JP6690436B2 - Discrimination device, discrimination method, and discrimination program - Google Patents

Description

  The present invention relates to a discrimination device, a discrimination method, and a discrimination program.

  Development of technology to attach sensors to robots and detect abnormal work is in progress. When determining an abnormality in the sensor signal, a technique is disclosed in which a section where the sensor signal is constant is discriminated as a steady area and a section where the sensor signal is dynamically changed as a transient area, and an abnormality is determined in each section (for example, See Patent Documents 1 and 2.

JP, 2005-85437, A JP, 2005-217486, A

  However, in the above technique, it is determined based on the magnitude of the statistical change amount (hereinafter, referred to as the transient determination amount) of the time-series data of the detection result of the sensor. In the section discrimination based on the magnitude of the transient discrimination amount, there is a possibility that the section discrimination cannot be performed with high accuracy.

  In one aspect, the present invention aims to provide a discriminating apparatus, a discriminating method, and a discriminating program capable of discriminating working sections with high accuracy.

  In one aspect, the discrimination device includes a ratio of a plurality of constituent elements that constitutes a statistical change amount of a time-series output value of a sensor that detects a force or a displacement of a working unit that works on an object, and the statistical value. Based on at least one of the calculation unit that calculates at least one of the increase and decrease values of the change amount, and the ratio of the plurality of components and the increase and decrease value of the statistical change amount, the working unit with respect to the object. And a discriminating section for discriminating between a transitional region where the work is performed and a steady region where the working unit does not perform work on the object.

  The work section can be discriminated with high accuracy.

FIG. 1 is a diagram for explaining an overall configuration of a work device according to a first embodiment. (A)-(d) is a figure which illustrates an example of a series of operation | movement of a work robot. (A)-(c) is a figure which illustrates an example of a series of operation | movement of a work robot. It is a figure which illustrates the calculation method of a statistical variation. (A)-(c) is a figure which illustrates the case where several different work continues. It is a block diagram of a determination device. (A)-(c) is a figure which illustrates the calculation method of a statistical variation. (A)-(c) is a figure which illustrates discrimination using the ratio of a component. It is a flow chart which illustrates processing which a judgment device performs. FIG. 6 is a block diagram of a determination device according to a second embodiment. (A)-(c) is a figure which illustrates the discrimination using the increase / decrease value of the transient discriminating amount. It is a flow chart which illustrates processing which a judgment device performs. FIG. 7 is a block diagram of a determination device according to a third embodiment. (A) to (g) are examples in which the steady region and the transient region are discriminated in consideration of the magnitude of the transient determination amount. It is a flow chart which illustrates processing which a judgment device performs. (A)-(d) is an example which extracts a transition area from a transition area candidate. It is a flow chart which illustrates processing. (A) to (c) are examples in which the window width is variable. It is a figure which illustrates the flowchart performed when setting a window width. (A) and (b) are other examples in which the window width is variable. It is a figure which illustrates the other flowchart performed when setting a window width. It is a figure which illustrates the flowchart performed when a window width is made variable and a transient discrimination amount is weighted. FIG. 9 is a block diagram of a determination device according to a fourth embodiment. (A) and (b) are examples of setting the threshold value of the transient discrimination amount. (A) And (b) is a figure which illustrates the flowchart which a determination device performs. (A) And (b) is a figure which illustrates the flowchart which a determination device performs. (A) is a block diagram for explaining the hardware configuration of the determination device, and (b) is a diagram illustrating a work system.

  FIG. 1 is a diagram for explaining the overall configuration of the work device 100 according to the first embodiment. As illustrated in FIG. 1, the work device 100 includes a work robot 10, a controller 20, a camera 30, a determination device 40, and the like.

  The work robot 10 includes a robot hand 11, a sensor 12, and the like. The robot hand 11 is a device that performs a predetermined work on an object. The sensor 12 is a sensor that detects the force, displacement, and the like of the robot hand 11, and is, for example, a strain gauge, a force sensor, an acceleration sensor, or the like. In this embodiment, a plurality of sensors 12 are provided. The controller 20 is a control device that gives a work instruction to the work robot 10 at a predetermined timing. The camera 30 is a device that images the work of the work robot 10. In this embodiment, a plurality of cameras 30 are provided.

  2A to 2D and FIGS. 3A to 3C are diagrams illustrating an example of a series of operations of the work robot 10. As illustrated in FIG. 2A, the work robot 10 includes a pair of robot hands 11 a and 11 b as the robot hand 11, and strain gauges 12 a and 12 b as the plurality of sensors 12. For example, the strain gauge 12a is provided at the tip of the robot hand 11a, and the strain gauge 12b is provided at the tip of the robot hand 11b. First, the robot hands 11a and 11b move toward the cable connector 14 that fits into the connector 13 (first moving step).

  Next, as illustrated in FIG. 2B, the robot hands 11a and 11b grip the cable connector 14 (gripping step). Next, as illustrated in FIG. 2C, the robot hands 11a and 11b lift the cable connector 14 to separate it from the connector 13 and move the cable connector 14 toward another connector (second movement). Process). Next, as illustrated in FIG. 2D, the robot hands 11a and 11b insert the cable connector 14 into the other connector 15 (inserting step).

  Next, as illustrated in FIG. 3A, the robot hands 11a and 11b release the grip of the cable connector 14 (release step). Next, as illustrated in FIG. 3B, the robot hands 11a and 11b move in a direction away from the cable connector 14 (third moving step). Finally, as illustrated in FIG. 3C, the robot hands 11a and 11b insert the cable connector 14 into the connector 15 (inserting step). In the example of FIG. 3C, the robot hand 11 b inserts the cable connector 14 into the connector 15. The work is completed by the above steps.

