Abnormality omen detection device, abnormality omen detection method, and program

Abstract

The detection method comprises the following steps: acquiring measurement values of a plurality of parameters representing the state of the production system; calculating a first statistic representing the similarity of the measured value with respect to the standard values of the plurality of parameters; selecting, from among the plurality of parameters, a plurality of effective parameters effective for improving the estimation accuracy of a T-method estimation model in which the plurality of parameters are explanatory variables and the first statistic is a target variable; calculating a second statistic representing the similarity of the measured value with respect to the standard values of the plurality of valid parameters; and detecting an abnormality indication of the production system on the basis of a comparison result between the second statistical amount and the management range.

Description

Technical Field

[0001] The present disclosure relates to an abnormality sign detection device, an abnormality sign detection method, and a program. Background Art

[0002] In order to improve production efficiency, it is desirable to detect signs of abnormality in the production system in advance. Japanese Patent Publication No. 2019-177783 (Patent Document 1) discloses a technology for detecting the deterioration of the ride comfort of a railway vehicle based on the increase in the Mahalanobis distance between characteristic data obtained from an acceleration sensor installed in a railway vehicle and pre-acquired unit data.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2019-177783

[0006] Non-patent literature

[0007] Non-patent document 1: Tamura Kishiomi, "The fifth MT system capable of direction determination - TS method, T method", Standardization and Quality Management, 2009, Vol. 62, No. 2

[0008] Non-patent document 2: Genichi Taguchi, "Target Function and Basic Function (6) - Comprehensive Prediction Based on T Method -", Quality Engineering, 2005, Vol. 13, No. 3, p. 5-10

[0009] Non-patent document 3: "Concepts and application of quality engineering in the development and design stage - System evaluation and improvement without trial production / testing -", [online], [retrieved on March 22, 2023], Internet <https: / / foundry.jp / bukai / wp-content / uploads / 2012 / 07 / e4806f10b0797ec0932d9317dd92a533.pdf> Summary of the invention

[0010] Problems to be solved by the invention

[0011] It is considered to apply the technology described in Japanese Patent Publication No. 2019-177783 (Patent Document 1) to the production system. However, the data obtained from the production system may contain many noise factors. Therefore, high accuracy cannot be obtained in the detection of abnormal signs using this data. Therefore, in the past, it was necessary to perform individual data screening based on the experience and knowledge of skilled people.

[0012] The present disclosure has been made in view of the above-mentioned problems, and an object of the present disclosure is to provide a predicted abnormality sign detection device, a predicted abnormality sign detection method, and a program that can detect a predicted abnormality sign of a monitoring object with high accuracy without relying on individual experience.

[0013] Means for solving problems

[0014] According to an example of the present disclosure, an abnormal sign detection device has an acquisition unit, a first calculation unit, a selection unit, a second calculation unit, and a detection unit. The acquisition unit acquires, for each unit interval, the measured values ​​of multiple parameters representing the state of the monitored object in the unit interval. The first calculation unit calculates a first statistic, which represents the similarity of the measured values ​​of the multiple parameters to the standard values ​​of the multiple parameters when the monitored object is operating normally. The selection unit selects from the multiple parameters a plurality of effective parameters that are effective for improving the estimation accuracy of the T-method estimation model that uses the multiple parameters as explanatory variables and the first statistic as the target variable. The second calculation unit calculates a second statistic, which represents the similarity of the measured values ​​of the multiple effective parameters to the standard values ​​of the multiple effective parameters. The detection unit detects abnormal signs of the monitored object based on the comparison result of the second statistic and the management scope.

[0015] According to the above disclosure, the influence of multiple effective parameters on the first statistic is greater than the influence of the remaining parameters on the first statistic. This means that the proportion of noise factors contained in the effective parameters is smaller than the proportion of noise factors contained in the remaining parameters. The second statistic calculated using such multiple effective parameters accurately represents the similarity of the state of the monitored object relative to normal operation. In addition, multiple effective parameters are selected based on the estimation accuracy of the T-method estimation model that uses multiple parameters as explanatory variables and the first statistic as the target variable. Therefore, multiple effective parameters can be selected without relying on individual experience, and these effective parameters accurately represent the similarity of the state of the monitored object relative to normal operation. According to the above disclosure, the abnormal sign detection device is able to detect abnormal signs of the monitored object with high accuracy without relying on individual experience.

[0016] In the above disclosure, the acquisition unit further acquires event information indicating the type of event that occurred in the monitored object in each unit interval. The second calculation unit provides a graph indicating the change over time of the second statistic calculated for each unit interval, wherein the graph indicates the type of event indicated by the event information for each unit interval.

[0017] According to the above disclosure, the user can understand the type of event that occurs at the timing when the second statistic changes by checking the graph, and can easily identify the cause of the abnormality.

[0018] In the above disclosure, the abnormality sign detection device further includes an analysis unit that performs a principal component analysis using a plurality of effective parameters and a second statistic for each unit section and provides an analysis result.

[0019] According to the above disclosure, the user can easily identify the cause of the abnormality by confirming the analysis result.

[0020] In the above disclosure, the analysis result includes a scatter plot of the principal component scores of the first principal component and the second principal component for each unit interval. In the scatter plot, vectors representing principal component loadings of a plurality of effective parameters and the second statistic are superimposed and displayed.

[0021] According to the above disclosure, the user can determine that an effective parameter having a relatively large component parallel to the vector direction of the second statistic is related to an abnormality from among a plurality of effective parameters. The user can easily identify the cause of the abnormality based on the effective parameter related to the abnormality.

[0022] In the above disclosure, the acquisition unit further acquires event information indicating the type of event occurring in the monitoring target in the unit section. The display format of the points of the scatter diagram differs according to the type of event indicated by the event information of the corresponding unit section.

[0023] According to the above disclosure, the user can grasp the types of events concentrated in the vector direction of the effective parameter related to the abnormality, thereby easily identifying the cause of the abnormality.

[0024] In the above disclosure, the selection unit provides the coefficient of determination of the T-method estimation model using the plurality of effective parameters as explanatory variables and the second statistic as the target variable.

[0025] According to the above disclosure, the user can determine whether a plurality of effective parameters are effective based on the determination coefficient.

[0026] In the above disclosure, the first statistic and the second statistic are standard SN ratios. Therefore, the first statistic and the second statistic properly represent the similarity of the measured value to the standard value.

[0027] In the above disclosure, the unit section has a predetermined time length. Alternatively, the monitored object repeatedly performs the target action. The unit section is the period during which the target action is performed.

[0028] In the above disclosure, the monitoring object includes a manufacturing device, and a plurality of parameters are extracted from a vibration waveform measured by an acceleration sensor provided in the manufacturing device.

[0029] The vibration waveform measured by the acceleration sensor is prone to contain a large number of noise factors. However, according to the above disclosure, from the multiple parameters extracted from the vibration waveform, multiple effective parameters with a relatively small proportion of noise factors are selected. As a result, it is possible to use multiple effective parameters to detect abnormal signs with high accuracy.

[0030] According to another example of the present disclosure, a method for detecting signs of abnormality has steps 1 to 5. The first step is the following step: according to each unit interval, the measured values ​​of multiple parameters representing the state of the monitored object in the unit interval are obtained. The second step is the following step: calculating the first statistic, the first statistic representing the similarity of the measured values ​​of the multiple parameters to the standard values ​​of the multiple parameters when the monitored object is operating normally. The third step is the following step: selecting multiple effective parameters from the multiple parameters that are effective for improving the estimation accuracy of the T-method estimation model that uses the multiple parameters as explanatory variables and the first statistic as the target variable. The fourth step is the following step: calculating the second statistic, the second statistic representing the similarity of the measured values ​​of the multiple effective parameters to the standard values ​​of the multiple effective parameters. The fifth step is the following step: detecting signs of abnormality of the monitored object based on the comparison result of the second statistic and the management scope.

[0031] According to another example of the present disclosure, the program causes a computer to execute the above-mentioned abnormality sign detection method. According to these disclosures, the abnormality sign detection method or program can also detect the abnormality sign of the monitoring object with high accuracy and stability without relying on individual experience.

[0032] Effects of the Invention

[0033] According to the present disclosure, an abnormality sign detection device, an abnormality sign detection method, or a program can detect an abnormality sign of a production system with high accuracy and stability without relying on individual experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram showing a method of detecting an abnormality sign according to an embodiment.

[0035] Figure 2 It is a schematic diagram showing the overall configuration of a system including the abnormality sign detection device according to the embodiment.

[0036] Figure 3 It is a schematic diagram showing an example of the hardware configuration of the abnormality sign detection device according to the embodiment.

[0037] Figure 4 It is a diagram showing an example of the functional configuration of the abnormality sign detection device according to the embodiment.

[0038] Figure 5 This is a diagram showing an example of a database included in the storage unit.

