 4 to 7 based on the virtual quality data (step S9). FIG. 6 is a graph obtained by projecting the composite vector of EV1 to EV3 in the virtual lot on the EV1-EV3 plane (a plane with EV1 on the horizontal axis and EV3 on the vertical axis) plotted on the above-described three-dimensional space. be.

Group 4 in FIG. 6 is a group to which virtual lots with no problem in NMP after purification belong, and group 5 is a group to which virtual lots in which quality of NMP after purification has a problem due to occurrence of weeping belongs. Group 6 is a group to which a virtual lot with a problem in the quality of NMP after purification due to the occurrence of Flooding belongs, and Group 7 is a group to which a virtual lot with a safety problem belongs. Note that FIG. 7 is an enlarged view of a part of the graph in which FIGS.

[Monitoring process prediction]

Next, the step of predicting the quality of the production process to be monitored (monitoring process) will be specifically described.

First, state data of the monitoring process in operation is collected (step S10). State data of the monitoring process are measured by various sensors and stored in the storage unit 14 .

Next, the determination unit 11d determines whether the monitoring process state data measured in step S10 belongs to any of groups 1 to 3 related to the operated lot divided in step S4 or groups 4 to 7 related to the virtual lot divided in step S9. (step S11).

Specifically, the analysis unit 11b retrieves the monitoring process state data stored in the storage unit 14, and derives the principal component loading amount and the principal component score in the same manner as in steps S2 and S3.

Then, the determination unit 11d decides which group the principal component score derived by the analysis unit 11b belongs to: groups 1 to 3 of operated lots divided in step S4 or groups 4 to 7 of virtual lots divided in step S9. or does not belong to any group.

Next, the procedure for determining whether or not the principal component score according to the state data of the monitoring process belongs to any of Groups 1 to 3 of operated lots or Groups 4 to 7 of virtual lots will be described in detail. explain.

First, let EV1, EV2, and EV3 be the principal component scores of the operated arbitrary lot in the three-dimensional space up to the first to third eigenvectors (EV1 to EV3), and X1, Given X2 and X3, identify the operated lot or virtual lot that is the closest composite vector. The distance D between the combined vector (X1, X2, X3) of the monitoring process in the three-dimensional space of EV1-3 and the combined vector (EV1, EV2, EV3) of the operated lot or virtual lot is calculated by Equation 8. be.

Then, the group (recent group) to which the lot with the shortest distance D (recent lot) belongs is determined as the group to which the monitoring process belongs.

In this way, the virtual state data of the virtual lot is set to be different from the actual state data of the operated lot, and the virtual quality data corresponding to the virtual state data of the virtual lot is collected by virtual processes such as experiments and simple simulations. By doing so, the quality of the monitoring process during operation (for example, it is predicted that the monitoring process will proceed in the same way as the past operated lot) based on the operated lot and the virtual lot set to complement this or the monitoring process deviates from the operated lot, but progresses in the same manner as the virtual lot for which the virtual quality data is grasped in advance, etc.). The procedure for determining whether or not the principal component score according to the state data of the monitoring process belongs to any of Groups 1 to 3 of operated lots or Groups 4 to 7 of virtual lots is the same as described above. It is not limited.

Next, the quality data of the monitoring process is predicted based on the quality data of the group to which the monitoring process belongs (step S12).

If the principal component score according to the state data of the monitoring process belongs to any of groups 1 to 3 of the operated lot classified in step S5, the monitoring process is performed based on the actual quality data of the group to which it belongs. It is possible to accurately predict the quality data of products produced in the past from the actual results of the production process.

On the other hand, if the principal component score according to the state data of the monitoring process belongs to any of the virtual lot groups 4 to 7 classified in step S9, the monitoring process is performed based on the quality data of the group to which it belongs. It is possible to predict the quality data of the products produced in the system without preparing an advanced simulation model.

Further, when it is determined that the monitoring process belongs to any one of groups 4 to 7, the control device 10 outputs the quality data of the monitoring process according to the virtual quality data of the virtual lot belonging to groups 4 to 7. Predetermined guidance may be displayed on the display unit 13 in order to stabilize the .

As such guidance, for example, if the monitoring process belongs to Group 4, the hypothetical quality data is good, so the quality data of the monitoring process is also expected to be good. be.

