Causal relationship estimation system and production support system

Abstract

To provide a causal relationship estimation system and a production support system with which the necessity of selecting a causal inference algorithm can be avoided and the accuracy of causal inference is improved.SOLUTION: A causal relationship estimation system 2 includes: a storage apparatus 2a which stores data DA to be analyzed including two variables, and a plurality of different causal inference algorithms CA; and a computing apparatus 2b which estimates causal directions of the two variables. The computing apparatus 2b is configured to: calculate causal direction values CV regarding the two variables by representing the causal direction of the two variables using positive and negative signs, for each of the different causal inference algorithms CA, based on a generated value GV obtained by applying the different causal inference algorithms CA for the two variables, and by representing the probability of the causal directions using the magnitude of numerical values; and estimates a causal direction of the two variables based on the sum ST of the causal direction values for the different causal inference algorithms CA.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a causal relationship estimation system and a production support system. [Background technology]

[0002] Conventionally, Patent Document 1 discloses a technique for estimating causal relationships within data using structural equation modeling means and generating a nonlinear regression model based on the estimation results. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2006-099482 Summary of the Invention [Problem to be solved by the invention]

[0004] Several causal inference algorithms have been proposed for structural equation modeling. Each causal inference algorithm assumes various conditions for the data. This makes it necessary to select an appropriate causal inference algorithm depending on the characteristics of the data, making the process of selecting a causal inference algorithm cumbersome. Furthermore, if an appropriate causal inference algorithm is not selected, there is a risk that the accuracy of the causal inference will decrease.

[0005] The present invention has been made in consideration of such problems, and aims to provide a causal relationship estimation system and a production support system that can omit the task of selecting a causal inference algorithm and that improves the accuracy of causal inference. [Means for solving the problem]

[0006] One aspect of the present invention is a storage device for storing data to be analyzed, the data including two variables, and a plurality of different causal inference algorithms; a calculation device that estimates a causal direction of the two variables using the plurality of different causal inference algorithms, The computing device calculating a causal direction value for the two variables based on values ​​obtained by applying the plurality of different causal inference algorithms to the two variables, by expressing the causal direction of the two variables as a positive or negative sign and expressing the likelihood of the causal direction as a numerical magnitude for each of the plurality of different causal inference algorithms; The causal relationship inference system estimates the causal direction of the two variables based on the sum of the causal direction values ​​for each of the plurality of different causal inference algorithms.

[0007] Another aspect of the present invention is The above causal relationship estimation system, a server or virtual server that stores a knowledge database that indicates the relationship between information or processing specifications related to a workpiece and a processing result that is the result of performing a predetermined processing on the workpiece; and a communication device that acquires data to be analyzed from the machine tool or the production system by communication. [Effects of the Invention]

[0008] According to one and other aspects of the present invention, it is possible to provide a causal relationship estimation system and a production support system that can omit the task of selecting a causal inference algorithm and that have improved accuracy in causal inference. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a block diagram showing a causal relationship estimation system and a production support system according to a first embodiment. [Figure 2] FIG. 1 is a diagram showing a network diagram representing a knowledge model according to the first embodiment. [Figure 3] 1 is a block diagram showing a storage device according to a first embodiment. [Figure 4] FIG. 2 is a diagram showing a main flow of the causal relationship estimation system according to the first embodiment. [Figure 5] 10 is a flowchart of a generated value calculation process according to the first embodiment. [Figure 6] 10 is a flowchart of a causal direction value calculation process according to the first embodiment. [Figure 7] 10 is a flowchart of a causal direction estimation process according to the first embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] (Embodiment 1) 1. Configuration of the causal relationship estimation system and production support system A production support system 1 according to this embodiment will be described with reference to Fig. 1. The production support system 1 includes a causal relationship estimation system 2, a server 3, a machine tool 4, and a production system 5. The causal relationship estimation system 2, the server 3, the machine tool 4, and the production system 5 are connected to a network 6 and configured to be able to communicate with each other. However, one of the machine tool 4 and the production system 5 may be omitted. Also, there may be two or more machine tools 4, and there may also be two or more production systems 5.