[0039] Figure 6 This is a diagram showing an example of the contribution of each parameter to the first standard SN ratio in the T-method estimation model using all the plurality of parameters.

[0040] Figure 7 This is a diagram showing an example of a plurality of valid parameters selected by the selection unit.

[0041] Figure 8 This is a flowchart showing an example of the processing flow of the abnormality sign detection device.

[0042] Fig. 9 This is a diagram showing an example of a scatter plot of estimated values ​​and actual values ​​of a target variable when a T method estimation model is used in which all of a plurality of parameters are used as explanatory variables and the first standard SN ratio is used as a target variable.

[0043] Fig.10 This is a diagram showing an example of a scatter plot of estimated values ​​and actual values ​​of a target variable when a T-method estimation model is used using a plurality of effective parameters as explanatory variables and the first standard SN ratio as a target variable.

[0044] Fig.11 This is a diagram showing an example of the change in the first standard SN ratio over time.

[0045] Fig.12 This is a diagram showing an example of the change in the second standard SN ratio over time.

[0046] Fig.13 : is a figure which shows an example of the analysis result provided in step S22.

[0047] Fig.14 It is a diagram showing an example of application of the abnormality sign detection device.

[0048] Fig.15 This is a diagram showing another example of a monitoring object.

[0049] Fig.16 This is a diagram showing changes in physical quantities measured by the sensor group during the execution of the injection molding operation.

[0050] Fig.17 This is a graph showing standard values ​​and measured values ​​of a plurality of parameters of the injection molding machine status.

[0051] Fig.18 This is a diagram showing the transition of the second statistic calculated for each injection molding operation.

[0052] Fig.19 This is a diagram showing the results of principal component analysis using a plurality of effective parameters and the second standard SN ratio for each injection molding operation.

[0053] Fig. 20 This is a diagram showing another example of a monitoring object.

[0054] Fig.21 This is a diagram showing the transition of the second statistic calculated for each joining operation. DETAILED DESCRIPTION

[0055] Embodiments of the present invention will be described in detail with reference to the accompanying drawings. In addition, the same or corresponding parts in the drawings are denoted by the same reference numerals and their description will not be repeated.

[0056] §1 Application Examples

[0057] Reference Figure 1 , an example of a scenario in which the present invention is applied is described. Figure 1 1 is a schematic diagram showing a method for detecting abnormal signs according to an embodiment of the present invention. Figure 1 As shown, the detection method has steps (1) to (5). Figure 1 1 shows a detection method in which the production system 2 is an example of a monitoring target of an abnormality sign.

[0058] Step (1) is a step of obtaining, for each unit period, measured values ​​of a plurality of parameters representing the state of the production system 2 during the unit period. The plurality of parameters represent, for example, the state of the manufacturing device included in the production system 2, the quality of the intermediate product or the final product manufactured by the production system 2, and the like.

[0059] Step (2) is a step of calculating a first statistic representing the similarity of the measured values ​​of a plurality of parameters relative to the standard values ​​of the plurality of parameters when the production system 2 is operating normally. Below, an example of calculating the first standard SN ratio as the first statistic is described. However, the first statistic is not limited to the first standard SN ratio, as long as it is a parameter representing the similarity between the standard values ​​and the actual values ​​of a plurality of variables. The first standard SN ratio represents the similarity relative to a first reference space defined by the standard values ​​of the plurality of parameters. The standard values ​​of the plurality of parameters when the production system 2 is operating normally are obtained in advance from the production system 2.

[0060] Step (3) is a step of selecting, from among a plurality of parameters, a plurality of effective parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first standard SN ratio as a target variable. Thus, a plurality of effective parameters that have a relatively large influence on the first standard SN ratio as a target variable can be selected.

[0061] Step (4) is a step of calculating a second statistic representing the similarity between the measured values ​​of a plurality of valid parameters and the standard values ​​of a plurality of valid parameters. Below, an example of calculating the second standard SN ratio as the second statistic is described. However, the second statistic is not limited to the second standard SN ratio, as long as it is a parameter representing the similarity between the standard values ​​and the actual values ​​of a plurality of variables. The second standard SN ratio represents the similarity relative to the second reference space defined by the standard values ​​of a plurality of valid parameters. The second standard SN ratio is calculated each time a measurement value is obtained.

[0062] Step (5) is a step of detecting an abnormality sign of the production system 2 based on the comparison result between the second standard SN ratio and the management range. For example, the abnormality sign of the production system 2 is detected based on the deviation of the moving average of the second standard SN ratio from the management range. The abnormality sign of the production system 2 includes abnormality signs of the manufacturing equipment included in the production system 2, abnormality signs of the quality of the intermediate products or final products manufactured by the production system 2, etc.

[0063] The influence of the plurality of effective parameters on the first standard SN ratio is greater than the influence of the remaining parameters on the first standard SN ratio. This means that the proportion of noise factors included in the effective parameters is smaller than the proportion of noise factors included in the remaining parameters. The second standard SN ratio calculated using such a plurality of effective parameters accurately represents the similarity to the state of the production system 2 during normal operation.

[0064] In addition, a plurality of effective parameters are selected based on the estimation accuracy of the T-method estimation model using a plurality of parameters as explanatory variables and the first standard SN ratio as the target variable. Therefore, a plurality of effective parameters are selected without relying on individual experience, and these effective parameters highly accurately represent the similarity of the state of the production system 2 relative to the normal operation.

[0065] As described above, according to the abnormality sign detection method of the present embodiment, it is possible to detect the abnormality sign of the production system 2 with high accuracy without relying on individual experience.

[0066] §2Concrete examples

[0067] <System Structure>

[0068] Figure 2 Schematic diagram showing the overall structure of a system including an abnormality sign detection device according to an embodiment. Figure 2 As shown, the system 1 includes a production system 2 and an abnormality sign detection device 100 .

[0069] The production system 2 includes a component supply device 2A, a stamping machine 2B, and an inspection device 2C. The component supply device 2A supplies metal components to the stamping machine 2B. The stamping machine 2B is a manufacturing device that manufactures products of a predetermined shape by performing shearing, drawing, and crimping on metal components. The inspection device 2C inspects the appearance and performance of the product.

[0070] The press working machine 2B is provided with an acceleration sensor 5. The acceleration sensor 5 measures a vibration waveform generated in the press working machine 2B and outputs a measurement signal indicating a measurement result.

[0071] The punching machine 2B outputs an event signal in response to an event occurring in any of the shearing process, the drawing process, and the crimping process. The event is, for example, an unfavorable situation that causes the temporary stop of the punching machine 2B. The event signal indicates which of the shearing process, the drawing process, and the crimping process the event occurred.

[0072] The abnormal sign detection device 100 performs the above-mentioned abnormal sign detection method. Specifically, the abnormal sign detection device 100 detects abnormal signs of the punching machine 2B using the measured values ​​of multiple parameters obtained from the vibration waveform measured by the acceleration sensor 5. The abnormal sign detection device 100 is composed of, for example, a programmable logic controller (PLC) or a general-purpose computer.

[0073] Furthermore, when detecting a sign of abnormality of the press machine 2B, the abnormality-sign detection device 100 provides auxiliary information for assisting in identifying the cause of the abnormality. For example, the abnormality-sign detection device 100 generates the auxiliary information based on an event signal output from the press machine 2B.

[0074] In addition, the abnormality-sign detection device 100 may directly receive the event signal from the press working machine 2B, or may receive the event signal from the press working machine 2B via a device (for example, a PLC) not shown in the figure.

[0075] <Hardware Configuration of Abnormality Sign Detection Device>

[0076] Figure 3 Schematic diagram showing an example of the hardware structure of the abnormality sign detection device according to the embodiment. Figure 3As shown, the abnormal sign detection device 100 typically has a structure that complies with a general computer architecture. Specifically, the abnormal sign detection device 100 includes a processor 101 such as a CPU (Central Processing Unit) or an MPU (Micro-Processing Unit), a memory 102, a storage 103, a display controller 104, an input interface 105, an amplifier 106, and a communication interface 107. These various parts are connected via a bus in a manner that allows data communication with each other.

[0077] The processor 101 implements various processes of the present embodiment by developing various programs stored in the storage 103 in the memory 102 and executing the programs.

[0078] The memory 102 is typically a volatile storage device such as a DRAM (Dynamic Random Access Memory), and stores programs read from the storage 103 , various data generated by executing the programs, and the like.

[0079] The memory 103 is typically a nonvolatile magnetic storage device such as a hard disk drive. The memory 103 stores a detection program 131 and a data storage program 132 executed by the processor 101. Various programs installed in the memory 103 are distributed in a state of being stored in a memory card or the like.

[0080] The memory 103 may also store various data generated by executing the detection program 131 and the data accumulation program 132 .

[0081] The display controller 104 is connected to the display device 70 , and outputs a signal for displaying various information to the display device 70 in accordance with an internal command from the processor 101 .

