System and method for supporting factor analysis of time series data

Abstract

To allow a countermeasure against an analyzed cause to be clarified and presented in factor analysis of time series data.SOLUTION: A factor analysis support system for time series data includes: a time series causal model storage unit 31 which holds a time series causal model for each of time series variables; a contribution degree calculation unit 15 which uses the time series causal model stored in the time series causal model storage unit 31 and time series data of an analysis object to calculate a contribution degree return rate obtained by evaluating how much a contribution degree of each time series variable at each time should be returned to a contribution degree of another time series variable at another time; a contribution degree return unit 17 which returns the contribution degree of each time series variable at each time to the contribution degree of another time series variable at another time on the basis of the contribution degree return rate calculated by the contribution degree calculation unit and calculates transition of contribution degrees due to a root factor to an objective variable; and an output unit 18 which outputs the contribution degree due to the root factor to the objective variable, which is calculated by the contribution degree return unit 17.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a time-series data factor analysis support system and a time-series data factor analysis support method.

Background Art

[0002] In quality improvement in the manufacturing industry, it is essential to perform factor analysis on product defects and consider and implement specific countermeasures based on the results from time-series data such as various sensor values in the manufacturing process. When performing factor analysis from time-series data such as various sensor values in the manufacturing process, it is known to use AI (Artificial Intelligence). That is, it is known to let AI learn the relationship between manufacturing data and the defect rate of products, and mechanically extract factors that contributed to the increase in the defect rate using XAI (Explainable AI) technology. Here, if the input to AI is a time variable (factor × time), it is possible to quantify and visualize the time-series transition of the influence on quality by each time and each factor.

[0003] Patent Document 1 describes a technique that enables the estimation of non-linear causal relationships between dimensions using time-series multivariate data obtained from a system. That is, Patent Document 1 describes learning a non-linear regression model that predicts data at a certain time from data at past times using data of an input time-series multi-dimensional numerical vector, and calculating the strength of causality of each dimension in the data of the time-series multi-dimensional numerical vector using the non-linear regression model.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] As described in Patent Document 1, it has been conventional practice to estimate the strength of causality when a certain state occurs. This makes it possible to determine, for example, which of the time-series data, such as various sensor values ​​in the manufacturing process, is linked to a defect that occurs during manufacturing. However, conventional analysis methods can only determine which of the time-series data, such as temperature and pressure of each part, is linked to the defect, and it is not always possible to determine how to control the system to prevent the defect.

[0006] For example, suppose that when temperature sensors measure the temperature at point A and point B, and pressure sensors measure the pressure at point C as time-series data, conventional analysis methods determine that the temperature information at point B is causing a decline in manufacturing quality. In this case, if the temperature at point B is a temperature that cannot be directly controlled on the manufacturing line, the analysis results cannot determine what measures are necessary to reduce the temperature at point B. Therefore, there was a problem in that concrete measures could not be taken to address the decline in manufacturing quality.

[0007] Furthermore, even if the temperature at point B could be directly controlled, there was a problem in that if it was not possible to estimate in advance how much quality improvement could be achieved by controlling the temperature at the time when the impact of quality deterioration was at its highest, it would be impossible to actually implement countermeasures.

[0008] In view of the above, the present invention aims to provide a time-series data factor analysis support system and a time-series data factor analysis support method that can clearly present countermeasures for the analyzed causes. [Means for solving the problem]

[0009] To solve the above problems, for example, the configuration described in the claims may be adopted. The present invention includes multiple means for solving the above problems, but to give one example, the time series data factor analysis support system of the present invention is a time series data factor analysis support system for analyzing the changes in the degree of contribution of time series variables to an objective variable, and comprises: a time series causal model storage unit that stores a time series causal model for each time series variable; a contribution calculation unit that uses the time series causal model stored in the time series causal model storage unit and the time series data to be analyzed to calculate a contribution reduction rate which evaluates how much the degree of contribution of each time series variable at each time should be reduced to the degree of contribution of other time series variables at other times; a contribution reduction unit that reduces the degree of contribution of each time series variable at each time to the degree of contribution of other time series variables at other times based on the contribution reduction rate calculated by the contribution calculation unit and calculates the changes in the degree of contribution due to root causes to the objective variable; and an output unit that outputs the degree of contribution due to root causes to the objective variable calculated by the contribution reduction unit. [Effects of the Invention]

[0010] According to the present invention, by calculating and outputting the reduction rate of the degree of contribution to the fundamental factors, it is possible to present to the user the processes that worsen quality, etc., focusing on factors that make it easier for the user to consider countermeasures. Other issues, configurations, and effects not mentioned above will be clarified by the following description of the embodiments. [Brief explanation of the drawing]

