JP6221414B2 - Determination apparatus, determination program, and determination method - Google Patents

Description

  The present invention relates to a determination device, a determination program, and a determination method, for example, a determination device, a determination program, and a determination method for determining whether a robot performs work.

  Work using robots such as industrial robots is widely performed. It is known that the quality of work performed by a robot is determined based on an average value and a variance value of feature quantities associated with quality of work (for example, Patent Document 1). It is known that the wear of the suction pad is determined based on the transient data of the suction pressure until the suction pressure of the suction pad that sucks the conveyed object is stabilized (for example, Patent Document 2). It is known to detect an abnormality in a device using a regression coefficient (for example, Patent Document 3).

JP 2011-230245 A JP 2012-152843 A JP-A-6-4789

  When determining the quality of work using the average value or variance value of certain data, if the data is in an excessive interval with non-stationarity, the average value or variance value may diverge, and correct / bad determination may not be possible. is there. When an operator or the like determines whether the data is in a data transition section in consideration of experience or intuition, it takes time because the worker or the like determines the transition section. If the quality of work is determined by the same determination method in a steady steady section and an unsteady transient section, the determination may not be appropriate. As described above, if the quality of work is determined for the sections having different properties by the same determination method, the determination may not be appropriate.

  An object of the present determination apparatus, determination program, and determination method is to appropriately perform work quality determination.

An estimated value obtained by estimating data in the data sequence using a stationary model is calculated from a time-series data sequence output by a sensor that detects the status of work performed by the robot, and a residual between the estimated value and the data is calculated. The model selection criterion is calculated. If the criterion is less than or equal to the threshold, the data is determined to belong to the first interval where the data is stationary. If the criterion is greater than the threshold, the data is transient A first determination unit that determines that the data belongs to the second section, and a method that determines whether the work corresponding to the data differs between when the data belongs to the first section and when the data belongs to the second section. A determination device comprising: a second determination unit for determining;

A computer calculates an estimated value obtained by estimating data in the data sequence using a stationary model from a time-series data sequence output by a sensor that detects a status of work performed by the robot , and the estimated value and the data A model selection criterion is calculated using a residual, and if the criterion is less than or equal to a threshold, the data is determined to belong to a first interval where the data is stationary, and if the criterion is greater than the threshold, the data is the data Is determined to belong to the second interval that is transitional, and the quality of the work corresponding to the data is determined using different methods depending on whether the data belongs to the first interval or the second interval. The determination program characterized by this is used.

A determination method executed by a computer , calculating an estimated value obtained by estimating data in the data sequence using a stationary model from a time-series data sequence output by a sensor that detects a status of work performed by a robot , A model selection criterion is calculated using a residual between the estimated value and the data. If the criterion is equal to or less than a threshold value, the data is determined to belong to a first interval in which the data is stationary. If the data is larger, the step of determining that the data belongs to the second interval in which the data is transient, and the operation corresponding to the data when the data belongs to the first interval and when the data belongs to the second interval And a step of determining using a different method.

  According to this determination apparatus, determination program, and determination method, it is possible to appropriately determine whether the work is good or bad.

FIG. 1A is a block diagram of a robot system in which the first embodiment is used, and FIG. 1B is a diagram illustrating an example of a robot in which the first embodiment is used. FIG. 2A is a block diagram of a computer that functions as a determination device according to the first embodiment, and FIG. 2B is a functional block diagram of the determination device. FIG. 3 is a flowchart showing the work performed by the computer. FIG. 4 is a flowchart showing processing performed by the determination apparatus. FIG. 5A to FIG. 5C are diagrams (part 1) illustrating a determination method according to the first embodiment. FIG. 6A and FIG. 6B are diagrams (part 2) illustrating the determination method in the first embodiment. FIG. 7 is a flowchart illustrating an example in which the first determination unit determines a section. FIG. 8A to FIG. 8D are diagrams showing a method for calculating the transient discrimination amount. FIG. 9A is a diagram showing a signal for a point, and FIG. 9B is a diagram showing a transient discrimination amount for the point. FIG. 10A to FIG. 10C are diagrams (part 1) showing the contents of work in the second embodiment. FIG. 11A to FIG. 11C are diagrams (part 2) showing the contents of work in the second embodiment. FIG. 12A is a diagram illustrating sensor signals for points in Example 2, and FIG. 12B is a diagram illustrating transient discrimination amounts for points.