[0082] The input interface 105 mediates data transmission between the processor 101 and an input device 75 such as a keyboard, a mouse, a touch panel, or a dedicated console. In other words, the input interface 105 receives instructions given by a user operating the input device 75 .

[0083] The amplifier 106 amplifies the measurement signal from the acceleration sensor 5. The communication interface 107 mediates data transmission between the processor 101 and an external device (e.g., the press machine 2B). The communication interface 107 typically includes Ethernet (registered trademark), USB (Universal Serial Bus), etc. In addition, various programs stored in the memory 103 can also be downloaded from a distribution server, etc. via the communication interface 107.

[0084] When using a computer having a structure conforming to a general computer architecture as described above, in addition to the application for providing the functions of the present embodiment, an OS (Operating System) for providing the basic functions of the computer may be installed. In this case, the program of the present embodiment may also call necessary modules among the program modules provided as part of the OS in a prescribed order and timing to execute processing. That is, the program of the present embodiment itself may sometimes not include the above-mentioned modules, but cooperate with the OS to execute processing.

[0085] Alternatively, part or all of the functions provided by the execution of the detection program 131 and the data storage program 132 may be implemented as a dedicated hardware circuit.

[0086] <Functional Structure of Abnormality Sign Detection Device>

[0087] Figure 4 FIG. 1 is a diagram showing an example of the functional structure of the abnormality sign detection device according to the embodiment. Figure 4 As shown, the abnormality sign detection device 100 includes a storage unit 10, a database construction unit 11, an acquisition unit 12, a first calculation unit 13, a selection unit 14, a second calculation unit 15, a detection unit 16, and an analysis unit 17. The storage unit 10 is implemented by a memory 102 and a storage 103. The database construction unit 11 is implemented by an amplifier 106, a communication interface 107, and a processor 101 that executes a data accumulation program 132. The acquisition unit 12, the first calculation unit 13, the selection unit 14, the second calculation unit 15, the detection unit 16, and the analysis unit 17 are implemented by the processor 101 executing the detection program 131.

[0088] The database construction unit 11 constructs a database 10 a included in the storage unit 10 based on the measurement signal from the acceleration sensor 5 and the event signal from the press working machine 2B.

[0089] Figure 5 FIG. 1 is a diagram showing an example of a database included in a storage unit. Figure 5 As shown, the database 10a is in the form of a table, and contains one or more records 10a1.

[0090] Record 10a1 is added for each unit period having a specified time length (for example, 1 minute, 5 minutes, 20 minutes, etc.). Each record 10a1 includes fields 10a2 to 10a5. The start time of the corresponding unit period is recorded in field 10a2. The end time of the corresponding unit period is recorded in field 10a3. The storage location of waveform data 10b representing the vibration waveform measured by the acceleration sensor 5 in the corresponding unit period is recorded in field 10a4. The type of event represented by the event signal received in the corresponding unit period (any one of "shearing event", "drawing event" and "crimping event") is recorded in field 10a5.

[0091] The database construction unit 11 adds a new record 10a1 to the database 10a every time a unit period ends. The database construction unit 11 describes the start time and end time of the corresponding unit period in the fields 10a2 and 10a3 of the new record 10a1, respectively.

[0092] The database construction unit 11 generates waveform data 10b representing the vibration waveform measured by the acceleration sensor 5 during each unit period, and stores the generated waveform data 10b in the storage unit 10. The database construction unit 11 describes the storage location of the generated waveform data 10b in the field 10a4 of the new record 10a1 added according to the unit period. For example, the storage location of the waveform data 10b is represented by a folder name and a file name.

[0093] When the unit period ends, the database construction unit 11 determines whether an event signal is received during the unit period. If the database construction unit 11 does not receive an event signal, it writes "null" in the field 10a5 of the new record 10a1 added according to the unit period. If the database construction unit 11 receives an event signal, it writes the type of event represented by the event signal in the field 10a5 of the new record 10a1 added according to the unit period.

[0094] Figure 4 The acquisition unit 12 shown acquires the measured values ​​of multiple parameters representing the state of the press machine 2B in each unit period. Specifically, the acquisition unit 12 determines the record 10a1 corresponding to the unit period from the database 10a for each unit period, and reads the waveform data 10b stored in the storage location described in the field 10a4 of the determined record 10a1. The acquisition unit 12 calculates the measured values ​​of multiple parameters representing the characteristics of the vibration waveform represented by the read waveform data 10b. For example, the multiple parameters represent information related to each frequency component of the vibration waveform.

[0095] Furthermore, the acquisition unit 12 acquires, for each unit period, event information indicating the type of event that occurred in the press machine 2B of the production system 2 during the unit period. Specifically, the acquisition unit 12 identifies, for each unit period, a record 10a1 corresponding to the unit period from the database 10a, and acquires event information indicating the type of event described in the field 10a5 of the identified record 10a1.

[0096] The first calculation unit 13 calculates the first standard SN ratio of the measured values ​​of the plurality of parameters relative to the standard values ​​of the plurality of parameters when the press machine 2B is operating normally, for each unit period. The first standard SN ratio is based on "The concept and flexible application of quality engineering in the development and design stage - System evaluation and improvement without trial production / testing -", [online], [retrieved on January 4, 2022], Internet<https: / / foundry.jp / bukai / wp-content / uploads / 2012 / 07 / e4806f10b0797ec0932d9317dd92a533.pdf> "(Non-patent document 3) is calculated. The standard values ​​of multiple parameters when the press working machine 2B is operating normally are represented by the normal data set 10c stored in the storage unit 10. The normal data set 10c is created in advance.

[0097] The selection unit 14 selects a plurality of effective parameters effective for improving the estimation accuracy of the T-method estimation model using the plurality of parameters as explanatory variables and the first standard SN ratio as a target variable from among the plurality of parameters.

[0098] The T method is one of the multivariate analysis methods provided by statistician Genichi Taguchi. It is a method of estimating the value of a target variable based on the values ​​of multiple explanatory variables (see "Tamura Kishishin, "The fifth MT system capable of direction determination - TS method, T method", Standardization and Quality Management, 2009, Vol. 62, No. 2" (Non-patent document 1)).

[0099] The selection unit 14 generates a T-method estimation model by performing T-method data processing on a data set representing the measured values ​​of multiple parameters and the value of the first standard SN ratio for each unit period. In the T-method data processing, a univariate regression is performed for each explanatory variable, and a weighted average is performed for each explanatory variable, thereby generating a T-method estimation model for calculating an estimated value of the target variable. The estimation model obtained by the T-method data processing is represented by the following formula (1).

[0100]

[0101] In formula (1), Xij is the standardized value of the jth explanatory variable corresponding to the i-th unit period. Standardization is the process of subtracting the average value of the measured value of the parameter represented by the data set. ηj represents the SN ratio of the j-th explanatory variable. The SN ratio ηj represents the linearity between the value of the j-th explanatory variable and the value of the target variable, indicating the estimation accuracy of the value of the target variable. βj represents the proportionality constant of the univariate regression. Mi is the estimated value of the target variable in the i-th unit period after standardized processing.

[0102] For example, the selection unit 14 uses the orthogonal table of the two-level system to select multiple effective parameters that are effective for improving the estimation accuracy of the target variable from multiple parameters. The selection unit 14 uses, for example, the method described in "Genichi Taguchi, "Target Function and Basic Function (6) - Comprehensive Prediction Based on T Method -", Quality Engineering, 2005, Vol. 13, No. 3, p. 5-10" (Non-Patent Document 2) to select multiple effective parameters. By using the selection of multiple effective parameters using the orthogonal table of the two-level system (also called "item selection"), it is possible to select explanatory variables that are effective for improving the estimation accuracy of the target variable for each explanatory variable (also called "item"), and exclude explanatory variables that have an adverse effect on the estimation accuracy of the target variable (reduce the estimation accuracy). Furthermore, by using the orthogonal table of the two-level system, it is possible to consider the impact when there is interaction between items. Specifically, according to the method described in "Genichi Taguchi, "Target Function and Basic Function (6) - Comprehensive Prediction Based on T Method -", Quality Engineering, 2005, Vol. 13, No. 3, p. 5-10" (Non-Patent Literature 2), an orthogonal table is created to which one of two levels of use or non-use is assigned for each of multiple explanatory variables (items), and a comprehensive SN ratio is calculated for each row of the orthogonal table. The selection unit 14 determines the estimated effect of the target variable of the item and whether there is any adverse effect on the estimation of the target variable based on the comprehensive SN ratio when each item is used and when each item is not used. The selection unit 14 selects multiple valid parameters from multiple parameters based on the judgment result.

[0103] Alternatively, the selection unit 14 may generate a plurality of patterns representing combinations of two or more parameters among the plurality of parameters instead of using the orthogonal table of the two-level system. Then, the selection unit 14 may generate a T-method estimation model using the actual values ​​of the two or more parameters represented by each pattern as explanatory variables and the first standard SN ratio as the target variable, and select a plurality of valid parameters based on the size of the comprehensive SN ratio of the generated T-method estimation model. For example, the selection unit 14 selects two or more parameters included in the pattern with the largest comprehensive SN ratio as valid parameters.

