Primary factor analysis device and primary factor analysis method

Abstract

A cause analysis device (100) is provided with: a data acquisition unit (110) that acquires a plurality of types of monitoring data relating to a monitoring target; a variable setting unit (120) that generates, on the basis of the monitoring data, monitoring data for learning, monitoring data for cause analysis, a data set for learning comprising an explanatory variable and a target variable, and a data set for cause analysis; a model learning unit (130) that, on the basis of the learning data set, generates a learning model that has been learned so as to input the explanatory variable and output the target variable; a contribution degree calculation unit (150) that calculates a contribution degree, which is the degree of influence of the explanatory variable on the target variable, on the basis of causal relationship data indicating a causal relationship in the plurality of types of monitoring data, the learning model, the monitoring data for learning, and the monitoring data for essential factor analysis; a cause analysis unit (160) that calculates a cause degree, which is the degree of the cause, for each explanatory variable on the basis of the contribution degree; and a cause analysis result output unit (170) that outputs cause analysis results on the basis of the cause analysis data set, the contribution degree, and the cause degree.

Description

Technical Field

[0001] The technology disclosed herein relates to cause analysis techniques. Background Technology

[0002] In addressing the necessary aspects of a problem, it is crucial to take appropriate measures to address the underlying issues in order to achieve the desired outcome. For example, regarding the need to improve the productivity or efficiency of equipment in a facility or factory, it is essential to implement appropriate countermeasures or actions against the factors that reduce productivity or efficiency. Thus, while identifying the problem is necessary to solve the problem, it is often the result of multiple overlapping conditions, making it difficult to pinpoint the exact issue. Therefore, a cause-and-effect analysis technique is known to exist, which uses learning models to quantitatively analyze the relationship between the problem (target variable) and several candidate issues (explanatory variables) considered related to it.

[0003] Patent document 1 describes a "method and apparatus for analyzing the causes of defects" that "extracts the causes of product quality variations" by "using a learning model that has undergone machine learning".

[0004] Specifically, Patent Document 1's "Method and Apparatus for Defect Cause Analysis" involves "composing a computer to perform the following steps: acquiring multiple monitoring data obtained by monitoring multiple manufacturing condition data or the operation of manufacturing equipment; predicting the quality of manufactured products by inputting the acquired multiple manufacturing condition data or monitoring data into a learning model that has been trained to output quality data representing the quality of products manufactured by the manufacturing equipment, when multiple manufacturing condition data or monitoring data have been input; and using the learning model to calculate the contribution of each of the multiple manufacturing condition data or monitoring data to the quality data or anomaly score data output from the learning model" (Patent Document 1: Abstract, Solution).

[0005] According to the "Method and Apparatus for Analysis of Defective Factors" in Patent Document 1, the contribution degree, which represents the degree of direct contribution of each factor (the explanatory variable in the learning model) to the quality of the product (the target variable in the learning model), is calculated.

[0006] Patent Document 1: Japanese Patent Application Publication No. 2022-043848

[0007] Sometimes, multiple factors have a dependency relationship where one factor influences the other factors.

[0008] Existing techniques fail to consider indirect contributions based on dependencies when multiple factors (explanatory variables) are interdependent, leading to incorrect analytical results based on erroneous contribution levels.

[0009] In such existing technologies, there are challenges such as the difficulty in improving the accuracy of factor analysis. Summary of the Invention

[0010] This disclosure addresses the aforementioned issues and aims to improve the accuracy of cause analysis.

[0011] The cause analysis apparatus disclosed herein includes:

[0012] The data acquisition department acquires various types of surveillance data related to the monitored objects.

[0013] The variable setting unit generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0014] The model learning unit generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0015] The contribution calculation unit calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0016] The causal analysis unit, based on the stated contribution level, calculates the degree to which each of the stated explanatory variables is a factor, i.e., the factor degree; and

[0017] The cause analysis results output unit outputs cause analysis results based on the cause analysis dataset, the contribution degree, and the cause degree.

[0018] According to this disclosure, it can improve the accuracy of factor analysis. Attached Figure Description

[0019] Figure 1 This is a diagram illustrating a structural example of the cause analysis apparatus 100 according to Embodiment 1 of this disclosure.

[0020] Figure 2 This is a diagram illustrating a structural example of a cause analysis system 10 including the cause analysis apparatus 100A of Embodiment 1 of this disclosure.

[0021] Figure 3 This is a flowchart illustrating an example of the processing of the cause analysis apparatus 100, 100A of this disclosure.

[0022] Figure 4 This is a diagram representing a first example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0023] Figure 5 This is a diagram illustrating a second example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0024] Figure 6 This is a diagram representing a third example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0025] Figure 7 This is a diagram illustrating the contribution values ​​output by the contribution calculation units 150 and 150A in the factor analysis apparatus 100 and 100A of this disclosure.

[0026] Figure 8 This is a diagram illustrating the factor degree output by the factor analysis units 160 and 160A in the factor analysis apparatus 100 and 100A of this disclosure.

[0027] Figure 9 This is a diagram showing a first example of the display of the cause analysis results output by the cause analysis result output units 170 and 170A in the cause analysis apparatus 100 and 100A of this disclosure.

[0028] Figure 10 This is a diagram showing a second example of the result of the cause analysis output by the cause analysis result output unit 170, 170A in the cause analysis apparatus 100, 100A of this disclosure.

[0029] Figure 11 This is a diagram illustrating a structural example of a factor analysis system 10 including the factor analysis apparatus 100B of Embodiment 2 of this disclosure.

[0030] Figure 12 This is a diagram showing an example of a factor analysis result output by the factor analysis result output unit 170B in the factor analysis apparatus 100B of Embodiment 2 of this disclosure.

[0031] Figure 13 This is a diagram illustrating a structural example of a cause analysis system 10 including the cause analysis apparatus 100C of Embodiment 3 of this disclosure.

[0032] Figure 14 This is a diagram illustrating the polymerization conditions of the factor analysis apparatus 100C used in Embodiment 3 of this disclosure.

[0033] Figure 15 This is a diagram illustrating a structural example of a cause analysis system 10 including the cause analysis apparatus 100D of Embodiment 4 of this disclosure.

[0034] Figure 16 This is a diagram showing a first display example of the cause analysis results output by the cause analysis result output unit 170B in the cause analysis apparatus 100D of Embodiment 4 of this disclosure.

[0035] Figure 17 This is a diagram showing a second example of the display of the cause analysis results output by the cause analysis result output unit 170B in the cause analysis apparatus 100D of Embodiment 4 of this disclosure.

[0036] Figure 18 This is a diagram illustrating a first example of a hardware structure used to implement the functionality based on the structure of this disclosure.

[0037] Figure 19 This is a diagram illustrating a second example of a hardware structure used to implement the functionality based on the structure of this disclosure. Detailed Implementation

[0038] Hereinafter, in order to provide a more detailed description of the present disclosure, embodiments of the present disclosure will be described with reference to the accompanying drawings.

[0039] This disclosure discloses a technique that uses causal relationship data such as causal graphs to estimate the influence of a variable on a target variable with high accuracy compared to existing techniques.

[0040] Implementation method 1.

[0041] Implementation 1 describes the basic structure of the cause analysis apparatus of this disclosure.

[0042] Figure 1 This is a diagram illustrating a structural example of the cause analysis apparatus 100 according to Embodiment 1 of this disclosure.

[0043] The cause analysis device 100 uses a causal graph (causal relationship data) related to the causal relationship or dependency relationship between the target variable, which is the target value, and the explanatory variable, which is the monitoring data, to quantify the magnitude of the impact of the monitoring data on the target.

[0044] The causal analysis device 100 is configured to include: a data acquisition unit 110, a variable setting unit 120, a model learning unit 130, a causal data acquisition unit 140, a contribution calculation unit 150, a causal analysis unit 160, and a causal analysis result output unit 170.

[0045] The data acquisition unit 110 acquires various types of surveillance data related to the monitored object.

[0046] Based on the aforementioned monitoring data, the variable setting unit 120 generates monitoring data for learning, monitoring data for cause analysis, a dataset for learning, and a dataset for cause analysis.

[0047] A learning dataset consists of explanatory variables and target variables.

[0048] The model learning unit 130 generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0049] The causal relationship data acquisition unit 140 acquires causal relationship data representing the causal relationships of various monitoring data.

[0050] The contribution calculation unit 150 calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0051] Based on the aforementioned contribution levels, the causal analysis unit 160 calculates the degree to which each of the aforementioned explanatory variables is a factor, i.e., the factor degree.

[0052] The cause analysis result output unit 170 outputs the cause analysis results based on the aforementioned cause analysis dataset, contribution rate, and cause degree.

[0053] The aforementioned factor analysis device 100 can be used for all production equipment in a factory that consumes certain resources such as electricity, gas, or oil.

