JP2021060723A - Monitoring condition determination device, monitoring condition determination method, and program - Google Patents

Abstract

To provide a monitoring condition determination device, a monitoring condition determination method, and a program capable of quantitatively determining a statistic that can indicate a peculiar change occurring before an anomaly occurs in time series data to be monitored.SOLUTION: The monitoring condition determination device is provided with: a time series data acquisition unit that acquires time series data showing changes over time in variables related to work on a work object; an anomaly information acquisition unit that acquires anomaly information showing the history of anomalies related to the work; an anomaly occurrence distance calculating unit that calculates an anomaly occurrence distance, which is a distance in a time space between the time series data and the time when the anomaly occurs, based on the time series data and the anomaly information; and a monitoring statistic type determining unit that determines a monitoring statistic type, which indicates the type of statistic to be monitored in the time series data, based on a relationship between the statistic for each time series data and the anomaly occurrence distance.SELECTED DRAWING: Figure 2

Description

The present invention relates to a monitoring condition determining device, a monitoring condition determining method, and a program.

In recent years, many efforts have been made to detect peculiar changes before an abnormality occurs by monitoring time-series data (temperature, current, etc.) recorded by a machine (processing device). In order to detect a peculiar change, it is common to monitor the threshold value for the time series data itself, but in some techniques, there is also a method of setting a threshold value (guard band) that fluctuates with time. In addition, some technologies introduce a machine learning method, train existing data, generate a trained model, and detect the presence or absence of peculiar changes from time series data by inference. (Patent Document 1).

Japanese Unexamined Patent Publication No. 2019-38074

However, the accuracy of detecting the peculiar changes before the anomaly depends on which statistic in the time series data is monitored. Among various statistics such as maximum value, minimum value, and average value in time series data, the statistic that can accurately detect the peculiar change before the abnormality cannot be known without knowledge and experience. Therefore, there is a problem that the accuracy of detecting the peculiar change before the abnormality depends on the ability of the analyst.

The present invention has been made to solve the above problems, and an object of the present invention is to quantitatively determine a statistic that can show a peculiar change that occurs before an abnormality occurs in the time series data to be monitored. The purpose is to provide a monitoring condition determination device, a monitoring condition determination method, and a program capable of the above.

In order to solve the above problem, one aspect of the present invention includes a time-series data acquisition unit that acquires time-series data indicating changes over time in variables related to work in a work object, and an abnormality that indicates a history of abnormalities related to the work. Anomalies that calculate the anomaly occurrence distance, which is the distance between the time-series data and the time when the anomaly occurred, based on the anomaly information acquisition unit that acquires information, the time-series data, and the anomaly information. Monitoring statistic type that determines the type of statistic to be monitored in the time-series data based on the relationship between the occurrence distance calculation unit, the statistic for each time-series data, and the abnormality occurrence distance. It is a monitoring condition determination device including a determination unit.

Further, one aspect of the present invention is the monitoring condition determination device described above, wherein the monitoring statistic type determination unit is based on the degree of correlation between the statistic for each time series data and the abnormality occurrence distance. The statistic having the highest degree of correlation may be determined for the monitoring statistic type.

Further, one aspect of the present invention is the monitoring condition determination device described above, wherein the monitoring statistic type determination unit is based on the degree of correlation between the statistic for each time series data and the abnormality occurrence distance. A statistic whose degree of correlation is greater than a predetermined threshold may be determined for the monitoring statistic type.

Further, in one aspect of the present invention, in the monitoring condition determining device described above, one value selected from the values of the monitoring statistic type for each time series data is determined as a monitoring threshold value which is a monitoring threshold. A monitoring threshold value determination unit may be further provided.

Further, in one aspect of the present invention, in the monitoring condition determination device described above, the monitoring threshold value determining unit arranges the values of the monitoring statistic type for each time series data according to the magnitude of the values. One value selected according to the changed order and the degree of correlation may be determined as the monitoring threshold value.