  Here, an example of a calculation procedure of a statistical variation amount (hereinafter, referred to as a transient determination amount) of time series data of the detection result of the sensor 12 will be described. For example, as illustrated in FIG. 4, at each data point of a predetermined sampling period on the time-series sensor signal, the “average difference” and the “standard deviation difference” of the sections before and after that are added up. The transient discrimination amount can be obtained. The front and rear sections (window width) at this time are fixed. For example, by taking an arithmetic mean of the transient discriminant amounts of the measurement data of the sensor 12 and cutting them at a predetermined threshold value, it is possible to discriminate the portion where the transient discriminant amount is large as the transient region and the portion where the transient discriminant amount is small as the steady region. The transient region is a time period in which the work robot 10 works on an object. The stationary region is a time period during which the work robot 10 does not work on an object. The simple movement of the robot hand 11 is included in the steady region. Assembly work and the like by the robot hand 11 are included in the transition area.

  For example, the operating state can be automatically discriminated into a steady region / transient region based on the magnitude of the transient determination amount. However, in this case, it is difficult to detect a measured data value of the sensor 12 or an operation with a small change (a portion with a small transient determination amount) as a transient region. On the other hand, if it is attempted to simply detect even such a portion having a small transient discrimination amount as a transient region, the transient region may increase, and the calculation amount of the identification processing may unnecessarily increase. If the calculation parameter is fixed in advance when calculating the transient discrimination amount or the threshold value thereof, it is difficult to execute the optimum section discrimination. In this respect as well, with the threshold value set in advance, a relatively small operation in the entire operation may be omitted from the detection as the transient region.

  Further, as illustrated in FIG. 5A, when a plurality of different works are continuous, it is difficult to correctly separate and discriminate each operation only by the magnitude of the transient determination amount. For example, in the work of inserting the connector, the details of the work change such as (1) contact of the connector, (2) insertion of the connector while making contact, and (3) butting when the insertion is completed. On the other hand, only a small change appears in the transient discrimination amount. The work of (1) to (3) corresponds to the work of FIGS. 2 (d) to 3 (c).

  Therefore, as illustrated in FIG. 5B, the entire connector insertion work may be discriminated as one transition area. In the abnormality determination system, the abnormality score is accumulated for each transient region, and the accuracy of the determination is higher when the determination is made for each individual operation than when the different types of operations are mixed and the abnormality is determined. Therefore, as illustrated in FIG. 5 (c), it is required to discriminate the transient region in each operation unit. Therefore, the work apparatus 100 according to the present embodiment discriminates the work section with high accuracy using an index other than the magnitude of the transient determination amount.

  FIG. 6 is a block diagram of the determination device 40. As illustrated in FIG. 6, the determination device 40 includes a data storage unit 50, a transient determination amount calculation unit 60, a discrimination unit 70, a pass / fail determination unit 80, and an output unit 90. The transient determination amount calculation unit 60 includes a window width setting unit 61, a constituent element calculation unit 62, and a constituent ratio calculation unit 63. The discrimination unit 70 includes a composition ratio acquisition unit 71 and a section calculation unit 72. The quality determination unit 80 includes a discriminator creation unit 81 and a state determination unit 82.

  The data storage unit 50 temporarily stores data. At the time of distinguishing between the steady region and the transient region, the data storage unit 50 temporarily stores the learning OK data. The learning OK data is a time-series output value waveform (measurement data) of each sensor 12 when the work robot 10 is caused to perform a work in advance and the work status is visually determined to be “normal”. When determining whether the work of the work robot 10 is good or bad, the data storage unit 50 temporarily stores the time-series output value waveform (measurement data) of the output value of the sensor 12.

  The window width setting unit 61 sets an attention point for the measurement data stored in the data storage unit 50 at a predetermined sampling cycle, and sets a predetermined window width for the attention point.

The component calculation unit 62 calculates the component of the time-series statistical change amount of the measurement data in each set window. FIG. 7A is a diagram illustrating the set window. As illustrated in FIG. 7B, for example, the component calculation unit 62 calculates the average value e _i1 and the standard deviation σ _i1 of the output values before (past) the attention point in each set window. Then, the average value e _i2 and the standard deviation σ _i2 of the output values after (in the future) the point of interest are calculated. "I" means i-th window. Next, the component calculation unit 62 calculates the transient discriminant amount E _i from the average values e _i1 , e _i2 and the standard deviations σ _i1 , σ _i2 . The transient determination amount E _i can be expressed by the following equation (1), for example. Further, Δe _i is an average change amount and can be expressed by the following equation (2). Δσ _i is the amount of change in standard deviation and can be expressed by the following equation (3).
E _i = (Δe _i ² + Δσ _i ² ) ^0.5 (1)
Δe _i = | e _i2 −e _i1 | (2)
Δσ = | σ _i2 −σ _i1 | (3)

  Next, as illustrated in FIG. 7C, the component calculation unit 62 sets the time after a predetermined time interval as the next point of interest, and will be described with reference to FIGS. 7A and 7B. The components of the next window are calculated by the procedure described above. By repeating the procedure of FIGS. 7A to 7C, time-series data of each component can be obtained.

Next, the composition ratio calculator 63 calculates the ratio of the constituent elements. As the ratio of the constituent elements, the ratio of the average change amount to the transient discriminant amount or the ratio of the standard deviation change amount to the transient discriminant amount can be used. In the present embodiment, as an example, the configuration ratio calculation unit 63 calculates the ratio of the average change amount to the transient determination amount. The component ratio m _i can be expressed by the following equation (4).
m _i = (average change amount) _i / (transient determination amount) _i (4)

  FIG. 8A illustrates time series data of the transient determination amount E. When the robot hand 11 works, the transient determination amount E increases. Therefore, a method may be considered in which a threshold value is set for the transient determination amount E and a section in which the transient determination amount E is equal to or greater than the threshold value is discriminated as a transient region. However, in this case, it becomes difficult to extract, as the transient region, a region in which the absolute value of the transient determination amount E is small, but a change in the transient determination amount E, such as the region A in FIG. 8A, is present. That is, discrimination leakage in the transient region may occur.

  Specifically, even if the total value of "change in average" and "change in standard deviation" is constant, the ratio of its constituent elements ("change in average" and "change in standard deviation") In some cases, changes depending on the progress of work. For example, in the connector insertion work, the "average change" becomes large at the first contact, the "standard deviation change" becomes the center during the insertion while making contact, and the "change in standard deviation" becomes again at the last contact. It is considered that the “change in average” will be large. If such a change in the ratio is used as a reference, it is possible to divide the transitional region into smaller pieces for each work content.

