JP2024070469A - Manufacturing support system and manufacturing support method - Google Patents

Abstract

[Problem] To provide information that serves as a guide to assist the judgment and operation of a user who performs an operation. [Solution] The system includes a plurality of sensors that measure state quantities or operation quantities of a manufacturing facility, a data collection and management unit that collects and stores time series data of a plurality of variables measured by the plurality of sensors, a variable definition unit that defines variables to be analyzed from the stored data and variables measured online, a data shaping unit and a principal component analysis unit that preprocess the variables and perform principal component calculations, a variable correlation display unit that executes multivariate analysis at every predetermined cycle, updates the analysis value and displays it as time series information, and displays the time series information that is a target index in accordance with the time series principal component value, a state change definition unit that defines state changes that are the targets of factor variable extraction from the multi-time series information, a change factor variable calculation unit that executes a process to extract factor variables that are highly correlated with the state-defined state changes, and a change factor variable display unit that displays the factor variables related to the extracted state changes. [Selected Figure] Figure 4

Description

The present invention relates to a system and method that provides guidance and assistance to users who operate manufacturing facilities and manufacturing equipment when adjusting the equipment status and device status.

Patent Document 1 discloses a technology that uses principal component analysis to process multidimensional time series data obtained based on detection signals output by a large number of sensors installed in manufacturing facilities and manufacturing equipment (hereinafter simply referred to as "manufacturing facilities"), and notifies the state of the manufacturing facilities in a visually easy-to-understand output format. In addition, a predictive diagnosis system that detects signs of abnormalities based on the analysis of multidimensional time series data is known, for example, from Patent Document 2.

These systems use multivariate analysis such as principal component analysis to diagnose the condition by combining information from a large number of sensors. By comprehensively and holistically monitoring the condition of the manufacturing equipment, signs of abnormalities in the manufacturing equipment can be detected and reported to the user. This reduces the rate of corrective maintenance and provides advanced support for the operation of manufacturing equipment. In addition, principal components that have an effect on the output variables are extracted, and a large amount of normal data is collected and learned in order to detect signs of abnormalities in the manufacturing equipment, and thresholds for normality and abnormality are determined using statistics, etc.

In addition, a system that displays factor variables that are highly correlated with the results of principal component analysis is known from Patent Document 3.

JP 2001-67117 A JP 2019-204342 A JP 2014-178844 A

For example, in manufacturing facilities such as chemical plants, productivity can fluctuate during operation, although this is not abnormal. Productivity refers to, for example, the basic unit, the rate of defective products, and the energy efficiency of manufacturing. There are various factors that cause this, such as environmental changes such as temperature, changes in the quality of catalysts and other materials used in the manufacturing process, and the effects of maintenance. One method of preventing a drop in productivity is to stop production and perform maintenance, but completely stopping production can lead to a significant drop in production volume and increased maintenance work costs, and there are also disadvantages. Therefore, while continuing production is an option, it is preferable to improve productivity even slightly.

As described in Patent Documents 1 to 3 above, there are many known techniques for predicting and identifying normal and abnormal conditions. However, in order to improve productivity even slightly within the range of normal conditions, there is still a need to rely on the experience of skilled technicians.

However, considering the possibility of experienced technicians retiring due to aging and a future shortage of personnel to handle operations due to a declining birthrate, there is a need to formalize the knowledge of experienced technicians on how to improve productivity, which could ultimately lead to automated control.

The object of the present invention is to provide a system and method for supporting manufacturing that extracts factors in response to various changes in the operating conditions that occur while manufacturing equipment is operating within the normal range, and provides information that serves as a guide to support the judgment and operation of users who perform operations.

In order to solve the above problems, the present invention is characterized by comprising: a plurality of sensors that measure the state quantities or operation quantities of a manufacturing facility; a data collection and management unit that collects and stores time-series data of a plurality of variables measured by the plurality of sensors; a variable definition unit that defines variables to be analyzed from the stored data and variables measured online; a data shaping processing unit and a principal component analysis processing unit that preprocess the variables defined by the variable definition unit and perform principal component calculations; a variable correlation display unit that executes multivariate analysis according to data for a predetermined period at each predetermined cycle, updates the analysis value based on the multivariate analysis for the predetermined period at each predetermined cycle and displays it as time-series information, and displays the time-series information that is the target index in accordance with the time-series principal component value; a state change definition unit that defines state changes that are the target of factor variable extraction from the time-series information based on the multivariate analysis; a change factor variable calculation processing unit that executes a process to extract a factor variable that is highly correlated with the state change defined by the state change definition unit; and a change factor variable display unit that displays the factor variables related to the state change extracted by the factor variable extraction process.

The present invention also includes the steps of: (a) executing a multivariate analysis according to data for a predetermined period at each predetermined cycle, updating an analysis value based on the multivariate analysis for the predetermined period at each predetermined cycle and displaying it as time-series information, and displaying the time-series information serving as a target indicator together with the time-series principal component values; (b) defining state changes from which factor variables are to be extracted from the time-series information based on the multivariate analysis; (c) executing a process to extract factor variables that are highly correlated with the defined state changes; and (d) displaying the factor variables related to the extracted state changes.

The present invention makes it possible to realize a system and method for supporting manufacturing that extracts factors in response to various changes in the operating conditions that occur while manufacturing equipment is operating within the normal range, and provides information that serves as a guide to support the judgment and operation of users who perform operations.