  Embodiments will be described below with reference to the drawings.

  FIG. 1A is a block diagram of a robot system in which the first embodiment is used, and FIG. 1B is a diagram illustrating an example of a robot in which the first embodiment is used. As shown in FIG. 1A, the robot system 100 includes a robot 22, a sensor 24, and a computer 10. The robot 22 is an industrial robot, for example. As shown in FIG. 1B, the robot 22 includes a stage 25 and a manipulator 26, for example. The manipulator 26 performs work using, for example, a tool 28. The tool 28 is tweezers, for example. The stage 25 supports the manipulator 26. The manipulator 26 is provided with a sensor 24. The sensor 24 detects the status of work performed by the robot 22. The sensor 24 is a strain sensor that detects strain of the manipulator 26, for example. As the sensor 24, for example, a 3-axis or 6-axis force sensor may be used. In this case, the sensor 24 detects the force vector and / or torque of the action point. The sensor 24 may be a load sensor, a pressure sensor, an acceleration sensor, or a microphone. Further, the sensor 24 may be a one-dimensional image sensor, or a two-dimensional image sensor such as a CCD (Charge Coupled Device) sensor or a CMOS (Complimentary Metal Oxide Semiconductor) sensor.

  FIG. 2A is a block diagram of a computer that functions as a determination device according to the first embodiment. The computer 10 executes a determination method and a determination program. The computer 10 includes a central processing unit (CPU) 11 that is a processor, a display device 12, an input device 13, and an output device 14. The computer 10 further includes a main storage device 15, a hard disk drive (HDD) 16, a storage medium drive 17, a communication interface 18, and an internal bus 19. The display device 12 includes a display panel such as a liquid crystal panel, for example, and displays processing results and the like. The input device 13 is, for example, a keyboard, a mouse, and a touch panel, and inputs processing data and the like. The output device 14 is a printer, for example, and outputs a processing result or the like. The main storage device 15 is a volatile memory such as a DRAM (Dynamic Random Access Memory), for example, and stores data being processed. The HDD 16 stores, for example, data during or after processing. The storage medium drive 17 is used when a program stored in the storage medium 21 is installed. Alternatively, the processed data is stored in the storage medium 21. The communication interface 18 is connected to other computers and transmits / receives data to / from other computers. The internal bus 19 connects each device in the computer 10.

  A portable storage medium can be used as the storage medium 21 readable by the computer 10 storing the program. As the portable storage medium, for example, a CD-ROM (Compact Disc Read Only Memory) disc, a DVD (Digital Video Disc) disc, a Blu-ray disc, or a USB (Universal Serial Bus) memory can be used. As the storage medium 21, a flash memory or an HDD may be used.

  FIG. 2B is a functional block diagram of the determination apparatus. The determination device 30 includes a first determination unit 32 and a second determination unit 34. The computer 10 functions as the first determination unit 32 and the second determination unit 34 in cooperation with the software.

  FIG. 3 is a flowchart showing the work performed by the computer. In this example, a series of operations performed by the robot 22 is divided into a plurality of operations. With reference to FIG. 3, the computer 10 acquires a time-series teaching data string of a series of operations (step S10). The teaching data string is data that causes the robot 22 to perform work in time series. The teaching data string may be acquired from the outside via the communication interface 18, for example. Further, it may be stored in a storage device such as the HDD 16 in advance. The computer 10 causes the robot 22 to perform an initial operation based on the teaching data string (step S12). The computer 10 determines whether or not the first work is good (step S14). In the case of No, the computer 10 makes the operation defective (step S17). Then, the process ends. In the case of good, the computer 10 determines whether or not the process is finished (step S18). For example, if a series of work has been completed, the computer 10 determines that the process has been completed. If yes, end. In No, the computer 10 moves to the next work (step S20). Return to step S12. Note that the teaching data string may be acquired for each operation. In this case, the process returns to step S10.