  Therefore, in this embodiment, attention is paid to the ratio of the constituent elements. The composition ratio acquisition unit 71 acquires the composition ratio m of each window from the composition ratio calculation unit 63. FIG. 8B illustrates time-series data of the ratio m. As illustrated in FIG. 8B, when a large change appears in the transient determination amount E, a change appears in the ratio m as in the area B. Further, as in the region C, even if the absolute value of the transient determination amount is small, a large change appears in the ratio m when the ratio between the “change in average” and the “change in standard deviation” changes.

  Therefore, the section calculation unit 72 detects the change point of the ratio m as the boundary of the transient range. For example, the section calculation unit 72 uses the degree of deviation from the average ratio. Specifically, the section calculation unit 72 extracts a section in which the deviation degree of the ratio m from the average value or the moving average value is equal to or more than a threshold value as a transient area. In FIG. 8B, the moving average value is represented by the dotted line. By extracting the transient range in this way, as illustrated in FIG. 8C, a section that is extracted as only one transient range by the magnitude of the transient determination amount can be extracted as a plurality of transient ranges. become able to. That is, the discrimination accuracy in the transient region is improved.

  FIG. 9 is a flowchart illustrating a process executed by the determination device 40. As illustrated in FIG. 9, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S1). Next, the window width setting unit 61 sets an attention point in the measurement data stored in the data storage unit 50 at a predetermined sampling cycle, and sets a window having a predetermined window width for each attention point (step S2).

Next, the component calculation unit 62 calculates the component of the statistical variation amount of the measurement data in each set window (step S3). As an example, the component calculation unit 62 calculates the average value e _i1 e _i2 , the standard deviations σ _i1 , σ _i2, and the transient discriminant amount E _i in each window. Next, the composition ratio calculation unit 63 calculates the ratio m _i (step S4). Next, the composition ratio acquisition unit 71 acquires the composition ratio m of each window from the composition ratio calculation unit 63 (step S5). Next, the section calculation unit 72 detects the change point of the ratio m as the boundary of the transient area (step S6). As an example, the section calculation unit 72 extracts a section in which the deviation degree of the ratio m from the moving average value is equal to or more than a threshold value as a transient area, and extracts the other areas as a steady area. The above processing completes the discrimination between the steady region and the transient region.

  The discriminator creating unit 81 calculates, for the steady region, the range of the average value ± 3σ of the learning OK data when the robot hand 11 performs a plurality of normal operations as a reference range. Further, the discriminator creating unit 81 creates a discriminator in the transient region. The discriminator is an individual subject that determines the quality of the work status in each transition region, and is, for example, an SVM (Support Vector Machine) discriminator. Providing a discriminator is synonymous with creating an individual standard (standard range) for determining the quality of work in each transient region. Each discriminator has an individual reference for the correlation of a plurality of feature quantities extracted from the pre-measurement data of the sensor 12 in each transient region, for example. Examples of the feature amount include an amplitude value, a differential waveform, an integral waveform, a frequency, and the like.

  When determining the quality of the work of the work robot 10, the state determination unit 82 determines the quality of the output value waveform (measurement data) of the output value of the sensor 12 in each steady region based on the 3σ value. That is, the state determination unit 82 determines normal if the output value of the measurement data is within the range of ± 3σ described above, and determines abnormal if the output value of the measurement data is outside the range of ± 3σ. The state determination unit 82 also digitizes the result of the quality determination as an abnormality score. In addition, the state determination unit 82 makes a pass / fail determination using a discriminator for each transition region. That is, the state determination unit 82 determines that the output value of the measurement data is normal if the output value of the measurement data is within the reference range of the discriminator, and determines the abnormality if the output value of the measurement data is outside the reference range. The state determination unit 82 also digitizes the result of the quality determination as an abnormality score. For example, the state determination unit 82 determines that the work of the work robot 10 is abnormal when the cumulative value of these abnormality scores exceeds the threshold value. The output unit 90 outputs the determination result of the state determination unit 82.

  According to the present embodiment, the steady region and the transient region are discriminated by using the ratio of the components of the statistical change amount of the time series data of the detection result of the sensor 12. In this case, it is possible to set the transition region more finely as compared with the discrimination based on only the magnitude of the statistical change amount. For example, in a connector insertion operation, it is possible to distinguish the first contact, the subsequent insertion, and the final butting as separate transient zones. Thereby, the work sections can be discriminated with high accuracy.

  In the present embodiment, the component calculation unit 62 uses the ratio of a plurality of components that make up the statistical change amount of the time-series output value of the sensor that detects the force or displacement of the work unit that works on the object. And functions as an example of a calculation unit that calculates The discriminator that the section calculator 72 discriminates, based on the ratio of a plurality of constituent elements, a transient region in which the working unit works on the object and a steady region in which the working unit does not work on the object. Function as an example.

  Although the first embodiment focuses on the ratio of the constituent elements, other indexes may be focused on. In the second embodiment, an example will be described in which a portion in which the transient determination amount E has changed (increased or decreased) is detected as the boundary of the transient region.

  FIG. 10 is a block diagram of the determination device 40a according to the second embodiment. As illustrated in FIG. 10, the determination device 40 a is different from the determination device 40 in FIG. 6 in that the transient determination amount calculation unit 60 a is provided instead of the transient determination amount calculation unit 60, and the discrimination unit 70 a is provided instead of the discrimination unit 70. The point is that. The transient discriminant amount calculation unit 60a is different from the transient discriminant amount calculation unit 60 in that an increase / decrease value calculation unit 64 is provided instead of the component ratio calculation unit 63. The discriminator 70a is different from the discriminator 70 in that an increase / decrease value acquisition unit 73 is provided instead of the component ratio acquisition unit 71. The components denoted by the same reference numerals as those in the first embodiment have the same functions as those in the first embodiment. Hereinafter, differences from the first embodiment will be described.