[0054] In addition, monitoring data is collected from sensors installed on the production equipment, as well as from monitoring the production equipment.

[0055] For example, the cause analysis device 100 collects monitoring data obtained by monitoring production equipment that is the object of analysis, i.e., production equipment that is the object of improving resource consumption efficiency and productivity.

[0056] The monitoring data consists of production management information such as resource consumption, temperature, humidity, product quality, production quantity, operation time, operators, and production conditions, collected by sensors installed on the equipment.

[0057] From the monitoring data, one or more data points representing the consumption efficiency or productivity of a certain resource are set as target variables, and two or more data points related to the target variable, that is, candidates that are considered to be factors causing changes in the target variable, are set as explanatory variables.

[0058] The cause analysis device 100 learns the learning model by taking explanatory variables as inputs and target variables as outputs.

[0059] The causal analysis device 100 calculates the degree of influence of the explanatory variable on the target variable for each data sample and explanatory variable based on the learning model, monitoring data, and causal graph between the monitoring data, namely, "the contribution of factors that cause changes in resource consumption efficiency and productivity".

[0060] The causal analysis device 100 statistically contributes to the analysis and calculates the causal degree for each explanatory variable.

[0061] The cause analysis device 100 estimates the factors that cause changes in resource consumption efficiency and productivity, and provides information related to the analysis results to users such as operators, resource managers, and on-site maintenance personnel of the target equipment.

[0062] Therefore, the cause analysis device 100 assists operators in improving resource consumption efficiency and productivity, and reduces the workload of operators.

[0063] The manner in which the cause analysis system, including the cause analysis device 100, is described.

[0064] Figure 2 This is a diagram illustrating a structural example of a cause analysis system 10 including the cause analysis apparatus 100A of Embodiment 1 of this disclosure.

[0065] Figure 3 This is a flowchart illustrating an example of the processing of the cause analysis apparatus 100, 100A of this disclosure.

[0066] Figure 4 This is a diagram representing a first example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0067] Figure 5 This is a diagram illustrating a second example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0068] Figure 6 This is a diagram representing a third example of causal relationship data used by the cause analysis apparatus 100, 100A of this disclosure.

[0069] The cause analysis system 10 is configured to include a cause analysis device 100A, a sensor 200, and a display device 300.

[0070] The cause analysis device 100A quantifies the impact of monitoring data on target values ​​by using a cause-effect graph (causal relationship data) related to the causal or dependency relationship between each target value (target variable) and monitoring data (explanatory variable), thereby assisting in the research of measures to improve resource consumption efficiency and productivity, and in the assignment of priorities.

[0071] Sensor 200 is installed on the production equipment that is being monitored. Sensor 200 outputs sensor data to the cause analysis device 100.

[0072] Sensor data, for example, is time-series data of sensor measurements over a specified time period obtained at specified intervals from a sensor 200 installed on the target device.

[0073] Sensor data may represent at least one sensor measurement, such as temperature, humidity, electricity, gas volume, oil volume, water volume, or steam flow rate.

[0074] Furthermore, this is just one example; sensor data may also include control values ​​such as command values ​​or reference values ​​obtained at specified intervals for a specified time period using multiple sensors 200.

[0075] The display device 300 is, for example, a display device.

[0076] Display device 300 displays the cause analysis results output by cause analysis device 100.

[0077] The display device 300 can also be configured as an input / output device with an input device. In this case, the display device 300 is composed of an information processing terminal or the like.

[0078] The structural examples of the cause analysis devices 100 and 100A are explained.

[0079] The causal analysis devices 100 and 100A are configured to include: data acquisition units 110 and 110A, variable setting units 120 and 120A, model learning units 130 and 130A, causal data acquisition units 140 and 140A, contribution calculation units 150 and 150A, causal analysis units 160 and 160A, and causal analysis result output units 170 and 170A.

[0080] Data acquisition units 110 and 110A acquire sensor data and production management information as monitoring data.

[0081] Data acquisition units 110 and 110A acquire various types of surveillance data related to the monitored object.

[0082] The data acquisition units 110 and 110A acquire multiple time-series sensor data collected by multiple sensors 200 installed on the target equipment, as well as production management information related to the production of the target equipment, and use it as monitoring data.

[0083] Data acquisition units 110 and 110A acquire sensor data from sensor 200.

[0084] In addition, data acquisition units 110 and 110A acquire production management information related to the production of the target equipment.

[0085] Production management information is time-series data obtained at specified intervals over a specified period of time.

[0086] Production management information includes at least one data point related to production, such as production conditions (production speed, product changeover, operators, etc.), production quantity, and operation time (value production time, adverse situation response time, preparation time, etc.).

[0087] Sensor data and production management information are obtained as monitoring data.

[0088] That is, the data acquisition units 110 and 110A acquire multiple time-series sensor data collected by multiple sensors installed on the monitored device, and production management information related to the production of the monitored device, as the monitoring data.

[0089] In cases where the sampling period varies depending on the data, resampling can also be performed based on the longest sampling period in the monitoring data.

[0090] For example, if n represents the number of monitored data, then the monitored data are represented as X1, X2, X3, X4, ... , Xn. Additionally, if at time 1, time 2, If data is collected at each time point t, the monitoring data is represented by a two-dimensional data frame with rows set to the time point t and columns set to the quantity n.

[0091] The value of the first monitoring data X1 at time 1 is represented as X11, the value of the second monitoring data X2 at time 1 is represented as X21, and the value of the first monitoring data X1 at time 2 is represented as X12.

[0092] The variable setting units 120 and 120A acquire monitoring data and, based on the monitoring data, generate monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0093] The variable setting units 120 and 120A generate monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis based on the monitoring data.

[0094] The variable setting units 120 and 120A set a data point representing the consumption efficiency or productivity of a certain resource as the target variable.

[0095] Here, the target variable can also directly utilize the monitoring data selected by the user. For example, the variable setting units 120 and 120A can also set data representing product quality as the target variable.

[0096] The variable setting units 120 and 120A can also use data obtained by preprocessing at least one monitoring data selected by the user as target variables.

[0097] The variable setting units 120 and 120A can, for example, calculate the original unit based on the production quantity and resource consumption, and set it as the target variable.

[0098] The target variable can also be discretized by more than one threshold (e.g., when the monitored data are continuous values).

[0099] At this point, the threshold can be set manually by the user, or statistical measures such as the mean, median, and standard deviation of the target variable can be used.

[0100] In addition, the variable setting units 120 and 120A can also be used by the user to select two or more data related to the target variable, that is, the data considered as the factors causing the target variable to change, from the monitoring data, and set them as explanatory variables.

[0101] However, the variable description does not include monitoring data that are affected by the target variable.

[0102] For example, in the cause-effect diagram described later, the variable setting units 120 and 120A do not include monitoring data that are located on the result side (sub-, downstream) of the target variable in the description variables.

[0103] Then, the variable setting units 120 and 120A generate a dataset that pairs the target variable with the explanatory variable.

[0104] The variable setting units 120 and 120A extract the dataset corresponding to a portion or all of the times when the dataset is generated, and output it as the learning dataset to the model learning units 130 and 130A.

[0105] In addition, when the variable setting units 120 and 120A extract the dataset corresponding to a portion of the time as the learning dataset, they output the dataset corresponding to the time other than the portion of the time as the factor analysis dataset to the factor analysis result output units 170 and 170A.

[0106] When all datasets are used as learning datasets, variable setting units 120 and 120A output all datasets as factor analysis datasets to factor analysis result output units 170 and 170A.

[0107] When the variable setting units 120 and 120A extract the dataset corresponding to a portion of the time as the learning dataset, they output the monitoring data corresponding to a portion of the time as the learning monitoring data and the monitoring data corresponding to other times as the factor analysis monitoring data to the contribution calculation units 150 and 150A.

[0108] With all datasets used as learning datasets, all monitoring data are output to contribution calculation units 150 and 150A as factor analysis monitoring data and learning monitoring data.

[0109] Model learning units 130 and 130A generate a learned model based on the learning dataset, using explanatory variables as inputs and outputting target variables.

[0110] Model learning units 130 and 130A generate learning models based on the learning dataset.

[0111] In detail, the model learning units 130 and 130A optimize the parameters constituting the learning model by taking the explanatory variables of the learning dataset as input, so that the values ​​output from the learning model are close to the target variables of the learning dataset.

[0112] The Model Learning Units 130 and 130A can use the following models as well-known learning models.

[0113] When the target variable is a continuous value, the model learning units 130 and 130A can use, for example, regression trees, random forest regression, neural networks, support vector regression, etc.

[0114] When the target variable is a discrete value, the model learning units 130 and 130A can use, for example, decision trees, random forests, neural networks, support vector machines, etc.

[0115] Learning models can also be constructed by combining at least one or more methods.