[0104] Figure 6 This is a diagram showing an example of the contribution of each parameter to the first standard SN ratio in the T-method estimation model using all the plurality of parameters. Figure 6 The graph 30 shown shows the contribution of 35 parameters. Parameters with relatively small contribution have little influence on the first standard SN ratio. This is because a large number of noise factors are included in the parameters with relatively small contribution. Therefore, it is preferred that the parameters with relatively small contribution are not selected as effective parameters.

[0105] Figure 7 This is a diagram showing an example of a plurality of valid parameters selected by the selection unit. Figure 7 The graph 32 shown shows the contribution of 15 effective parameters selected using the orthogonal array of the two-level system. Figure 7 The vertical axis represents the contribution of each parameter in the T-method estimation model with the 15 effective parameters as explanatory variables and the first standard SN ratio as the target variable. The influence of the 15 effective parameters on the first standard SN ratio is greater than that of the remaining parameters. In other words, the proportion of noise factors in the 15 effective parameters is smaller than that in the remaining parameters.

[0106] The selection unit 14 provides information related to the selection result. Specifically, the selection unit 14 causes the display device 70 to display a screen showing information related to the selection result.

[0107] The second calculation unit 15 calculates the second standard SN ratio of the measured values ​​of the multiple valid parameters relative to the standard values ​​of the multiple valid parameters for each unit period. The second standard SN ratio is calculated based on "Genichi Taguchi, "Target Function and Basic Function (6) - Comprehensive Prediction Based on T Method -", Quality Engineering, 2005, Vol. 13, No. 3, p. 5-10". As described above, the proportion of noise factors in the multiple valid parameters selected by the selection unit 14 is smaller than the proportion of noise factors in the remaining parameters. Therefore, the second standard SN ratio calculated using multiple valid parameters represents the similarity of the state of the stamping machine 2B during normal operation with higher accuracy than the first standard SN ratio.

[0108] The second calculation unit 15 provides a graph showing the temporal change of the second standard SN ratio for each unit period. Specifically, the second calculation unit 15 displays the graph on the display device 70 .

[0109] The detection unit 16 detects the abnormal sign of the stamping machine 2B based on the comparison result of the second standard SN ratio and the management range. The management range is predetermined. The detection unit 16 can also detect the abnormal sign of the stamping machine 2B according to the deviation of the second standard SN ratio from the management range in each unit period. Among them, it is preferred that the detection unit 16 considers the deviation of the second standard SN ratio and detects the abnormal sign of the stamping machine 2B according to the deviation of the moving average of the second standard SN ratio from the management range.

[0110] The analysis unit 17 performs a principal component analysis using a plurality of effective parameters and a second standard SN ratio for each unit period based on the abnormality sign detected by the detection unit 16, and provides the analysis result. Specifically, the analysis unit 17 causes the display device 70 to display a screen showing the analysis result. The analysis result is used to identify the cause of the abnormality.

[0111] <Processing Flow of Abnormality Sign Detection Device>

[0112] Figure 8 FIG. 1 is a flowchart showing an example of the processing flow of the abnormality sign detection device. Figure 8 As shown, the process of the abnormality sign detection device 100 includes a preliminary preparation stage, a monitoring stage, and an analysis stage.

[0113] The preliminary preparation phase is executed during the trial operation of the production system 2 or for a certain period after the start of operation of the production system 2. Alternatively, the preliminary preparation phase may be executed for a certain period after the maintenance of the production system 2.

[0114] In the preliminary preparation stage, first, the processor 101 of the abnormality-sign detection device 100 reads the normal data set 10c indicating the standard values ​​of a plurality of parameters when the press working machine 2B to be monitored is operating normally (step S1).

[0115] Next, the processor 101 obtains the measured values ​​of the multiple parameters for each unit period (step S2). Specifically, the processor 101 reads the waveform data 10b of the storage location described in the field 10a3 of each record 10a1 of the database 10a. The processor 101 calculates the measured values ​​of the multiple parameters based on the read waveform data 10b.

[0116] Next, the processor 101 calculates a first standard SN ratio of the measured values ​​of the plurality of parameters to the standard values ​​of the plurality of parameters for each unit period (step S3 ).

[0117] Next, the processor 101 selects a plurality of effective parameters effective for improving the estimation accuracy of the T-method estimation model using the plurality of parameters as explanatory variables and the first standard SN ratio as the target variable from among the plurality of parameters (step S4 ).

[0118] Next, the processor 101 provides information related to the selection result of step S4 (step S5). The user confirms the validity of the selection result based on the provided information. The details of the information provided in step S5 will be described later.

[0119] When the user determines that the selection result is appropriate, the user inputs an instruction to shift to the monitoring phase into the input device 75 .

[0120] If the user determines that the selection result is inappropriate, the user re-evaluates the normal data set 10 c or waits until the amount of data in the database 10 a reaches the desired amount. Then, the user inputs a retry instruction prepared in advance into the input device 75 .

[0121] Next, processor 101 checks an input instruction to input device 75 (step S6). If a retry instruction is input to input device 75, the process returns to step S1. If a transition instruction to the monitoring phase is input to input device 75, the process transitions to the monitoring phase.

[0122] In the monitoring phase, first, the processor 101 obtains the measured values ​​of multiple parameters in the new unit period (step S11). Specifically, the processor 101 reads the waveform data 10b of the storage location described in the field 10a4 of the record 10a1 newly added to the database 10a, and calculates the measured values ​​of multiple parameters based on the read waveform data 10b.

[0123] Furthermore, when an event occurs in the new unit period, the processor 101 obtains event information indicating the type of the event (step S12). Specifically, the processor 101 obtains event information indicating the type of the event described in the field 10a5 of the record 10a1 newly added to the database 10a.

[0124] Next, the processor 101 calculates the second standard SN ratio of the measured values ​​of the plurality of effective parameters obtained in step S11 relative to the standard values ​​of the plurality of effective parameters (step S13). Furthermore, the processor 101 provides a graph showing the change of the second standard SN ratio over time (step S14). Specifically, the processor 101 causes the display device 70 to display the graph. Thus, the user can confirm the change of the second standard SN ratio.

[0125] Next, the processor 101 determines whether there is a sign of abnormality of the stamping machine 2B based on the comparison result between the second standard SN ratio calculated in step S12 and the management range (step S15). Specifically, the processor 101 determines that there is no sign of abnormality of the stamping machine 2B when the value of the second standard SN ratio (or the moving average of the second standard SN ratio) is within the management range. The processor 101 detects a sign of abnormality of the stamping machine 2B when the value of the second standard SN ratio (or the moving average of the second standard SN ratio) is outside the management range.

[0126] When there is no sign of abnormality in the press working machine 2B (NO in step S15), the process returns to step S11.

[0127] When there is a sign of abnormality in the press working machine 2B (YES in step S15), the process shifts to the analysis phase.

[0128] In the analysis phase, first, the processor 101 performs a principal component analysis using a plurality of effective parameters and a second standard SN ratio for each unit period (step S21). The object of the principal component analysis is a data set representing the measured values ​​of a plurality of effective parameters and a value of the second standard SN ratio for each of a plurality of unit periods included in the target period including the timing at which the abnormality is detected.

[0129] Next, the processor 101 provides the analysis result (step S22). Specifically, the processor 101 causes the display device 70 to display a screen showing the analysis result. The details of the analysis result provided in step S22 will be described later.

[0130] The user determines the cause of the abnormality based on the analysis result. The user performs maintenance of the press working machine 2B to eliminate the determined cause.

[0131] When the maintenance of the punching machine 2B is performed, the values ​​of the multiple parameters of the punching machine 2B when it is in normal operation may change. Therefore, after the maintenance, the user confirms the difference between the reference space defined by the values ​​of the multiple parameters when the punching machine 2B is in normal operation and the reference space defined by the standard values ​​of the multiple parameters shown in the current normal data set 10c. If the difference is less than the threshold value, the user determines that it is not necessary to change the normal data set 10c, and inputs the continuous indication of monitoring to the input device 75.

[0132] On the other hand, if the difference is greater than the threshold, the user determines that the normal data set 10c needs to be changed. Then, the user creates a new normal data set 10c representing the standard values ​​of multiple parameters when the press machine 2B is operating normally, and replaces the normal data set 10c stored in the abnormal sign detection device 100 with the new normal data set 10c. Then, the user inputs a transition instruction to the preliminary preparation stage into the input device 75.

[0133] Next, processor 101 checks an input instruction to input device 75 (step S23). If an instruction to continue monitoring is input to input device 75, the process returns to step S11. If an instruction to shift to the preliminary preparation phase is input to input device 75, the process returns to step S1.