Problems, configurations, and advantages other than those described above will become clear from the description of the embodiment of the invention below.

FIG. 1 is a diagram illustrating an example of time-series data on productivity of a manufacturing facility. 1 is a block diagram showing a schematic configuration of a chemical plant management system according to a first embodiment of the present invention; 2 is a diagram illustrating an example of a manufacturing facility 200 and a sensor group 300. FIG. 1 is a functional block diagram showing the configuration and processing flow of a plant monitoring and control support system 100. 11A and 11B are diagrams showing examples of display on a state change display unit 108. 11A and 11B are diagrams showing examples of display on a state change display unit 108. 11A and 11B are diagrams showing examples of display on a state change display unit 108. 11A and 11B are diagrams showing examples of display on a state change display unit 108. 11A to 11C are diagrams illustrating examples of operations in the state change definition unit 110. FIG. 13 is a diagram showing an example in which the inner product values of all variables are displayed in the order of steps. FIG. 13 is a diagram showing an example of displaying inner products in order of magnitude. FIG. 11 is a diagram showing an example of data of each part of a chemical plant management system according to a second embodiment of the present invention. FIG. 6 is a diagram illustrating an example of a predictive diagnosis system and anomaly detection system 600. FIG. 11 is a diagram showing a change amount of a sensor value. FIG. 13 is a diagram conceptually showing the difference in user behavior expected depending on whether the present invention is applied or not.

The following describes embodiments of the present invention with reference to the drawings. The examples are illustrative for explaining the present invention, and some parts have been omitted or simplified as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

In addition, the position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc., in order to facilitate understanding of the invention. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings.

With reference to Figures 1 to 9, a system for supporting manufacturing and a method for supporting manufacturing according to a first embodiment of the present invention will be described.

First, the basic concept of the present invention will be explained using Figure 1. Figure 1 shows an example of time-series data on the productivity of manufacturing equipment. The horizontal axis shows time, and the vertical axis shows productivity indicators such as the operating rate of the manufacturing equipment and the quality rate of products.

In Figure 1, productivity refers to indicators such as the operating rate of manufacturing equipment, the rate of quality products, and production volume. Additionally, anomaly detection refers to the detection of serious problems such as breakdowns or stoppages of manufacturing equipment, or a significant drop in productivity.

For example, in manufacturing facilities such as chemical plants, annual production plans may be formulated based on an assumed benchmark productivity. If productivity falls below the plan and there is room for improvement, an attempt is made to continue operations while suppressing the decline in productivity (B). In this case, there is a possibility that this could lead to sudden breakdowns or shutdowns of the manufacturing equipment in the future, or unexpected declines in product quality.

On the other hand, in case C: where production is stopped when an abnormality or a sign of an abnormality is detected, and then restarted after the cause of the abnormality is removed, stopping the equipment for maintenance sacrifices the operating rate and ultimately the total productivity.

Therefore, in this invention, A: By continuing to try to improve production efficiency within the normal range, it contributes to improving the manufacturing capacity of the target equipment itself. To this end, it extracts factors according to various changes in the operating conditions that occur while the manufacturing equipment is operating within the normal range, and provides information that serves as a guide to support the judgment and operation of the user who operates the equipment.

The system for supporting manufacturing in this embodiment will be described with reference to FIG. 2. FIG. 2 is a block diagram showing the schematic configuration of a chemical plant management system 1 including the plant monitoring and control support system 100 in this embodiment.

As shown in FIG. 2, the plant monitoring and control support system 100 of this embodiment is a system that monitors the operation of the manufacturing equipment 200 based on detection signals from a sensor group 300 consisting of multiple sensors, extracts slight state changes and their fluctuating factors, and supports efficient user work and equipment operation of the manufacturing equipment 200.

The chemical plant management system 1 in FIG. 2 is, as an example, configured to include a productivity data management system 400, a manufacturing management system 500, a predictive diagnosis system and anomaly detection system 600, a monitoring and control unit 700, and a data collection and management system 800.

In the specification, the productivity data management system 400, the manufacturing management system 500, the predictive diagnosis system and anomaly detection system 600, the monitoring and control unit 700, and the data collection and management system 800 are described as separate functions, but since the present invention relates to the plant monitoring and control support system 100, all or part of these may be configured within the same device as the plant monitoring and control support system 100.

Various data based on the detection signals output from the sensor group 300 are input to the monitoring and control unit 700 and transmitted to the productivity data management system 400, the predictive diagnosis system and anomaly detection system 600, and the plant monitoring and control support system 100 via the data collection and management system 800.

The plant monitoring and control support system 100 is a system that diagnoses the status of the manufacturing equipment 200 and fluctuations in the manufacturing environment, and displays and outputs information that serves as a guide for parameter control of the manufacturing equipment 200. Its detailed functions and structure will be described later.

The productivity data management system 400 manages data on the productivity of products that are measured and inspected, for example, hourly or daily, or for each raw material lot, or for each production lot. Productivity data includes, for example, the product yield rate, the basic unit of raw materials consumed to make a certain amount of products, and the energy efficiency consumed per unit production volume during the manufacturing process.

As a specific example, the productivity data management system 400 has a function of importing and managing analysis results from an analysis device (not shown) that analyzes the quality of raw materials or final products. The productivity data management system 400 also receives data related to raw material lots and production lots from the production management system 500, and manages productivity data for each period, raw material lot, or production lot in accordance with this data.