Further, in one aspect of the present invention, the time-series data acquisition unit acquires time-series data indicating changes in variables related to the work in the work object over time, and the abnormality information acquisition unit indicates the history of the abnormality related to the work. Abnormality information is acquired, and the abnormality occurrence distance calculation unit calculates the abnormality occurrence distance, which is the distance between the time-series data and the time when the abnormality occurs, based on the time-series data and the abnormality information. After calculation, the monitoring statistic type determination unit determines the monitoring statistic type indicating the type of statistic to be monitored in the time-series data based on the relationship between the statistic for each time-series data and the abnormality occurrence distance. This is a monitoring condition determination method characterized by the fact that

Further, one aspect of the present invention is to cause a computer to acquire time-series data acquisition means for acquiring time-series data indicating changes in variables related to work in a work object over time, and abnormality information indicating a history of abnormalities related to the work. Anomaly information acquisition means, an anomaly occurrence distance calculation means for calculating an anomaly occurrence distance, which is a distance between the time series data and the time when the anomaly occurred, based on the time series data and the anomaly information. It functions as a monitoring statistic type determining means for determining a monitoring statistic type indicating the type of statistic to be monitored in the time series data based on the relationship between the statistic for each time series data and the abnormality occurrence distance. It is a program for.

According to the present invention, it is possible to quantitatively determine a statistic that can show a peculiar change that occurs before an abnormality occurs in the time series data to be monitored.

It is a block diagram which shows the configuration example of the monitoring system 1 of embodiment. It is a block diagram which shows the structural example of the monitoring condition determination apparatus 50 of embodiment. It is a figure explaining the alarm occurrence distance of an embodiment. It is a figure which shows the structural example of the abnormality occurrence distance information 550 of embodiment. It is a figure which shows the structural example of the statistic information 551 of an embodiment. It is a figure which shows the structural example of the correlation coefficient information 552 of embodiment. It is a figure which shows the structural example of the monitoring statistic type information 553 of embodiment. It is a figure which shows the structural example of the monitoring threshold value information 554 of an embodiment. It is a sequence diagram which shows the flow of the process performed by the monitoring system 1 of an embodiment. It is a figure explaining the modification of embodiment.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

The monitoring system 1 is a system applied in the process of performing work such as processing at a manufacturing site or the like. FIG. 1 is a block diagram showing a configuration example of the monitoring system 1 of the embodiment. The monitoring system 1 includes, for example, a work object 10, a work device 20, an abnormality detection device 30, a monitoring condition determination device 50, a monitoring device 60, and an update device 70. Each component of the monitoring system 1 (work object 10, work device 20, abnormality detection device 30, monitoring condition determination device 50, monitoring device 60, and update device 70) is communicably connected via the network 80. .. Alternatively, each component of the monitoring system 1 may be communicatively connected to each other by a communication cable or the like.

The work object 10 is a material to be worked on in the work of manufacturing, for example, metal in the work of manufacturing screws, wood in the work of manufacturing chairs, and the like. The work here is some kind of work performed on the work object 10, for example, cutting, crimping, shaving, making a hole, and polishing.

The work device 20 is a device that performs work on the work object 10, and is, for example, a cutting device, a crimping device, a polishing device, and the like. The working device 20 includes a sensor unit 21. The sensor unit 21 measures variables related to work by the work device 20. The variables here are, for example, the temperature, humidity, voltage, torque, rotation speed of the shaft, etc., which are measured in association with the work by the work apparatus 20. The working device 20 outputs time-series data indicating changes over time of variables measured by the sensor unit 21.

The abnormality detection device 30 detects an abnormality in the process of performing the work, and outputs information indicating the result of detecting the abnormality. The abnormality here is that the working device 20 is in a state different from the state in which it is normally operated. For example, the temperature inside the device rises, parts need to be replaced, or the work device 20 needs to be replaced. The product manufactured by performing the work on the work object 10 becomes a defective product. For example, the abnormality detection device 30 detects an abnormality based on the relationship between the amount of current flowing through the work device 20 and the amount of drive in a drive unit such as a motor. Alternatively, the abnormality detecting device 30 may detect the abnormality based on the sound, vibration, temperature, or the like of the driving unit. In this case, the abnormality detection device 30 outputs alarm information indicating whether or not an abnormality has occurred in the work device 20 as information indicating the result of detecting the abnormality. The alarm information is an example of "abnormality information".