[0134] <Example of information related to selection results>

[0135] The information related to the selection result provided in step S5 is explained. As the information related to the selection result, the processor 101 provides, for example, a scatter plot of the estimated value and the measured value of the target variable when the T method estimation model using multiple valid parameters as explanatory variables and the first standard SN ratio as the target variable is used. Furthermore, in order to confirm the validity of the valid parameters, the processor 101 may also provide a scatter plot of the estimated value and the measured value of the target variable when the T method estimation model using all the multiple parameters as explanatory variables and the first standard SN ratio as the target variable is used as a comparison object.

[0136] Fig. 9 This is a diagram showing an example of a scatter plot of estimated values ​​and actual values ​​of a target variable when a T method estimation model is used in which all of a plurality of parameters are used as explanatory variables and the first standard SN ratio is used as a target variable. Fig.10 This is a diagram showing an example of a scatter plot of estimated values ​​and actual values ​​of a target variable when a T-method estimation model is used using a plurality of effective parameters as explanatory variables and the first standard SN ratio as a target variable.

[0137] like Fig. 9 and Fig.10 As shown, the deviation of the scatter diagram 36 when multiple effective parameters are used as explanatory variables is smaller than that of the scatter diagram 34 when all multiple parameters are used as explanatory variables. This is because the proportion of noise factors included in the multiple effective parameters is smaller than the proportion of noise factors included in the remaining parameters. Therefore, the multiple effective parameters accurately represent the similarity with respect to the normal state of the punching machine 2B.

[0138] Furthermore, if Fig. 9 and Fig.10 As shown, the processor 101 preferably provides a coefficient of determination R indicating the degree to which the explanatory variable explains the target variable. 2 The user can determine the coefficient of determination R 2 To evaluate the estimation accuracy of the T method estimation model.

[0139] The user can confirm the scatter plot and the coefficient of determination R 2 , and can judge whether the selection result is appropriate.

[0140] Furthermore, the processor 101 may also provide Figure 6 , Figure 7 The graphs 30 and 32 shown are information related to the selection result. Thus, the user can understand the contribution of each parameter.

[0141] <Advantages of using the second standard SN ratio>

[0142] Fig.11 This is a diagram showing an example of the temporal change of the first standard SN ratio. Fig.12This is a diagram showing an example of the change over time of the second standard SN ratio. Fig.11 The following table shows the temporal changes in the first standard SN ratio for each unit period from May 31 to June 3, 2021. Fig.12 The following table shows the temporal changes in the second standard SN ratio for each unit period from May 31 to June 3, 2021.

[0143] exist Fig.11 and Fig.12 In the display format of the points representing the standard SN ratio corresponding to each unit period corresponds to the type of event that occurred in the unit period. That is, the points corresponding to the unit period in which no event occurred are represented by circles, the points corresponding to the unit period in which an event occurred during the shearing process are represented by triangles, the points corresponding to the unit period in which an event occurred during the drawing process are represented by quadrilaterals, and the points corresponding to the unit period in which an event occurred during the crimping process are represented by plus signs. Fig.11 and Fig.12 As shown, multiple crimping processing events occurred on May 31, 2021, and crimping processing events occurred again on June 1, 2021. At the end of the same day, the poor condition of crimping processing also spread to drawing processing and a drawing processing event occurred. Then, on June 3, 2021, the stamping machine 2B broke down.

[0144] like Fig.11 As shown by the frame line 20, it can be seen that the first standard SN ratio changed significantly on June 1, before the failure of the press machine 2B on June 3. Therefore, even if the first standard SN ratio is used, the abnormal sign of the press machine 2B can be detected to some extent before the failure occurs.

[0145] In contrast, Fig.12 As shown in the frame line 22 of , it can be seen that the second standard SN ratio changed significantly on May 31, which is earlier than June 1. That is, the change of the second standard SN ratio on May 31 is a sign of the failure on June 3. In this way, by using multiple effective parameters with a relatively small proportion of noise factors, it is possible to detect the abnormal signs of the stamping machine 2B earlier.

[0146] The processor 101 operating as the second calculation unit 15 makes the Fig.12 The graph 38 shown, that is, the graph 38 showing the temporal change of the second standard SN ratio calculated for each unit period, is displayed on the display device 70. Thus, the user can see the change of the second standard SN ratio.

[0147] like Fig.12As shown, the processor 101 shows the type of event represented by the event information obtained in step S12 in the graph 38 representing the change of the second standard SN ratio over time. That is, as described above, the processor 101 makes the display form of the points representing the standard SN ratio corresponding to each unit period different according to the type of event represented by the event information corresponding to the unit period in the graph 38. As a result, the user can identify the type of event that occurred in the time period when the second standard SN ratio changes. As a result, the user can easily identify the cause of the abnormality of the stamping machine 2B.

[0148] <Example of analysis results>

[0149] When an abnormality of the punching machine 2B is detected, the processor 101 selects a target period as a target of the principal component analysis. Specifically, the processor 101 selects a target period including the timing of the abnormality detection. Specifically, the processor 101 selects a certain period before the second standard SN ratio starts to change and a period from the start of the second standard SN ratio change to the current time as the target period.

[0150] The processor 101 performs a principal component analysis on the measured values ​​of the plurality of effective parameters and the value of the second standard SN ratio for each unit period included in the target period, and provides the analysis result.

[0151] Fig.13 FIG. 4 is a diagram showing an example of the analysis result provided in step S22. Fig.13 5 shows a biplot 50 obtained by principal component analysis. The biplot 50 includes a scatter plot of the principal component scores of the first principal component and the second principal component for each unit period included in the target period, and a vector representing the principal component load of each of the plurality of effective parameters and the second standard SN ratio displayed superimposed on the scatter plot. Fig.13 In the example shown, a vector v0 corresponding to the second standard SN ratio and vectors v1 to v6 corresponding to some of the effective parameters among the plurality of effective parameters are shown, and display of vectors corresponding to the remaining effective parameters is omitted.

[0152] The processor 101 may also make the display format of each point of the scatter diagram of the principal component score different depending on the type of event that occurred in the unit period corresponding to the point. Fig.13 In the example shown, points corresponding to unit periods in which no event occurs are represented by circles, points corresponding to unit periods in which an event occurs during the shearing process are represented by triangles, points corresponding to unit periods in which an event occurs during the drawing process are represented by quadrilaterals, and points corresponding to unit periods in which an event occurs during the crimping process are represented by plus signs.

[0153] exist Fig.13In , the frame line 40 represents the confidence interval of the points corresponding to the unit period in which no event occurs. The frame line 41 represents the confidence interval of the points corresponding to the unit period in which an event occurs during the shearing process. The frame line 42 represents the confidence interval of the points corresponding to the unit period in which an event occurs during the drawing process. The frame line 43 represents the confidence interval of the points corresponding to the unit period in which an event occurs during the crimping process. Fig.13 In the biplot 50 shown, it can be seen that the confidence intervals for each event type are separated.

[0154] The user can understand the variables that affect the reduction of the second standard SN ratio by checking the biplot 50. That is, the user can identify the variables corresponding to the vectors with large components parallel to the vector v0 representing the principal component load of the second standard SN ratio among the vectors representing the principal component load of each of the plurality of effective parameters as the variables that affect the reduction of the standard SN ratio. Fig.13 In the biplot 50 shown, the user can identify variables fp10, fp19, fp13, ip02 corresponding to vectors v1, v2, v4, v5 having relatively large components parallel to the direction of vector v0 as variables that affect the reduction of the second standard SN ratio. On the other hand, the user can identify variables ip01, fp03 corresponding to vectors v3, v6 having relatively small components parallel to the direction of vector v0 as variables that do not have a large effect on the reduction of the second standard SN ratio.

[0155] Furthermore, the user can estimate that the type of event corresponding to the confidence interval existing in the vector direction of the variable that affects the decrease in the second standard SN ratio is related to the cause of the abnormality. Fig.13 In the example shown, there is a confidence interval indicated by a frame line 43 in the direction of the vector v2. Therefore, the user can list components related to the crimping process as candidate causes of the abnormality.

[0156] The user can determine the cause of the abnormality based on the characteristics of the vibration waveform represented by the variable that affects the reduction of the second standard SN ratio. For example, when the variable representing the high-frequency component characteristics of the vibration waveform affects the reduction of the second standard SN ratio, the user can list a metal part as a candidate cause of the abnormality.

[0157] <Application example>

[0158] Fig.14 2 is a diagram showing an example of the use of an abnormality sign detection device. Fig.14 In the example, the horizontal axis represents time. Fig.14In the example shown, at timing T1, the moving average of the second standard SN ratio falls outside the management range, and a sign of abnormality of the press machine 2B is detected. Based on the detection of the sign of abnormality, the user identifies the cause of the abnormality based on the analysis result and performs maintenance on the press machine 2B.