The manufacturing management system 500 is a system that manages the raw materials used in manufacturing at the manufacturing facility 200, the management of raw material lots and manufacturing lots, inventory management, management of final products, and management of information related to the control of the manufacturing facility 200, and manages manufacturing at the manufacturing facility 200 through the monitoring and control unit 700.

The monitoring and control unit 700 has the function of monitoring the detection signals from the sensor group 300 and controlling the manufacturing equipment 200 based on the data. The monitoring and control unit 700 also controls the manufacturing equipment 200 according to various information managed by the manufacturing management system 500.

Figure 3 shows an example of a manufacturing facility 200 and a sensor group 300. As an example, the manufacturing facility 200 includes a raw material tank 201, a storage tank 202, a reaction tank 203, a reaction tank 204, and a product storage tank 205, and is a manufacturing facility for manufacturing a final product C from raw material groups S and M.

Tanks 201-205 are connected to each other by pipes, through which the raw materials, intermediate products, and final products are transported. In addition to the raw materials, pipes for supplying nitrogen and oxygen are also connected. Furthermore, a large number of heaters, valves, etc. (not shown) are installed either independently or in combination with sensors, and adjustment of these also changes the detection value of sensor group 300. In reaction tanks 203 and 204, raw materials S, M and gas are stirred, heated, pressurized, etc. to cause a reaction, and the final product C is produced.

The sensor group 300 includes, for example, a flow sensor that measures the flow rate of raw materials, etc., a temperature sensor that measures the temperature of the reaction tank 203, etc., a pressure sensor that measures the pressure of the reaction tank 204, etc., a level sensor that measures the storage amount of the product in the product storage tank 205, etc., and a concentration sensor that measures the concentration of the product.

The sensor group 300 also includes flow sensors that measure the supply amount of gases such as nitrogen and oxygen, but the types of sensors are not limited to these, and their placement is of course only one example. For example, temperature sensors can be installed not only in the reaction tank 203, but also in multiple locations within the manufacturing equipment 200. Furthermore, the objects that the temperature sensors measure can include not only raw materials, intermediate products, and final products, but also oil, cooling water, environmental temperature, etc.

In such a manufacturing facility 200, each of the multiple sensors in the sensor group 300 outputs a detection signal, and the monitoring and control unit 700 detects whether the value of each detection signal is a normal value within a predetermined range (upper limit, lower limit) or an abnormal value that exceeds this range.

The detected values of the sensor group 300, including data not directly related to control, are collected and stored by the data collection and management system 800, and become databases in the productivity data management system 400, the predictive diagnosis system and anomaly detection system 600, and the plant monitoring and control support system 100. In addition to storing data, the data collection and management system 800 may also have a function for shaping data, such as standardizing the timestamp information of each detected value of the sensor group.

The output values of the predictive diagnosis system and anomaly detection system 600 are displayed in the same way as the detection signals of the sensor group 300, and the degree of anomaly that is monitored in an integrated manner for the manufacturing equipment 200 can be displayed.

In the present invention, the plant monitoring and control support system 100 further emphasizes and displays relatively small state changes. It is similar to the process monitoring and diagnosis device of Patent Document 3 in that it performs principal component analysis on multidimensional time series data obtained from the sensor group 300, but instead of distinguishing between normal and abnormal, this system monitors and displays information about state changes based on principal component analysis to support control.

Therefore, the present invention does not have a learning model. As it is not a learning-based system, data for learning is not required. Not relying on learning has the advantage that it can respond flexibly and perform analysis based on the most recent data, even when the state of the manufacturing equipment changes due to, for example, weather changes. However, this does not deny the use of the present invention in combination with a monitoring system that has a learning model. As an example, a predictive diagnosis system or anomaly detection system that has a learning model can be incorporated as shown in Figure 2.

Figure 4 shows the configuration and processing flow of the plant monitoring and control support system 100. To execute the above processing, the plant monitoring and control support system 100 is, as an example, composed of an offline data accumulation unit 101, an online data acquisition unit 102, a variable definition unit 103, a data shaping processing unit 104, a principal component analysis processing unit 105, an index definition unit 106, a clustering processing unit 107, a state change display unit 108, a variable correlation display unit 109, a state change definition unit 110, a change factor variable calculation processing unit 111, and a change factor variable display unit 112. In addition, it has an adjustable variable processing unit 113 as a function of imposing constraint conditions that control the state of the manufacturing equipment 200.

The processing in the monitoring processing unit 114, which is composed of a variable definition unit 103, a data shaping processing unit 104, and a principal component analysis processing unit 105, is a general principal component analysis such as that described in Patent Document 1, for example.

The data accumulation unit 101 is a storage unit that stores (accumulates) the detection signals output from the sensor group 300 as raw data for each time. The data acquisition unit 102 acquires the data of the sensor group 300 online at the time being monitored. Depending on the system configuration, the offline data accumulation unit 101 and the online data acquisition unit 102 can be considered as functions of the data collection and management system 800, but in this embodiment they are described as one of the functional blocks in FIG. 4.

A detection signal group to be used for analysis is selected from the detection signal groups of the variable definition unit 103, data accumulation unit 101, and data acquisition unit 102. For example, data from all sensors can be selected, or only the sensor group related to the reaction tank 203 can be selected. In addition, the period to be used for analysis is defined from the time series data of the data accumulation unit 101.