For example, the abnormality detection device 30 detects a defective product by inspecting the quality of the work object 10 after the work by the work device 20 is performed by irradiating the work object 10 with X-rays or the like. Alternatively, the abnormality detection device 30 may detect a defective product based on information indicating an inspection result or the like input by the inspector. In this case, the abnormality detection device 30 outputs non-defective product information indicating whether or not a defective product has occurred as information indicating the result of detecting the abnormality. Non-defective product information is an example of "abnormality information".

The monitoring condition determination device 50 determines the conditions for monitoring a peculiar change before an abnormality occurs. The peculiar change here is like a sign before the abnormality occurs, and it is considered that the probability that the abnormality will occur increases if the peculiar change is left untreated. Is. By capturing the peculiar changes that occur before the abnormality actually occurs, it is possible to prevent the abnormality.

In the present embodiment, it is premised that the time series data is selected in advance for each abnormality detected by the abnormality detecting device 30. That is, it is premised that it is predetermined that the time-series data of the current value is to be monitored for a specific abnormality of the working device 20. The monitoring condition determination device 50 determines in advance which statistic to monitor with respect to the time series data. The statistic here is an index showing the tendency and characteristics of the behavior of time-series data in a certain time division, and is, for example, the maximum value, the minimum value, the average value, the median value, the first quartile, and the third. Quartile, standard deviation, kurtosis, skewness, etc. In the following description, the type of statistic to be monitored in the time series data is referred to as "monitoring statistic type".

The monitoring condition determination device 50 determines the monitoring statistic type based on the time series data accumulated in the past fixed period and the history of alarm information and non-defective product information. The monitoring condition determination device 50 determines a threshold value (hereinafter, referred to as a monitoring threshold value) in the monitoring statistic type. The monitoring condition determination device 50 outputs information indicating the determined monitoring statistic type and the monitoring threshold value. The method by which the monitoring condition determination device 50 determines the monitoring statistic type and the monitoring threshold value will be described in detail later.

The monitoring device 60 monitors the presence or absence of a peculiar change before the abnormality by using the monitoring statistic type determined by the monitoring condition determining device 50 and the monitoring threshold value. The monitoring device 60 includes, for example, an acquisition unit 61, a calculation unit 62, a comparison unit 63, an output unit 64, and a storage unit 65. The acquisition unit 61 acquires time series data. The calculation unit 62 calculates the statistic corresponding to the monitoring statistic type in the time series data. The comparison unit 63 compares the calculated statistic with the monitoring threshold. The output unit 64 outputs information indicating the presence or absence of a peculiar change before the abnormality, which is determined based on the comparison result, as a monitoring result. The storage unit 65 stores the monitoring statistic type, the monitoring threshold value, and the like.

The update device 70 updates the monitoring statistic type and the monitoring threshold value determined by the monitoring condition determination device 50. The update device 70 includes, for example, a determination unit 71, an output unit 72, and a storage unit 73. The determination unit 71 determines whether or not to update the monitoring statistic type and the monitoring threshold value. For example, when a certain period (for example, one week, one month, or one year) has passed since the monitoring statistic type and the monitoring threshold were determined, the determination unit 71 determines the monitoring statistic type and the monitoring threshold. Is determined to be updated. When the determination unit 71 determines that the monitoring statistic type and the monitoring threshold value are to be updated, the output unit 72 outputs information (update instruction) for instructing the update. The storage unit 73 stores information indicating the monitoring statistic type and the date and time when the monitoring threshold is determined.

Based on the update instruction, the monitoring condition determination device 50 updates the monitoring statistic type and the monitoring threshold value. At this time, when the monitoring condition determination device 50 is accumulated for a certain period of time (for example, one week, one month, or one year) after determining the monitoring statistic type and the monitoring threshold value last time, for example. The monitoring statistic type and the monitoring threshold are updated based on the series data and the history of alarm information and non-defective product information.