[0011] The causal relationship estimation system 2 includes a storage device 2a, a calculation device 2b, an input device 2c, an output device 2d, a communication device 2e, and an update device 2f.

[0012] The server 3 stores a knowledge database DB, but the server 3 may be a virtual server on the Internet or an intranet.

[0013] The machine tool 4 performs predetermined processing on a workpiece (not shown). The machine tool 4 is not particularly limited, and may be, for example, a lathe, a machining center, a milling machine, a gear processing machine, or a boring machine.

[0014] The production system 5 comprises a management device 5a, a data processing device 5b, and a machine tool 5c. However, the configuration of the production system 5 is not limited to the above and can be selected as appropriate. The management device 5a stores, for example, a production plan. The management device 5a may also have a function for generating a production plan. The production plan includes management data such as the production deadline (delivery date) for each workpiece, process information for the workpiece, and an execution schedule for each process. The data processing device 5b determines command information for the operation of the machine tool 4 based on the production plan in the management device 5a, and outputs the command information to the machine tool 5c. The machine tool 5c in the production system 5 is similar to the machine tool 4 described above, so a redundant description will be omitted.

[0015] The server 3 stores a knowledge database DB. Generally, in the field of machining, an operator determines machining conditions such as cutting speed and cutting depth per unit time, taking into consideration various information such as the material of the workpiece, the material of the tool, the quality of the workpiece, and the machining cycle time. In this case, the knowledge database DB is a model of the operator's thought process when the operator acquires various input information and determines the machining conditions.

[0016] In other words, in addition to the material of the workpiece, the material of the tool, the quality of the workpiece, the processing cycle time, the cutting speed, the depth of cut, etc., the knowledge database DB defines each of the industrial technology elements that appear in the worker's thought process as factors, and defines the relationships between the factors.

[0017] The knowledge database DB is used, for example, as follows: When an operator inputs necessary information about the input factors, such as the material of the workpiece, the material of the tool, and the machining requirements (machining accuracy, machining cycle time, etc.), information about the cutting speed, cutting depth, etc., as output factors is output.

[0018] A knowledge database DB is conceptually expressed in the form of a network. An example of a knowledge network diagram 10 that expresses a knowledge database DB as a network diagram will be described with reference to FIG. 2. In this example, a knowledge network diagram 10 relating to a knowledge database DB in the machining field will be used as an example. However, the knowledge network diagram 10 is not limited to FIG. 2.

[0019] As shown in FIG. 2, the knowledge network diagram 10 comprises a plurality of node graphics 11 and link graphics 12 connecting the node graphics 11 together. The node graphics 11 are represented by any graphic such as a box, a graphic including text, an icon, etc. The node graphics 11 represent factors in the knowledge database DB. The link graphics 12 are represented by straight lines, curves, elbow lines, etc. In this example, the link graphics 12 are represented by arrows to define the directionality of the relationships. The link graphics 12 represent the relationships connecting factors in the knowledge database DB. In the knowledge network diagram 10 shown in FIG. 2, all of the node graphics 11 are represented by boxes in which text can be written, and the link graphics 12 are represented by arrows.

[0020] Here, factors are defined by technical terms. Multiple factors may have a technical inclusive relationship (also called a hierarchical relationship, parent-child relationship, or master-slave relationship), or they may have a technical heterogeneous relationship. In other words, the relationships between factors are classified into the above two types.

[0021] For example, workpiece specifications have an inclusive relationship with workpiece thermal properties, workpiece hardness, workpiece elongation, etc. In other words, as factors having a technical inclusive relationship, workpiece specifications are considered to be superordinate conceptual factors, and workpiece thermal properties, workpiece hardness, workpiece elongation, etc. are considered to be subordinate conceptual factors. For example, factors having a technically heterogeneous relationship include workpiece thermal properties and required tool heat resistance, etc. In the following, the relationship between two factors having a technical inclusive relationship will be simply referred to as an inclusive relationship, and the relationship between two factors having a technically heterogeneous relationship will be simply referred to as a heterogeneous relationship.