In the present invention, the period of time series data to be analyzed is shorter than that of conventional techniques for constructing learning models, etc. In this embodiment, there is data from more than 200 days prior to the analysis period, but the variable definition unit 103 uses a total of 20 days of data, consisting of one day of online data and the most recent 19 days of offline data. A large amount of data is effective for constructing a learning model, but conditions such as the weather and the state of consumables may differ from those in the most recent analysis period. Therefore, in order to capture slight changes in the most recent conditions, the analysis period in this embodiment is fixed at 20 days, and the analysis data is updated each time an analysis is performed once a day.

The data shaping processing unit 104 performs preprocessing to input data to the downstream principal component analysis unit 105. For example, data groups acquired at different sampling times are standardized to sampling times of one minute or one hour. In addition, intensity normalization processing is performed for each detection signal, and obviously abnormal data is removed. After the detection signal group is preprocessed, principal component analysis processing is performed in the principal component analysis processing unit 105.

By principal component analysis, n sensor data at each time are transformed into principal component space. The direction of principal component axis 1 with the largest variance is expressed by equation (1). s _k is the sensor value after shaping, and P _1,k is the coefficient for the kth variable to transform into principal component axis 1 obtained by principal component analysis.

Similarly, calculate the second principal component score Score2 that maximizes the variance in the direction perpendicular to the first principal component. Plotting Score1 on the horizontal axis and Score2 on the vertical axis allows you to plot two-dimensional principal component scores. Expressing the Score at each time t as an n-dimensional vector Z gives us Equation (2).

The indicator definition unit 106 defines the productivity indicators to be analyzed. Examples include basic units, non-defective product rates, and energy efficiency. Alternatively, analysis can be performed using the output of the predictive diagnosis system and anomaly detection system 600.

The clustering processing unit 107 classifies the indices defined by the index definition unit 106 into multiple classes. Although it is not an essential element, it is used in cases where it is more appropriate to express the productivity index as a numerical value as is. For example, for a manufacturing facility where the yield rate is a productivity index and the target is 80-90%, it may be easier to recognize the productivity index by classifying 90% or more as good, 80-90% as medium, and less than 80% as bad, rather than by numerical value. In such cases, a clustering process is applied to the productivity index.

Next, the state change display unit 108 will be described. As an example of the display, a two-dimensional plot of the principal component scores will be described. 20 days' worth of data is acquired in the variable definition unit 103. There are n principal component scores, the same number as the variables, but it is appropriate to display them in two or three dimensions for display to the operator, and in this embodiment, two dimensions will be described.

As mentioned above, a typical method is to plot two-dimensional principal component scores by plotting Score1 as PC1 on the horizontal axis and Score2 as PC2 on the vertical axis, and to distinguish between normal and abnormal by focusing on these coordinates. The present invention is characterized by focusing on the coordinate change between two points, i.e., vectors, rather than on individual coordinates, and displaying this as information on the change in the state of the manufacturing equipment 200.

Figures 5A and 5B show a representative example of the display of the status change display unit 108 in this embodiment.

Figure 5A is an example of a two-dimensional plot, and the scores of the principal component analysis are plotted as coordinate information, just like in conventional methods. In the present invention, the plots are further connected in chronological order with lines to indicate the direction. Alternatively, a method of adding information such as time and date next to the plot points may be used. In addition, the output from the clustering processing unit 107 is displayed in an overlapping manner. Specifically, in the case of the above-mentioned non-defective rate, if it is 90% or more and is in good condition, it is plotted with a circle, if it is 80-90%, it is in an intermediate state, it is plotted with a triangle, and if it is less than 80%, it is plotted with an x.

In this display, instead of the shape of the plot, the plot may be distinguished by color, such as blue, yellow, or red, or a rank, such as A, B, or C, may be displayed next to the plot points. In addition, the productivity index values may be displayed directly without passing through the clustering processing unit 107, and it is desirable to change the display according to the convenience of the user.

A display method other than the two-dimensional plot will be described with reference to FIG. 5B. The more variables the plot of the principal component axis contains, the more difficult it is to recognize the relationship with the actual equipment. The present invention is characterized in that the principal component analysis process is performed while sequentially inputting online data, and in this embodiment, the analysis is performed using the most recent 20 days of data including the time of online data acquisition. Therefore, the principal component axis changes every time the analysis and display are updated every day. In other words, P _1,k in formula (1) and P _n in formula (2) change every time the analysis and display are performed. If the display is updated every day, the vertical and horizontal axes currently displayed are different from the vertical and horizontal axes displayed on the previous day.

This display with varying axes can be difficult for users to understand and can be hard to use. Therefore, instead of the two-dimensional plot of Figure 5A, the distance between two points on the two-dimensional plot (Euclidean distance) and the direction can be plotted against the time axis.

The top part of Figure 5B is an example of a time series plot of the distance between two points (Euclidean distance). The middle part of Figure 5B plots the change in direction of two vectors defined between three points. For the middle part of Figure 5B, a method is known in which principal component analysis is performed while sequentially updating data. This method focuses on the amount of fluctuation in the principal component axis, and as in Patent Document 3, a threshold value is set for the amount of fluctuation to achieve anomaly detection.