[0116] Causal relationship data acquisition units 140 and 140A acquire causal relationship data representing causal relationships in various surveillance data.

[0117] Causal relationship data is data that represents the dependency relationship between data.

[0118] Causal relationship data is data that represents the causal and dependency relationships between multiple monitoring data.

[0119] Causal relationship data refers to the causal relationships and dependencies between the data input into the learning model.

[0120] Causal relationship data is data created manually by users.

[0121] Alternatively, causal relationship data can be inferred from surveillance data.

[0122] Alternatively, causal data is data that is given in advance using both user-generated data and data inferred from monitoring data.

[0123] Causal relationship data is, for example, a causal graph described later. Figure 4 , Figure 5 or Figure 6 The data presented.

[0124] In the following explanation, the causal relationship data will also be recorded as a causal diagram.

[0125] The contribution calculation units 150 and 150A calculate the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on the causal relationship data representing the causal relationship in various monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0126] The contribution calculation units 150 and 150A acquire the causal graph, the learning model, the monitoring data for learning, and the monitoring data for causal analysis, and calculate the contribution of each data sample, which quantifies the influence of the explanatory variable on the target variable. Here, the causal graph may be pre-stored in a causal graph storage unit, or it may be configured to output from the causal graph storage unit to the contribution calculation units 150 and 150A. In the case where the contribution calculation units 150 and 150A directly receive the causal graph from the causal graph storage unit, it may also be configured without the aforementioned causal data acquisition unit.

[0127] The contribution calculation units 150 and 150A calculate the contribution of each data sample to the degree of influence of the target variable, based on the input causal graph, learning model, learning monitoring data, and causal analysis monitoring data.

[0128] In addition to the conventional contribution calculation methods, the contribution calculation units 150 and 150A are extended by taking into account cause-effect graphs.

[0129] The contribution calculation units 150 and 150A have the function of calculating the magnitude of the influence of the explanatory variable on the target variable based on the dependency relationship between data shown in the cause-effect diagram, while also considering indirect influences.

[0130] For example, as a way to quantify the magnitude of the impact, XAI (Explainable AI) such as SHAP value and LIME can be used.

[0131] The contribution calculation units 150 and 150A use the learning model at the end of the learning process, monitoring data for causal analysis, monitoring data for learning, and causal graphs (causal relationship data) to calculate the contribution of the explanatory variable to the target variable.

[0132] The contribution calculated by the contribution calculation units 150 and 150A is a contribution that considers indirect contributions in addition to direct contributions, based on the causal graph (causal relationship data).

[0133] For example, contribution is a SHAP value that can be calculated using a learning model.

[0134] The SHAP value of a specific variable in a given data sample is the difference between the output value of the learning model when that variable is included in the input of the learning model and the output value of the learning model when that variable is not included in the input of the learning model. Here, a data sample refers to data corresponding to a specific time point.

[0135] In addition, the sum of the SHAP values ​​of all explanatory variables for a given data sample is equal to the difference between the output value of the learning model for that data sample and the average output value of the learning model across all data samples.

[0136] Therefore, the SHAP value is a quantitative measure of the impact of the explanatory variable on the target variable for each data sample; that is, the value of the target variable that changes to the extent that the explanatory variable is present or absent.

[0137] In detail, for example, the SHAP value is represented by equations (1) and (2) as follows.

[0138] The contribution calculation units 150 and 150A use the SHAP value to calculate the contribution based on the difference between the expected values ​​of the learning model output values ​​in the presence or absence of variable i.

[0139]

[0140] Φ i : Describes the SHAP value of variable i

[0141] N: The set of indices for all declared variables

[0142] S: A partial set that does not include the set N containing the variable i.

[0143] :N / S

[0144] : The set of indexes of monitoring data not included in the set

[0145] x S In the monitoring data used for cause analysis, set S contains the monitoring data (description variables).

[0146] In the monitoring data used for cause analysis, the set Included monitoring data (description of variables)

[0147] In the monitoring data used for cause analysis, the set The included monitoring data

[0148] Value function

[0149] Learning Model

[0150] Additionally, in this explanation, a label will be added above the text. The marker used to describe this is recorded as " "bar". For example, Recorded as "Sbar" Recorded as "Nbar" Recorded as "Xbar" Recorded as "X" S bar Recorded as "X" N bar.

[0151] In equation (2), term A represents the learning model f.

[0152] In the calculation of v(S) shown in Equation (2), the input to the model f is the explanatory variable selected by the user.

[0153] In addition, the probability distribution P (the B term of Equation (2)) considering the dependency relationship is always defined based on all monitoring data, including those other than the explanatory variables.

[0154] Therefore, whether the explanatory variables are pre-determined by user judgment or by receiving analysis results, factor analysis can be performed while taking into account the dependencies between the monitored data.

[0155] That is, the ability to effectively utilize user insights to identify key factors.

[0156] For example, when there are 5 variables, the relationship between the output value of the learning model and the SHAP value is shown in the following equation (3).

[0157] However, in equation (3), “Φ0” is the average value of the output value of the learning model, and “x” is all the explanatory variables.

[0158]

[0159] In the contribution calculation units 150 and 150A, when the explanatory variables contained in the set S are given, the probability distribution P of the explanatory variables used to calculate the value function v(S) is defined based on the causal graph (the B term of equation (2)).

[0160] The cause-effect graph is given. It can be created manually by the user, or it can be inferred from the data using well-known cause-effect graph inference methods (PC algorithm, GES, LiNGAM, etc.), or both. The cause-effect graph is given in the following form.

[0161] Here, the causal graph can also be pre-stored in the causal graph storage unit, for example.

[0162] Explain the cause-and-effect diagram.

[0163] Figure 4 This is a diagram representing a first example of causal relationship data (causal diagram) used in the cause analysis apparatus 100, 100A of this disclosure.

[0164] Figure 5 This is a diagram representing a second example of causal relationship data (causal diagram) used in the cause analysis apparatus 100, 100A of this disclosure.

[0165] Figure 6 This is a diagram representing a third example of causal relationship data (causal diagram) used in the cause analysis apparatus 100, 100A of this disclosure.

[0166] A cause-effect graph is a directed acyclic graph (DAG) that represents the relationships between monitored data. Cause-effect graphs can be specified in the form of charts, matrices, parent-child relationships, etc.

[0167] Chart: Nodes represent monitoring data Xi, and arrows represent the relationship between monitoring data Xi and Xj. The starting point of the arrow is set as the parent monitoring data Xi, and the ending point is set as the child monitoring data Xj to represent the parent-to-child relationship. Figure 4 )

[0168] Matrix: Rows represent parent monitoring data Xi, and columns represent child monitoring data Xj. If a parent-to-child relationship exists, the element in row i and column j is represented as 1; otherwise, it is represented as 0. Figure 5 )

[0169] Parent-child relationship: Let pa(Xj) represent the set of monitoring data (Xj being the child) that are the parent of the monitoring data (Xj). Figure 6 )

[0170] For example, for 5 monitoring data X1, X5, cause-effect diagram as follows Figure 4 , Figure 5 , Figure 6 That's how it's specified. Here, Figure 4 , Figure 5 , Figure 6 All represent the same cause-effect graph.

[0171] Contribution calculation units 150 and 150A, for example, when giving a causal graph representing the relationship between monitoring data, express the probability distribution P of the explanatory variable (term B of equation (2)) as in the following equations (4) and (5).

[0172] In the contribution calculation department, for example, when presenting a causal graph representing the relationship between monitoring data, the probability distribution of the explanatory variables can be represented as follows. .

[0173]

[0174] The parent variable of the declarative variable Xj

[0175] In the monitoring data used for cause analysis, describe the monitoring data (describe the variable) contained in the parent set S of variable Xj.

[0176] In the monitoring data used for cause analysis, describe the parent set of variable Xj. Included monitoring data (description of variables)

[0177] In the monitoring data used for cause analysis, the parent and child variables of variable Xj are described. The monitoring data contained in the collection

[0178] Term C in equation (4) represents the relationship between monitoring data. In the relationship between monitoring data, the result (child) depends only on the cause (parent). (When the parent is given, the child is independent of the parent.)

[0179] For example, for 5 monitoring data X1, X5, as shown in the following cause-effect graph.

[0180]

[0181] For example, if the target variable is set to X5, and the variables are X1, X2, and X4, i.e., N = {1, 2, 4}, In the case of , the probability distribution P (the B term of equation (2)) can be represented as follows.

[0182]

[0183] For example, if the target variable is set to X5, and the variables are X1, X2, and X4, i.e., N = {1, 2, 4}, In the case of P (the B term in equation (2)), P can be expressed as follows.

[0184]

[0185] For example, if the target variable is set to X5, and the variables are X1, X3, and X4, i.e., N = {1, 3, 4}, In the case of P (the B term in equation (2)), P can be expressed as follows.