[0159] At the time T2 when the maintenance is completed, the user determines the similarity of the reference space defined by the values ​​of multiple parameters when the punching machine 2B is operating normally before and after the maintenance. The user determines whether it is necessary to update the normal data set 10c based on the similarity of the reference space before and after the maintenance. Then, based on the judgment result, the normal data set 10c is updated. By updating the normal data set 10c, Figure 8 In the preliminary preparation phase shown, a number of valid parameters are also updated.

[0160] Furthermore, it is also possible to periodically determine whether the normal data set 10c needs to be updated. For example, the user obtains the values ​​of multiple parameters when the stamping machine 2B is operating normally at a timing T3. Then, the user determines the similarity between the reference space defined by the values ​​of the multiple parameters obtained and the reference space defined by the standard values ​​of the multiple parameters shown in the normal data set 10c. The user determines whether the normal data set 10c needs to be updated based on the determined similarity, and implements the update of the normal data set 10c as needed. By updating the normal data set 10c, the normal data set 10c is implemented. Figure 8 In the preliminary preparation phase shown, a number of valid parameters are also updated.

[0161] <Modification 1>

[0162] In the above description, it is assumed that the acquisition unit 12 acquires the measured values ​​of multiple parameters representing the state of the monitored object in each unit period (unit interval) having a specified time length (for example, 1 minute, 5 minutes, 20 minutes, etc.). However, the acquisition unit 12 may also acquire the measured values ​​of multiple parameters representing the state of the monitored object in each unit interval that does not have a specified time length. For example, the unit interval may also be the period during which the object action is performed in the stamping machine 2B that is the monitored object. The object action is, for example, a series of actions of shearing, drawing, and crimping on a metal part. Alternatively, the object action may also be a processing action selected from shearing, drawing, and crimping on a metal part.

[0163] <Modification 2>

[0164] In the abnormality sign detection device 100 of the above embodiment, the monitoring target of the abnormality sign is the production system 2 (particularly, the press working machine 2B). However, the monitoring target is not limited to the production system 2.

[0165] (Another example of a surveillance target)

[0166] Fig.15 FIG. 1 is a diagram showing another example of a monitoring object. Fig.15 , an injection molding machine 6 is shown as a monitoring target. The injection molding machine 6 includes a molding die 61, a cylinder 62, a screw 63, a hopper 64, a screw drive device 65, and a sensor group 66.

[0167] The cylinder 62 is a cylindrical member having an internal space to which resin is supplied. The hopper 64 supplies resin to the internal space of the cylinder 62. A screw 63 is inserted into the internal space of the cylinder 62. A screw drive device 65 is connected to the base end of the screw 63. The screw 63 is rotated by the control of the screw drive device 65 and can move along the length direction of the cylinder 62. The sensor group 66 measures various physical quantities indicating the status of the injection molding machine 6.

[0168] The injection molding machine 6 repeatedly performs the injection molding operation. The injection molding operation is performed by moving the screw 63 from Fig.15 The upper state is moved to the lower state, and the resin accommodated in the internal space of the cylinder 62 is injected into the molding die 61 .

[0169] Fig.16 : is a graph showing the change of the physical quantity measured by the sensor group during the period of performing the injection molding operation. Hereinafter, the period of performing the injection molding operation is referred to as the "injection molding period". Fig.16 In the graph shown, the horizontal axis represents the elapsed time from the start of the injection molding operation, and the vertical axis represents the values ​​of various physical quantities measured by the sensor group 66. Fig.16 As shown, the sensor group 66 measures the screw position, injection pressure and injection speed as physical quantities representing the status of the injection molding machine 6. The screw position is, for example, determined by the distance D between the front end of the cylinder 62 and the front end of the screw 63 (see Fig.15 ) indicates. Fig.16 In FIG. 6 , line 67 indicates the change of the screw position during injection molding, line 68 indicates the change of the injection pressure, and line 69 indicates the change of the injection speed.

[0170] When the monitoring object is the injection molding machine 6, the normal data set 10c represents the standard values ​​of multiple parameters representing the status of the injection molding machine 6 during the injection molding period when the injection molding machine 6 is operating normally. The normal data set 10c is created in advance and stored in the storage unit 10. The multiple parameters include, for example, the screw position, injection pressure, and injection speed at the time t1, t2, ..., tn respectively from the start of the injection molding period. In this case, the multiple parameters include 3×n parameters.

[0171] When the monitoring object is the injection molding machine 6, the acquisition unit 12 acquires the measured values ​​of multiple parameters indicating the status of the injection molding machine 6 during each injection molding period. The acquisition unit 12 may acquire the measured values ​​of the multiple parameters from the sensor group 66, or may acquire the measured values ​​of the multiple parameters from a control device (e.g., PLC) that controls the injection molding machine 6 (not shown).

[0172] Fig.17 It is a graph showing the standard values ​​and measured values ​​of a plurality of parameters of the condition of the injection molding machine. Line 67a shows the standard value of the screw position (n variables) at the time t1, t2, ..., tn from the start of the injection molding period. Line 68a shows the standard value of the injection pressure (n variables) at the time t1, t2, ..., tn. Line 69a shows the standard value of the injection speed (n variables) at the time t1, t2, ..., tn. Line 67b shows the measured value of the screw position (n variables) at the time t1, t2, ..., tn from the start of the injection molding period. Line 68b shows the measured value of the injection pressure (n variables) at the time t1, t2, ..., tn. Line 69b shows the measured value of the injection speed (n variables) at the time t1, t2, ..., tn.

[0173] The first calculation unit 13, the selection unit 14, the second calculation unit 15, the detection unit 16, and the analysis unit 17 use Fig.17 The standard values ​​and measured values ​​of the 3×n parameters shown are subjected to the processing described in the above embodiment.

[0174] Specifically, if Fig.17 As shown, the first calculation unit 13 calculates the first standard SN ratio as the first statistic representing the similarity of the measured values ​​of 3×n parameters to the standard values ​​of the 3×n parameters. The selection unit 14 selects a plurality of effective parameters from the 3×n parameters that are effective for improving the estimation accuracy of the T-method estimation model using the 3×n parameters as explanatory variables and the first standard SN ratio as the target variable. In addition, the second calculation unit 15 calculates the second standard SN ratio of the measured values ​​of the plurality of effective parameters to the standard values ​​of the plurality of effective parameters. The detection unit 16 detects the abnormal sign of the injection molding machine 6 based on the comparison result of the second standard SN ratio with the management range. The analysis unit 17 performs a principal component analysis using the plurality of effective parameters and the second standard SN ratio during each injection molding period based on the abnormal sign detected by the detection unit 16, and provides the analysis result.

[0175] Fig.18 : is a graph showing the transition of the second statistic calculated for each injection molding operation. Fig.18, a graph is shown with the number of the injection molding action as the horizontal axis and the second statistic (second standard SN ratio) calculated by the second calculation unit 15 as the vertical axis. The 7-digit number corresponding to each point in the graph represents the number of the product manufactured by the injection molding action. The second standard SN ratio of the points in the area surrounded by the frame lines 50a and 50b is outside the management range. The product manufactured by the injection molding action corresponding to the point in the area surrounded by the frame line 50a is a defective product. On the other hand, the product manufactured by the injection molding action corresponding to the point in the area surrounded by the frame line 50b is a qualified product. The injection molding action corresponding to the point in the area surrounded by the frame line 50b is performed before the injection molding action corresponding to the point in the area surrounded by the frame line 50a. Thus, the user can suppress the generation of defective products in advance based on the abnormal signs detected by the detection unit 16.

[0176] Fig.19 : is a graph showing the results of principal component analysis using a plurality of effective parameters for each injection molding operation and the second standard SN ratio. Fig.19 The biplot of the two principal components PC1 and PC2 selected by the analysis unit 17 is shown in FIG. Fig.19 In the biplot shown, points corresponding to the principal component scores of the two principal components obtained from the measured values ​​of the plurality of variable parameters of each injection molding action are plotted. Points corresponding to the injection molding actions for which the second standard SN ratio falls within the management range are concentrated within the frame line 51a. In contrast, points corresponding to the injection molding actions for which the second standard SN ratio falls outside the management range are concentrated within the frame line 51b. Thus, the user can confirm the Fig.19 The biplot shown makes it easy to identify injection molding actions that may cause product anomalies.

[0177] Then, the user confirms Fig.19 The biplot shown in Figure 2 allows us to examine the mechanism of abnormality. Fig.19 In the biplot shown, vectors v10, v11, v12, v13, and v14 corresponding to the variables "Speed_time_315 (injection speed at time t315)", "Pressure_time_315 (injection pressure at time t315)", "Speed_time_316 (injection speed at time t316)", "Speed_time_317 (injection speed at time t317)", and "Cylinder_time_318 (screw position at time t318)" respectively face directions different from those corresponding to other variables. Therefore, the user can understand that the injection speed and screw position near time t315 to t318 contribute greatly to the abnormality of the injection molding machine 6.