[0011] [Figure 1] This is a configuration diagram showing an example of a factor analysis support system according to the first embodiment of the present invention. [Figure 2] This is a configuration diagram showing an example of the hardware of a factor analysis support system according to an embodiment of the first embodiment of the present invention. [Figure 3] This figure shows an example of a time-series data table of actual results obtained by a factor analysis support system according to the first embodiment of the present invention. [Figure 4] This figure shows an example of a non-time-series data table of actual results obtained using a factor analysis support system according to the first embodiment of the present invention. [Figure 5]This figure shows an example of a target time-series data table using a factor analysis support system according to a first embodiment of the present invention. [Figure 6] This figure shows an example of a target non-time series data table using a factor analysis support system according to a first embodiment of the present invention. [Figure 7] This is a configuration diagram showing an example of a predictor generation unit in a factor analysis support system according to a first embodiment of the present invention. [Figure 8] This figure shows an example of actual time variable data obtained by a factor analysis support system according to a first embodiment of the present invention. [Figure 9] This flowchart shows an example of processing by the time variable generation unit of the factor analysis support system according to the first embodiment of the present invention. [Figure 10] This flowchart shows an example of processing by the predictor generation unit of the factor analysis support system according to the first embodiment of the present invention. [Figure 11] This figure shows an example of generating a time-series causal model using a factor analysis support system according to a first embodiment of the present invention. [Figure 12] This figure shows an example of lag variable data obtained by a factor analysis support system according to the first embodiment of the present invention. [Figure 13] This figure shows an example of time-series dependent variable group data obtained using a factor analysis support system according to the first embodiment of the present invention. [Figure 14] This flowchart shows an example of processing by the lag variable generation unit of the factor analysis support system according to the first embodiment of the present invention. [Figure 15] This is a configuration diagram showing an example of a contribution calculation unit by a factor analysis support system according to a first embodiment of the present invention. [Figure 16] This figure shows an example of target time variable data obtained by a factor analysis support system according to a first embodiment of the present invention. [Figure 17] This figure shows an example of the contribution of each time variable by the factor analysis support system according to the first embodiment of the present invention. [Figure 18]It is a diagram showing an example of contribution degrees by non-time series variables by a factor analysis support system according to a first embodiment example of the present invention. [Figure 19] It is a flowchart showing an example of the processing of a contribution degree calculation unit by a factor analysis support system according to a first embodiment example of the present invention. [Figure 20] It is a configuration diagram showing an example of reduction processing of contribution degrees by time variables by a factor analysis support system according to a first embodiment example of the present invention. [Figure 21] It is a diagram showing an example of target lag variable data by a factor analysis support system according to a first embodiment example of the present invention. [Figure 22] It is a diagram showing an example of contribution degree reduction rate data by a factor analysis support system according to a first embodiment example of the present invention. [Figure 23] It is a flowchart showing an example of reduction processing of contribution degrees by time variables by a factor analysis support system according to a first embodiment example of the present invention. [Figure 24] It is a flowchart showing an example of the processing of a reduction rate calculation unit by a factor analysis support system according to a first embodiment example of the present invention. [Figure 25] It is a flowchart showing an example of the processing of a contribution degree reduction unit by a factor analysis support system according to a first embodiment example of the present invention. [Figure 26] It is a diagram showing an example of a factor analysis screen in a result output unit by a factor analysis support system according to a first embodiment example of the present invention. [Figure 27] It is a diagram showing an example of a countermeasure consideration screen by a factor analysis support system according to a second embodiment example of the present invention. [Figure 28] It is a configuration diagram showing an example of pre-effect verification processing including countermeasure consideration by a factor analysis support system according to a second embodiment example of the present invention. [Figure 29] It is a diagram showing an example of change content data by a factor analysis support system according to a second embodiment example of the present invention. [Figure 30] It is a flowchart showing an example of the processing of a change content extraction unit by a factor analysis support system according to a second embodiment example of the present invention. [Figure 31] This flowchart shows an example of processing by the change data generation unit of the factor analysis support system according to a second embodiment of the present invention. [Modes for carrying out the invention]

[0012] <Example of the first embodiment> Hereinafter, a time-series data factor analysis support system and time-series data factor analysis support method according to a first embodiment of the present invention will be described with reference to Figures 1 to 26.

[0013] [Overall structure of the factor analysis support system] Figure 1 shows the overall configuration of the factor analysis support system in this embodiment. As shown in Figure 1, the factor analysis support system consists of multiple computers 10, 20, and 30, and a terminal 40 connected to each computer 10-30 via a network NW. However, the configuration of multiple computers 10-30 and terminal 40 is just one example; for example, computers 10-30 could be configured as a single computer, and terminal 40 does not have to be configured separately from the computers.

[0014] The computer 10 comprises a predictor generation unit 11, a time series causal model generation unit 12, a time variable generation unit 13, a lag variable generation unit 14, a contribution calculation unit 15, a reduction rate calculation unit 16, a contribution reduction unit 17, and a result output unit 18.

[0015] The computer 20 includes a real-time series data storage unit 21, a real-time non-time series data storage unit 22, a target time series data storage unit 23, and a target non-time series data storage unit 24. The computer 30 includes a time-series causal model storage unit 31.

[0016] Figure 2 shows example hardware configurations for computers 10, 20, and 30. While computer 10 is used as an example, computers 20 and 30 have similar configurations. Furthermore, terminal 40 also has essentially the same configuration as computer 10. The computer 10 comprises a processor 1, main memory 2, secondary memory 3, input device 4, output device 5, and network interface 6. Processor 1 controls the execution of programs stored in main memory 2 or sub-memory 3 in main memory 2. By executing the programs, Processor 1 configures the processing units and storage units shown in Figure 1 in main memory 2 or sub-memory 3.