[0186]

[0187] The contribution calculation units 150 and 150A can, for example, use a normal distribution, Dirichlet distribution, t-distribution, empirical distribution, etc., as the probability distribution P (term B of equation (2)). When using a distribution with parameters as the probability distribution, each parameter can be determined using the learning monitoring data. For example, when using a normal distribution as the probability distribution, the expected value and variance-covariance matrix can be calculated using the learning monitoring data.

[0188] The contribution calculation units 150 and 150A calculate the contribution of the explanatory variable i of the nth data sample of the monitoring data used for cause analysis by setting it as Φin, and output it to the cause analysis units 160 and 160A and the cause analysis result output units 170 and 170A.

[0189] The contribution rate is obtained and statistically analyzed by the 160 and 160A components of the causality analysis department, and the causality degree is calculated.

[0190] Factor analysis sections 160 and 160A calculate the degree to which each explanatory variable is a factor, or factor degree, based on contribution. The factor degree calculated for each explanatory variable can also be described as a value representing the extent to which that explanatory variable may be a factor influencing the target variable.

[0191] Figure 7 This is a diagram illustrating the contribution values ​​output by the contribution calculation units 150 and 150A in the factor analysis apparatus 100 and 100A of this disclosure.

[0192] In detail, the contribution Φin of each data sample for explanatory variable i is calculated, and the factor degree Ci of explanatory variable i is calculated. Figure 7 )

[0193] For example, the average of the absolute values ​​of Φin can be set as the factor Ci of the variable i.

[0194]

[0195] The factor analysis units 160 and 160A output the calculated factor degree to the factor analysis result output units 170 and 170A.

[0196] The causal analysis output units 170 and 170A output causal analysis results based on the causal analysis dataset, contribution, and causal degree.

[0197] The cause analysis output units 170 and 170A acquire the data set used for cause analysis, contribution rate, and cause degree, and output information related to the cause analysis results.

[0198] Figure 8 This is a diagram illustrating the factor degree output by the factor analysis units 160 and 160A in the factor analysis apparatus 100 and 100A of this disclosure.

[0199] In the cause analysis result output sections 170 and 170A, the factor degree is used to output information related to the factor degree.

[0200] Figure 9 This is a diagram showing a first example of the display of the cause analysis results output by the cause analysis result output units 170 and 170A in the cause analysis apparatus 100 and 100A of this disclosure.

[0201] Figure 9 This shows an example of the output screen related to factor degree when the number of explanatory variables is set to 6.

[0202] The display shows information describing variable i and information describing factor Ci. Figure 9 )

[0203] In the initial state, the variables are displayed in the order they were entered or in the order they were pre-stored.

[0204] Based on the factor, a sorting button is displayed to arrange the sorted items in descending order.

[0205] When the sort button is pressed, the items are sorted in descending order, and the sort button is highlighted in black.

[0206] When the sort button is pressed again, the variables will be arranged in the order they were entered or stored.

[0207] In the initial state, the elements can also be sorted in descending order of importance. At this time, the sort button will be blacked out.

[0208] In the causal analysis results output sections 170 and 170A, the contribution rate and the causal analysis dataset are used to output information representing the relationship between the contribution rate, explanatory variables, and target variables.

[0209] Figure 10 This is a diagram showing a second example of the result of the cause analysis output by the cause analysis result output unit 170, 170A in the cause analysis apparatus 100, 100A of this disclosure.

[0210] Figure 10 This shows an example of an output screen showing the relationship between the explanatory variable 1, the explanatory variable, the contribution, and the target variable.

[0211] A scatter plot showing the values ​​of variables configured on the horizontal axis and their contributions configured on the vertical axis.

[0212] Alternatively, the axis representing the values ​​of the explanatory variables can be reversed with the axis representing the contribution, displaying a scatter plot with the contribution on the horizontal axis and the values ​​of the explanatory variables on the vertical axis.

[0213] In a scatter plot, it is shown as a dashed line representing a contribution of 0.

[0214] In a scatter plot, each point corresponds to one data sample.

[0215] The concentration at points represents the magnitude relationship of the target variable values ​​in the dataset used for causal analysis.

[0216] When the expand button is pressed, a list of explanatory variables is displayed. When an explanatory variable is selected, a scatter plot corresponding to the selected explanatory variable is displayed.

[0217] When the target variable is discretized data, i.e. categorical data, you can change the shape of the points rather than their density.

[0218] In addition to the structure described above, the cause analysis devices 100 and 100A may also be configured to include a control unit (not shown) and a storage unit (not shown).

[0219] A control unit (not shown) controls the entirety of the cause analysis devices 100 and 100A. For example, the control unit (not shown) may activate the cause analysis devices 100 and 100A, or shut them down or put them into a sleep state according to external commands.

[0220] The storage unit (not shown) stores various data used in the cause analysis devices 100 and 100A. The storage unit (not shown) stores, for example, the outputs (output data) performed by each structural unit in the cause analysis devices 100 and 100A, and outputs data requested by each structural unit to the structural unit of the request source.

[0221] The same applies to other implementations.

[0222] The processing examples of the cause analysis devices 100 and 100A are explained.

[0223] Figure 3The processing shown is a cause-effect analysis method based on cause-effect analysis devices 100 and 100A, including: a data acquisition step, where the data acquisition units 110 and 110A of the cause-effect analysis devices 100 and 100A acquire various monitoring data related to the monitored object; a variable setting step, where the variable setting units 120 and 120A of the cause-effect analysis devices 100 and 100A generate, based on the monitoring data, monitoring data for learning, monitoring data for cause-effect analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause-effect analysis; a model learning step, where the model learning units 130 and 130A of the cause-effect analysis devices 100 and 100A generate a learning model based on the learning dataset, learning in a manner that takes explanatory variables as inputs and outputs target variables; and a causal relationship data acquisition step, where the causal relationship data of the cause-effect analysis devices 100 and 100A... The data acquisition units 140 and 140A acquire causal relationship data representing causal relationships among various monitoring data. In the contribution calculation step, the contribution calculation units 150 and 150A of the causal analysis devices 100 and 100A calculate the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on the causal relationship data representing causal relationships among various monitoring data, the learning model, the monitoring data for learning, and the monitoring data for causal analysis. In the causal analysis step, the causal analysis units 160 and 160A of the causal analysis devices 100 and 100A calculate the degree of influence of each of the explanatory variables, i.e., the causal degree, based on the contribution. In the causal analysis result output step, the causal analysis result output units 170 and 170A of the causal analysis devices 100 and 100A output the causal analysis result based on the dataset for causal analysis, the contribution, and the causal degree.

[0224] Furthermore, if the contribution calculation units 150 and 150A are configured to directly receive the causal graph from the causal graph storage unit, the cause analysis method may also include the step of the contribution calculation units 150 and 150A receiving the causal graph from the causal graph storage unit, instead of the aforementioned step of obtaining causal relationship data.

[0225] When the cause analysis devices 100 and 100A start Figure 3 During the processing shown, firstly, the cause analysis devices 100 and 100A perform monitoring data acquisition processing. (Step ST110)

[0226] In the monitoring data acquisition and processing, the data acquisition units 110 and 110A of the cause analysis devices 100 and 100A acquire various monitoring data related to the monitored object.

[0227] The data acquisition units 110 and 110A acquire sensor data and production management information from the sensor 200 as monitoring data and output them to the variable setting units 120 and 120A.

[0228] Next, the cause analysis devices 100 and 100A perform variable setting processing. (Step ST120)

[0229] In the variable setting process, the variable setting units 120 and 120A of the cause analysis devices 100 and 100A generate learning monitoring data, cause analysis monitoring data, learning dataset, and cause analysis dataset based on the aforementioned monitoring data.

[0230] The variable setting units 120 and 120A perform variable setting processing on the monitoring data output by the data acquisition units 110 and 110A in step ST110, output the learning dataset to the model learning unit 130 and 130A, output the learning monitoring data and the causal analysis monitoring data to the contribution calculation unit 150 and 150A, and output the causal analysis dataset to the causal analysis result output unit 170 and 170A.

[0231] Next, the cause analysis devices 100 and 100A perform model learning processing. (Step ST130)

[0232] In the model learning process, the model learning units 130 and 130A of the factor analysis devices 100 and 100A generate a learned model based on the learning dataset, in a manner that takes explanatory variables as inputs and outputs target variables.

[0233] The model learning units 130 and 130A learn the learning dataset output by the variable setting units 120 and 120A in step ST120 by taking the explanatory variables as input and outputting the target variables, and output the learned model to the contribution calculation units 150 and 150A.

[0234] Next, the cause analysis devices 100 and 100A perform causal relationship data acquisition processing. (Step ST140)

[0235] In the causal relationship data acquisition and processing, the causal relationship data acquisition units 140 and 140A of the causal analysis devices 100 and 100A acquire causal relationship data representing causal relationships in various monitoring data.