[0022] The link graphics 12 are displayed in such a way that a first link graphic 12a representing an inclusive relationship and a second link graphic 12b representing a heterogeneous relationship are distinguished from each other. That is, the first link graphic 12a and the second link graphic 12b are displayed in different ways.

[0023] 2, the first link graphic 12a representing the inclusion relationship is represented by a frame line indicating the area of ​​the superordinate concept factor, and the subordinate concept factors are arranged within the frame line representing the first link graphic 12a. Note that the first link graphic 12a is not limited to a frame line.

[0024] 2, the second link graphic 12b representing the heterogeneous relationship is represented by a straight line, a broken line, or the like, connecting the node graphics 11 placed at any positions (up, down, left, right) apart. The second link graphic 12b is represented by an arrow line indicating the direction of the definition of the factors.

[0025] The following are examples of factors that are effective when the causal relationship estimation system 2 according to this embodiment is applied to the production system 5. The factors include the power value and torque of the spindle motor. Other factors include thermal displacement of the machine tool 4 and shape error due to thermal displacement. However, the factors that are effective in the causal relationship estimation system 2 according to this embodiment are not limited to those mentioned above.

[0026] 2. Configuration of Causal Relationship Inference System 2 The configuration of the causal relationship estimation system 2 will be described with reference to FIGS.

[0027] (1) Storage device 2a As shown in FIG. 3, the storage device 2a stores analysis target data DA, a causal inference algorithm CA, generated values ​​GV, causal direction values ​​CV, and a sum total of the causal direction values ​​ST.

[0028] The data DA to be analyzed includes machine setting conditions or machining conditions of the machine tool 4 or production system 5 that executes a predetermined process on the workpiece, and data related to the processing results of executing the predetermined process on the workpiece. The data DA to be analyzed includes two variables. In this embodiment, one of the two variables may be designated X and the other Y. In this embodiment, a causal relationship is inferred between one of the two variables (X) as the cause and the other (Y) as the result, and also between the other variable (Y) as the cause and the one variable (X) as the result.

[0029] The causal inference algorithm CA is an algorithm for estimating causal relationships in structural equation modeling. The causal inference algorithm CA is not particularly limited and may be appropriately selected from, for example, Addettive Noise Model (ANM), Regression Error Based Causal Inference (RECI), Linear Non-Gaussian Acyclic Model (LiNGAM), Information-Geometric Causal Inference (IGCI), Post Nonlinear Causal model (PNL), etc. In this embodiment, LiNGAM, ANM, IGCI, and RECI are selected as the causal inference algorithm CA.

[0030] A predetermined assumption is set for data to which each causal inference algorithm CA can be applied. In other words, when a causal inference algorithm CA in which an assumption corresponding to data having predetermined properties is set is applied, highly accurate causal inference can be performed, but when a causal inference algorithm CA in which an assumption not corresponding to the properties of the data is set is applied, the accuracy of the causal inference may decrease.

[0031] LiNGAM is based on assumptions such as that the data is linear and that explanatory variables and residuals are independent.

[0032] ANM is based on assumptions such as that the data is three times differentiable and that explanatory variables and residuals are independent.

[0033] The IGCI is based on assumptions such as nonlinearity and the absence of noise in the data.

[0034] RECI is based on assumptions such as that the data is twice differentiable.

[0035] The generated value GV is obtained by applying multiple different causal inference algorithms CA to two variables included in the analysis target data DA. The generated value GV is not particularly limited, and any value can be appropriately selected, such as a score in the Hilbert-Schmidt independence criterion, mutual information, mean squared error, p-value, chi-square value, etc. The generated value GV may be calculated as a different generated value for each different causal inference algorithm CA, or the same generated value may be calculated for each different causal inference algorithm CA.

[0036] For each causal inference algorithm CA, the generated value GV is calculated for each of the causal directions of the two variables included in the analysis target data DA. As a result, two generated values ​​GV are calculated for each causal inference algorithm CA.