[0017] Input device 4 consists of input devices such as a keyboard and mouse, and accepts input operations from the operator. Output device 5 consists of a display device and an audio output device, and displays processing results and communicates the processing results by voice. The network interface 6 transmits and receives data with other computers 20, 30 and terminal 40 via the network NW.

[0018] [Examples of each data set] Next, we will describe an example of the data handled by the system in this example. Figure 3 shows the actual time-series data of temperature A, pressure B, etc., measured by sensors installed on the manufacturing line used in this embodiment, which are stored in the actual time-series data storage unit 21. As shown in Figure 3, the actual time-series data is stored in the actual time-series data storage unit 21 in chronological order for each manufacturing ID, starting with the actual values ​​of temperature A, pressure B, ... at time 0, the actual values ​​of temperature A, pressure B, ... at time 1, and the actual values ​​of temperature A, pressure B, ... at time 2.

[0019] Figure 4 shows the non-time-series data of actual performance measured in the inspection process of sensors installed on the manufacturing line used in this embodiment, and stored in the non-time-series data storage unit 22. Here, the hardness, turbidity, amount of substances added in the manufacturing process, and reaction time required for manufacturing of the products manufactured on the manufacturing line are stored. As shown in Figure 4, the actual non-time-series data is stored as a single record for each manufacturing ID, including hardness, turbidity, additive amount, reaction time, etc.

[0020] Figure 5 shows the time-series data of temperature A, pressure B, ..., which are the factors for factor analysis of defective products and are stored in the target time-series data storage unit 23. Figure 5 shows the actual values ​​of temperature A, pressure B, etc., at time 0 for the target time series data, which are the manufacturing IDs being analyzed.

[0021] Figure 6 shows non-time-series data such as hardness, turbidity, amount of additive, and reaction time, which are stored in the target non-time-series data storage unit 24 and are used for factor analysis of defective products. Figure 6 shows the non-time-series data for which the analysis was conducted, including hardness, turbidity, amount added, and reaction time for the manufacturing ID of the subject. In this context, polymer products are the focus, and the more cloudy they are, the more defective they are considered. "Cloudiness" is used as an indicator of defect.

[0022] [Processing in the predictor generation unit] Next, we will explain the process of supporting factor analysis of time series data using such time series and non-time series data, as shown in the configuration of Figure 1. Figure 7 shows the configuration of the predictor generation unit 11. The predictor generation unit 11 uses the actual time-series data stored in the actual time-series data storage unit 21 (Figure 3) and the actual non-time-series data stored in the actual non-time-series data storage unit 22 (Figure 4) to perform prediction processing that learns the relationship between past actual data (time-series data and non-time-series data) and the degree of deterioration (turbidity). For this learning, for example, learning using a machine learning model may be applied, but learning using a simple regression model may also be used.

[0023] First, the time variable generation unit 13 generates actual time variable data D1 from the actual time series data stored in the actual time series data storage unit 21 in order to generate a predictor. Figure 8 shows an example of actual time variable data D1 generated by the time variable generation unit 13. The actual time variable data D1 shows the values ​​of temperature A and pressure B extracted for each manufacturing ID from time 0 (t=0) to time 232 (t=232: the last time). The actual time variable data D1 consists of one row of data per manufacturing ID, and since a variable is generated with the number of time-series variables (such as temperature A) multiplied by the number of minutes of the time, the data is often quite wide. The actual time variable data D1 shown in Figure 8 and the actual non-time series data (Figure 4) are supplied to the predictor generation unit 11.

[0024] Figure 9 is a flowchart showing the procedure of the processing performed by the time variable generation unit 13. First, the time variable generation unit 13 acquires the actual time series data stored in the actual time series data storage unit 21 (step S11). The actual time series data here is, for example, data such as temperature A and pressure B for each manufacturing ID, as shown in Figure 3. Then, the time variable generation unit 13 divides the acquired time series data into data with the same manufacturing ID to obtain divided data (step S12).

[0025] Next, the time variable generation unit 13 starts loop processing of the divided data (step S13a). Once the loop processing starts, the time variable generation unit 13 flattens all the time data for each time series change into one record and adds it to the time variable data (step S14). The time variable generation unit 13 then finishes loop processing of the divided data (step S13b). By repeating the loop start in step S13a and the loop end in step S13b for the number of data points, the time variable generation unit 13 obtains the actual time variable data shown in Figure 8. Then, the time variable generation unit 13 outputs the acquired actual time variable data (step S15).

[0026] Figure 10 is a flowchart showing the processing steps performed by the predictor generation unit 11. The predictor generation unit 11 acquires the actual time variable data shown in Figure 8 and the actual non-series data shown in Figure 4 (step S21). Then, the predictor generation unit 11 combines the acquired actual time variable data and actual non-series data to generate training data (step S22). Here, the actual time variable data is arranged in one row for each manufacturing ID, as shown in Figure 8. Therefore, using the actual time variable data as a key, the actual non-series data and manufacturing ID are combined with this data to generate training data, and the training process is performed. Then, the target variable is specified from the training data, and the trained predictor 19 is output (step S23).