Compared to the present invention, this is a method of monitoring by setting normal/abnormal thresholds on a time series plot similar to the middle part of Figure 5B. On the other hand, the present invention visualizes fluctuations in status, regardless of whether they are normal or abnormal, including the time series plot in the middle part of Figure 5B, and visualizes indicators for extracting the causes of status changes.

The distance in the upper part of Figure 5B is the amount of change in the sensor group detection value, and the direction in the middle part of Figure 5B is closely related to the type of variable that changed during the relevant period. Therefore, if the direction of the plot changes and the distance is large, it is possible to notify the user (issue an alert) that the state of the manufacturing equipment 200 may have changed significantly.

For example, the transition distance from 5/10 to 5/11 exceeds 3, which is larger than the average value of 2 over the 20 days when the data was collected. Furthermore, the state change from 5/10 to 5/11 has a different direction from the state change from 5/9 to 5/10. Here, as an example, 90°±20° is defined as a large change. Around 180°, the variables are positive and negative, but the type of variable to be extracted is similar to 0°, so this is a small angle change.

Based on the above transition length and angle change conditions, the lower part of Figure 5B shows a time series plot in which a large state change is assigned a value of 1 and a small state change is assigned a value of 0. For example, (2) the change from 5/19 to 5/20 has a small transition distance. The angle change is 135°, which is not small, but since the method of the present invention allows for slight state changes to be plotted as angle changes, the state change from 5/19 to 5/20 was considered to be relatively small.

If the condition that both the transition distance and the angle change are large is imposed, when analyzing the status changes over the past 20 days from 5/20, 5/7, 11, 17, and 19 will be displayed as days with large status changes. Clustered productivity symbols may be overlaid on this display. The above is the display of the status change display unit 108 of this embodiment on 5/20. Two-dimensional plots and one-dimensional time series plots may be displayed side by side, or only one of them may be displayed.

Next, we will explain the variable correlation display unit 109. It has a function to display the calculation results of the principal component analysis processing unit 105, and although it is not a required element of the present invention, it may be displayed depending on the application.

Figure 6A is a two-dimensional plot of the same principal component scores as Figure 5A. Figure 6B is a factor plot of the variables used to plot Figure 6A. This is equivalent to showing a vector (P1 _,k, P2 _,k ) consisting of the elements in the first and second rows for the n-dimensional matrix _Pn shown in equation (2). When there are only a few variables, there is no problem with visibility even if it is displayed three-dimensionally as ( _P1,k, P2 _,k, P3 _,k ).

Figure 6B makes it possible to grasp the correlation between variables in the principal component space. For example, it can be seen that variables plotted at the same point behave in the same way. Furthermore, it can be inferred from this figure that for the 5/20 data plotted in the lower left of Figure 6A, the factor in the lower left of Figure 6B is relatively large, and the factor in the upper right is large as a negative value. Figure 6B is effective for grasping the status if there are only a few types of variables, but when dealing with tens or hundreds of variables, it is difficult to recognize important variables, and it is not suitable as a method of supporting user operations.

Therefore, in the present invention, by providing a state change definition unit 110 and a change factor variable calculation processing unit 111, variables that are analyzed from among many variables as being important for the state change focused on by the user are output to a change factor variable display unit 112.

The function of the state change definition unit 110 in this embodiment is to select the date on which the factor variables are to be displayed. For example, there is a method in which 5/11, which is suggested to be a date with a large state change from the time series plot in FIG. 5B, is automatically selected. Alternatively, a method in which 5/20, which is the most recent date, is automatically selected may be used. Alternatively, a method in which the user checks the time series plot screen in FIG. 5A or FIG. 5B and selects a date on the screen may be used.

Another method of operation in the state change definition unit 110 is shown in Figure 7. The user may visually confirm the state change to be analyzed in Figure 5A, which is a two-dimensional plot displayed in the state change display unit 108, and select it from the monitor screen by clicking the pointer, etc. For example, when focusing on the state change from 5/10 to 5/11, this method selects the line connecting 5/10 and 5/11 as shown by arrow (A), and can be expressed by equation (3). In this example, t = 5/11, k = 1.

Another method for defining state changes is to insert an arrow into a two-dimensional plot diagram. As shown in the variable correlation display unit 109, highly correlated variables can be displayed as factor vectors in the principal component space. Therefore, in order for the user to focus on the change from 5/10 to 5/11 and perform an analysis, the direction of the state change can be defined by defining an arrow (B) parallel to the transition direction from 5/10 to 5/11. For example, this method may be performed by operating the mouse of the monitoring device, or may be defined on a touch panel screen.

The advantage of this method is that you can define any direction. Selecting data based on dates or plot points only extracts factor variables for about 19 types of state changes in this example. On the other hand, by inserting arrows into a two-dimensional plot, you can examine factor variables in directions that do not exist in the data. Alternatively, you can select the start and end points of the arrow.

The operation of the state change definition unit 110 is useful not only for data analysis but also for prior consideration of adjusting plant parameters. For example, if the production efficiency is relatively good at the time of 5/20, it can be assumed that the production efficiency will decrease if the state moves to the right of the two-dimensional plot. Therefore, it is possible to use the operation of the state change definition unit 110 to grasp in advance the variables with high correlation in the left-right arrow (C), i.e., the PC1 axis direction, and prepare to adjust the variables that move the state to the left if there is a tendency for the state to move to the right. In addition, the reverse operation of the normalization process performed by the data shaping processing unit 104 may be performed from the length and direction of the arrow defined during the principal component display, and an example of the sensor detection value based on the actual physical quantity may be displayed.