  FIG. 4 is a flowchart showing processing performed by the determination apparatus. It is an example of the step which performs pass / fail stability of step S16 of FIG. With reference to FIG. 4, the first determination unit 32 acquires a time-series data string of the output signal of the sensor 24 of the determination work (step S <b> 30). The 1st determination part 32 determines the area in 1 or several data among data strings (step S32). For example, a section where one or a plurality of data is stationary is a first section, and a section where data is transient is a second section. The first determination unit 32 determines whether one or more pieces of data belong to the first section or the second section (step S34). In the case of the first section, the second determination unit 34 determines the first quality of work based on one or more data (step S36). The second determination unit 34 determines whether or not it is good (step S38). In No, the 2nd determination part 34 complete | finishes a work as defect and complete | finishes (step S40). In the case of Yes, it progresses to step S48.

  In step S34, in the case of the second section, the second determination unit 34 determines the second quality of the work based on one or a plurality of data (step S42). The second determination unit 34 determines whether or not it is good (step S44). In No, the 2nd determination part 34 complete | finishes work as defect and complete | finishes (step S46). In the case of Yes, it progresses to step S48. The second determination unit 34 determines whether or not it is finished (step S48). For example, if it is the last data in the work to be determined, the process ends. In No, the 1st determination part 32 moves to the following 1 or several data (step S50), and returns to step S32. In addition, you may acquire a data row | line for every one or several data. In this case, the process returns to step S30. In the case of Yes in step S48, the second determination unit 34 determines that the determination operation is good (step S52). Then, the process ends.

  FIG. 5A to FIG. 6B are diagrams illustrating a determination method according to the first embodiment. FIGS. 5A and 5C to 6B show sensor output signals with respect to time. The output signal is, for example, a voltage. FIG. 5B is a diagram illustrating the amount of transient discrimination with respect to time. For example, when the sensor 24 outputs a signal at a constant interval Δt, the point interval of the data string corresponds to Δt.

  With reference to FIG. 5A, the signal output from the sensor 24 is a time-series data string 40. In the work, there are a steady section in which the data string 40 is stable and a transition section in which the data string 40 changes greatly. With reference to FIG. 5B, a time-series transient discrimination amount sequence 42 is calculated based on the data sequence 40. The transient discrimination amount sequence 42 is small at the time when the data sequence 40 is stable, and is large at the time when the data sequence 40 is not stable. A threshold value 44 is provided for the transient discrimination amount sequence 42. With reference to FIG. 5C, the data when the transient discrimination amount sequence 42 is greater than or equal to the threshold 44 is determined as the second interval T2, and the data when the transient discrimination amount sequence 42 is smaller than the threshold is determined as the first interval T1. To do. Thereby, the section in which the data string 40 is stable can be set as the first section T1, and the section in which the data string 40 is not stable can be set as the second section T2.

  With reference to Fig.6 (a), the determination method of the quality of the work in 1st area T1 is demonstrated. The same operation is performed a plurality of times in advance, and the variance of the output signal (each data string) at each time is calculated. The lower limit 46 indicates (average−A × standard deviation) of a data string of work performed a plurality of times. The upper limit 48 indicates (average + A × standard deviation) of a data string of work performed a plurality of times. A is a positive natural number, for example, 3. In step S <b> 36, the second determination unit 34 determines the quality of the work based on whether the data string 40 is within the upper limit 48 and the lower limit 46. For example, when all of the data string 40 in the first section T1 is between the upper limit 48 and the lower limit 46, the work is determined to be good. Alternatively, when a part of the data string 40 in the first section T1 is between the upper limit 48 and the lower limit 46, the work may be determined as good.