[0236] Next, the cause analysis devices 100 and 100A perform contribution calculation processing. (Step ST150)

[0237] In the contribution calculation process, the contribution calculation units 150 and 150A of the causal analysis devices 100 and 100A calculate the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on the causal relationship data representing the causal relationship in various monitoring data, the learning model, the monitoring data for learning, and the monitoring data for causal analysis.

[0238] The contribution calculation units 150 and 150A obtain the cause-effect graph, and perform contribution calculation processing on the learning model output by the model learning units 130 and 130A in step ST130, the learning monitoring data output by the variable setting units 120 and 120A in step ST120, and the causal analysis monitoring data. The contribution is then output to the causal analysis units 160 and 160A and the causal analysis result output units 170 and 170A.

[0239] Next, the cause analysis devices 100 and 100A perform cause analysis processing. (Step ST160)

[0240] In the cause analysis process, the cause analysis units 160 and 160A of the cause analysis devices 100 and 100A calculate the degree to which the above-mentioned explanatory variables are considered as causes, i.e., the cause degree, based on the above-mentioned contribution.

[0241] The cause analysis units 160 and 160A perform cause analysis processing on the contribution values ​​output by the contribution calculation units 150 and 150A in step ST150, and output the cause values ​​to the cause analysis result output units 170 and 170A.

[0242] Next, the cause analysis devices 100 and 100A perform cause analysis result output processing. (Step ST170)

[0243] In the cause analysis result output processing, the cause analysis result output units 170 and 170A of the cause analysis devices 100 and 100A output the cause analysis results based on the aforementioned cause analysis dataset, the aforementioned contribution degree, and the aforementioned cause degree.

[0244] The cause analysis result output units 170 and 170A acquire the cause analysis dataset output by the variable setting units 120 and 120A in step ST120, the contribution degree output by the contribution calculation units 150 and 150A in step ST150, and the cause degree output by the cause analysis units 160 and 160A in step ST160, and display information related to the cause analysis results.

[0245] Next, the cause analysis devices 100 and 100A are completed. Figure 3 The processing shown.

[0246] The cause analysis apparatus disclosed herein is configured as follows.

[0247] A factor analysis device, comprising:

[0248] The data acquisition department acquires various types of surveillance data related to the monitored objects.

[0249] The variable setting unit generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0250] The model learning unit generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0251] The contribution calculation unit calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0252] The causal analysis unit, based on the stated contribution level, calculates the degree to which each of the stated explanatory variables is a factor, i.e., the factor degree; and

[0253] The cause analysis results output unit outputs cause analysis results based on the cause analysis dataset, the contribution degree, and the cause degree.

[0254] Therefore, this disclosure provides a factor analysis apparatus that can improve the accuracy of factor analysis.

[0255] The factor analysis method disclosed herein is structured as follows.

[0256] A cause-effect analysis method, performed by a cause-effect analysis device, comprises the following steps:

[0257] In the data acquisition step, the data acquisition unit of the cause analysis device acquires various monitoring data related to the monitored object;

[0258] In the variable setting step, the variable setting unit of the cause analysis device generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0259] In the model learning step, the model learning unit of the factor analysis device generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0260] In the contribution calculation step, the contribution calculation unit of the factor analysis device calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the learning monitoring data, and the factor analysis monitoring data.

[0261] The causal analysis step involves the causal analysis unit of the causal analysis device calculating the degree to which each explanatory variable is a factor, i.e., the causal degree, based on the contribution level; and

[0262] In the cause analysis result output step, the cause analysis result output unit of the cause analysis device outputs the cause analysis result based on the cause analysis dataset, the contribution degree, and the cause degree.

[0263] Therefore, this disclosure provides a factor analysis method that can improve the accuracy of factor analysis.

[0264] The factor analysis system disclosed herein is structured as follows.

[0265] A factor analysis system, comprising:

[0266] sensor;

[0267] The data acquisition unit acquires various monitoring data related to the monitored object measured by the sensor.

[0268] The variable setting unit generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0269] The model learning unit generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0270] The contribution calculation unit calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0271] The factor analysis department calculates the degree to which each explanatory variable is a factor, i.e., the factor degree, based on the aforementioned contribution.

[0272] The causal analysis results output unit outputs causal analysis results based on the causal analysis dataset, the contribution rate, and the causal degree; and

[0273] The display device displays the cause analysis results output by the cause analysis result output unit.

[0274] Therefore, this disclosure provides a factor analysis system that can improve the accuracy of factor analysis.

[0275] The factor analysis procedure disclosed herein is structured as follows.

[0276] A factor analysis procedure, characterized in that,

[0277] To enable the computer to function as a factor analysis device.

[0278] The cause analysis device includes:

[0279] The data acquisition department acquires various types of surveillance data related to the monitored objects.

[0280] The variable setting unit generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis.

[0281] The model learning unit generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables.

[0282] The contribution calculation unit calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis.

[0283] The causal analysis department, based on the stated contribution level, calculates the degree to which each explanatory variable is a factor, i.e., the factor degree; and

[0284] The cause analysis results output unit outputs cause analysis results based on the cause analysis dataset, the contribution degree, and the cause degree.

[0285] Therefore, this disclosure provides a factor analysis procedure that can improve the accuracy of factor analysis.

[0286] The cause analysis apparatus disclosed herein is further configured as follows.

[0287] Based on the aforementioned factor analysis apparatus, the characteristic of the factor analysis apparatus is that...

[0288] The data acquisition unit acquires multiple time-series sensor data collected by multiple sensors installed on the monitored device, and production management information related to the production of the monitored device, as the monitoring data.

[0289] Therefore, this disclosure provides a factor analysis apparatus that can improve the accuracy of factor analysis related to production equipment.

[0290] Furthermore, this disclosure achieves the same effect as described above by applying the above structure to the above cause analysis method, the above cause analysis system, or the above cause analysis procedure.

[0291] The cause analysis apparatus disclosed herein is further configured as follows.

[0292] Based on the aforementioned factor analysis apparatus, the characteristic of the factor analysis apparatus is that...

[0293] The contribution calculation unit uses XAI (Explainable AI) to calculate the contribution.

[0294] Therefore, this disclosure can use XAI to calculate contribution, thus providing a factor analysis apparatus that can further improve the accuracy of factor analysis.

[0295] In addition, this disclosure provides a cause analysis apparatus that allows users to easily interpret the cause analysis process.

[0296] Furthermore, this disclosure achieves the same effect as described above by applying the above structure to the above cause analysis method, the above cause analysis system, or the above cause analysis procedure.

[0297] Implementation method 2.

[0298] Implementation method 2 describes a method that can reflect the evaluation results related to the reasonableness of the explanatory variables.

[0299] In the description of Embodiment 2, sometimes the same content as that already described in Embodiment 1 is appropriately omitted.

[0300] An example of the structure of a cause analysis system including cause analysis device 100B will be described.

[0301] Figure 11 This is a diagram illustrating a structural example of a factor analysis system 10 including the factor analysis apparatus 100B of Embodiment 2 of this disclosure.

[0302] In addition, Figure 11 In this embodiment, only the characteristic structure of embodiment 2 is shown, relative to the structure shown in embodiment 1. In embodiment 2, additional structures may be added based on the structure shown in embodiment 1. Figure 11 Structures not shown in the diagram.

[0303] The cause analysis device of the cause analysis system can also provide an evaluation related to the rationality of variables accepted by the user in the output screen related to the degree of cause analysis results output from the cause analysis result output unit.

[0304] The cause analysis system 10 is configured to include a cause analysis device 100B, a sensor 200, and a display device 300.

[0305] The cause analysis device 100B is configured to include a data acquisition unit 110B, a variable setting unit 120B, a model learning unit 130B, a causal relationship data acquisition unit 140B, a contribution calculation unit 150B, a cause analysis unit 160B, and a cause analysis result output unit 170B.

[0306] In the above structure, Figure 11 The diagram shows a variable setting unit 120B, a model learning unit 130B, a causal data acquisition unit 140B, a contribution calculation unit 150B, a causal analysis unit 160B, and a causal analysis result output unit 170B. The following mainly describes the contents not described in the above embodiment.

[0307] The cause analysis result output unit 170B outputs information related to at least one or more explanatory variables that have been judged to be unreasonable through evaluations related to reasonableness as explanatory variable removal information to the variable setting unit 120B.

[0308] Figure 12 This is a diagram showing an example of a factor analysis result output by the factor analysis result output unit 170B in the factor analysis apparatus 100B of Embodiment 2 of this disclosure.

[0309] For example, in the removal decision column of the output screen related to the degree of cause analysis, the cause analysis result output unit 170B sets a checkbox and accepts user input.