The change factor variable calculation processing unit 111 finds variables that are highly correlated with the state change defined in the state change definition unit 110. In this embodiment, the n factor plots calculated in the principal component analysis processing unit 105 and, in some cases, output to the variable correlation display unit 109 are treated as vectors, and the inner product of the vectors defined in the state change definition unit 110 is calculated as shown in equation (4).

It also includes the implementation of a process for sorting the n dot product values and displaying them in the change factor variable display unit 112. There is also a process for sorting in order of the dot product value or the absolute value of the dot product. Alternatively, there may be a process for selecting the 10 largest dot product values.

The change factor variable display unit 112 displays the results of the change factor variable calculation processing unit 111.

Figure 8 is an example of the inner product values of all variables shown in equation (4) displayed in the order of steps. Figure 9 is an example of displaying the inner products in order of magnitude. The white parts of the bar graph indicate that they behave in the same way as the variable to the left of them and can therefore be considered to be the same variable. Alternatively, 10 to 20 variables with large absolute values of inner products can be displayed as variables with high correlation, and variables with small absolute values of inner products can be displayed as variables with low correlation. Figure 9 shows the display as a bar graph, but a table and correlation values can also be displayed.

Furthermore, as an example of a method for extracting factor variables, the multiplication result of the inner product value shown in equation (5) and the fluctuation of each sensor value of the sensor group 300 may be displayed.

As described above, the system for supporting manufacturing in this embodiment includes a plurality of sensors (sensor group 300) that measure state quantities or operation quantities of the manufacturing equipment 200, a data collection and management unit (data collection and management system 800) that collects and holds time series data of a plurality of variables measured by the plurality of sensors, a variable definition unit 103 that defines variables to be analyzed from the held data and variables measured online, a data shaping processing unit 104 and a principal component analysis processing unit 105 that preprocess the variables defined in the variable definition unit 103 and perform principal component calculations, and a calculation unit 106 that calculates the time series data according to the data for a predetermined period at a predetermined cycle. It is equipped with a variable correlation display unit 109 that executes multivariate analysis, updates the analysis values based on the multivariate analysis for a specified period at each specified cycle and displays them as time-series information, and displays the time-series information that is the target index in accordance with the time-series principal component values, a state change definition unit 110 that defines state changes that are the subject of factor variable extraction from the time-series information based on the multivariate analysis, a change factor variable calculation processing unit 111 that executes a process to extract factor variables that are highly correlated with the state changes defined by the state change definition unit 110, and a change factor variable display unit 112 that displays the factor variables related to the state changes extracted by the factor variable extraction process.

The time series information that serves as the target indicator is actual data on the productivity indicator or a predicted value based on predictive diagnosis.

The manufacturing support system and manufacturing support method of this embodiment visualize slight changes in the state of manufacturing equipment that is operating normally. In addition, by extracting the causal variables of this state change, it is possible to provide guidelines for adjusting equipment parameters to improve the production efficiency of the manufacturing equipment. It is also possible to formalize parameter adjustments that previously relied on the experience of skilled engineers.

With reference to Figures 10 to 13, a system for supporting manufacturing and a method for supporting manufacturing according to a second embodiment of the present invention will be described.

Figure 10 shows an example of data for each part of the chemical plant management system of this embodiment. Figure 10 shows data for the period 3/1 to 3/20.

Productivity is displayed in the productivity data management system 400, and fluctuates around 0.9 during the target period. Information from the predictive diagnosis system and anomaly detection system 600 based on principal component analysis can also be displayed. The monitoring and control unit 700 monitors the detection signals of each of the multiple sensors in the sensor group 300 and determines the abnormality of each individual sensor. If it is easier to recognize the overall picture by displaying the productivity data management system 400, predictive diagnosis system and anomaly detection system 600, monitoring and control unit 700, and plant monitoring and control support system 100 on the same display device, they may be displayed side by side as shown in FIG. 10.

The predictive diagnosis system and anomaly detection system 600 are not essential components of the present invention, but are described in detail below for comparison with the configuration shown in Figure 4 of the present invention.

Figure 11 shows an example of a predictive diagnosis system and anomaly detection system 600 that determines whether there is an anomaly in the behavior of the detection signals of the sensor group 300 obtained through the data collection and management system 800.

The data accumulation unit 601 accumulates past detection values of the sensor group 300 in order to construct a learning model. In addition, data related to productivity collected and analyzed by the productivity data management system 400 is also accumulated as offline data. Note that the functions of the data collection and management system 800 may be used as the data accumulation unit 601.

The variable definition unit 602 selects items necessary for constructing a learning model for process monitoring from the sensor detection values accumulated in the data accumulation unit 601, and defines n input variables by synthesizing new variables such as management indicators as necessary.

The data reforming processing unit 603 performs normalization processing and removes outliers and data not used for learning on the input variables defined in the variable definition unit 602, and performs preprocessing to input the data to the subsequent principal component analysis processing unit 604.