  With reference to FIG.6 (b), the determination method of the quality of the work in 2nd area T2 is demonstrated. The second section T2 determines the quality of work using a method different from the first section T1. For example, the 2nd determination part 34 determines the quality of the work in a 2nd area based on the content learned using machine learning.

  FIG. 7 is a flowchart illustrating an example in which the first determination unit determines a section. FIG. 8A to FIG. 8D are diagrams showing a method for calculating the transient discrimination amount. Referring to FIG. 7, the first determination unit 32 sets a window width 52 for calculating the transient determination amount (step S60). Referring to FIG. 8 (a), a data string 40 is illustrated with respect to time. For example, the window width 52 is defined between the times t1 and t3 with the data 50 at the time t2 as the center. Referring to FIG. 7, the first determination unit 32 calculates an estimated value using a mathematical model (step S62). With reference to FIG. 8B, the estimated value sequence 56 is calculated corresponding to the data in the window width 52 centered on the data 50. For example, the estimated value 56 corresponding to the data 50a at time t1 is calculated using data having a width 54 from time t4 to immediately before time t1. The estimated value 56 corresponding to the data 50 at time t2 is calculated using data with a width 54 from time t5 to immediately before time t2. The estimated value 56 corresponding to the data 50b at time t3 is calculated using data having a width 54 from time t6 to immediately before time t3. In this way, the estimated value sequence 56 can be calculated in time series. The width 54 is, for example, a certain period.

ここで、ｘ´_ｔは各時刻ｔにおけるデータの推定値、ｘ_ｔは各時刻ｔにおけるデータ、ε_ｔはノイズ項、ａ_ｉはパラメータであり、定常性の条件として｜ａ_ｉ｜＜１である。 For example, when a steady autoregressive model is used as an estimated value, the estimated value can be calculated by the following equation.

Here, x ′ _t is an estimated value of data at each time t, x _t is data at each time t, ε _t is a noise term, a _i is a parameter, and | a _i | <1 is there.

  Thus, for example, when the estimated value corresponding to the data 50a at the time t1 is obtained, the estimated value is calculated using the data string 40 immediately before the time t4 from the time t4 that is the past time. The dotted line in FIG.

Referring to FIG. 7, the first determination unit 32 calculates a residual between the estimated value sequence 56 and the data sequence 40 (step S64). Referring to FIG. 8C, the difference between the estimated value sequence 56 and the data sequence 40 at each time from time t1 to time t3 is a residual 58. The residual 58 (l) at each time t can be calculated by the following equation.

ここで、σは、時刻ｔ１からｔ３の残差ｌの標準偏差、Ｎは時刻ｔ１からｔ３のデータ列４０のポイント数、ｐはモデルの次数である。各時刻における過渡判別量Ｃを時系列とした列が過渡判別量列４２である。 Referring to FIG. 7, the first determination unit 32 calculates a transient determination amount C based on the residual 58 (step S66). For example, an AIC (Akaike's Information Criterion) is calculated using the residual 58 from time t1 to time t3 as the transient discrimination amount C of the data 50 at time t2. The transient discrimination amount C can be calculated by the following equation.

Here, σ is the standard deviation of the residual l from time t1 to t3, N is the number of points in the data string 40 from time t1 to t3, and p is the order of the model. A column in which the transient discrimination amount C at each time is in time series is the transient discrimination amount column 42.

  Referring to FIG. 7, the first determination unit 32 determines whether the section to which the data belongs is a stationary first section or a transient second section (step S68). As shown in FIGS. 5B and 5C, the determination is made based on whether or not the transient determination amount C is equal to or greater than a threshold value.

  Referring to FIG. 8D, when the determination of data 50 is completed, the process moves to data 50c at time t7, which is the next data, in step S48 of FIG.