[0037] By comparing the magnitude of the two generated values ​​GV calculated for different causal directions, one of the two variables (X and Y) included in the data DA to be analyzed is selected: the causal direction in which X is the cause and Y is the result (X → Y), or the causal direction in which Y is the cause and X is the result (Y → X).

[0038] In the following explanation, the generated value GV of two variables (X and Y) related to the causal direction (X → Y) where X is the cause and Y is the result is called the generated value X→Y GV X→Y and the generated value GV relating to the causal direction (Y → X) where Y is the cause and X is the result is the generated value Y→X GV Y→X It may be written as follows.

[0039] Generated Value X→Y GV X→Y and the generated valueY→X GV Y→X The relationship between the magnitude relationship and the causal direction may be different for each causal inference algorithm CA, or may be the same for all causal inference algorithms CA.

[0040] For example, among a plurality of different causal inference algorithms CA, one causal inference algorithm CA generates X→Y GV X→Y is the generated value Y→X GV Y→X If it is greater than , then it is assumed that the causal direction is (X → Y) where X is the cause and Y is the result. In other causal inference algorithms, CA, the generated value X→Y GV X→Y is the generated value Y→X GV Y→X If it is greater than , the causal direction may be determined as Y → X, where Y is the cause and X is the effect.

[0041] Also, for example, for all causal inference algorithms CA, the generated values X→Y GV X→Y is the generated value Y→X GV Y→X If the value is larger than , the causal direction (X → Y) may be set such that X is the cause and Y is the result. X→Y GV X→Y is the generated value Y→X GV Y→X If it is greater than , the causal direction may be determined to be Y → X, where Y is the cause and X is the result.

[0042] The causal direction value CV represents the causal direction of two variables contained in the analysis target data DA as a positive or negative sign, and the likelihood of the causal direction as a numerical value. One causal direction value CV is calculated for one causal inference algorithm CA.

[0043] The sign of the causal direction value CV is arbitrary for the causal direction and can be selected appropriately. In this embodiment, when, of two variables (X and Y) included in the analysis target data DA, X is the cause and Y is the result (X → Y), it is considered positive, and when X is the result because of Y (Y → X), it is considered negative. However, when, of two variables (X and Y) included in the analysis target data DA, X is the cause and Y is the result (X → Y), it may also be considered negative, and when X is the result because of Y (Y → X), it may also be considered positive.

[0044] As described above, the magnitude (absolute value) of the numerical value of the causal direction value CV represents the likelihood of the causal direction between two variables included in the analysis target data DA. Therefore, when one causal direction is inferred for one causal inference algorithm CA among a plurality of different causal inference algorithms CA and the inferred causal direction is likely to be accurate, the numerical value (absolute value) of the causal direction value CV calculated for this causal inference algorithm CA is relatively large. On the other hand, when one causal direction is inferred for one causal inference algorithm CA among a plurality of different causal inference algorithms CA and the inferred causal direction is unlikely to be accurate, the numerical value (absolute value) of the causal direction value CV calculated for this causal inference algorithm CA is relatively small.

[0045] The magnitude (absolute value) of the numerical value of the causal direction value CV is calculated based on the difference between the generated values ​​GV calculated for both directions of the causal directions of the two variables included in the analysis target data DA. In other words, the magnitude (absolute value) of the numerical value of the causal direction value CV is calculated based on the difference between the generated values ​​GV calculated for both directions of the causal directions of the two variables included in the analysis target data DA. X→Y GV X→Y and the generated value Y→X GV Y→X It is calculated based on the difference between

[0046] In this embodiment, the causal direction value CV has, in the numerator, the difference between the generated values ​​GV calculated for both directions of the causal directions of the two variables included in the analysis target data DA. In other words, the causal direction value CV has, in the numerator, the generated value X→Y GV X→Y and the generated value Y→X GV Y→X It has the difference between.