[0027] [Processing of the time-series causal model generation unit] Figure 11 shows the processing performed by the time-series causal model generation unit 12. The lag variable generation unit 14 generates actual lag variable data D2 from the actual time series data stored in the actual time series data storage unit 21. Figure 12 shows an example of actual lag variable data. Here, as shown in Figure 12, with a time window length T=3, the data from the period 3 time points before to 1 time point before becomes the respective variables, and a lag variable for time t(4,5,...) is obtained. In other words, when the time window length T=3, data for negative time points where no data exists below time t=3 is required, so it starts from t=4.

[0028] Returning to the explanation of Figure 11, the lag variable generation unit 14 generates time series target variable data D3 from actual time series data. Figure 13 shows an example of the time-series target variable data D3 generated by the lag variable generation unit 14. As shown in Figure 13, the time-series dependent variable data D3 represents the values ​​of each time-series variable at time t (for each time point from t=4 onwards). For example, the time-series dependent variable data D3 includes values ​​such as temperature A and pressure B for manufacturing ID=id01 at time t=4.

[0029] The time-series causal model generation unit 12 generates a time-series causal model from the actual lag variable data D2 and the time-series target variable group data D3, and the generated time-series causal model is stored in the time-series causal model storage unit 31. The process of storing this generated time-series causal model in the time-series causal model storage unit 31 is called the time-series causal model storage process. A time series model is generated for each time series variable, and the value of each variable at a given time t is predicted from the values ​​of all variables in the preceding period T. In the example shown, the number of time points prior to the prediction is determined by allowing the user to specify the time window length T on the screen, but it could also be determined automatically for each time series variable based on the prediction accuracy of the time series causal model trained with various time window lengths T.

[0030] Figure 14 is a flowchart showing the procedure of the processing performed by the lag variable generation unit 14. The lag variable generation unit 14 acquires time-series data (step S31). Then, the lag variable generation unit 14 divides the acquired time-series data into data for each manufacturing ID (step S32). Then, the lag variable generation unit 14 starts loop processing (step S33a). When the loop processing starts, the lag variable generation unit 14 sets i to 0 (step S34) and sets (T+1+i) to t (step S35).

[0031] Then, the lag variable generation unit 14 adds the record from row t of the divided data to the time series target variable group data (step S36). Furthermore, the lag variable generation unit 14 adds the values ​​of each time series variable from row t-1 to tT of the divided data to the lag variable data (step S37).

[0032] Subsequently, the lag variable generation unit 14 determines whether the current value of i is the number of records in the divided data (step S38). If, in step S38, the value of i is not the number of records in the divided data (No. in step S38), the lag variable generation unit 14 increments the value of i by 1 (step S39) and returns to the process in step S35.

[0033] Furthermore, if step S38 determines that the value of i is the number of records in the split data (Yes in step S38), the loop processing from step S33a is terminated (step S33b). Then, the lag variable generation unit 14 outputs the lag variable data and the time series target variable group data (step S40).

[0034] [Processing of the contribution calculation unit] Figure 15 shows the process performed by the contribution calculation unit 15. The time variable generation unit 13 obtains target time variable data D4 from the target time series data stored in the target time series data storage unit 23. The contribution calculation unit 15 performs a contribution calculation process to calculate the contribution of each input variable to the output of the predictor 19. The contributions are obtained in two types: time variable contribution D5 and non-time series variable contribution D6. For example, if there are two input variables, X1 and X2, and the contribution of X1 is +150 and the contribution of X2 is -50, then the sum of these will be the output value of predictor 19, which is 100.

[0035] Figure 16 shows an example of the target time variable data. The target time variable data is obtained by converting the target time series data into a time variable, which is the input format for the predictor. For example, as shown in Figure 16, the target time variable data is obtained by converting the temperature A and pressure B of a certain manufacturing ID (id97) into data (time variables) from time 0 (t=0) to time 232 (t=232).

[0036] Figure 17 shows an example of the time variable-specific contribution, which is one of the outputs of the contribution calculation unit 15. The time variable contribution is the result of calculating the contribution of each variable in the target time variable data to the output of the predictor 19. For example, as shown in Figure 17, the contribution of temperature A for a certain manufacturing ID (id97) is calculated as +0.01 at time t=0 and as -0.03 at time t=1, and so on, with the contribution calculated for each time variable such as temperature A and pressure B at each time.

[0037] Figure 18 shows an example of the contribution by non-time-series variable, which is one of the outputs of the contribution calculation unit 15. The contribution of each non-time-series variable is the result of calculating the contribution of each variable in the non-time-series variable data, excluding the target variable, to the predictor's output. For example, the contribution of each variable, such as hardness, additive amount, and reaction time, to the predictor's output is calculated for each of them, excluding the turbidity, which is the target variable for a certain manufacturing ID (id97).