FIG. 2 is a block diagram showing a configuration example of the monitoring condition determination device 50 of the embodiment. The monitoring condition determination device 50 includes, for example, an acquisition unit 51, a calculation unit 52, an update control unit 53, an output unit 54, and a storage unit 55.

The acquisition unit 51 includes, for example, a time-series data acquisition unit 510, an abnormality information acquisition unit 511, and an update instruction acquisition unit 512. The time-series data acquisition unit 510 acquires the time-series data accumulated in the past fixed period, and outputs the acquired time-series data to the calculation unit 52. The abnormality information acquisition unit 511 acquires the history of alarm information and non-defective product / defective product information accumulated in the past fixed period, and outputs the acquired time-series data to the calculation unit 52. The update instruction acquisition unit 512 acquires the update instruction and outputs the acquired update instruction to the update control unit 53.

The calculation unit 52 includes, for example, an abnormality occurrence distance calculation unit 520, a statistic calculation unit 521, a monitoring statistic type determination unit 522, and a monitoring threshold value determination unit 523.

The abnormality occurrence distance calculation unit 520 calculates the abnormality occurrence distance. The abnormality occurrence distance is an index showing the time interval between the time series data and the occurrence of the abnormality. For example, when the time when the time series data is measured and the time when the abnormality occurs are separated from each other, the abnormality occurrence distance is an index showing a large value as compared with the case where the abnormality occurs.

In the present embodiment, the abnormality occurrence distance calculation unit 520 calculates the alarm occurrence distance and the defective product occurrence distance as the abnormality occurrence distance. The alarm occurrence distance is the distance between the time series data and the alarm occurrence in time and space. The defective product occurrence distance is the distance between the time series data and the occurrence of defective products in time and space. That is, the alarm generation distance is an example of the “abnormality generation distance”. The defective product occurrence distance is an example of the “abnormality occurrence distance”.

In the following, a case where the abnormality occurrence distance calculation unit 520 calculates the alarm generation distance as the abnormality occurrence distance will be described as an example. The defective product generation distance can also be calculated by applying the same method as the method for calculating the alarm generation distance.

The alarm generation distance will be described with reference to FIG. FIG. 3 is a diagram for explaining the alarm generation distance of the embodiment. Waveforms W1 to W10 in FIG. 3 show changes over time in measured values such as the current measured per work by the work device 20. That is, the waveform is an example of "time series data".

The alarm occurrence shown in FIG. 3 indicates that an alarm has occurred due to an abnormality occurring in the work device 20 or the like. In the example of FIG. 3, it is shown that in the work of one day, the waveform W1 was acquired in the first work, the waveform W2 was acquired in the second work, and the waveform W3 was acquired in the third work. .. After that, it indicates that an alarm was generated in the fourth work. It shows that the waveform W4 was acquired in the fifth work after the countermeasure was taken against the alarm occurrence of the work device 20. That is, in this example, the waveform corresponding to the work performed without the alarm occurring and the relationship with the timing at which the alarm occurs are shown in chronological order.

As shown in FIG. 3, the abnormality occurrence distance calculation unit 520 has generated an abnormality with a waveform corresponding to the work performed without generating an alarm based on the history of the waveform for each work and the alarm information. Extract the relationship with timing in chronological order. The abnormality occurrence distance calculation unit 520 calculates the alarm generation distance according to the time-series relationship between the extracted waveform and the timing at which the abnormality occurs. The alarm generation distance is the distance between the waveform for each work and the alarm generated after that in time and space. For example, when one waveform is used as the unit of the alarm generation distance, the alarm generation distance s (W1) of the waveform W1 is “3”. This means that an alarm was generated during the third operation after the operation corresponding to the waveform W1 was performed. Applying the same concept, the alarm generation distance s (W2) of the waveform W2 is “2”, and the alarm generation distance s (W3) of the waveform W3 is “1” (see FIG. 4).