  Next, determination was performed by the method shown in FIG. 7 using an artificially created data string. FIG. 9A is a diagram showing a signal for a point, and FIG. 9B is a diagram showing a transient discrimination amount for the point. The point interval corresponds to 10 milliseconds. The signal in the section T3 was created assuming steady data. The signal in the section T3 is random numbers according to a normal distribution, and is data in which the average is constant at 0 and σ (standard deviation) is constant at 10. The signal in the section T4 was created assuming transient data. The signal in the section T4 is a random number according to a normal distribution, and the average is changed from 100 to 1 and σ is changed from 1 to 100. The signal in the section T5 was created assuming steady data. The signal in the section T5 is a random number according to a normal distribution, and is data in which the average is constant at 50 and σ is constant at 10. The data string 40 is stable in the sections T3 and T5. However, in the section 4, the data string 40 varies greatly.

  FIG. 9B is a diagram showing a result of calculating the transient discrimination amount by the method described in FIGS. 7 to 8D. In the sections T3 and T5, the transient discrimination amount is as small as 200 or less. In section 4, the transient discrimination amount is 250 or more. Therefore, for example, about 220 is set as the threshold 44. Accordingly, it is possible to determine the sections T3 and T5 as the first section and the section T4 as the second section. As described above, the determination method described in FIGS. 7 to 8D is effective for determining the first section and the second section.

  According to the first embodiment, as in step S32 of FIG. 4, the first determination unit 32, based on the data string 40, the data in the data string 40 belongs to the first section T1 or the second section T2. Determine whether. As in steps S34, S36, and S42 of FIG. 4, the second determination unit 34 determines the quality of work corresponding to data depending on whether the data belongs to the first section T1 or the second section T2. Determine using. Accordingly, an appropriate quality determination method can be applied to sections having different properties, such as a section where data is stationary and a section where data is transitional.

  Further, as in step S <b> 62 of FIG. 7, the first determination unit 32 calculates an estimated value 56 of data from the data string 40 using a steady model. As in step S64, a residual 58 between the estimated value 56 and the data 50 is calculated. As in step S66, based on the residual 58, it is determined whether the data 50 belongs to the first section T1 or the second section T2. Accordingly, an appropriate quality determination method can be applied depending on whether the data is a stationary section or the data is a transition section. In addition to the stationary autoregressive model, a moving average model, an autoregressive moving average model, and the like can be used as the stationary model.

  As illustrated in FIG. 8A to FIG. 8D, the first determination unit 32 determines whether the data belongs to the first section T <b> 1 based on the residual 58 of the certain section (width 52) including the data 50. It is determined whether it belongs to the section T2. Thereby, an appropriate quality determination method can be applied.

  As shown in FIGS. 5B and 5C, the first determination unit 32 determines that the data belongs to the first section T1 when the transient determination amount 42 (for example, AIC) is equal to or greater than the threshold value 44. When the transient determination amount 42 is smaller than the threshold value 44, the first determination unit 32 determines that the data belongs to the second section T2. Thereby, an appropriate quality determination method can be applied. Note that, as the transient discrimination amount 42, a statistical model selection criterion such as a residual, a Bayesian Information Criterion (BIC), a minimum description length (MDL), or the like can be used.

  When the first section T1 is a stationary section, the first pass / fail determination can be performed using a statistical method. Statistical methods are suitable for the determination of stationary intervals. As a statistical method, for example, an average value and a standard deviation can be used.

  When the second section T2 is a transitional section, it is preferable to use a method other than the statistical method for the second pass / fail determination. For example, a machine learning method can be used as the second pass / fail determination. The machine learning method is suitable for determining a transitional section. As the machine learning method, for example, a support vector machine method, a naive Bayes classifier, a k nearest neighbor method, or a Parse proton can be used.