[0310] For example, a checkbox is initially unselected. When an input action (pressing or pushing) is performed on the checkbox while it is unselected, it becomes selected. Conversely, when an input action (pressing or pushing) is performed on the checkbox while it is selected, it becomes unselected.

[0311] For example, it is possible to set the description variable corresponding to the selected checkbox as the description variable to remove information.

[0312] In the figure, explanatory variable 3 and explanatory variable 4 are set as explanatory variables to remove information.

[0313] When at least one of the selected states is selected, and the cause analysis is executed again, the cause analysis result output unit 170B outputs explanatory variable removal information to the variable setting unit 120B, and performs update-related processing based on the user's evaluation related to rationality.

[0314] The variable setting unit 120B uses the explanatory variable removal information output by the cause analysis result output unit 170B and the information of the monitoring data set as explanatory variables in the monitoring data to reset the monitoring data not included in the "monitoring data included in the explanatory variable removal information" in the "monitoring data set as explanatory variables" as the updated explanatory variables.

[0315] Here, the information of the monitoring data set as the description variable is stored in the variable setting section 120B.

[0316] The cause analysis device 100B is configured to replace the explanatory variables with updated explanatory variables and perform the same processing as in Implementation Method 1 to obtain the updated learning model, updated contribution, and updated cause degree, and output information related to the updated cause analysis.

[0317] The processing in the above structure will be explained.

[0318] The cause analysis result output unit 170B outputs information related to at least one or more explanatory variables that have been judged to be unreasonable through evaluations related to reasonableness as explanatory variable removal information to the variable setting unit 120B.

[0319] The variable setting unit 120B uses the explanatory variable removal information output by the cause analysis result output unit 170B and the information of the monitoring data set as explanatory variables in the monitoring data to reset the monitoring data not included in the "monitoring data included in the explanatory variable removal information" in the "monitoring data set as explanatory variables" as the updated explanatory variables.

[0320] In the following, the explanatory variables will be replaced with the updated explanatory variables in the processing described in Implementation 1, and the same processing as in Implementation 1 will be performed to obtain the updated learning model, the updated contribution degree, and the updated factor degree, and output information related to the updated factor analysis.

[0321] In addition, the above-mentioned processing related to updating based on user evaluations of reasonableness can be repeated.

[0322] In Implementation 2, the following structure is described: The user evaluates the cause-effect analysis results based on the information related to cause-effect analysis output to the display device, performs a phase difference analysis on the candidate causes, removes the candidate causes that are determined to be unnecessary, and performs cause-effect analysis processing again. As a result, the accuracy of cause-effect analysis is improved.

[0323] The cause analysis apparatus disclosed herein is further configured as follows.

[0324] Compared to the cause analysis apparatus of other embodiments, the cause analysis apparatus is further configured as follows:

[0325] The factor analysis output unit, based on the factor analysis results, accepts an evaluation related to the reasonableness of the explanatory variables.

[0326] The variable setting unit sets the explanatory variable based on the evaluation results related to the reasonableness of the explanatory variable, in a manner that does not use the monitoring data of the explanatory variable that has been evaluated as having been removed.

[0327] Therefore, this disclosure provides a factor analysis apparatus that can further improve the accuracy of factor analysis.

[0328] Furthermore, this disclosure achieves the same effect as described above by applying the above structure to the above cause analysis method, the above cause analysis system, or the above cause analysis procedure.

[0329] Implementation method 3.

[0330] Implementation method 3 describes a method that can output results that satisfy aggregation conditions that represent conditions related to the target variable and the explanatory variable.

[0331] In the description of Embodiment 3, sometimes the same content as that already described in Embodiment 1 and Embodiment 2 is appropriately omitted.

[0332] Figure 13 This is a diagram illustrating a structural example of a factor analysis system 10 including the factor analysis apparatus 100C of Embodiment 3 of this disclosure. Furthermore, in Figure 13 In this embodiment, only the characteristic structure of embodiment 3 is shown, relative to the structures shown in embodiments 1 and 2. In embodiment 3, additional structures may be added based on the structures shown in embodiments 1 and 2. Figure 13 Structures not shown in the diagram.

[0333] Figure 14 This is a diagram illustrating the polymerization conditions of the factor analysis apparatus 100C used in Embodiment 3 of this disclosure.

[0334] The cause analysis device 100C can also be configured such that, in the cause analysis unit 160C, based on aggregated condition information representing conditions related to the target variable and explanatory variable input by the user, the cause degree is calculated for the contribution of satisfying the conditions.

[0335] The cause analysis system 10 is configured to include a cause analysis device 100C, a sensor 200, and a display device 300.

[0336] The causal analysis device 100C is configured to include a data acquisition unit 110C, a variable setting unit 120C, a model learning unit 130C, a causal relationship data acquisition unit 140C, a contribution calculation unit 150C, a causal analysis unit 160C, a causal analysis result output unit 170C, and an aggregation condition acquisition unit 180C.

[0337] In the above structure, Figure 13 The diagram shows a variable setting unit 120C, a contribution calculation unit 150C, a cause analysis unit 160C, a cause analysis result output unit 170C, and a aggregation condition acquisition unit 180C. Hereinafter, we will mainly explain the contents not described in the above embodiments.

[0338] The polymerization conditions are obtained at 180°C.

[0339] Aggregation condition information, representing aggregation conditions, can be, for example, expressed as... Figure 14 It is given in the form shown.

[0340] Figure 14 The format shown is for setting lower and upper limits for each target variable and descriptive variable.

[0341] The cause analysis unit 160C obtains the aggregation conditions and outputs the conditional contribution degree, conditional cause degree, and conditional cause analysis dataset corresponding to the monitoring data shown by the aggregation conditions in the aforementioned cause analysis dataset, so as to satisfy the aggregation conditions.

[0342] The cause analysis result output unit 170C uses the conditional contribution degree, the conditional cause degree, and the conditional cause analysis dataset output by the cause analysis unit 160C to output the cause analysis results.

[0343] An example of the processing of the cause analysis device 100C will be explained.

[0344] First, the polymerization conditions are obtained by the polymerization condition acquisition section 180C of the factor analysis apparatus 100C.

[0345] Next, the cause analysis unit 160C obtains the aggregation conditions and outputs the conditional contribution, conditional cause degree, and conditional cause analysis dataset corresponding to the monitoring data shown by the aggregation conditions in the aforementioned cause analysis dataset, so as to satisfy the aggregation conditions.

[0346] The factor analysis unit 160C of the factor analysis device 100C uses the aggregation condition information obtained by the aggregation condition acquisition unit 180C to calculate the factor degree for the contribution degree of satisfying the specified conditions in the same way as in Embodiment 1.

[0347] For example, the contribution level that meets the specified conditions can be the data sample numbers extracted from the target variable and explanatory variables respectively into the causal analysis datasets above the lower limit and below the upper limit, and set as the contribution level corresponding to the data sample numbers extracted from all target variables and explanatory variables. Additionally, it is possible to set the causal analysis datasets corresponding to the data sample numbers extracted from all target variables and explanatory variables as causal analysis datasets that meet the specified conditions.

[0348] If only an upper or lower limit is set, extract the data sample numbers that are below the upper limit or above the lower limit. If neither an upper nor lower limit is set, extract all data sample numbers.

[0349] For example, such as Figure 14 As shown, extract the contribution value corresponding to the data sample number that meets all of the following conditions.

[0350] The target variable is 2 or more and 4 or less.

[0351] This indicates that variable 1 is greater than or equal to 0.

[0352] This indicates that variable 3 is below 10.

[0353] This indicates that variable 4 is between 0.5 and 0.7.

[0354] This indicates that variable 5 is greater than or equal to 5.

[0355] Next, the cause analysis unit 160C obtains the aggregation conditions and outputs the conditional contribution, conditional cause degree, and conditional cause analysis dataset corresponding to the monitoring data shown by the aggregation conditions in the aforementioned cause analysis dataset, so as to satisfy the aggregation conditions.

[0356] Specifically, the cause analysis unit 160C takes the cause degree calculated based on the contribution degree that meets the specified conditions as the conditional cause degree, the contribution degree that meets the specified conditions as the conditional contribution degree, and the cause analysis dataset that meets the specified conditions as the conditional cause analysis dataset, and outputs it to the cause analysis result output unit 170C.

[0357] More specifically, the cause analysis unit 160C calculates the contribution degree output by the contribution degree calculation unit 150C, the cause analysis dataset output by the variable setting unit 120C, and the aggregation condition information according to the prescribed conditions, and outputs the calculated conditional cause degree, conditional contribution degree, and conditional cause analysis dataset to the cause analysis result output unit 170C.

[0358] The cause analysis result output unit 170C uses the conditional contribution, conditional cause degree, and conditional cause analysis dataset output by the cause analysis unit 160C to output the cause analysis results.