Further, the alarm generation distance s (W4) of the waveform W4 is “6”. This means that an alarm was generated during the sixth work after the work corresponding to the waveform W4 was performed. Similarly, the alarm generation distance s (W5) of the waveform W5 is "5", the alarm generation distance s (W6) of the waveform W6 is "4", ..., The alarm generation distance s (W9) of the waveform W9 is "1". .. The alarm generation distance s (W10) of the waveform W10 is not calculated (see FIG. 4). This is because in this example, it is not possible to know when the alarm has occurred after the work corresponding to the waveform W10 has been performed.

The abnormality occurrence distance calculation unit 520 stores the calculated alarm generation distance in the abnormality occurrence distance information 550 of the storage unit 55. The abnormality occurrence distance information 550 is information in which the waveform ID and the alarm generation distance s are associated with each other. The waveform ID is identification information that uniquely identifies the waveform (time series data).

Returning to FIG. 2, the statistic calculation unit 521 calculates the statistic of the time series data. The statistic calculation unit 521 calculates a statistic for each of the waveforms W1 to Wn (n is a natural number of 2 or more), for example. For example, the statistic calculation unit 521 calculates the maximum values max (W1) to max (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the minimum values min (W1) to min (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the mean values mean (W1) to mean (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the median values med (W1) to med (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the first quartiles q1 (W1) to q1 (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the third quartiles q3 (W1) to q3 (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the standard deviations std (W1) to std (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the kurtosis (W1) to kurt (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 calculates the skewness skew (W1) to skew (Wn) of the waveforms W1 to Wn. The statistic calculation unit 521 may calculate a statistic of a type different from the above-mentioned statistic (for example, the result of frequency conversion). The statistic calculation unit 521 stores the calculated statistic in the storage unit 55 as statistic information 551. The statistic information 551 is information in which the waveform ID and the statistic are associated with each other.

The monitoring statistic type determination unit 522 determines the monitoring statistic type based on the degree of correlation between the statistic for each waveform and the alarm generation distance for each waveform. The degree of correlation is an index showing the strength of the relationship between the two variables over time, and is, for example, a correlation coefficient. The correlation coefficient is an index indicated by a real value from -1 to +1. When the correlation coefficient is positive (positive correlation), the larger the correlation coefficient, the stronger the tendency for both variables to synchronize compared to the case where the correlation coefficient is small. For example, when one variable increases. The other variable also increases, and when one variable decreases, the other variable also tends to decrease. When the correlation coefficient is negative (negative correlation), the larger the absolute value of the correlation coefficient, the stronger the tendency for both variables to change in opposite directions compared to the case where the absolute value of the correlation coefficient is small. Become. For example, when one variable increases, the other variable tends to decrease, and when one variable decreases, the other variable tends to increase.

The monitoring statistic type determination unit 522 calculates the correlation coefficient between the alarm generation distance and the statistic for each waveform. For example, the monitoring statistic type determination unit 522 calculates the correlation coefficient r (max, s) between the maximum values max (W1) to max (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (min, s) between the minimum values min (W1) to min (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (mean, s) between the mean values mean (W1) to mean (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (med, s) between the median med (W1) to med (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (q1, s) between the first quartile q1 (W1) to q1 (Wn) and the alarm generation distances s (W1) to s (Wn). To do. The monitoring statistic type determination unit 522 calculates the correlation coefficient r (q3, s) between the third quartile q3 (W1) to q3 (Wn) and the alarm generation distances s (W1) to s (Wn). To do. The monitoring statistic type determination unit 522 calculates the correlation coefficient r (std, s) between the standard deviations std (W1) to std (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (curt, s) between the kurtosis kurt (W1) to kurt (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 calculates the correlation coefficient r (skew, s) between the skewnesses skew (W1) to skew (Wn) and the alarm generation distances s (W1) to s (Wn). The monitoring statistic type determination unit 522 stores the calculated correlation coefficient in the storage unit 55 as the correlation coefficient information 552. The correlation coefficient information 552 is information indicating the correlation coefficient between various statistics and the alarm generation distance.