[0038] Figure 19 is a flowchart showing the procedure of the processing performed by the contribution calculation unit 15. First, the contribution calculation unit 15 acquires the output of the predictor 19, the target time variable data, and the target non-time series data (step S51). Then, the contribution calculation unit 15 removes the target variable from the acquired target non-time series data and combines it with the target time variable data (step S52). Next, the contribution calculation unit 15 calculates the contribution of each variable in the combined data to the output value of the predictor 19 (step S53).

[0039] Then, the contribution calculation unit 15 outputs the contribution of each variable in the target time variable data as the contribution by time variable, as shown in Figure 17 (step S54). Furthermore, as shown in Figure 18, the contribution calculation unit 15 outputs the contribution of each variable in the target non-time series data as the contribution by target non-time series variable (step S55).

[0040] [Reducing the contribution of each time variable] Figure 20 shows the process of reducing the contribution of each time variable. In the process shown in Figure 20, the contribution calculated for each time variable (time series variable × time) is reduced to the root cause. Specifically, as shown in Figure 20, the reduction rate calculation unit 16 obtains the target lag variable data D7 generated by the lag variable generation unit 14 and the time series causal model stored in the time series causal model storage unit 31. The reduction rate calculation unit 16 then calculates which time variable, or in other words, which time series variable at which time point, should be reduced by how much, and obtains the contribution reduction rate data D8. The contribution reduction unit 17 then obtains the contribution reduction rate data D8 calculated by the reduction rate calculation unit 16 and the time variable-specific contribution D9. The reduction rate calculation unit 16 then performs a process to actually reduce the contribution based on the reduction rate and obtains the reduced contribution data D10. The reduced contribution data D10 obtained in this manner is output from the result output unit 18.

[0041] Figure 21 shows an example of target lag variable data, and Figure 22 shows an example of contribution reduction rate data. The target lag variable data shown in Figure 21 is obtained by converting the target time series data into lag variables. For example, when the time window length T=3, for manufacturing ID=id97 at time t=4, data for temperature A, pressure B, etc., is obtained for t-3, data for t-2, and data for t-1. Similarly, data is obtained for times t=5, t=6, ... up to the last time t=232.

[0042] The contribution reduction rate data shown in Figure 22 indicates, as a reduction rate, how much of the contribution (source) of a variable at a given time is reduced to a variable (destination) at a different time. For example, the contribution of temperature A (the source of reduction) at time t=232 is shown as contribution of pressure B (the target of reduction) at time t=230, with a reduction rate of +75%, as shown in the contribution-reduction rate data.

[0043] Figure 23 is a flowchart showing the procedure for reducing the contribution of each time variable as shown in Figure 20. First, the lag variable generation unit 14 acquires the target time series data (step S61). Then, the lag variable generation unit 14 performs the lag variable generation process and outputs the target lag variable data (step S62). The target lag variable data is the data shown in Figure 21.

[0044] Next, the reduction rate calculation unit 16 performs a reduction rate calculation process using the target lag variable data generated by the lag variable generation unit 14 and the time series causal model stored in the time series causal model storage unit 31, and outputs contribution reduction rate data (step S63). Details of the reduction rate calculation process in the reduction rate calculation unit 16 are explained in Figure 24.

[0045] Then, the contribution reduction unit 17 performs a contribution reduction process based on the contribution reduction rate data and the time variable-specific contribution calculated by the reduction rate calculation unit 16, and outputs the reduced contribution data (step S64). Details of the contribution reduction process in the reduction rate calculation unit 16 are explained in Figure 25. Then, the reduced contribution data obtained by the contribution reduction unit 17 is output from the result output unit 18 (step S65).

[0046] Figure 24 is a flowchart showing the procedure of the processing performed by the reduction rate calculation unit 16. First, the reduction rate calculation unit 16 acquires the target lag variable data (step S71). The target lag variable data is the data explained in Figure 21. The reduction rate calculation unit 16, which has acquired the target lag variable data, first sets the number of records of the target lag variable data to j (step S72). Then, the reduction rate calculation unit 16 starts a loop processing based on the time series variable (step S73a).

[0047] When the loop processing starts, the reduction rate calculation unit 16 obtains the time series causal model for each time series variable (step S74). The reduction rate calculation unit 16 then calculates the contribution of each variable in the jth row of the target lag variable data to the output value of the time series causal model (step S75). The reduction rate calculation unit 16 adds the calculated contribution to the lag variable-specific contribution data (step S76). The data with this contribution added to the lag variable-specific contribution data is shown in the middle right section of Figure 24, and is the result of analyzing how much each lag variable contributes to the predicted value of each variable at each time point by the predictor.

[0048] Once steps S74 to S76 are completed, the loop processing based on the time series variable is terminated (step S73b), and the reduction rate calculation unit 16 determines whether the value of j is "1" or not (step S77).

[0049] If the value of j in step S77 is not "1" (NO in step S77), the reduction rate calculation unit 16 subtracts one from the value of j (step S78) and returns to the loop processing in step S73a. Furthermore, if the value of j is "1" in step S77 (YES in step S77), the reduction rate calculation unit 16 generates contribution reduction rate data from the lag variable contribution data and outputs it to the contribution reduction unit 17 (step S79). The contribution reduction rate data is as shown in Figure 22.