  The increase / decrease value calculation unit 64 calculates the increase / decrease value of the transient determination amount E. In the present embodiment, as an example, the increase / decrease value calculation unit 64 calculates the differential value of the transient determination amount E, the smoothing differential value, and the like. FIG. 11A illustrates time series data of the transient determination amount E. As described with reference to FIG. 8A, it is difficult to extract a region such as the region A in which the absolute value of the transient determination amount is small but the change in the transient determination amount appears as the transient region. That is, discrimination leakage in the transient region may occur.

  Therefore, in this embodiment, attention is paid to the increase / decrease value of the transient determination amount E. The increase / decrease value acquisition unit 73 acquires the differential value D of the transient determination amount E from the increase / decrease value calculation unit 64. FIG. 11B illustrates the time series data of the differential value. As illustrated in FIG. 11B, when a large change appears in the transient determination amount E, a change appears in the differential value D. Further, as in the area A of FIG. 11A, even if the absolute value of the transient determination amount E is small, the differential value D changes if the increase degree is large.

  Therefore, the section calculation unit 72 detects the change point of the differential value D as the boundary of the transient region. For example, the section calculation unit 72 extracts a combination in which the areas of the positive region and the negative region continuous with the positive region exceed the threshold value as the transient region. In addition, the section calculation unit 72 may divide the transient region at the valley portion of the transient determination amount E. By extracting the transient range in this way, as illustrated in FIG. 11C, a section that is extracted as one transient range only by the magnitude of the transient determination amount can be extracted as a plurality of transient ranges. become able to. That is, the discrimination accuracy in the transient region is improved.

  FIG. 12 is a flowchart illustrating a process executed by the determination device 40a. As illustrated in FIG. 12, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S11). Next, the window width setting unit 61 sets a point of interest in the measurement data stored in the data storage unit 50 at a predetermined sampling period, and sets a window having a predetermined window width for the point of interest (step). S12).

Next, the component calculation unit 62 calculates the component of the statistical variation amount of the measurement data in each set window (step S13). As an example, the component calculation unit 62 calculates the average value e _i1 e _i2 , the standard deviations σ _i1 , σ _i2, and the transient discriminant amount E _i in each window. Next, the increase / decrease value calculation unit 64 calculates the differential value D as the increase / decrease value of the transient determination amount E (step S14). Next, the increase / decrease value acquisition unit 73 acquires the differential value D of each window from the increase / decrease value calculation unit 64 (step S15). Next, the section calculation unit 72 detects the change point of the differential value D as the boundary of the transient area (step S16). As an example, the section calculation unit 72 extracts a combination in which the area of the positive area of the differential value and the area of the negative area connected to the positive area exceed the threshold value as the transient area. The above processing completes the discrimination between the steady region and the transient region.

  According to the present embodiment, the steady range and the transient range are discriminated by using the increase / decrease value of the statistical change amount of the time series data of the detection result of the sensor 12. In this case, it is possible to set the transitional range of the discrimination more finely as compared with the discrimination based on only the magnitude of the statistical change amount. For example, in a connector insertion operation, it is possible to distinguish the first contact, the subsequent insertion, and the final butting as separate transient zones. Thereby, the work sections can be discriminated with high accuracy.

  In the present embodiment, the increase / decrease value calculation unit 64 is a calculation unit that calculates an increase / decrease value of the statistical change amount of the time-series output value of the sensor that detects the force or displacement of the working unit that works on the object. It serves as an example. The section calculation unit 72 discriminates, based on the increase / decrease value of the statistical change amount, a transitional region in which the working unit works on the object and a steady region in which the working unit does not work on the object. It functions as an example of a section.

  The steady range and the transient range may be discriminated in consideration of the magnitude of the transient determination amount. FIG. 13 is a block diagram of the determination device 40b according to the third embodiment. As illustrated in FIG. 13, the determination device 40b is different from the determination device 40 of FIG. 6 in that the transient discriminant amount calculation unit 60b is provided instead of the transient discriminant amount calculation unit 60, and the discriminator 70b is provided instead of the discriminator 70. The point is that. The transient discriminant amount calculation unit 60b differs from the transient discriminant amount calculation unit 60 in that an increase / decrease value calculation unit 64 is further provided. The discriminator 70b is different from the discriminator 70 in that an increase / decrease value acquisition unit 73, a transient determination amount acquisition unit 74, and a weight setting unit 75 are further provided. The components designated by the same reference numerals as those in the first and second embodiments fulfill the same functions as those in the first and second embodiments. Hereinafter, differences from the first and second embodiments will be described.

  FIG. 14A is a diagram illustrating the transient determination amount E calculated by the component calculation unit 62. FIG. 14B is a diagram illustrating the ratio m calculated by the composition ratio calculation unit 63. The dotted line in FIG. 14B is the moving average value of the ratio m. FIG. 14C is a diagram exemplifying the differential value D calculated by the increase / decrease value calculation unit 64.

  The section calculator 72 calculates a normalized value of the transient discriminant amount E. FIG. 14D is a normalized value of the transient determination amount E. The section calculator 72 calculates the deviation degree of the ratio m from the average value, and calculates the normalized value of the deviation degree. FIG.14 (e) is a figure which illustrates the normalization value of the deviation degree of the ratio m from a moving average value. The section calculation unit 72 calculates a normalized value of the integrated amount of the differential value D of the transient determination amount E. FIG. 14F is a diagram illustrating a normalized value of the integrated amount of the differential value D of the transient determination amount E. Next, the section calculation unit 72 integrates the normalized value of the transient determination amount E, the normalized value of the deviation degree of the ratio m from the moving average value, and the differential value D according to the weight set by the weight setting unit 75. Weighted addition with the normalized value of the quantity. FIG. 14G illustrates the weighted addition value. In the present embodiment, this weighted addition value is used as the composite discriminant amount C. The section calculation unit 72 extracts a section in which the combined determination amount C is equal to or more than a threshold value as a transitional area.

  FIG. 15 is a flowchart illustrating a process executed by the determination device 40b. As illustrated in FIG. 15, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S21). Next, the window width setting unit 61 sets an attention point in the measurement data stored in the data storage unit 50 at a predetermined sampling cycle, and sets a predetermined window width for each attention point (step S22). .