The monitoring statistic type determination unit 522 determines the monitoring statistic type based on the correlation coefficient between various statistics and the alarm occurrence distance. The monitoring statistic type determination unit 522 sets the type of the statistic corresponding to the correlation coefficient having the largest absolute value among the correlation coefficients between various statistics and the alarm occurrence distance as the monitoring statistic type. .. For example, the correlation coefficients r (max, s), r (min, s), r (med, s), r (q1, s), r (q3, s), r (std, s), r (curt). , S), r (skew, s), when the absolute value of the correlation coefficient r (std, s) is the largest, the standard deviation std is set as the monitoring statistic type. The monitoring statistic type determination unit 522 stores the determined information indicating the monitoring statistic type in the storage unit 55 as the monitoring statistic type information 553. The monitoring statistic type information 553 is information regarding the monitoring statistic type.

The monitoring statistic type determination unit 522 may set the lower limit value of the correlation coefficient for the statistic as the monitoring statistic type. Specifically, the monitoring statistic type determination unit 522 determines that the absolute value of each of the correlation coefficients between various statistics and the alarm generation distance is larger than a predetermined threshold value (for example, 0.5). It may be extracted as a candidate for the monitoring statistic type. When the absolute value is smaller than the predetermined threshold, there is not much strong correlation between both variables. If the largest correlation coefficient among the absolute values of the correlation coefficients between various statistics and the alarm occurrence distance is smaller than a predetermined threshold, any of the statistics of the time series data will generate an alarm. The correlation with the distance is weak. In such a case, the monitoring statistic type determination unit 522 may determine that there is no statistic to be monitored.

The monitoring threshold value determination unit 523 determines the threshold value to be monitored (monitoring threshold value) in the monitoring statistic type of the time series data. The monitoring threshold value determination unit 523 sorts (sorts) the statistics corresponding to the monitoring statistic type according to the sign of the correlation coefficient. The monitoring threshold value determination unit 523 sorts the statistics corresponding to the monitoring statistic type in ascending order when the sign of the correlation coefficient is + (plus), that is, when the correlation is positive. In the case of positive correlation, the smaller the statistic, the smaller the alarm generation distance. In this case, it is highly likely that an alarm will occur (that is, an abnormality will occur) during the work following the work corresponding to the first statistic after sorting. Similarly, after the work corresponding to the k-th statistic after sorting has been performed, it is highly probable that an abnormality will occur during the k-th work. However, k is an arbitrary integer from 2 to n.

When the sign of the correlation coefficient is − (minus), that is, when the correlation coefficient is negative, the monitoring threshold value determination unit 523 sorts the statistics corresponding to the monitoring statistic type in descending order. In the case of negative correlation, the larger the statistic, the smaller the alarm generation distance. In this case, there is a high possibility that an alarm will occur (that is, an abnormality will occur) after the work corresponding to the first statistic after sorting. In the same manner thereafter, it is considered that there is a high possibility that an abnormality will occur while the k-th work is being performed after the work corresponding to the k-th statistic after sorting. The monitoring threshold value determination unit 523 stores information indicating the statistics after sorting according to the sign of the correlation coefficient in the storage unit 55 as the monitoring statistic type information 553.

The monitoring threshold value determination unit 523 refers to the setting information included in the monitoring threshold value information 554. In the setting information, the timing of issuing an alarm as a monitoring result is set. For example, a value corresponding to the alarm generation distance is set in the setting information. In this case, if "10" is set in the setting information, an alarm is issued at the timing that there is a high possibility that an abnormality will occur when the work is performed 10 times. The setting information may be set to any value. For example, as setting information, a time sufficiently possible for the work manager to rush to the installation location of the work device 20 is set. By doing so, the work manager can rush after the warning is issued and before the actual abnormality occurs, and measures such as stopping the work device 20 and replacing worn parts are taken. It becomes possible.

The monitoring threshold value determination unit 523 selects a statistic according to the value shown in the setting information from the sorted statistic included in the monitoring statistic type information 553. For example, when "10" is set in the setting information, the monitoring threshold value determination unit 523 selects the statistic located at the 10th position among the sorted statistic. The monitoring threshold value determination unit 523 determines the selected statistic as the monitoring threshold value.