The principal component analysis processing unit 604 calculates the direction from the center of gravity where the variance of the data is maximum (first principal component), and then calculates the location where the variance is maximum in the direction perpendicular to the first principal component (second principal component), repeating this process for the number of data dimensions. The above process is the same as in the present invention, except that only offline data is used.

P _n in equation (2) is fixed as P _n(learning) when learning is completed, and this P _n(learning) is used when processing online data.

The principal component score extraction processing unit 606, as an example, creates a two-dimensional graph showing the relationship between the score of the first principal component analysis and the score of the second principal component analysis based on the results of the principal component analysis processing. This corresponds to the display of the predictive diagnosis system and anomaly detection system 600 in FIG. 10. Information on the productivity index defined in the index definition unit 605 may be added to the two-dimensional score graph.

If it is possible to separate the plots of normal values from the plots of abnormal values for the defined productivity, it is possible to diagnose abnormalities. For example, in the display of the predictive diagnosis system and anomaly detection system 600 in FIG. 10, if all "x: bad" can be plotted in the abnormal region, it means that normal/abnormal judgments can be made accurately. In addition, the extent to which the observed data is away from the boundary between normal and abnormal can be quantified as the degree of abnormality or normality. Indices such as Q statistics and T2 statistics can be used as a quantification method. These statistics are calculated by the statistics calculation unit 607, and the abnormality threshold is defined by the threshold definition unit 608, completing the model.

During monitoring, a variable definition unit 610 selects detection values to be used for analysis from the online data acquired from the sensor group 300 by the data acquisition unit 609 through the data collection and management system 800, and calculates principal component scores from preprocessing and Pn _(learning) using the conditions at the time of learning. Furthermore, after shaping the data in a data shaping processing unit 611, a statistics calculation processing unit 612 calculates statistics such as the degree of abnormality and the degree of normality. An abnormality determination processing unit 613 determines whether the statistics exceed the threshold value defined in the threshold definition unit 608, and displays the diagnosis result of normality or abnormality on a status determination display unit 614.

For the predictive diagnosis system and anomaly detection system 600 in Figure 10, data for the period of analysis (3/1 to 3/20) is plotted in the area surrounded by a dotted line. Based on learning from past data, the area at the bottom right is determined to be abnormal, and there are no signs of moving into the abnormal area in the following period. The sensor values are displayed on the screen of the monitoring and control unit 700. In reality, the detected values of approximately 100 sensors are monitored, but here, one detected value each from the flow sensor, pressure sensor, concentration sensor, and temperature sensor attached to the reaction tank 203 is displayed.

Information from the status change display section 108 is displayed on the screen of the plant monitoring and control support system 100 in FIG. 10. In a principal component analysis based on the data for these 20 days, a large status change was detected from 3/14 to 3/15. Furthermore, it is possible to distinguish the status of 3/19 and 3/20, which were good productivity, from the two-dimensional plot, and it can be inferred that the status change from 3/14 to 3/15 had a positive effect. Furthermore, by extracting and understanding the factors behind the status change from 3/14 to 3/15, the user can obtain guidelines for parameter adjustment from 3/20 onwards.

By selecting 3/15 on the screen of the plant monitoring and control support system 100 in Figure 10, the state change to be analyzed is defined.

Figure 12 shows the amount of change in the sensor values. The upper graph in Figure 12 shows the results of extracting the largest amount of normalized change in the detection values of the group of sensors 300 installed in the raw material tank 201 and the reaction tank 203 from 3/14 to 3/15. Furthermore, the lower graph in Figure 12 shows the results of multiplying the inner product value of the factor variables extracted from the principal component analysis using equation (5), which shows that although the amount of change in the detection values of A_Upper Pressure 3 and A_Upper Pressure 4 was large in this state change, their contribution to the state change was small.

The table on the right of Figure 12 shows the top 10 variables that caused state changes from March 14th to March 15th, extracted from the inner product of the sensor value change and the principal component analysis. From this table, the user can interpret the state changes from March 14th to March 15th, and obtain information that can serve as a guide for improving productivity in the future. Note that the table on the right of Figure 12 shows normalized values, but it is also possible to perform the inverse operation of the normalization process performed by the data reforming processing unit 104 and display an example of a sensor detection value based on the actual physical quantity.

Figure 13 conceptually shows the expected difference in user behavior depending on whether or not the present invention is applied.

The change in state from 3/14 to 3/15 was an expected result for an experienced worker, and adjustments to the manufacturing equipment 200 were possible based on past knowledge. However, the conventional operation by the monitoring and control unit 700 shown in the left diagram of Figure 13 may not be fully understood by users with little operational experience.

On the other hand, as shown in the right diagram of FIG. 13, application of the plant monitoring and control support system 100 of the present invention can contribute to improving the sense of satisfaction and understanding by using data when explaining to inexperienced users, for example, for the purpose of passing on experience. Alternatively, the manufacturing equipment 200 can be operated while multiple users share the hypothesis that the change in condition from 3/14 to 3/15 had a positive effect.

The present invention is not limited to the above-described embodiments, but includes various modified examples. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. It is also possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. It is also possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

Furthermore, the above-mentioned configurations, functions, processing units, processing means, etc. may be realized in hardware, in part or in whole, for example by designing them as integrated circuits. Furthermore, the above-mentioned configurations, functions, etc. may be realized in software, by a processor interpreting and executing a program that realizes each function. Information on the programs, tables, files, etc. that realize each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card, SD card, or DVD.