[0359] Specifically, the cause analysis result output unit 170C replaces the contribution degree with a conditional contribution degree, the cause degree with a conditional cause degree, and the cause analysis dataset with a conditional cause analysis dataset, performing the same processing as in Implementation Method 1.

[0360] Implementation method 3 illustrates a structure that sets conditions for the target variable and explanatory variables, calculates the contribution of each variable under a specific condition, and then calculates the causal degree. This further improves the accuracy of causal analysis.

[0361] The cause analysis apparatus disclosed herein is further configured as follows.

[0362] Compared to the cause analysis apparatus of other embodiments, the cause analysis apparatus is further configured as follows:

[0363] The factor analysis unit obtains aggregation conditions and outputs the conditional contribution degree, conditional factor degree, and conditional factor analysis dataset corresponding to the monitoring data represented by the aggregation conditions in the factor analysis dataset in a manner that satisfies the aggregation conditions.

[0364] The factor analysis result output unit uses the conditional contribution degree, the conditional factor degree, and the conditional factor analysis dataset output by the factor analysis unit to output the factor analysis results.

[0365] Therefore, this disclosure provides a factor analysis apparatus that can further improve the accuracy of factor analysis.

[0366] Furthermore, this disclosure achieves the same effect as described above by applying the above structure to the above cause analysis method, the above cause analysis system, or the above cause analysis procedure.

[0367] Implementation method 4.

[0368] Implementation method 4 describes a way to evaluate the suitability of the learning model for the monitoring data.

[0369] In the description of Embodiment 4, sometimes the same content as that already described in Embodiments 1, 2 and 3 is appropriately omitted.

[0370] Figure 15 This is a diagram illustrating a structural example of a factor analysis system 10 including the factor analysis apparatus 100D of Embodiment 4 of this disclosure. Furthermore, in Figure 15 In this embodiment, only the characteristic structure of embodiment 4 is shown, relative to the structures shown in embodiments 1, 2, and 3. In embodiment 4, additional structures may be added based on the structures shown in embodiments 1, 2, and 3. Figure 15 Structures not shown in the diagram.

[0371] Figure 16 This is a diagram showing a first display example of the cause analysis results output by the cause analysis result output unit 170D in the cause analysis apparatus 100D of Embodiment 4 of this disclosure.

[0372] Figure 17 This is a diagram showing a second example of the display of the cause analysis results output by the cause analysis result output unit 170D in the cause analysis apparatus 100D of Embodiment 4 of this disclosure.

[0373] The cause analysis system 10 is configured to include a cause analysis device 100D, a sensor 200, and a display device 300.

[0374] The cause analysis device 100D can also have a model fit evaluation unit 190D for evaluating the fit of the learning model to the monitoring data.

[0375] The causal analysis device 100D is configured to include a data acquisition unit 110D, a variable setting unit 120D, a model learning unit 130D, a causal data acquisition unit 140D, a contribution calculation unit 150D, a causal analysis unit 160D, a causal analysis result output unit 170D, a aggregation condition acquisition unit 180D, and a model fitness evaluation unit 190D.

[0376] In the above structure, Figure 15 The diagram shows the variable setting unit 120D, the model learning unit 130D, the cause-effect analysis result output unit 170D, and the model fitness evaluation unit 190D. The following mainly describes the contents not described in the above embodiment.

[0377] The Model Fit Evaluation Unit 190D uses the dataset for factor analysis and the learning model described above to calculate the fit of the learning model.

[0378] The cause analysis result output unit 170D also uses the fitness calculated by the model fitness evaluation unit 190D to output the cause analysis results.

[0379] An example of the processing of the cause analysis device 100D will be explained.

[0380] The model fitness evaluation unit 190D of the cause analysis device 100D uses the cause analysis dataset output by the variable setting unit 120D and the learning model output by the model learning unit 130D to generate the learning model output value, which is the output when the explanatory variables of the cause analysis dataset are input into the learning model.

[0381] The Model Fit Evaluation Unit 190D calculates the fit using the output values ​​of the generated learning model and the target variables of the dataset used for factor analysis.

[0382] The model fitness evaluation unit 190D outputs the generated learning model output value and the calculated fitness to the factor analysis result output unit 170D.

[0383] The cause analysis result output unit 170D of the cause analysis device 100D uses the fitness score output by the model fitness evaluation unit 190D to output information related to the fitness score. Additionally, the cause analysis result output unit 170D uses the learning model output value output by the model fitness evaluation unit 190D and the cause analysis dataset output by the variable setting unit 120D to output information related to the learning model output.

[0384] The Model Fit Evaluation Unit 190D inputs the explanatory variables from the causal analysis dataset into the learning model and generates the learning model output values. It then compares the learning model output values ​​with the target variables from the causal analysis dataset and calculates the fit.

[0385] Here, the fitness rating uses at least one accuracy evaluation metric set by the user.

[0386] When the target variable is continuous, it is possible to use, for example, mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and coefficient of determination.

[0387] When the target variable is binary classification data, metrics such as accuracy, fit, recall, and F1 score can be used.

[0388] When dealing with categorical data where the target variable has 3 or more values, metrics such as accuracy and average F1 score can be used.

[0389] However, when the training dataset is the same as the causal analysis dataset, the fitness can also be calculated through cross-validation. In this case, the learning model does not use the learning model output from the model learning unit 130D, but instead uses a newly learned model for cross-validation.

[0390] Here, cross-validation can use methods such as leave-one-out cross-validation, k-fold cross-validation, and hierarchical k-fold cross-validation.

[0391] When calculating more than two fitness values ​​for a precision metric through cross-validation, such as using the average, maximum, minimum, or median values, a representative value is calculated for the precision metric.

[0392] The model fitness evaluation unit 190D outputs the fitness calculated using the generated learning model output value and at least one accuracy evaluation index to the factor analysis result output unit 170D.

[0393] The cause analysis result output unit 170D uses the fitness score output by the model fitness evaluation unit 190D to output information related to the fitness score.

[0394] For example, such as Figure 16 In that case, the types of accuracy evaluation indicators and their corresponding suitability can be displayed in tabular form.

[0395] exist Figure 16 The table shows categorical data with a target variable of 2 values, used as accuracy evaluation metrics, with the user selecting F1 score, accuracy, and precision.

[0396] The cause analysis result output unit 170D uses the learning model output value output by the model fitness evaluation unit 190D and the cause analysis dataset output by the variable setting unit 120D to output information related to the learning model output.

[0397] For example, such as Figure 17 That way, a scatter plot is displayed, showing the data sample numbers of the dataset used for factor analysis on the horizontal axis and the target variable and model output values ​​on the vertical axis. Alternatively, it can be displayed as a line graph instead of a scatter plot.

[0398] exist Figure 17 In the diagram, black dots represent the values ​​of the target variable, and white dots represent the output values ​​of the learning model.

[0399] Implementation method 4 illustrates a structure capable of evaluating the fitness of a learning model. This improves the reliability of factor analysis by determining the quality of the learning model. For example, in cases of poor fitness, the specification of explanatory variables can be reconsidered.

[0400] The cause analysis apparatus disclosed herein is further configured as follows.

[0401] Compared to the cause analysis apparatus of other embodiments, the cause analysis apparatus is further configured as follows:

[0402] It includes a model fitness evaluation unit that uses the factor analysis dataset and the learning model to calculate the fitness of the learning model.

[0403] The factor analysis result output unit also uses the fitness calculated by the model fitness evaluation unit and outputs the factor analysis results.

[0404] Therefore, this disclosure provides a factor analysis apparatus that can further improve the reliability of factor analysis.

[0405] Furthermore, this disclosure achieves the same effect as described above by applying the above structure to the above cause analysis method, the above cause analysis system, or the above cause analysis procedure.

[0406] Hereinafter, the hardware structure for implementing the functions based on the structure of this disclosure described above will be explained.

[0407] Figure 18 This is a diagram illustrating a first example of a hardware structure used to implement the functionality based on the structure of this disclosure. Figure 18 The hardware structure shown implements the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D.

[0408] Figure 19 This is a diagram illustrating a second example of a hardware structure used to implement the functionality based on the structure of this disclosure. Figure 19 The hardware structure shown executes software that implements the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D.

[0409] The auxiliary storage device 1010 is a storage device having a storage area for reading and writing data. The storage unit (not shown) can be implemented using the auxiliary storage device 1010 or other memory (not shown).

[0410] The information input device 1030 is a device that inputs data into the cause analysis devices 100, 100A, 100B, 100C, and 100, such as a touch panel, mouse, and keyboard.

[0411] Data input using the information input device 1030 is input to the cause analysis device via the information input IF 1020.

[0412] Display IF1040 is the IF that relays the data input from cause analysis devices 100, 100A, 100B, 100C, and 100D to display 1050.

[0413] The monitor 1050 displays the data.