The monitoring threshold value determination unit 523 sets the type of the determined monitoring threshold value, that is, whether the monitoring threshold value is used as the upper limit value or the lower limit value of monitoring. The monitoring threshold value determination unit 523 sets the monitoring threshold value to the lower limit value of monitoring when the sign of the correlation coefficient in the monitoring statistic type is + (plus). In this case, an alarm is issued when the statistic in the time series data is smaller than the monitoring threshold value. The monitoring threshold value determination unit 523 sets the monitoring threshold value to the upper limit value of monitoring when the sign of the correlation coefficient in the monitoring statistic type is − (minus). In this case, an alarm is issued when the statistic in the time series data is larger than the monitoring threshold value. The monitoring threshold value determination unit 523 stores the determined monitoring threshold value and the type in the monitoring threshold value information 554 of the storage unit 55.

The output unit 54 outputs the monitoring statistic type and the monitoring threshold value determined by the calculation unit 52.
The update control unit 53 updates the monitoring statistic type and the monitoring threshold value based on the update instruction. When the update control unit 53 acquires the update instruction, for example, the acquisition unit 51 is made to acquire the actual result (time series data for a certain period and the abnormality history information) used for the update, and the calculation unit 52 is made to acquire the monitoring statistics based on the actual result. Let the type and monitoring threshold be determined.
The storage unit 55 stores the abnormality occurrence distance information 550, the statistic information 551, the correlation coefficient information 552, the monitoring statistic type information 553, and the monitoring threshold information 554.

FIG. 4 is a diagram showing a configuration example of the abnormality occurrence distance information 550 of the embodiment. As shown in FIG. 4, the abnormality occurrence distance information 550 includes items of the waveform ID and the alarm generation distance s. The waveform ID is identification information that uniquely identifies the waveform (time series data). The alarm generation distance s indicates the alarm generation distance with respect to the waveform corresponding to the waveform ID. In this example, the abnormality occurrence distance information 550 calculated from the time series data and the alarm information shown in FIG. 3 is shown.

FIG. 5 is a diagram showing a configuration example of the statistic information 551 of the embodiment. As shown in FIG. 5, the statistic information 551 includes items of waveform ID and statistic. The waveform ID is identification information that uniquely identifies the waveform (time series data). In the statistic, various statistics (maximum value max, minimum value min, ... In this example) are shown for each waveform.

FIG. 6 is a diagram showing a configuration example of the correlation coefficient information 552 of the embodiment. As shown in FIG. 6, the correlation coefficient information 552 includes an item of the correlation coefficient. The correlation coefficient shows the correlation coefficient between various statistics (in this example, the maximum value max, the minimum value min, ...) And the alarm occurrence distance.

FIG. 7 is a diagram showing a configuration example of the monitoring statistic type information 553 of the embodiment. As shown in FIG. 7, the monitoring statistic type information 553 includes items of data ID, monitoring statistic type, correlation, sorting direction, and monitoring threshold candidate. The data ID indicates identification information that uniquely identifies the type of time-series data to be monitored. The monitoring statistic type indicates the monitoring statistic type determined by the monitoring statistic type determination unit 522. The correlation indicates whether the correlation between the statistic corresponding to the monitoring statistic type and the alarm occurrence distance is a positive correlation or a negative correlation. The sort direction indicates whether the order in which the statistics are sorted is descending or ascending. In the monitoring candidate threshold, the sorted statistics are shown in order.

FIG. 8 is a diagram showing a configuration example of the monitoring threshold value information 554 of the embodiment. As shown in FIG. 8, the monitoring threshold information 554 includes items of a set number of times, a monitoring threshold, and a type. The set number of times indicates how many more operations should be performed before issuing a warning at the timing that an abnormality may occur. The monitoring threshold value indicates a threshold value corresponding to the set number of times. The type indicates whether the monitoring threshold value is the lower limit threshold value indicating the lower limit value or the upper limit threshold value indicating the upper limit value.