1... Chemical plant management system 100... Plant monitoring and control support system 101, 601... Data storage unit (offline)
102,609...Data acquisition section (online)
103, 602, 610...variable definition section 104, 603, 611...data shaping processing section 105, 604...principal component analysis processing section 106...index definition section 107...clustering processing section 108...state change display section 109...variable correlation display section 110...state change definition section 111...change factor variable calculation processing section 112...change factor variable display section 113...adjustable variable processing section 114...monitoring processing section 200...manufacturing equipment 201...raw material tank 202...storage tank 203, 204...reaction tank 205...product storage tank 300...sensor group 400...productivity data management system 500...manufacturing management system 600...predictive diagnosis system and anomaly detection system 605...index definition section 606...principal component score extraction processing section 607...statistics calculation section 608...threshold definition section 612...statistics calculation processing section 613: Abnormality determination processing unit 614: Status determination display unit 700: Monitoring and control unit 800: Data collection and management system (data collection system)

Claims (13)

A plurality of sensors for measuring state quantities or operation quantities of the manufacturing equipment;
a data collection and management unit that collects and stores time-series data of a plurality of variables measured by the plurality of sensors;
a variable definition unit that defines variables to be analyzed based on the stored data and variables measured online;
a data shaping unit and a principal component analysis unit that preprocess the variables defined in the variable definition unit and perform principal component calculations;
a variable correlation display unit that executes a multivariate analysis according to data for a predetermined period at each predetermined cycle, updates an analysis value based on the multivariate analysis for the predetermined period at each predetermined cycle and displays it as time-series information, and displays the time-series information serving as a target index together with the time-series principal component values;
a state change definition unit that defines state changes to be targets for factor variable extraction from the time-series information based on the multivariate analysis;
a change factor variable calculation processing unit that executes a process of extracting a factor variable having a high correlation with the state change defined by the state change definition unit;
a change factor variable display unit that displays the factor variables related to the state change extracted by the factor variable extraction process;
A manufacturing support system comprising: 2. The system for supporting manufacturing according to claim 1,
A manufacturing support system characterized in that analytical values based on multivariate analysis for a predetermined period are updated for each predetermined cycle and displayed as time-series information, and variable relationships are displayed as factor plots. 2. The system for supporting manufacturing according to claim 1,
A manufacturing support system, characterized in that the numerical data of the time-series information serving as the target index is displayed in accordance with the time-series principal component values as classes classified by clustering processing. 2. The system for supporting manufacturing according to claim 1,
a manufacturing support system comprising: when defining a state change that is to be the subject of factor variable extraction from the time-series information based on the multivariate analysis, selecting any two points from a predetermined period from analysis values based on the multivariate analysis for the predetermined period for each predetermined cycle. 2. The system for supporting manufacturing according to claim 1,
a manufacturing support system, characterized in that, when defining a state change that is to be the subject of factor variable extraction from the time-series information based on the multivariate analysis, the system defines the state change by inputting a direction of the state change on a screen that updates an analysis value based on the multivariate analysis for a predetermined period at each predetermined cycle and displays the analysis value as time-series information. 2. The system for supporting manufacturing according to claim 1,
A manufacturing support system characterized in that, in a process of extracting factor variables having a high correlation with state changes, the system calculates, for each variable, the inner product value of a factor vector of variable relationships in a principal component space and a state change vector that is the subject of factor variable extraction defined from the time-series information. A system for supporting manufacturing according to claim 6, comprising:
A manufacturing support system, characterized in that in a process of extracting factor variables having a high correlation with state changes, the system multiplies the inner product of a factor vector of variable relationships in a principal component space and a state change vector that is the subject of factor variable extraction defined from the time-series information by the amount of change in the preprocessed variable. 2. The system for supporting manufacturing according to claim 1,
The manufacturing support system is characterized in that, for time series information in which analytical values based on multivariate analysis for a specified period are updated for each specified cycle, the analytical values based on multivariate analysis display the Euclidean distance between plots of principal component analysis at the relevant time and the time one cycle prior. The system for supporting manufacturing according to claim 8,
A manufacturing support system that issues an alert when the Euclidean distance is greater than a predetermined value. 2. The system for supporting manufacturing according to claim 1,
A manufacturing support system characterized by displaying a factor variable having a high correlation with the extracted state change, or displaying a controllable variable when plotting a variable relationship. 2. The system for supporting manufacturing according to claim 1,
A manufacturing support system, wherein the time-series information serving as the target index is performance data of a productivity index or a predicted value based on a predictive diagnosis. The system for supporting manufacturing according to claim 2,
A manufacturing support system characterized by converting the magnitude of correlation based on principal component analysis and the defined amount of state change for factor variables that are highly correlated with the extracted state change into a physical quantity of the sensor detection value and displaying it. A method for assisting in manufacturing, comprising the steps of:
(a) executing a multivariate analysis according to data for a predetermined period at each predetermined cycle, updating an analysis value based on the multivariate analysis for the predetermined period at each predetermined cycle and displaying the updated analysis value as time-series information, and displaying the time-series information serving as a target index together with time-series principal component values;
(b) defining state changes to be subjected to factor variable extraction from the time-series information based on the multivariate analysis;
(c) executing a process of extracting factor variables having a high correlation with the defined state change;
(d) displaying the factor variables related to the extracted state changes.

Images