[0047] As described above, the generated value GV can be any value, such as a score in the Hilbert-Schmidt independence criterion, mutual information, mean square error, p-value, chi-square value, etc., for each different causal inference algorithm CA. Therefore, the magnitude of the generated value GV may differ for each causal inference algorithm CA. Therefore, in this embodiment, the generated value GV has, in the denominator, the sum of the generated values ​​GV calculated for both causal directions of two variables. In other words, the causal direction value CV has, in the denominator, the generated value X→Y GV X→Y and the generated value Y→X GV Y→X It has the sum of and.

[0048] However, the product value GV may have, in its denominator, the product value GV of the causal direction inferred for the two variables, or the product value GV may have, in its denominator, the geometric mean of the product values ​​GV calculated for both directions of the causal direction of the two variables, or the harmonic mean of the product values ​​GV calculated for both directions of the causal direction of the two variables.

[0049] The sum of causal direction values ​​ST is the sum of the causal direction values ​​CV calculated for each causal inference algorithm CA. In this embodiment, if the sum of causal direction values ​​ST is positive, the causal direction is estimated to be that X is the cause and Y is the result, if the sum of causal direction values ​​ST is negative, the causal direction is estimated to be that Y is the cause and X is the result, and if the causal direction value is 0, it is determined that inference of the causal direction is impossible.

[0050] However, with regard to the sign of the causal direction value CV, if, of the two variables (X and Y) included in the data DA to be analyzed, X is the cause and Y is the result (X → Y), it is considered negative, and if X is the result and X is caused by Y (Y → X), it is considered positive. If the sum of the causal direction values ​​ST is negative, the causal direction is presumed to be X is the cause and Y is the result; if the sum of the causal direction values ​​ST is positive, the causal direction is presumed to be Y is the cause and X is the result; and if the causal direction value is 0, it is deemed impossible to infer the causal direction.

[0051] In this embodiment, the causal direction is estimated using the sum ST of the causal direction values, but this is not limiting, and the causal direction may be estimated using the average value of the causal direction values. The average value may be an arithmetic mean, a geometric mean, or a harmonic mean.

[0052] (2) Arithmetic unit 2b 1 calculates a causal direction value CV for the two variables based on a generated value GV obtained by applying a plurality of different causal inference algorithms CA to the two variables, by expressing the causal direction of the two variables as a positive or negative sign and expressing the likelihood of the causal direction as a numerical magnitude for each of the plurality of different causal inference algorithms CA. The calculation device 2b estimates the causal direction of the two variables based on the sum of the causal direction values ​​CV for each of the plurality of different causal inference algorithms CA.

[0053] The arithmetic device 2b may be, for example, a central processing unit (CPU) or a graphics processing unit (GPU).

[0054] (3) Input device 2c The input device 2c shown in FIG. 1 is not particularly limited, and any input device 2c can be selected, such as a keyboard, a mouse, a touch panel, a joystick, or a voice input device.

[0055] (4) Output device 2d 1 is not particularly limited, and any output device 2d can be selected, such as a liquid crystal display, an audio output device, a tablet terminal, a printer, etc. The output device 2d outputs the causal relationship estimated by the causal relationship estimation system 2 and notifies the worker of it.

[0056] (5) Communication device 2e 1 acquires analysis target data DA through communication from the machine tool 4 or the production system 5. In addition, the communication device 2e transmits update information of the knowledge database DB to the server 3.

[0057] (6) Update device 2f When the causal relationship estimation system 2 updates the causal relationships related to the analysis target data DA, the update device 2f shown in FIG. 1 updates the corresponding portion of the knowledge database DB.

[0058] 3. Operation of Causality Inference System 2 Next, the operation of the causal relationship estimation system 2 will be described with reference to Figs. 4 to 7. Fig. 4 shows a main flow of the causal relationship estimation system 2 according to this embodiment. When the causal relationship estimation system 2 is started, a generated value calculation process (S1) is executed. Next, a causal direction value calculation process (S2) is executed. Next, a causal direction estimation process (S3) is executed. With the above, the operation of the causal relationship estimation system 2 is completed.