Also, if the monitoring process belongs to Group 5, there is a possibility that "Weeping" will occur, so it is conceivable to display guidance to the effect that the recirculation amount will be increased.

Also, if the monitoring process belongs to group 6, there is a possibility that "flooding" will occur, so it is conceivable to display guidance to reduce the input amount of raw materials.

Furthermore, if the monitoring process belongs to Group 7, there is a risk that substances related to safety will concentrate at the bottom of the column, so it is conceivable to display guidance to increase the amount of high-concentration components extracted from the bottom of the column. be done.

In this way, even if the monitoring process progresses in an inexperienced state, by providing the deviating factors and countermeasures as guidance, the operator of the distillation column 1 can proceed with the operation. It can give you a sense of security.

Furthermore, by displaying the graph shown in FIG. 7 on which the principal component scores of the monitoring process are superimposed on the display unit 13, the user can visually recognize the quality of the refined NMP.

On the other hand, for reasons such as the distance D calculated by Equation 8 being significantly longer than the size of each group 1 to 7, the principal component scores according to the state data of the monitoring process are If it is determined that the virtual lot does not belong to any of Groups 1 to 3 or Groups 4 to 7 of the virtual lot divided in step S9, the monitoring process may differ from both the operated lot and the virtual lot. highly sexual. In this case, the control device 10 may display guidance for calling attention on the display unit 13 .

In addition, in the present embodiment, the case where the quality data of the monitoring process is predicted based on the experienced lot condition data and quality data and the virtual lot condition data and quality data has been described as an example. The quality and quality of the monitoring process may be predicted based on the determination of whether or not it belongs to any of a plurality of groups classified based on the virtual state data and the virtual quality data.

It should be noted that the present invention can be modified in various ways without departing from the spirit of the present invention, and it is a matter of course that the present invention extends to the modified ones.

1: Distillation column 2: Recovery unit 3: Concentration unit 4: Heating device 5: Cooling device 10: Control device 11: Device control unit 11a: Control unit 11b: Analysis unit 11c: Verification unit 11d: Determination unit 12: Input/output unit 13: display unit 14: storage unit

Claims (4)

A method of monitoring a production process for producing a product or service, comprising:
a step of collecting performance status data, which is status data indicating the status of a plurality of operated lots in an operated process, which is a production process performed in the past, and performance quality data, which is quality data indicating the quality of the product or service; ,
Setting virtual state data, which is state data indicating states of a plurality of virtual lots corresponding to inexperienced states different from the plurality of operated lots in the virtual process for verification set by simulating the production process. and collecting virtual quality data, which is quality data indicating the quality of the product or service when the virtual process is executed with the virtual state data;
dividing the operated lot and the virtual lot into a plurality of groups on multi-dimensional coordinates in which a plurality of principal component scores derived from the state data are set on each coordinate axis;
determining whether the monitoring process belongs to any of the groups or not, based on status data indicating the status of the monitoring process, which is the production process to be monitored;
A method for monitoring a production process, comprising: 2. The production process monitoring method according to claim 1, further comprising the step of predicting quality data indicating the quality of said product or service in said monitoring process from the quality data of said group to which said monitoring process belongs. A method of monitoring a production process for producing a product or service, comprising:
setting virtual state data, which is state data indicating states of a plurality of virtual lots corresponding to different states in a virtual process for verification set by simulating the production process, and using the virtual state data to set the virtual process; collecting virtual quality data, which is quality data indicative of the quality of the product or service when performing
a step of dividing the virtual lot into a plurality of groups on multidimensional coordinates in which a plurality of principal component scores derived from the state data are set on each coordinate axis;
determining whether the monitoring process belongs to any of the groups or not, based on status data indicating the status of the monitoring process, which is the production process to be monitored;
A method for monitoring a production process, comprising: a step of collecting performance status data, which is status data indicating the status of a plurality of operated lots in an operated process, which is a production process performed in the past, and performance quality data, which is quality data indicating the quality of the product or service; further includes
The plurality of virtual lots are set corresponding to inexperienced states different from the plurality of operated lots,
4. The production process monitoring method according to claim 3, wherein the operated lot and the virtual lot are classified into a plurality of groups on the multidimensional coordinates.

Images