Next, the component calculation unit 62 calculates the component of the statistical variation amount of the measurement data in each set window (step S23). As an example, the component calculation unit 62 calculates the average value e _i1 e _i2 , the standard deviations σ _i1 , σ _i2, and the transient discriminant amount E _i in each window. Next, the composition ratio calculation unit 63 calculates the ratio m _i (step S24). Next, the increase / decrease value calculation unit 64 calculates the differential value D as the increase / decrease value of the transient determination amount E (step S25).

  Next, the section calculation unit 72 sets the normalized value of the transient discriminant amount E, the normalized value of the deviation degree of the ratio m from the average value, and the normalized value of the integrated amount of the differential value D of the transient discriminant amount E. Is calculated (step S26). Next, the section calculation unit 72 integrates the normalized value of the transient determination amount E, the normalized value of the deviation degree of the ratio m from the moving average value, and the differential value D according to the weight set by the weight setting unit 75. The composite discriminant amount C is calculated by performing weighted addition with the normalized value of the amount (step S27). Next, the section calculation unit 72 extracts a section in which the combined determination amount C exceeds the threshold value as a transient area (step S28). The above processing completes the discrimination between the steady region and the transient region.

  According to this embodiment, in addition to the magnitude of the statistical change amount of the time series data of the detection result of the sensor 12, the ratio of the components of the statistical change amount and the increase / decrease value of the statistical change amount are used. Discrimination between the steady region and the transient region is performed. In this case, it is possible to set the transitional range of the discrimination more finely as compared with the discrimination based on only the magnitude of the statistical change amount. That is, the work sections can be discriminated with high accuracy. Note that only one of the ratio of the components of the statistical change amount and the increase / decrease value of the statistical change amount may be used. For example, when only the ratio of the components of the statistical change amount is used, the increase / decrease value calculation unit 64 and the increase / decrease value acquisition unit 73 are unnecessary. When only the increase / decrease value of the statistical change amount is used, the composition ratio calculation unit 63 and the composition ratio acquisition unit 71 are unnecessary.

(Modification 3-1)
The transient range candidate may be extracted from each of the transient determination amount E, the ratio m, and the differential value D, and the transient range may be extracted from the transient range candidate. The section calculation unit 72 extracts a section in which the transient determination amount E exceeds a threshold value as a transient area candidate. FIG. 16A is a diagram exemplifying the transition region candidates extracted from the transition determination amount E. Further, the section calculation unit 72 extracts a section in which the deviation degree of the ratio m from the moving average value exceeds the threshold value as a transition area candidate. FIG. 16B is a diagram exemplifying the transition region candidates extracted from the ratio m. In addition, the section calculation unit 72 extracts a combination in which the area of the positive area of the differential value D and the area of the negative area connected to the positive area exceed the threshold value as the transient area. FIG. 16C is a diagram exemplifying the transition region candidates extracted from the differential value D. As illustrated in FIG. 16D, the section calculation unit 72 extracts the transient range by using the logical sum or logical product of these transient range candidates. In this modification, the weight setting unit 75 may not be provided.

  FIG. 17 is a flowchart illustrating the process according to this modification. As illustrated in FIG. 17, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S31). Next, the window width setting unit 61 sets an attention point in the measurement data stored in the data storage unit 50 at a predetermined sampling cycle, and sets a predetermined window for each attention point (step S32).

Next, the component calculation unit 62 calculates the component of the statistical variation amount of the measurement data in each set window (step S33). As an example, the component calculation unit 62 calculates the average value e _i1 e _i2 , the standard deviations σ _i1 , σ _i2, and the transient discriminant amount E _i in each window. Next, the configuration ratio calculation unit 63 calculates the ratio m _i (step S34). Next, the increase / decrease value calculation unit 64 calculates the differential value D as the increase / decrease value of the transient determination amount E (step S35).

  Next, the section calculation unit 72 extracts a section in which the transient determination amount E exceeds the threshold as a transient area candidate, extracts a section in which the deviation degree of the ratio m from the moving average exceeds the threshold as a transient area candidate, and differentiates the same. A combination in which the area of the positive area of the value D and the area of the negative area connected to the positive area exceeds the threshold value is extracted as the transient area (step S36). Next, the section calculation unit 72 extracts the transient range by using the logical sum, the logical product, and the like of these transient range candidates (step S37).

(Modification 3-2)
In each of the above examples, the window width setting unit 61 may change the window width. For example, the window width setting unit 61 selects a window width that maximizes the obtained transient determination amount E. The width before (past) the attention point of each window is L1, and the width after (future) the attention point is L2. The window width setting unit 61 sets L1 = L2. As illustrated in FIGS. 18A to 18C, the window width setting unit 61 gradually widens the window width L1 = L2 stepwise. The component calculation unit 62 calculates the transient determination amount at each stage. The window width setting unit 61 sets, as the window width, the window width L1 + L2 when the transition determination amount calculated by the component calculation unit 62 is maximum. For example, the magnitude of the transient determination amount can be determined by comparing the peak values of the transient determination amount.

  FIG. 19 is a diagram exemplifying a flowchart executed when the window width setting unit 61 sets the window width. As illustrated in FIG. 19, the window width setting unit 61 sets the window width L1 = L2 to an unset value (step S41). Next, the window width setting unit 61 selects an unprocessed data point of the measurement data (step S42). Next, the component calculation unit 62 calculates the transient determination amount E (step S43). Next, the window width setting unit 61 determines whether or not the calculation of the transient determination amount E at all data points has been completed (step S44). When it is determined as "No" in step S44, the process is repeated from step S42.

  When it is determined to be “Yes” in step S44, the window width setting unit 61 calculates the total of the transient determination amounts (step S45). Next, the window width setting unit 61 determines whether the window width has been set to all the set values (step S46). When it is determined as "No" in step S46, the process is executed again from step S41. When it is determined to be “Yes” in step S46, the window width setting unit 61 compares the transient determination amount total values (step S47). Next, the window width setting unit 61 determines the window width to be the window width when the transient discrimination amount total value is the highest (step S48).

  When the window width L1 and the window width L2 are independently determined, steps S41 to S46 may be individually executed for the window width L1 and the window width L2.