[0050] Figure 25 is a flowchart showing the procedure of the processing performed by the contribution reduction unit 17. First, the contribution reduction unit 17 acquires time variable-specific contribution data and contribution reduction rate data for the target time series data (step S81). The time variable-specific contribution data for the target time series data is shown in the upper right of Figure 25, and the contribution reduction rate data is shown in the lower right of Figure 25.

[0051] Then, the contribution reduction unit 17 starts loop processing based on the contribution reduction rate data (step S82a).

[0052] When the loop processing starts, the contribution reduction unit 17 reduces each contribution of the time variable-specific contribution data from the source to the destination according to the contribution reduction rate data (step S83). Then, the contribution reduction unit 17 terminates the loop processing based on the contribution reduction rate data (step S82b). When the loop processing is complete, the contribution reduction unit 17 outputs the reduced contribution data to the result output unit 18 (step S84).

[0053] [Example of a factor analysis screen] Figure 26 shows an example of a factor analysis screen displayed on the display unit, which serves as the result output unit 18. The factor analysis screen in Figure 26 provides fields for selecting time-series performance data, non-time-series performance data, time-series data to be analyzed, and non-time-series data to be analyzed. In the example in Figure 26, the user specifies the storage location (folder) of each type of data to indicate the data to be analyzed.

[0054] The factor analysis screen also displays a data loading button, a section for selecting the dependent variable, a section for selecting the time window length, and a button to start the analysis. The data loading button is a button that accepts the user's operation to start loading the data that is the target of the specified factor analysis. The section for selecting the dependent variable is where the user can select the dependent variable from which to obtain the contribution. Figure 26 shows an example where "turbidity" is selected as the dependent variable. The time window length is the section where the user can select the time window length T, as explained in Figure 12 and other diagrams. The "Start Analysis" button is a button that accepts the command to start the analysis.

[0055] Then, when the operation to start the analysis is performed, the factor analysis results will be displayed as shown in Figure 26. In other words, as shown in the lower part of Figure 26, the factor analysis screen displays a graph of the time-series pattern of the data under analysis, a graph of the changes in the contribution to the dependent variable "turbidity," and a graph of the changes in the contribution focused on the root cause. The time-series pattern graph of the data being analyzed visualizes the target time variable data, showing the time changes of the target time variable data such as A temperature, P pressure, etc. The graph showing the change in contribution to the dependent variable "turbidity" visualizes the contribution data for each time-series variable that has not been reduced, and shows the change in the contribution of target time-variable data such as A temperature, P pressure, ... over time. The graph showing the contribution trends focused on the root causes visualizes the reduced time-series contribution data for each variable, and shows the time changes in the contributions of the root causes of "turbidity," such as A temperature, P pressure, etc.

[0056] As shown in Figure 26, the display of factor analysis results allows for the visualization of the reduced contribution, focusing on the root cause of the change in "turbidity." Therefore, it is possible to trace back to the origin, showing the user which factor was controlled at what point in time and in what way, and reducing the influence of intermediate factors to that root cause. This allows us to present the contribution of root causes during factor analysis, unlike simply hiding the influence of intermediate factors that cannot be directly controlled. Furthermore, it becomes possible to efficiently consider feasible countermeasures. In addition, even if the identified root cause is something that cannot be controlled by humans, it can be used to consider countermeasures such as design changes. Furthermore, to display the changes in the contribution to turbidity, as shown in Figure 26, it is possible to simultaneously display a graph showing the changes in the contribution of the root cause, as well as a graph showing the changes in the contribution data for each time-series variable that has not been reduced. This makes it easy to see from the display whether the relevant root cause is indeed the cause. In this case, once it is determined that it is the root cause, the subsequent investigation can be carried out appropriately by switching to a contribution trend focused on the root cause.

[0057] <Second Embodiment Example> Next, a time-series data factor analysis support system and time-series data factor analysis support method according to a second embodiment of the present invention will be described with reference to Figures 27 to 31. The second embodiment applies the factor analysis support system described in the first embodiment to consider countermeasures based on the results obtained from the result output unit 18 of the factor analysis support system and simulate their effects. The factor analysis processing itself performed by the factor analysis support system is the same as the configuration and processing described in Figures 1 to 26.

[0058] [Example of a countermeasure consideration screen] Figure 27 shows an example where the result output unit 18, which functions as a display unit, displays the countermeasures consideration screen. The countermeasures consideration screen shown in Figure 27 displays a graph detailing the changes in "T temperature," a graph showing the results of the change effect verification, and a graph showing the changes in the contribution to "turbidity." The graph detailing the change in "T temperature" shows the characteristic Tx, which indicates the change in T temperature over time, and the change in its contribution α. ​​In this example in Figure 27, the contribution α becomes higher after about 100 hours. Furthermore, the graph showing the detailed trend of "T temperature" displays the baseline value (dashed line) for T temperature. Furthermore, near the graph showing the detailed changes in "T temperature," there are buttons for changing the data, resetting changes, and executing the effect verification.