  Example 2 is an example in which the operation of peeling the release paper of the double-sided tape is performed. The configuration of the robot system and the determination method are the same as those in the first embodiment, and a description thereof will be omitted. FIG. 10A to FIG. 11C are diagrams illustrating work contents in the second embodiment. The robot 22 performs work using tweezers 76. The tweezers 76 are operated by the manipulator 26. The sensor 24 detects distortion of the manipulator 26. Referring to FIG. 10A, a double-sided tape 75 is affixed to the stage 70. The double-sided tape 75 has an adhesive portion 72 and a release paper 74. The bonding part 72 is bonded to the stage 70. The stage 70 and the double-sided tape 75 correspond to the object. With reference to FIG. 10B, the robot 22 places the tip of the tweezers 76 against the stage 70. With reference to FIG. 10C, the robot 22 inserts the tweezers 76 between the release paper 74 and the adhesive portion 72.

  Referring to FIG. 11A, the robot 22 picks the release paper 74 using the tweezers 76. Referring to FIG. 11B, the robot 22 peels the release paper 74 from the adhesive portion 72 using the tweezers 76. Referring to FIG. 11C, the robot 22 moves the peeled release paper 74 to a designated position.

  FIG. 12A is a diagram illustrating sensor signals for points in Example 2, and FIG. 12B is a diagram illustrating transient discrimination amounts for points. The point interval is 10 milliseconds. Referring to FIG. 12A, the section 60 is a section of work in which the tip of the tweezers 76 hits the stage 70 as shown in FIG. 10B. The section 62 is a work section in which the tweezers 76 is inserted between the release paper 74 and the adhesive portion 72 as shown in FIG.

  The section 64 is a section of the work of picking the release paper 74 using the tweezers 76 as shown in FIG. The section 66 is an operation section in which the release paper 74 is peeled off from the adhesive portion 72 using the tweezers 76 as shown in FIG. The section 68 is a work section in which the peeled release paper 74 is moved to a designated position as shown in FIG.

  Referring to FIG. 12B, the transient discrimination amount sequence 42 was calculated by the method described in FIGS. 7 to 8D. The threshold 44 is about 95. A section in which the transient discrimination amount sequence 42 is greater than or equal to the threshold 44 is referred to as a second interval T2, and a segment in which the transient discrimination amount sequence 42 is smaller than the threshold 44 is referred to as a first interval T1.

  As in the second embodiment, the first section T1 and the second section T2 can be appropriately determined from the data string 40.

  Although the embodiments of the present invention have been described in detail above, the present invention is not limited to such specific embodiments, and various modifications and changes can be made within the scope of the gist of the present invention described in the claims. It can be changed.