  The window width setting unit 61 may set the window width for each point of interest. As illustrated in FIG. 20A, the window width setting unit 61 gradually widens the window width L1 = L2 step by step for each target point. The component calculation unit 62 calculates the transient determination amount at each stage. The window width setting unit 61 sets, as the window width, the window width L1 + L2 when the transition determination amount calculated by the component calculation unit 62 is maximum.

  FIG. 20B is a diagram illustrating a flowchart executed by the window width setting unit 61 when setting the window width. As illustrated in FIG. 20B, the window width setting unit 61 selects an unprocessed data point of the measurement data (step S51). Next, the window width setting unit 61 sets the window width L1 = L2 to an unset value (step S52). Next, the component calculation unit 62 calculates the transient determination amount E (step S53). Next, the window width setting unit 61 determines whether the window width has been set to all the set values (step S54). When it is determined as "No" in step S54, the process is repeated from step S52.

  When it is determined to be “Yes” in step S54, the transient determination amounts are compared (step S55). The window width setting unit 61 determines the window width to be the window width when the transient discrimination amount total value is the highest (step S56). Next, the window width setting unit 61 stores the transient determination amount in the determined window width (step S57). Next, the window width setting unit 61 determines whether the processing of all data points is completed (step S58). When it is determined as "No" in step S58, the process is repeated from step S51. When it is determined to be “Yes” in step S58, the execution of the flowchart ends.

  When the window width L1 and the window width L2 are set independently, steps S52 to S57 may be executed individually. This modification can be applied to the first and second embodiments.

(Modification 3-3)
In each of the above examples, the constituent element calculation unit 62 may weight each constituent element when calculating the transient determination amount. For example, the component calculation unit 62 sets weights on “change in average” and “change in standard deviation”. Specifically, the component calculation unit 62 calculates “change in average” and “change in standard deviation” according to the above equation (2) at all the points of interest in the measurement data. Next, the constituent element calculation unit 62 calculates the cumulative addition value of the square of “change in average” and “change in standard deviation”. The square of the cumulative addition value of "change in average" and A _E, the square of the cumulative value of the "change of standard deviation" in the case of the A _S, then the coefficient k _{e =} 1, the coefficient k _{s =} A _E / A _S. Next, the component calculation unit 62 calculates the weighted transient determination amount M according to the following equation (5).
_{M i = (k e Δe i} 2 + k s Δσ i 2) 0.5 (5)

  FIG. 21 is a diagram illustrating a flowchart executed in this case. As illustrated in FIG. 21, the component calculation unit 62 selects an unprocessed data point of the measurement data (step S61). Next, the component calculation unit 62 calculates "change in average" and "change in standard deviation" (step S62). Next, the component calculation unit 62 cumulatively adds the squares of the “change in average” and the “change in standard deviation” (step S63). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S64). When it is determined as "No" in step S64, the process is repeated from step S61.

When it determines with "Yes" at step S64, the component calculation part 62 determines a weighting coefficient based on the ratio of a cumulative addition value (step S65). The square of the cumulative addition value of "change in average" and A _E, the square of the cumulative value of the "change of standard deviation" in the case of the A _S, then the coefficient k _{e =} 1, the coefficient k _{s =} A _E / A _S. Next, the component calculation unit 62 selects an unprocessed data point of the measurement data of the sensor 12 (step S66). Next, the component calculation unit 62 calculates the transient determination amount M according to the above equation (5) (step S67). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S68). When it is determined as "No" in step S68, the process is repeated from step S66. When it is determined to be “Yes” in step S68, the execution of the flowchart ends.

  FIG. 22 is a diagram exemplifying a flowchart that is executed when the window width is variable and the transient determination amount is weighted. As illustrated in FIG. 22, the window width setting unit 61 sets the window width L1 = L2 to an unset value (step S71). Next, the component calculation unit 62 selects an unprocessed data point of the measurement data (step S72). Next, the component calculation unit 62 calculates "change in average" and "change in standard deviation" (step S73). Next, the component calculation unit 62 cumulatively adds the squares of the “change in average” and the “change in standard deviation” (step S74). Next, the component calculation unit 62 determines whether or not the processing of all the data points has been completed (step S75). When it is determined as "No" in step S75, the process is executed again from step S72.

When it is determined to be “Yes” in step S75, the constituent element calculation unit 62 calculates the weighting coefficient based on the ratio of the cumulative addition value (step S76). The square of the cumulative addition value of "change in average" and A _E, the square of the cumulative value of the "change of standard deviation" in the case of the A _S, then the coefficient k _{e =} 1, the coefficient k _{s =} A _E / A _S. Next, the component calculation unit 62 selects an unprocessed data point of the measurement data (step S77). Next, the component calculation unit 62 calculates the transient determination amount M according to the above equation (5) (step S78). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S79). When it is determined as "No" in step S79, the process is executed again from step S77.

  When it is determined to be “Yes” in step S79, the window width setting unit 61 calculates the total of the transient determination amount M (step S80). Next, the window width setting unit 61 determines whether the window width has been set to all the set values (step S81). When it is determined to be “No” in step S81, the window width setting unit 61 is executed again from step S71. When it is determined to be “Yes” in step S81, the window width setting unit 61 compares the total value of the transient determination amount M (step S82).

  Next, the window width setting unit 61 determines the window width to a set value that maximizes the total value of the transient determination amount M (step S83). Next, the component calculation unit 62 sets a weighting factor corresponding to the window width (step S84). Next, the component calculation unit 62 selects an unprocessed data point of the measurement data of the sensor 12 (step S85). Next, the constituent element calculating unit 62 calculates the transient determination amount M (step S86). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S87). When it is determined as "No" in step S87, the process is repeated from step S85. When it is determined to be “Yes” in step S87, the execution of the flowchart ends.