[0414] For example, the cause analysis result output units 170, 170A, 170B, 170C, and 170D output the results to the display 1050 via the display IF1040.

[0415] The display 1050 displays the estimation result input via the display IF1040 on the screen.

[0416] When the aforementioned processing circuit is dedicated hardware, the processing circuit 1060 is, for example, equivalent to a single circuit, a composite circuit, a programmable processor, a parallel programmable processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0417] The aforementioned processing circuit, through the hardware described above, implements the functions of data acquisition units 110, 110A, 110B, 110C, 110D; variable setting units 120, 120A, 120B, 120C, 120D; model learning units 130, 130A, 130B, 130C, 130D; causal relationship data acquisition units 140, 140A, 140B, 140C, 140D; contribution calculation units 150, 150A, 150B, 150C, 150D; causal analysis units 160, 160A, 160B, 160C, 160D; causal analysis result output units 170, 170A, 170B, 170C, 170D; aggregation condition acquisition units 180C, 180D; model fitness evaluation unit 190D; and a control unit (not shown).

[0418] When the aforementioned processing circuit is a processor 1070, the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D are implemented through software, firmware, or a combination of software and firmware.

[0419] The software or firmware is described as a program and stored in memory 1080.

[0420] The processor 1070 implements the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D by reading and executing the program stored in the memory 1080.

[0421] These programs are processes or methods by which the computer performs the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D.

[0422] The processor 1070 executes software that implements the functions of the data acquisition units 110, 110A, 110B, 110C, 110D, the variable setting units 120, 120A, 120B, 120C, 120D, the model learning units 130, 130A, 130B, 130C, 130D, the causal relationship data acquisition units 140, 140A, 140B, 140C, 140D, the contribution calculation units 150, 150A, 150B, 150C, 150D, the causal analysis units 160, 160A, 160B, 160C, 160D, the causal analysis result output units 170, 170A, 170B, 170C, 170D, the aggregation condition acquisition units 180C, 180D, the model fitness evaluation unit 190D, and a control unit (not shown).

[0423] The memory 1080 may also be a computer-readable storage medium storing programs for functioning as factor analysis devices 100, 100A, 100B, 100C, and 100D.

[0424] The memory 1080 includes, for example, non-volatile or volatile semiconductor memories such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), and EEPROM (Electrically Erasable Programmable Read Only Memory), as well as disks, floppy disks, optical disks, compressed optical disks, mini disks, DVDs, etc.

[0425] Furthermore, the functions of the cause analysis devices 100, 100A, 100B, 100C, and 100D can be partially implemented by dedicated hardware or by software or firmware.

[0426] The functions of the data acquisition units 110, 110A, 110B, 110C, 110D, the variable setting units 120, 120A, 120B, 120C, 120D, the model learning units 130, 130A, 130B, 130C, 130D, the causal data acquisition units 140, 140A, 140B, 140C, 140D, the contribution calculation units 150, 150A, 150B, 150C, 150D, the causal analysis units 160, 160A, 160B, 160C, 160D, the causal analysis result output units 170, 170A, 170B, 170C, 170D, the aggregation condition acquisition units 180C, 180D, the model fitness evaluation unit 190D, and the control unit (not shown) can be implemented by separate processing circuits or by processing circuits together.

[0427] Alternatively, it can be data acquisition units 110, 110A, 110B, 110C, 110D; variable setting units 120, 120A, 120B, 120C, 120D; model learning units 130, 130A, 130B, 130C, 130D; causal relationship data acquisition units 140, 140A, 140B, 140C, 140D; and contribution calculation units 150, 150A, 150B, 150C, 150... D. The functions of the cause analysis units 160, 160A, 160B, 160C, 160D, the cause analysis result output units 170, 170A, 170B, 170C, 170D, the aggregation condition acquisition units 180C, 180D, the model fitness evaluation unit 190D, and the control unit (not shown) are partly implemented by the processor 10001 and the memory 10002, and the remaining functions are implemented by the processing circuit 20001.

[0428] Thus, the processing circuit can implement the above functions through hardware, software, firmware, or a combination thereof.

[0429] Furthermore, this disclosure allows for free combination of various embodiments, arbitrary modification of the constituent elements of each embodiment, or omission of any constituent elements of each embodiment.

[0430] Industrial availability

[0431] This disclosure improves the accuracy of factor analysis by calculating the contribution of each factor by considering the dependency or causal relationship between multiple factors. Therefore, it is suitable for factor analysis apparatus, factor analysis method, factor analysis system, and factor analysis program, for example, for factor analysis of production equipment.

[0432] Explanation of reference numerals in the attached figures

[0433] 10...Cause analysis system; 100, 100A, 100B, 100C, 100D...Cause analysis device; 110, 110A, (110B, 110C, 110D)...Data acquisition unit; 120, 120A, 120B, 120C, 120D...Variable setting unit; 130, 130A, 130B, 130C, 130D...Model learning unit; 140, 140A, 140B, (140C, 140D)...Causal relationship data acquisition unit; 150, 150A, 150B, 150C, (150D)...Contribution calculation unit; 160, 160A, 160B... 160C, (160D)... Cause Analysis Unit; 170, 170A, 170B, 170C, 170D... Cause Analysis Result Output Unit; 180C, (180D)... Aggregation Condition Acquisition Unit; 190D... Model Suitability Evaluation Unit; 200... Sensor; 300... Display Device; 1010... Auxiliary Storage Device; 1020... Information Input Interface (Information Input IF); 1030... Information Input Device; 1040... Display Interface (Display IF); 1050... Display; 1060... Processing Circuit; 1070... Processor; 1080... Memory.

Claims

1. A factor analysis device, characterized in that, have: The data acquisition department acquires various types of surveillance data related to the monitored objects. The variable setting unit generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis. The model learning unit generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables. The contribution calculation unit calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the monitoring data for learning, and the monitoring data for factor analysis. The factor analysis unit calculates the degree to which each of the explanatory variables is a factor, i.e., the factor degree, based on the contribution degree. as well as The cause analysis results output unit outputs cause analysis results based on the cause analysis dataset, the contribution degree, and the cause degree.

2. The cause analysis device according to claim 1, characterized in that, The data acquisition unit acquires multiple time-series sensor data collected by multiple sensors installed on the monitored device, and production management information related to the production of the monitored device, as the monitoring data.

3. The cause analysis apparatus according to claim 1 or 2, characterized in that, The contribution calculation unit uses XAI (Xarl AI) to interpret artificial intelligence to calculate the contribution.

4. The factor analysis apparatus according to any one of claims 1 to 3, characterized in that, The factor analysis output unit, based on the factor analysis results, accepts an evaluation related to the reasonableness of the explanatory variables. The variable setting unit sets the explanatory variable based on the evaluation results related to the reasonableness of the explanatory variable, in a manner that does not use the monitoring data of the explanatory variable that has been evaluated as having been removed.

5. The factor analysis apparatus according to any one of claims 1 to 4, characterized in that, The factor analysis unit obtains aggregation conditions and outputs the conditional contribution degree, conditional factor degree, and conditional factor analysis dataset corresponding to the monitoring data represented by the aggregation conditions in the factor analysis dataset in a manner that satisfies the aggregation conditions. The factor analysis result output unit uses the conditional contribution degree, the conditional factor degree, and the conditional factor analysis dataset output by the factor analysis unit to output the factor analysis results.

6. The cause analysis apparatus according to any one of claims 1 to 5, characterized in that, It includes a model fitness evaluation unit that uses the factor analysis dataset and the learning model to calculate the fitness of the learning model. The factor analysis result output unit also uses the fitness calculated by the model fitness evaluation unit and outputs the factor analysis results.

7. A factor analysis method, which is a factor analysis method performed by a factor analysis device, characterized in that, The following steps are required: In the data acquisition step, the data acquisition unit of the cause analysis device acquires various monitoring data related to the monitored object; In the variable setting step, the variable setting unit of the cause analysis device generates, based on the monitoring data, monitoring data for learning, monitoring data for cause analysis, a learning dataset consisting of explanatory variables and target variables, and a dataset for cause analysis. In the model learning step, the model learning unit of the factor analysis device generates a learning model based on the learning dataset, which learns by taking explanatory variables as inputs and outputting target variables. In the contribution calculation step, the contribution calculation unit of the factor analysis device calculates the degree of influence of the explanatory variable on the target variable, i.e., the contribution, based on causal relationship data representing the causal relationship of multiple monitoring data, the learning model, the learning monitoring data, and the factor analysis monitoring data. In the cause analysis step, the cause analysis unit of the cause analysis device calculates the degree to which each explanatory variable is a cause, i.e., the cause degree, based on the contribution degree. as well as In the cause analysis result output step, the cause analysis result output unit of the cause analysis device outputs the cause analysis result based on the cause analysis dataset, the contribution degree, and the cause degree.

Images