[0059] A flowchart of the generated value calculation process (S1) is shown in Figure 5. When the generated value calculation process (S1) is executed, the calculation device 2b selects one causal inference algorithm CA from among a plurality of different causal inference algorithms CA stored in the storage device 2a (S10).

[0060] Next, the calculation device 2b uses the selected one causal inference algorithm CA to calculate the generated value GV for both of the causal directions of the two variables included in the analysis target data DA (S11). X→Y GV X→Y and the generated value Y→X GV Y→X and are calculated.

[0061] Next, the calculation device 2b determines whether the generated values ​​GV have been calculated for all of the multiple different causal inference algorithms CA stored in the storage device 2a (S12). If the generated values ​​GV have been calculated for all of the multiple different causal inference algorithms CA stored in the storage device 2a (S12: Y), the generated value calculation process ends. On the other hand, if the generated values ​​GV have not been calculated for all of the multiple different causal inference algorithms CA stored in the storage device 2a (S12: N), the processes of S10 to S12 are repeated.

[0062] A flowchart of the causal direction value calculation process (S2) is shown in Figure 6. When the causal direction value calculation process (S2) is executed, the calculation device 2b selects one causal inference algorithm CA from among a plurality of different causal inference algorithms CA stored in the storage device 2a (S20).

[0063] Next, the calculation device 2b calculates the generated value GV calculated using the selected causal inference algorithm CA. X→Y GV X→Y and the generated value Y→X GV Y→X The magnitude relationship between and is compared (S21).

[0064] The calculation unit 2b calculates the generated value compared in S21. X→Y GV X→Y and the generated value Y→X GV Y→X Based on the magnitude relationship between and , the causal direction of the two variables included in the analysis target data DA is inferred (S22). As a result, one causal direction is inferred for the selected one causal inference algorithm CA.

[0065] Next, the calculation device 2b calculates the numerical value of the causal direction value CV (S23). In detail, the calculation device 2b calculates the numerical value of the causal direction value CV based on the following formula (1).

[0066]

number

[0067] Next, the calculation device 2b acquires the causal direction inferred in S22 of FIG. 6, and determines whether this causal direction is a direction (X → Y) in which X is the cause and Y is the result of two variables (X and Y) included in the data DA to be analyzed (S24).

[0068] If the causal direction is such that, of the two variables (X and Y) included in the data to be analyzed DA, X is the cause and Y is the result (X → Y) (S24: Y), the calculation device 2b sets the sign of the causal direction value CV to positive (S25).

[0069] On the other hand, if the causal direction is not such that X is the cause and Y is the result (X → Y) of the two variables (X and Y) included in the analysis target data DA (S24: N), the calculation device 2b makes the sign of the causal direction value CV negative (S26).

[0070] Next, the calculation device 2b determines whether the causal direction values ​​CV have been calculated for all of the different causal inference algorithms CA stored in the storage device 2a (S27). If the causal direction values ​​CV have been calculated for all of the different causal inference algorithms CA stored in the storage device 2a (S27: Y), the calculation device 2b adds up all of the calculated causal direction values ​​CV to calculate the sum ST of the causal direction values ​​CV (S26). This completes the causal direction value calculation process (S2).

[0071] On the other hand, if the causal direction values ​​CV have not been calculated for all of the different causal inference algorithms CA stored in the storage device 2a (S27: N), the processes of S20 to S26 are repeated.

[0072] 7 shows a flowchart of the causal direction estimation process (S3). When the causal direction estimation process (S3) is executed, the calculation device 2b determines whether the sum ST of the causal direction values ​​CV is 0 (S30). If the sum ST is 0 (S30: Y), the calculation device 2b determines that it is impossible to infer the causal direction (S31). This completes the causal direction estimation process (S3).

[0073] On the other hand, if the sum ST is not 0 (S30: N), the calculation device 2b determines whether the sum ST is greater than 0 (S32). If the sum ST is greater than 0 (S32: Y), the calculation device 2b estimates that the direction in which X is the cause and Y is the result (X → Y) of the two variables (X and Y) included in the analysis target data DA is the causal direction (S33). This completes the causal direction estimation process (S3).