  FIG. 23 is a block diagram of the determination device 40c according to the fourth embodiment. As illustrated in FIG. 23, the determination device 40c is different from the determination device 40 of FIG. 6 in that the transient discriminant amount calculation unit 60c is provided instead of the transient discriminant amount calculation unit 60, and the discriminator 70c is provided instead of the discriminator 70. The point is that. The difference between the transient determination amount calculation unit 60c and the transient determination amount calculation unit 60 is that an increase / decrease value calculation unit 64 is further provided. The discriminator 70b is different from the discriminator 70 in that an increase / decrease value acquisition unit 73, a transient determination amount acquisition unit 74, a weight setting unit 75, and a threshold value setting unit 76 are further provided. Hereinafter, differences from the first embodiment will be described. The components denoted by the same reference numerals as those in the first to third embodiments fulfill the same functions as those in the first to third embodiments. Hereinafter, points different from the first to third embodiments will be described.

  When the threshold for detecting the magnitude of the transient discrimination amount E is fixed, it becomes difficult to detect a particularly small transient discrimination amount E. Therefore, it is preferable to determine the optimum threshold value by the learning OK data and perform the discrimination between the steady region / transient region based on the optimum threshold value. As illustrated in FIG. 24A, the threshold setting unit 76 creates a histogram of the transient discriminant amount E and binarizes Otsu for it. In this case, the parameter setting can be completely automated. Since the transient discrimination amount E is small in the steady region, it is automatically excluded from the transient region. If the blank area in the transient area is long after the transient area is detected by the automatic threshold setting, the threshold is set by recursively applying the Otsu binarization to the interval, and the transient discrimination amount E It is also possible to detect the portion of the transient discrimination amount E that is smaller than the above as the steady region.

  FIG. 24B shows a method in which only the portion where the transient determination amount E is locally large is detected as the transient region. By setting the discrimination threshold value as the ratio of the transient discrimination amount E to the local peak, the portion where the transient discrimination amount E is small can be detected as the transient region. In this case, it is necessary to set some parameters such as a peak value dropped as noise, a ratio of a threshold value to a peak, and an interval size for searching a local peak. It should be noted that this method can also be applied to the combined discriminant amount.

  25 (a) and 25 (b) are diagrams illustrating a flowchart executed by the determination device 40c. As illustrated in FIG. 25A, the constituent element calculation unit 62 executes the calculation process of the total transient determination amount (step S91). FIG. 25B is a flowchart illustrating the details of step S91. As illustrated in FIG. 25B, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S101). Next, the window width setting unit 61 selects an unprocessed data point of the measurement data (step S102).

  Next, the window width setting unit 61 sets the window width by executing the flowchart of FIG. 19 or FIG. 20 (b) (step S103). Next, the constituent element calculation unit 62 calculates "change in average" and "change in standard deviation" (step S104). Next, the constituent element calculation unit 62 calculates the constituent ratio of the transient determination amount E (step S105). Next, the component calculation unit 62 calculates the transient determination amount E of each data point (step S106). Next, the component calculation unit 62 stores each calculated value (step S107). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S108). When it is determined as "No" in step S108, the process is executed again from step S102. When it is determined to be “Yes” in step S108, the execution of the flowchart ends.

  Next, the section calculation unit 72 selects an unprocessed data point of the measurement data (step S92). Next, the transient determination amount acquisition unit 74 acquires the transient determination amount E (step S93). Further, the composition ratio acquisition unit 71 acquires the ratio m (step S94). Next, the section calculation unit 72 normalizes the transient determination amount E and the ratio m, and performs weighted addition (step S95). Next, the threshold setting unit 76 automatically sets the binarization threshold as described with reference to FIG. 24A (step S96). Next, the section calculation unit 72 discriminates between the steady range and the transient range (step S97). Next, the section calculation unit 72 determines whether or not the processing of all data is completed (step S98). When it is determined as "No" in step S98, the process is executed again from step S92. When it is determined to be “Yes” in step S98, the execution of the flowchart ends.

  26A and 26B are diagrams illustrating another flowchart executed by the determination device 40c. As illustrated in FIG. 26A, the constituent element calculation unit 62 executes the calculation process of the total transient determination amount (step S111). FIG. 26B is a flowchart illustrating the details of step S111. As illustrated in FIG. 26B, the data storage unit 50 temporarily stores the learning OK data as the measurement data (step S121). Next, the window width setting unit 61 selects an unprocessed data point of the measurement data (step S122).

  Next, the window width setting unit 61 sets the window width by executing the flowchart of FIG. 19 or FIG. 20 (b) (step S123). Next, the component calculation unit 62 calculates "change in average" and "change in standard deviation" (step S124). Next, the increase / decrease value calculation unit 64 calculates the differential value D as the increase / decrease value of the transient determination amount (step S125). Next, the component calculation unit 62 calculates the transient determination amount of each data point (step S126). Next, the component calculation unit 62 stores each calculated value (step S127). Next, the component calculation unit 62 determines whether or not the processing of all data points has been completed (step S128). When it is determined as “No” in step S128, the process is repeated from step S122. When it is determined to be “Yes” in step S128, the execution of the flowchart ends.

  Next, the section calculation unit 72 selects an unprocessed data point of the measurement data (step S112). Next, the transient determination amount acquisition unit 74 acquires the transient determination amount E (step S113). Next, the threshold setting unit 76 performs threshold setting using a local peak, as described with reference to FIG. 24 (b) (step S114). Next, the section calculation unit 72 discriminates between the steady region and the transient region by applying the threshold value set in step S114 to the transient determination amount E (step S115). After the execution of step S112, the increase / decrease value acquisition unit 73 acquires the differential value D as the increase / decrease value from the increase / decrease value calculation unit 64 (step S116). Next, the section calculation unit 72 discriminates between the steady region and the transient region based on the increase / decrease value (step S117).

  After executing steps S115 and S117, the section calculation unit 72 integrates the steady region and the transient region discriminated in step S115 with the steady region and the transient region discriminated in step S117 (step S118). For example, they can be integrated using logical sum and logical product. Next, the section calculation unit 72 determines whether or not the processing of all data points is completed (step S119). When it is determined as “No” in step S119, the process is repeated from step S112. When it is determined to be “Yes” in step S119, the execution of the flowchart ends.