[0178] (Another example of a surveillance target)

[0179] Fig. 20 FIG. 1 is a diagram showing another example of a monitoring object. Fig. 20 , the mounting machine 8 is shown as a monitoring object. The mounting machine 8 includes a chuck 84, a driving device 85, and a sensor group 86.

[0180] The chuck 84 holds the chip 82 to which the bumps 83 are attached. The drive device 85 controls the generation of ultrasonic waves, the lifting and lowering movement of the chuck 84, and the holding force of the chuck 84 on the chip 82. The sensor group 86 measures the values ​​of various physical quantities indicating the status of the mounting machine 8.

[0181] like Fig. 20 As shown, the driving device 85 lowers the chuck 84 holding the chip 82. When the bump 83 contacts the substrate 81, the mounting machine 8 performs a joining action to join the bump 83 to the substrate 81. Specifically, the driving device 85 generates ultrasonic waves to promote the joining of the bump 83 to the substrate 81, and further lowers the chuck 84 while applying a load. When the bump 83 is deformed to a specified amount, the driving device 85 stops the generation of ultrasonic waves and reduces the holding force of the chuck 84 on the chip 82. Thus, the joining action ends. Then, the driving device 85 raises the chuck 84. Each time the chuck 84 holds a new chip 82, the mounting machine 8 performs the above-mentioned joining action.

[0182] The sensor group 86 measures the current and voltage applied to the driving device 85 and the deformation amount of the bump 83 as physical quantities indicating the status of the mounting machine 8. The deformation amount of the bump 83 is indicated by the distance between the chip 82 and the substrate 81.

[0183] When the monitoring object is the mounting machine 8, the normal data set 10c shows the standard values ​​of multiple parameters indicating the status of the mounting machine 8 during the bonding operation when the mounting machine 8 is operating normally. The normal data set 10c is created in advance and stored in the storage unit 10. The multiple parameters include, for example, the voltage, current, and deformation amount at each of the elapsed times t1, t2, ..., tm from the start of the bonding operation. In this case, the multiple parameters include 3×m parameters.

[0184] When the monitoring object is the mounting machine 8, the acquisition unit 12 acquires the measured values ​​of multiple parameters indicating the status of the mounting machine 8 during each bonding operation. The acquisition unit 12 may acquire the measured values ​​of multiple parameters from the sensor group 86, or may acquire the measured values ​​of multiple parameters from a control device (e.g., PLC) that controls the mounting machine 8 (not shown).

[0185] The first calculation unit 13 , the selection unit 14 , the second calculation unit 15 , the detection unit 16 , and the analysis unit 17 execute the processing described in the above embodiment using the standard values ​​and the measured values ​​of the 3×m parameters.

[0186] Specifically, the first calculation unit 13 calculates the first standard SN ratio as the first statistic representing the similarity of the measured values ​​of the 3×m parameters relative to the standard values ​​of the 3×m parameters. The selection unit 14 selects a plurality of effective parameters from the 3×m parameters that are effective for improving the estimation accuracy of the T-method estimation model that uses the 3×m parameters as explanatory variables and the first standard SN ratio as the target variable. Furthermore, the second calculation unit 15 calculates the second standard SN ratio of the measured values ​​of the plurality of effective parameters relative to the standard values ​​of the plurality of effective parameters. The detection unit 16 detects signs of abnormality of the installation machine 8 based on the comparison result of the second standard SN ratio with the management range. The analysis unit 17 performs a principal component analysis using the plurality of effective parameters and the second standard SN ratio during each joining action, based on the detection of signs of abnormality by the detection unit 16, and provides the analysis results.

[0187] Fig.21 : is a graph showing the transition of the second statistic calculated for each joining action. Fig.17 , a graph is shown with the number of the joining action (installation number) as the horizontal axis and the second statistic (second standard SN ratio) calculated by the second calculation unit 15 as the vertical axis. Some of the products manufactured by the joining action with the second standard SN ratio outside the management range are defective products. Thus, the user can suppress the generation of defective products in advance based on the abnormal signs detected by the detection unit 16.

[0188] (Another example of a unit section)

[0189] In the above example, the unit interval is set to a period of a specified time length, or a period during which the action of the monitored object (injection molding action or joining action) is performed. However, the unit interval is not limited to this. For example, the unit interval may also be each of a plurality of intervals included in the range of the physical quantity of the manufacturing device.

[0190] For example, when monitoring the Fig.15 In the case of the injection molding machine 6 shown, the screw position can be varied from Fig.15 The position of the screw 63 shown in the upper part (indicated by the distance D1) is Fig.15 The range of the position of the screw 63 shown in the lower part (indicated by the distance D2). In this case, a plurality of intervals included in the range of the distance D1 to the distance D2 are respectively set as unit intervals. For example, the intervals of the distance D from D1 to D3 and the intervals of the distance D from D3 to D2 are set as unit intervals.

[0191] Alternatively, the interval in which the physical quantity obtained from the tensile tester or compression tester installed at the production site falls within a specific range may be set as the unit interval. For example, the interval in which the strain observed in the stress-strain diagram falls within a specified range may be set as the unit interval. Alternatively, in the relationship between the cooling rate and hardness of the metal heated at the production site, the interval in which the cooling rate falls within a specified range may be set as the unit interval.

[0192] §3 Notes

[0193] As described above, this embodiment includes the following disclosure.

[0194] (Structure 1)

[0195] The abnormal sign detection device (100) comprises:

[0196] an acquisition unit (101, 12) which acquires, for each unit period, measured values ​​of a plurality of parameters representing the state of the production system (2) in the unit period;

[0197] a first calculation unit (101, 13) for calculating a first statistic indicating the similarity of the measured values ​​of the plurality of parameters to standard values ​​of the plurality of parameters when the production system is operating normally;

[0198] A selection unit (101, 14) that selects a plurality of effective parameters from the plurality of parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first statistic as a target variable;

[0199] a second calculation unit (101, 15) for calculating a second statistic indicating a similarity between the measured values ​​of the plurality of effective parameters and the standard values ​​of the plurality of effective parameters; and

[0200] A detection unit (101, 16) detects a sign of abnormality in the production system (2) based on a comparison result between the second statistic and a management range.

[0201] (Structure 2)

[0202] The abnormal sign detection device (100) comprises:

[0203] an acquisition unit (101, 12) that acquires, for each unit section, measurement values ​​of a plurality of parameters representing the state of a monitoring object (2, 2B, 6, 7) in the unit section;

[0204] a first calculation unit (101, 13) for calculating a first statistic indicating the similarity of the measured values ​​of the plurality of parameters to standard values ​​of the plurality of parameters when the monitored object (2, 2B, 6, 7) is operating normally;

[0205] A selection unit (101, 14) that selects a plurality of effective parameters from the plurality of parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first statistic as a target variable;

[0206] a second calculation unit (101, 15) for calculating a second statistic indicating a similarity between the measured values ​​of the plurality of effective parameters and the standard values ​​of the plurality of effective parameters; and

[0207] A detection unit (101, 16) detects a sign of abnormality of the monitoring object (2, 2B, 6, 7) based on a comparison result between the second statistic and a management range.

[0208] (Structure 3)

[0209] In the abnormality sign detection device (100) described in configuration 1 or 2,

[0210] The acquisition unit (101, 12) further acquires event information for each of the unit sections (or the unit periods), the event information indicating the type of event occurring in the monitoring object (2, 2B, 6, 7) (or the production system (2)) in the unit section (or the unit period),

[0211] The second calculation unit (101, 15) provides a graph (38) showing the change over time of the second statistic calculated for each of the unit intervals (or the unit periods),

[0212] The graph (38) shows the type of event represented by the event information in each of the unit sections (or the unit periods).

[0213] (Structure 4)

[0214] In the abnormality sign detection device (100) described in any one of the configurations 1 to 3,

[0215] An analyzing unit (101, 17) is further provided, and the analyzing unit (101, 17) performs a principal component analysis using the plurality of effective parameters and the second statistic for each of the unit sections (or the unit periods), and provides an analysis result.

[0216] (Structure 5)

[0217] In the abnormality sign detection device (100) described in Configuration 4,

[0218] The analysis result includes a scatter plot of the principal component scores of the first principal component and the second principal component for each of the unit intervals (or the unit periods),

[0219] In the scatter plot, vectors representing the principal component loading amounts of the plurality of effective parameters and the second statistic are superimposed and displayed.

[0220] (Structure 6)

[0221] In the abnormality sign detection device (100) described in configuration 5,

[0222] The acquisition unit (101, 12) further acquires event information, wherein the event information indicates a type of event occurring in the monitoring object (2, 2B, 6, 7) (or the production system (2)) in the unit section (or the unit period),

[0223] The display format of the points of the scatter diagram differs according to the type of the event represented by the event information of the corresponding unit section (or the unit period).