[0074] On the other hand, if the sum ST is not greater than 0 (S32: N), the calculation device 2b estimates that the direction in which Y is the cause and X is the result (Y → X) of the two variables (X and Y) included in the analysis target data DA is the causal direction (S34). This completes the causal direction estimation process (S3).

[0075] 4. Effects of this form Next, the effects of this embodiment will be described. A causal relationship estimation system 2 according to this embodiment includes a storage device 2a that stores analysis target data DA including two variables and a plurality of different causal inference algorithms CA, and a calculation device 2b that estimates the causal direction of the two variables using the plurality of different causal inference algorithms CA. The calculation device 2b calculates a causal direction value CV for the two variables based on a generated value GV obtained by applying the plurality of different causal inference algorithms CA to the two variables, by expressing the causal direction of the two variables as a positive or negative sign and expressing the likelihood of the causal direction as a numerical magnitude for each of the plurality of different causal inference algorithms CA, and estimates the causal direction of the two variables based on the sum of the causal direction values ​​CV for each of the plurality of different causal inference algorithms CA.

[0076] According to this embodiment, it is possible to omit the task of selecting one causal inference algorithm CA from among a plurality of causal inference algorithms CA. In addition, since it is not necessary to consider the assumptions set for each causal inference algorithm CA, it is possible to improve the accuracy of causal inference.

[0077] For example, as described above, ANM and RECI have different assumptions for the analysis target data DA. Therefore, when ANM and RECI are applied to one analysis target data DA to estimate a causal relationship, one of them may be correct and the other may be incorrect.

[0078] For example, while ANM is able to estimate correct causal relationships, examples of causal relationships estimated by RECI that are incorrect include data relating to altitude and temperature, data relating to the decomposition of fine roots, data relating to the content of high-performance plasticizers and the compressive strength of concrete, data relating to the content of fine aggregate (sand) and the compressive strength of concrete, data relating to the number of days of use and the compressive strength of concrete, and data relating to room size and rent.

[0079] Furthermore, while RECI is able to estimate correct causal relationships, examples of causal relationships estimated by ANM that are incorrect include data relating to age and hourly wages, data relating to latitude and temperature, data relating to ozone levels and temperature, data relating to the population with access to running water and infant mortality rates, data relating to clay content and organic carbon content in soil, and data relating to precipitation and runoff in river basins.

[0080] According to this embodiment, in all of the above examples, the process of selecting a causal inference algorithm CA can be omitted and the accuracy of causal inference can be improved.

[0081] Furthermore, according to this embodiment, the magnitude of the numerical value of the causal direction value CV is calculated based on the difference between the generated values ​​GV calculated for both causal directions of the two variables. Because the magnitude of the numerical value of the causal direction value CV can be calculated by such a simple method, the processing can be made more efficient.

[0082] Furthermore, the causality direction value CV according to this embodiment has, in its denominator, the sum of the generated values ​​GV calculated for both causal directions of the two variables, thereby enabling the magnitude of the causality direction value CV, which varies for each causal inference algorithm CA, to be standardized.

[0083] Furthermore, the causal direction value CV according to this embodiment includes the generated value GV of the causal direction inferred for the two variables in the denominator, which allows the magnitude of the causal direction value CV, which varies depending on the causal inference algorithm CA, to be standardized.

[0084] In addition, in this embodiment, the multiple different causal inference algorithms CA include at least two selected from the Linear Non-Gaussian Acyclic Model (LiNGAM), Addettive Noise Model (ANM), Information-Geometric Causal Inference (IGCI), and Regression Error Based Causal Inference (RECI).

[0085] According to this embodiment, the accuracy of causal inference can be improved by using highly reliable LiNGAM, ANM, IGCI, and RECI as the causal inference algorithm CA.