[0059] Here, let's assume that the user instructs a data change via the data change button, and that in the interval with a high contribution, the change Ty (thick line) after the change in temperature T is indicated. This allows the effects of the changes to be displayed in the graph of the change effect verification results. In other words, the graph of the change effect verification results displays the characteristic Tx, which shows the time change of temperature T before the change, and the change in its contribution α, as well as the characteristic Ty, which shows the time change of temperature T after the change, and the change in its contribution β. Furthermore, the graph displaying the results of the change effect verification shows the total contribution value and a selection field for the display variable. Figure 27 shows an example where "T temperature" is selected in the display variable selection field.

[0060] The factor analysis support system of this embodiment generates time-series data that reflects the content of Ty (in this case, T temperature) where the data has changed in a graph of the changes in the trend, calculates the contribution, and visualizes and displays the results of the prediction of turbidity by the predictor.

[0061] Furthermore, a graph showing the changes in the contribution of various factors to "turbidity" is displayed, including A) temperature, B) pressure, etc. A button to switch to a specific contributing factor is also displayed in the graph. When a user clicks the button to switch to a contribution scale focused on the root cause, the graph showing the progression of contributions switches to a graph showing the result of reducing the contribution scale to the root cause of the turbidity.

[0062] [Processing of pre-implementation effectiveness verification that reflects the content of the countermeasures being considered] Figure 28 shows the configuration of the time-series data factor analysis support system for performing pre-effect verification processing that reflects the countermeasures considered in Figure 27. First, the change content extraction unit 41 extracts the operations associated with the display in the result output unit 18 (operations associated with the display in Figure 27) and obtains the change content data D11. This change data D11 is supplied to the change data generation unit 42.

[0063] The change data generation unit 42 obtains the time series causal model from the time series causal model storage unit 31, and also obtains the target time series data from the target time series data storage unit 23. The change data generation unit 42 then generates modified time series data D12 based on the change content data D11 and supplies the generated modified time series data D12 to the time variable generation unit 13. The time variable generation unit 13 generates the modified time variable data D13 and supplies it to the contribution calculation unit 15. The contribution calculation unit 15 calculates the contribution of the modified time variable data D13 and obtains the modified time variable-specific contribution data D14. This modified time variable-specific contribution data D14 is output by the result output unit 18.

[0064] Figure 29 shows an example of change data D11. This change data D11 is a detailed record of the changes made by the user to a specific variable (in this case, temperature T) within a specified range (time 102-232) on the screen shown in Figure 27. Figure 30 is a flowchart showing the processing procedure in the change extraction unit 41. The change content extraction unit 41 extracts the content of the data changed by the user as change content data in the transition details section of the countermeasure consideration screen (screen shown in Figure 27) (step S91). Next, the change content extraction unit 41 supplies the change content data extracted in step S91 to the change data generation unit 42 (step S92).

[0065] Figure 31 is a flowchart showing an example of processing by the change data generation unit 42. This change data generation unit 42 uses the change data to estimate the values ​​of other time series variables that change as a result using a time series causal model, and then generates the target time series data after the change (modified time series data). In other words, the change data generation unit 42 first acquires the change details data and the target time series data (step S101). Then, the change data generation unit 42 updates the changed parts of the target time series data based on the change details data (step S102).

[0066] Next, the change data generation unit 42 sets t to the change start time + 1 (step S103). Then, the change data generation unit 42 generates a lag variable for time t from the target time series data (step S104). Then, the change data generation unit 42 starts a loop processing based on the time series variables excluding the variable to be changed (step S105a). When the loop processing starts, the modified data generation unit 42 retrieves the causal model for each time series variable from the time series causal model storage unit 31 (step S106).

[0067] Then, the modified data generation unit 42 takes the lag variable for time t as input and estimates the values ​​of each time series variable at time t in each time series causal model (step S107). Once the estimated values ​​are obtained, the modified data generation unit 42 updates the values ​​of each time series variable at time t in the target time series data with those estimated values ​​(step S108), and terminates the loop processing (step S105b).

[0068] Subsequently, the change data generation unit 42 determines whether the current time t is the final time of the target time series data (step S109). If, in step S109, the current time t is not the final time of the target time series data (NO in step S109), the change data generation unit 42 increments the time t value by one and returns to the process in step S104.

[0069] Furthermore, in step S109, if the current time t is the final time of the target time series data (YES in step S109), the modified data generation unit 42 outputs the updated target time series data as modified time series data to the time variable generation unit 13 (step S111).

[0070] This process allows for a preliminary effectiveness verification that reflects the proposed countermeasures, as shown in Figure 27. Furthermore, the ability to perform a preliminary effectiveness verification that reflects the proposed countermeasures allows users to easily verify the effectiveness of the proposed countermeasures in advance, making it possible to find appropriate solutions to the root causes. In other words, when considering countermeasures, it becomes possible to eliminate the influence of noise factors and focus only on factors that can be directly controlled by humans to present the quality deterioration process. This significantly reduces the factors that need to be interpreted when considering countermeasures, and allows for the consideration of countermeasures using factors that can actually be controlled.

[0071] <Variation> It should be noted that the embodiments described above are detailed explanations provided to facilitate understanding of the present invention, and are not necessarily limited to those comprising all the described configurations. Furthermore, the configurations and processes described in the above embodiments can be modified or altered in various ways. For example, in the embodiments described above, examples were shown where temperature and other factors were controlled as the root cause of improving turbidity in the manufacturing line. However, the present invention may also be applied to manufacturing lines other than such.