[0224] (Structure 7)

[0225] In the abnormality sign detection device (100) described in any one of Configurations 1 to 6,

[0226] The selection unit (101, 14) provides a coefficient of determination of a T-method estimation model using the plurality of effective parameters as explanatory variables and the second statistic as a target variable.

[0227] (Structure 8)

[0228] In the abnormality sign detection device (100) described in any one of Configurations 1 to 7,

[0229] The first statistic and the second statistic are standard SN ratios.

[0230] (Structure 9)

[0231] In the abnormality prediction device (100) according to any one of configurations 1 to 8,

[0232] The unit section (or the unit period) has a predetermined time length.

[0233] (Structure 10)

[0234] In the abnormality prediction device according to any one of structures 1 to 8,

[0235] The monitoring object (2, 2B, 6, 7) (or the production system (2)) repeatedly performs the target action,

[0236] The unit section (or the unit period) is a period during which the target action is performed.

[0237] (Structure 11)

[0238] In the abnormality sign detection device (100) described in any one of Configurations 1 to 8,

[0239] The monitoring object (or the production system (2)) includes a manufacturing device (2B),

[0240] The plurality of parameters are extracted from a vibration waveform measured by an acceleration sensor (5) provided in the manufacturing device (2B).

[0241] (Structure 12)

[0242] The method for detecting abnormal signs has the following steps:

[0243] Step (S2, S11), obtaining, for each unit section (or unit period), measurement values ​​of a plurality of parameters representing the state of the monitoring object (2, 2B, 6, 7) (or the production system (2)) in the unit section (or unit period);

[0244] Step (S3), calculating a first statistic, wherein the first statistic represents the similarity of the measured values ​​of the plurality of parameters to standard values ​​of the plurality of parameters when the monitored object (2, 2B, 6, 7) (or the production system (2)) is operating normally;

[0245] Step (S4), selecting a plurality of effective parameters from the plurality of parameters that are effective in improving the estimation accuracy of the T-method estimation model using the plurality of parameters as explanatory variables and the first statistic as the target variable;

[0246] Step (S13), calculating a second statistic, wherein the second statistic represents the similarity between the measured values ​​of the plurality of effective parameters and the standard values ​​of the plurality of effective parameters; and

[0247] In step (S15), based on the comparison result between the second statistic and the management range, a sign of abnormality of the monitoring object (2, 2B, 6, 7) (or the production system (2)) is detected.

[0248] (Structure 13)

[0249] A program that causes a computer to execute a method for detecting an abnormality sign, wherein:

[0250] The detection method comprises the following steps:

[0251] Step (S2, S11), obtaining, for each unit section (or unit period), measurement values ​​of a plurality of parameters representing the state of the monitored object (2, 2B, 6, 7) (or the production system (2)) in the unit section (or the unit period);

[0252] Step (S3), calculating a first statistic, wherein the first statistic represents the similarity of the measured values ​​of the plurality of parameters to standard values ​​of the plurality of parameters when the monitored object (2, 2B, 6, 7) (or the production system (2)) is operating normally;

[0253] Step (S4), selecting a plurality of effective parameters from the plurality of parameters that are effective in improving the estimation accuracy of the T-method estimation model using the plurality of parameters as explanatory variables and the first statistic as the target variable;

[0254] Step (S13), calculating a second statistic, wherein the second statistic represents the similarity between the measured values ​​of the plurality of effective parameters and the standard values ​​of the plurality of effective parameters; and

[0255] In step (S15), based on the comparison result between the second statistic and the management range, a sign of abnormality of the monitoring object (2, 2B, 6, 7) (or the production system (2)) is detected.

[0256] Although the embodiments of the present invention have been described, the embodiments disclosed this time should be considered to be illustrative in all aspects and not restrictive. The scope of the present invention is indicated by the claims, and it is intended to include all modifications within the meaning and scope equivalent to the claims.

[0257] Description of symbols

[0258] 1: system; 2: production system; 2A: component supply device; 2B: stamping machine; 2C: inspection device; 5: acceleration sensor; 6: injection molding machine; 7: mounting machine; 10: storage unit; 10a: database; 10a1: record; 10a2-10a4: field; 10b: waveform data; 10c: normal data set; 11: database construction unit; 12: acquisition unit; 13: first calculation unit; 14: selection unit; 15: second calculation unit; 16: detection unit; 17: analysis unit; 30, 32, 38: graph; 34, 36: scatter plot ; 50: biplot; 61: molding die; 62: cylinder; 63: screw; 64: hopper; 65: screw drive device; 66, 86: sensor group; 70: display device; 75: input device; 81: substrate; 82: chip; 83: bump; 84: chuck; 85: drive device; 100: abnormal sign detection device; 101: processor; 102: memory; 103: storage; 104: display controller; 105: input interface; 106: amplifier; 107: communication interface; 131: detection program; 132: data accumulation program.

Claims

1. An abnormal sign detection device, wherein: The abnormal sign detection device has: an acquisition unit that acquires, for each unit section, measured values ​​of a plurality of parameters indicating a state of a monitoring object in the unit section; a first calculation unit that calculates a first statistic that indicates a degree of similarity between the measured values ​​of the plurality of parameters and standard values ​​of the plurality of parameters when the monitored object operates normally; a selection unit that selects, from the plurality of parameters, a plurality of effective parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first statistic as a target variable; a second calculation unit that calculates a second statistic that indicates a degree of similarity between the measured values ​​of the plurality of effective parameters and the standard values ​​of the plurality of effective parameters; as well as The detection unit detects a sign of abnormality of the monitoring object based on a comparison result between the second statistic and a management range.

2. The abnormality sign detection device according to claim 1, wherein: The acquisition unit further acquires, for each of the unit sections, event information indicating a type of an event occurring in the monitoring target in the unit section. The second calculation unit provides a graph showing a temporal change of the second statistic calculated for each of the unit intervals, The graph shows the type of event represented by the event information for each of the unit sections.

3. The abnormality sign detection device according to claim 1, wherein: The predicted abnormality sign detection device further includes an analysis unit that performs a principal component analysis using the plurality of effective parameters and the second statistic for each unit section and provides an analysis result.

4. The abnormality sign detection device according to claim 3, wherein: The analysis result includes a scatter plot of the principal component scores of the first principal component and the second principal component for each of the unit intervals. In the scatter plot, vectors representing the principal component loading amounts of the plurality of effective parameters and the second statistic are superimposed and displayed.

5. The abnormality sign detection device according to claim 4, wherein: The acquisition unit further acquires event information indicating a type of event occurring in the monitoring target in the unit section. The display format of the points of the scatter diagram differs according to the type of the event represented by the event information of the corresponding unit section.

6. The abnormality sign detection device according to claim 1, wherein: The selection unit provides a coefficient of determination of a T-method estimation model using the plurality of effective parameters as explanatory variables and the second statistic as a target variable.

7. The abnormality sign detection device according to claim 1, wherein: The first statistic and the second statistic are standard SN ratios.

8. The abnormality sign detection device according to any one of claims 1 to 7, wherein: The unit interval has a prescribed time length.

9. The abnormality sign detection device according to any one of claims 1 to 7, wherein: The monitored object repeatedly performs the object action, The unit section is a period during which the target action is performed.

10. The abnormality sign detection device according to any one of claims 1 to 7, wherein: The monitoring object includes a manufacturing device, The plurality of parameters are extracted from a vibration waveform measured by an acceleration sensor provided in the manufacturing device.

11. A method for detecting abnormal signs, comprising the following steps: acquiring, for each unit section, measurement values ​​of a plurality of parameters indicating a state of a monitoring object in the unit section; calculating a first statistic indicating a degree of similarity between the measured values ​​of the plurality of parameters and standard values ​​of the plurality of parameters when the monitored object operates normally; Selecting, from the plurality of parameters, a plurality of effective parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first statistic as a target variable; calculating a second statistic, the second statistic representing a similarity between the measured values ​​of the plurality of valid parameters and the standard values ​​of the plurality of valid parameters; as well as Based on the comparison result between the second statistic and the management range, a sign of abnormality of the monitored object is detected.

12. A program causing a computer to execute a method for detecting an abnormality sign, wherein: The detection method comprises the following steps: acquiring, for each unit section, measurement values ​​of a plurality of parameters indicating a state of a monitoring object in the unit section; calculating a first statistic indicating a degree of similarity between the measured values ​​of the plurality of parameters and standard values ​​of the plurality of parameters when the monitored object operates normally; Selecting, from the plurality of parameters, a plurality of effective parameters that are effective for improving the estimation accuracy of a T-method estimation model that uses the plurality of parameters as explanatory variables and the first statistic as a target variable; calculating a second statistic, the second statistic representing a similarity between the measured values ​​of the plurality of valid parameters and the standard values ​​of the plurality of valid parameters; as well as Based on the comparison result between the second statistic and the management range, a sign of abnormality of the monitored object is detected.

Images