[0086] Furthermore, according to this embodiment, the analysis target data DA includes machine setting conditions or machining conditions of the machine tool 4 or production system 5 that executes a predetermined process on the workpiece, and data related to the processing results of executing the predetermined process on the workpiece, and the calculation device 2b infers the causal direction between the machine setting conditions or machining conditions and the processing results on the workpiece by the machine tool 4 or production system 5. This makes it possible to apply the causal relationship estimation system 2 to the machine tool 4 or production system 5.

[0087] Moreover, the production support system 1 according to this embodiment includes the above-described causal relationship estimation system 2, a server 3 or a virtual server (not shown) that stores a knowledge database DB that indicates the relationship between information about the workpiece or processing specifications and a processing result that is the result of performing a predetermined process on the workpiece, and a communication device 2e that acquires analysis target data DA by communication from the machine tool 4 or the production system 5. According to this embodiment, the causal relationship estimation system 2 can be applied to the production support system 1 that includes the knowledge database DB.

[0088] The production system 5 according to this embodiment further includes an update device 2f that updates the knowledge database DB based on the estimated causal direction, thereby enabling the knowledge database DB to be updated based on the causal relationship of the analysis target data DA estimated using the causal estimation system.

[0089] The present invention is not limited to the above-described embodiments, and can be applied to various embodiments within the scope of the present invention. [Explanation of symbols]

[0090] 1: Production support system, 2: Causality estimation system, 2a: Storage device, 2b: Calculation device, 2c: Input device, 2d: Output device, 2e: Communication device, 2f: Update device, 3: Server, 4, 5c: Machine tool, 5: Production system, 6: Network, ANM: Addettive Noise Model, CA: Causal inference algorithm, CV: Causal direction value, DA: Data to be analyzed, DB: Knowledge database, GV: Generated value, IGCI: Information-Geometric Causal Inference, LiNGAM: Linear Non-Gaussian Acyclic Model, RECI: Regression Error Based Causal Inference, ST: Sum of causal direction values

Claims

1. a storage device for storing analysis target data including two variables and a plurality of different causal inference algorithms; a calculation device that estimates a causal direction of the two variables using the plurality of different causal inference algorithms, The computing device calculating a causal direction value for the two variables based on values ​​obtained by applying the plurality of different causal inference algorithms to the two variables, by expressing the causal direction of the two variables as a positive or negative sign and expressing the likelihood of the causal direction as a numerical magnitude for each of the plurality of different causal inference algorithms; a causal relationship inference system that estimates a causal direction of the two variables based on a sum of the causal direction values ​​for each of the plurality of different causal inference algorithms;

2. The causal relationship estimation system according to claim 1 , wherein the magnitude of the numerical value of the causal direction value is calculated based on a difference between the generated values ​​calculated for both causal directions of the two variables.

3. The causal relationship estimation system according to claim 1 , wherein the causal direction value has, in a denominator, a sum of the product values ​​calculated for both causal directions of the two variables.

4. The causal relationship inference system of claim 1 , wherein the causal direction value comprises, in a denominator, the product of the inferred causal direction for the two variables.

5. The plurality of different causal inference algorithms include at least two selected from the Linear Non-Gaussian Acyclic Model (LiNGAM), Additive Noise Model (ANM), Information-Geometric Causal Inference (IGCI), and Regression Error Based Causal Inference (RECI). The causal relationship estimation system according to claim 1.

6. the analysis target data includes machine setting conditions or machining conditions of a machine tool or production system that performs a predetermined process on a workpiece, and data related to a processing result of performing the predetermined process on the workpiece; The causal relationship estimation system according to any one of claims 1 to 3, wherein the calculation device infers a causal direction between the machine setting conditions or the machining conditions and the processing results on the workpiece by the machine tool or the production system.

7. The causal relationship estimation system according to claim 6 ; a server or virtual server that stores a knowledge database that indicates the relationship between information or processing specifications related to a workpiece and a processing result that is the result of performing a predetermined processing on the workpiece; a communication device that acquires data to be analyzed from the machine tool or the production system via communication.

8. 8. The production support system according to claim 7, further comprising an update device that updates the knowledge database based on the estimated causal direction.

Images