In addition, the following additional notes are disclosed regarding the above description.
(Additional remark 1) The 1st which determines whether the data of the said data row | line | column belong to a 1st area or a 2nd area based on the time series data string which the sensor which detects the condition of the operation | work which a robot performs is output. A determination unit; and a second determination unit that determines whether the work corresponding to the data is good or bad by using different methods depending on whether the data belongs to the first interval or the second interval. The determination apparatus characterized by this.
(Supplementary Note 2) The first determination unit determines whether the data belongs to the first section based on a residual between the estimated value obtained by estimating the data using a steady model from the data string and the data. The determination device according to supplementary note 1, wherein the determination device determines whether the image belongs to the above.
(Additional remark 3) The said 1st determination part determines whether the said data belongs to the said 1st area or the said 2nd area based on the said residual of the fixed area containing the said data. The determination apparatus described.
(Supplementary Note 4) When the model selection criterion calculated using the residual is equal to or greater than a threshold, the first determination unit determines that the data belongs to the first section, and the criterion is smaller than the threshold. The determination apparatus according to appendix 2 or 3, wherein the data is determined to belong to the second section.
(Supplementary Note 5) The determination apparatus according to Supplementary Note 4, wherein the model selection criterion is AIC, BIC, or MDL.
(Additional remark 6) When the said data belongs to the said 1st area, the 2nd determination part determines the quality of the operation | work corresponding to the said data using the standard deviation and average of the data of the said previous work, It is characterized by the above-mentioned. The determination apparatus according to any one of appendices 1 to 5.
(Supplementary note 7) When the data belongs to the second section, the second determination unit determines whether the work corresponding to the data is good or bad using machine learning from the data of the past work. The determination device according to any one of supplementary notes 1 to 5.
(Supplementary Note 8) When the data belongs to the first section, the second determination unit determines whether the work corresponding to the data is good or bad using a standard deviation and an average of the data of the past work, and the data 6 is determined using the machine learning from the data of the past work, when the work belongs to the second section, the quality of the work corresponding to the data is determined. Judgment device.
(Supplementary Note 9) Based on a time-series data sequence output from a sensor that detects the status of work performed by the robot, the computer determines whether the data in the data sequence belongs to the first section or the second section And determining whether the work corresponding to the data is good or bad by using different methods depending on whether the data belongs to the first section or the second section.
(Additional remark 10) It is the determination method which a computer performs, Comprising: Based on the time series data sequence which the sensor which detects the condition of the operation | work which a robot performs outputs, whether the data of the said data sequence belong to a 1st area. A step of determining whether the data belongs to two sections, a step of determining whether the work corresponding to the data is good or bad using a different method depending on whether the data belongs to the first section or the second section;
The determination method characterized by including.

DESCRIPTION OF SYMBOLS 10 Computer 22 Robot 24 Sensor 30 Determination apparatus 32 1st determination part 34 2nd determination part 40 Data sequence 42 Transient discrimination amount sequence 44 Threshold value 56 Estimated value

Claims (6)

An estimated value obtained by estimating data in the data sequence using a stationary model is calculated from a time-series data sequence output by a sensor that detects the status of work performed by the robot, and a residual between the estimated value and the data is calculated. The model selection criterion is calculated. If the criterion is less than or equal to the threshold, the data is determined to belong to the first interval where the data is stationary. If the criterion is greater than the threshold, the data is transient A first determination unit that determines to belong to the second section ,
A second determination unit that determines whether the work corresponding to the data is good or bad using a different method depending on whether the data belongs to the first section or the second section;
The determination apparatus characterized by comprising.   The stationary model is a stationary autoregressive model, a moving average model or an autoregressive moving average model,
  The determination apparatus according to claim 1, wherein the reference is an AIC, BIC, or MDL reference. The residuals determination apparatus according to claim 1 or 2, wherein the said estimated value of a predetermined section including the data to be residual of the data. When the data belongs to the first section, the second determination unit determines the quality of the work corresponding to the data using a standard deviation and an average of the data of the past work, and the data is the second when belonging to the interval, said quality of work corresponding to the data, the determination apparatus according to any one claim of claims 1 to 3, characterized in that determined using machine learning from the data of the past the work. On the computer,
An estimated value obtained by estimating data in the data sequence using a stationary model is calculated from a time-series data sequence output from a sensor that detects a status of work performed by the robot, and a residual between the estimated value and the data is calculated. If the criterion is less than a threshold, the data is determined to belong to the first interval where the data is stationary, and if the criterion is greater than the threshold, the data is transient To belong to the second section ,
A determination program for determining whether the work corresponding to the data is good or bad by using different methods depending on whether the data belongs to the first section or the second section. A determination method executed by a computer,
An estimated value obtained by estimating data in the data sequence using a stationary model is calculated from a time-series data sequence output by a sensor that detects the status of work performed by the robot, and a residual between the estimated value and the data is calculated. The model selection criterion is calculated. If the criterion is less than or equal to the threshold, the data is determined to belong to the first interval where the data is stationary. If the criterion is greater than the threshold, the data is transient Determining that it belongs to the second section ,
Determining whether the work corresponding to the data is good or bad using a different method depending on whether the data belongs to the first section or the second section;
The determination method characterized by including.

Images