  According to the present embodiment, even if the change in the output value of the sensor 12 is small, an operation with a small transient determination amount E can be accurately detected as a transient region by appropriately setting the threshold value. On the other hand, it is possible to suppress the generation of an unnecessary transient region from the reaction of the noise level in the steady region. As a result, even a motion that cannot be generated by the transient classifier can be determined by the complex classifier, and the abnormality determination performance is improved.

  FIG. 27A is a block diagram for explaining the hardware configuration of the determination devices 40 to 40c. With reference to FIG. 27A, the determination devices 40 to 40c include a CPU 101, a RAM 102, a storage device 103, a display device 104, and the like. A CPU (Central Processing Unit) 101 is a central processing unit.

  The CPU 101 includes one or more cores. A RAM (Random Access Memory) 102 is a volatile memory that temporarily stores a program executed by the CPU 101, data processed by the CPU 101, and the like. The storage device 103 is a non-volatile storage device. As the storage device 103, for example, a ROM (Read Only Memory), a solid state drive (SSD) such as a flash memory, a hard disk driven by a hard disk drive, or the like can be used. The storage device 103 stores a discrimination program. The display device 104 is a liquid crystal display, an electroluminescence panel, or the like, and displays the determination result. Although each unit of the determination devices 40 to 40c is realized by executing a program in the present embodiment, hardware such as a dedicated circuit may be used.

  FIG. 27B is a diagram illustrating a work system. In each of the above examples, the determination devices 40 to 40c acquire the measurement data from the sensor 12 and the image data from the camera 30. On the other hand, the server 202 having the functions of the determination devices 40 to 40c may acquire data from the sensor 12 and the camera 30 via the telecommunication line 201 such as the Internet.

  In each of the above examples, the sum of “change in average” and “change in standard deviation” is used as the transient discriminant amount, but other transient discriminant amounts may be used. For example, the median value, the range from the minimum value to the maximum value, the interquartile range (first quartile to third quartile), skewness, kurtosis, and the like can be used as the transition discriminant amount. When the median value is used as the transient discriminating amount, the ratio of the change in the median value and the change in the standard deviation can be used as the ratio of the constituent elements. When the range from the minimum value to the maximum value is used as the transient determination amount, the ratio of the change of the average and the change of (maximum value-minimum value) can be used as the ratio of the constituent elements. When the quartile range is used as the transient discriminant amount, the ratio of the change in the first quartile and the change in the third quartile can be used as the ratio of the constituents. When the skewness is used as the transient discrimination amount, the ratio of the change in the standard deviation and the change in the skewness can be used as the ratio of the constituent elements. When kurtosis is used as the transient discriminant, the ratio of the change in standard deviation to the change in kurtosis can be used as the ratio of the constituent elements.

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

10 Working Robot 11 Robot Hand 12 Sensor 20 Controller 30 Camera 40 Judgment Device 50 Data Storage Section 60 Transient Discrimination Amount Calculation Section 61 Window Width Setting Section 62 Constituent Calculation Section 63 Composition Ratio Calculation Section 64 Increase / Decrease Calculation Section 70 Discrimination Section 71 Composition Ratio acquisition unit 72 Section calculation unit 73 Increase / decrease value acquisition unit 74 Transient determination amount acquisition unit 75 Weight setting unit 76 Threshold setting unit 80 Pass / fail judgment unit 81 Discriminator creation unit 82 State determination unit 90 Output unit 100 Working device

Claims (9)

At least one of a ratio of a plurality of constituent elements constituting a statistical change amount of a time-series output value of a sensor that detects a force or a displacement of a working unit that performs work on an object and an increase / decrease value of the statistical change amount. A calculation unit for calculating one,
Based on at least one of the ratio of the plurality of constituent elements and the increase / decrease value of the statistical change amount, a transitional region in which the working unit works on the object and the working unit works on the object. And a discriminating section for discriminating between a stationary region in which the operation is not performed.   The discriminator according to claim 1, wherein the discriminator further discriminates between the transient region and the steady region based on the magnitude of the statistical change amount.   The discriminator is configured to perform the transition based on a determination value obtained by weighting and adding the ratio of the constituent elements, the increase / decrease value of the statistical change amount, and the normalized value of the statistical change amount. The discriminating device according to claim 1, wherein the discriminating device discriminates a region from the steady region.   The discrimination unit is a transient region candidate extracted based on the ratio of the constituent elements, a transient region candidate extracted based on the increase / decrease value of the statistical change amount, and a transient extracted based on the magnitude of the statistical change amount. The discriminator according to claim 1 or 2, wherein the transitional region and the steady region are discriminated by using a logical sum or a logical product of the region candidates. The calculation unit calculates the statistical variation amount in each window having a predetermined window width at a predetermined sampling period with respect to the time-series output value of the sensor,
The discriminator according to any one of claims 1 to 4, wherein the window width is changed so that a total value of the statistical change amounts becomes maximum.   The discrimination apparatus according to claim 2, further comprising a threshold value setting unit that creates a histogram of the statistical change amount and sets a threshold value for the magnitude of the statistical change amount by binarizing Otsu for the histogram.   The discriminating apparatus according to claim 2, further comprising a threshold setting unit that sets a threshold for the magnitude of the statistical variation based on a local peak of the statistical variation. At least one of a ratio of a plurality of constituent elements constituting a statistical change amount of a time-series output value of a sensor that detects a force or a displacement of a working unit that performs work on an object and an increase / decrease value of the statistical change amount. Calculate one,
Based on at least one of the ratio of the plurality of constituent elements and the increase / decrease value of the statistical change amount, a transitional region in which the working unit works on the object and the working unit works on the object. The discrimination method is characterized by discriminating between a stationary region in which no operation is performed. On the computer,
At least one of a ratio of a plurality of constituent elements constituting a statistical change amount of a time-series output value of a sensor that detects a force or a displacement of a working unit that performs work on an object and an increase / decrease value of the statistical change amount. A process of calculating one,
Based on at least one of the ratio of the plurality of constituent elements and the increase / decrease value of the statistical change amount, a transitional region in which the working unit works on the object and the working unit works on the object. A discrimination program for discriminating between a stationary region in which is not performed, and a discrimination program.

Images