[0072] Furthermore, while Figure 1 shows an example of a time-series data factor analysis support system composed of three computers 10, 20, and 30, this configuration of three computers is just one example and is not limited to the configuration shown in Figure 1. Furthermore, each computer may, for example, be a general-purpose computer on which a program that performs the processing described in each embodiment is implemented to constitute a time-series data factor analysis support system. In this case, the program to be implemented on the computer may be stored on an external recording medium such as memory, an IC card, an SD card, or an optical disc, and then transferred to the computer.

[0073] Furthermore, in configuration diagrams such as Figure 1, only control lines and information lines deemed necessary for explanation are shown, and not all control lines and information lines are necessarily shown in the actual product. In reality, it can be assumed that almost all components are interconnected. [Explanation of symbols]

[0074] 1…Processor, 2…Main memory, 3…Secondary memory, 4…Input device, 5…Output device, 6…Network interface, 10,20,30…Computer, 11…Predictor generation unit, 12…Time series causal model generation unit, 13…Time variable generation unit, 14…Lag variable generation unit, 15…Contribution calculation unit, 16…Return rate calculation unit, 17…Contribution return unit, 18…Result output unit, 19…Predictor, 21…Actual time series data storage unit, 22…Actual non-time series data storage unit, 23…Target time series data storage unit, 24…Target non-time series data storage unit, 31…Time series Causal model storage unit, 40... Terminal, 41... Change content extraction unit, 42... Change data generation unit, D1... Actual time variable data, D2... Actual lag variable data, D3... Time series target variable group data, D4... Target time variable data, D5... Contribution by time variable, D6... Contribution by non-time series variable, D7... Target lag variable data, D8... Contribution reduction rate data, D9... Contribution by time variable, D10... Reduced contribution data, D11... Change content data, D12... Post-change time series data, D13... Post-change time variable data, D14... Post-change time variable contribution data

Claims

1. A time series data factor analysis support system that analyzes the changes in the contribution of time series variables to the dependent variable, A time-series causal model storage unit that stores a time-series causal model for each of the aforementioned time-series variables, A contribution calculation unit calculates a contribution reduction rate by using the time series causal model stored in the time series causal model storage unit and the time series data to be analyzed, which evaluates how much the contribution of each time series variable at each time point should be reduced to the contribution of other time series variables at other time points. Based on the contribution reduction rate calculated by the contribution calculation unit, the contribution of each time series variable at each time point is reduced to the contribution of other time series variables at other time points, and the contribution reduction unit calculates the change in the contribution of the root cause to the target variable. The system includes an output unit that outputs the contribution of the root cause to the target variable calculated by the contribution reduction unit. A system to support factor analysis of time series data.

2. The time-series causal model stored in the time-series causal model storage unit is a model that generates actual lag variable data and time-series target variable group data from actual time-series data, and predicts the model from the values ​​of the generated actual lag variable data and time-series target variable group data for a period from a specific time to a predetermined time. A time-series data factor analysis support system according to claim 1.

3. Furthermore, it includes a predictor that learns the relationship between past performance time-series data and / or performance non-series data and the state to be analyzed. The contribution calculation unit uses the results learned by the predictor to calculate the contribution reduction rate. A time-series data factor analysis support system according to claim 2.

4. When the root cause is identified based on the contributions reduced by the aforementioned contribution reduction unit, the modified contribution is calculated for the content change data when that root cause is changed. A time-series data factor analysis support system according to claim 3.

5. The output unit displays a screen for considering countermeasures. The system now obtains data on content changes when a user modifies the root cause on the aforementioned countermeasure consideration screen. A time-series data factor analysis support system according to claim 4.

6. The output unit displays a factor analysis screen. The aforementioned factor analysis screen displays a graph of the time-series pattern of the data under analysis, which visualizes the time-series variables, and a graph of the trend of the contribution of the root cause, which visualizes the reduced contribution. A time-series data factor analysis support system according to claim 1.

7. Furthermore, the aforementioned factor analysis screen displays a graph of the contribution trend, which visualizes the contribution data for each time-series variable that has not yet been reduced. A time-series data factor analysis support system according to claim 6.

8. A method for supporting factor analysis of time series data, in which a computer analyzes the changes in the contribution of time series variables to the dependent variable, The computer performs a time-series causal model storage process that stores a time-series causal model for each of the time-series variables, The computer performs a contribution calculation process that uses the time series causal model stored by the time series causal model storage process and the time series data to be analyzed to calculate a contribution reduction rate, which evaluates how much the contribution of each time series variable at each time point should be reduced to the contribution of other time series variables at other time points. The computer performs a contribution reduction process that, based on the contribution reduction rate calculated in the contribution calculation process, reduces the contribution of each time series variable at each time point to the contribution of other time series variables at other time points, and calculates the change in the contribution of the root cause to the target variable. The computer includes an output process that outputs the contribution of the root cause to the target variable calculated by the contribution reduction process. Methods to support factor analysis of time series data.

Images