FIG. 9 is a sequence diagram showing a flow of processing performed by the monitoring system 1 of the embodiment. The process shown in steps S10 to S18 is a monitoring condition determination phase in which the monitoring statistic type and the monitoring threshold value are determined based on the time-series data and the alarm information for a certain period, which are past results (reference numeral 100). The processes shown in steps S19 to S20 are monitoring phases in which monitoring is performed using the monitoring statistic type and monitoring threshold value determined in the monitoring condition determination phase (reference numeral 200).

First, the flow of processing in the monitoring condition determination phase will be described.
The monitoring condition determination device 50 acquires time-series data and alarm information for a certain period of time (steps S10 and S11).
The monitoring condition determination device 50 calculates the alarm generation distance for each waveform (time series data) using the time series data and the alarm information in a certain period (step S12).
The monitoring condition determination device 50 calculates various statistics for each waveform (step S13).
The monitoring condition determination device 50 calculates the correlation coefficient between the statistic and the alarm generation distance for each of the various statistics (step S14).
The monitoring condition determination device 50 determines the monitoring statistic type from the correlation coefficient for each of the various statistics (step S15), and outputs the determined monitoring statistic type to the monitoring device 60. The monitoring device 60 acquires the monitoring statistic type and stores the acquired monitoring statistic type (step S16).
The monitoring condition determination device 50 determines the monitoring threshold value by rearranging the statistics corresponding to the monitoring statistic type (step S17), and outputs the determined monitoring threshold value to the monitoring device 60. The monitoring device 60 acquires the monitoring threshold value and stores the acquired monitoring threshold value (step S18).

Next, the processing flow in the monitoring phase will be described.
The monitoring device 60 acquires time series data (step S19). The monitoring device 60 performs monitoring processing based on the time series data (step S20). In the monitoring process, the monitoring device 60 calculates the value of the monitoring statistic type in the time series data. The monitoring device 60 compares the calculated statistic with the monitoring threshold. When the monitoring threshold is the lower limit threshold, the monitoring device 60 outputs a monitoring result to warn if the statistic is smaller than the monitoring threshold. When the monitoring threshold is the upper limit threshold, the monitoring device 60 outputs a monitoring result to warn if the statistic is larger than the monitoring threshold.

The use case of the embodiment will be described with reference to FIG. FIG. 10 is a diagram illustrating a use case of the embodiment. As shown in FIG. 10, the work object 10 includes a work WK and a pallet PT on which the pole PL is arranged. The work WK penetrates, for example, a cylindrical portion, a disk portion having the same central axis as the central axis of the cylinder and being formed so as to be continuous from the lower end portion of the cylinder, and the cylindrical portion and the disk portion along the central axis. It is a component having a through portion KT in which a hole is formed as described above. The work WK is stored on the pallet PT by penetrating the work WK in a skewered manner so that the central axis of the penetrating portion KT is along the central axis of the pole PL. Here, a plurality of work WKs are stored on the pallet PT so as to overlap in the height direction.

The work device 20 is a picking device that grips and takes out the work WK stored on the pallet PT by the servo hand SH. For example, the working device 20 moves the servo hand SH relatively directly above the pole PL, and then lowers the servo hand SH to grip the work WK. The work device 20 takes out the work WK by lifting the servo hand SH holding the work WK to a position higher than the upper end of the pole PL while holding the work WK.

In such picking work, the servo hand SH is repeatedly raised and lowered, the shape of the inner peripheral surface of the penetrating portion KT is irregular, and the coordinates of the picking position of the servo hand SH and the central axis of the pole PL are The inner peripheral surface of the penetrating portion KT and the side surface (outer peripheral surface) of the pole PL may rub against each other due to an error or the like that occurs between the coordinates. In such a case, since the work WK or the pole PL may be scratched, an alarm is generated by the abnormality detecting device 30, and the raising and lowering of the servo hand SH is stopped. When the penetrating portion KT and the side surface of the pole PL rub against each other, the current value flowing through the elevating motor for raising and lowering the servo hand SH becomes larger than that when the servo hand SH does not rub. This is because the friction generated by the rubbing acts in a direction that hinders the raising and lowering of the servo hand. In such a use case, it is said that the current value flowing through the elevating motor gradually changes to a large current value from the state where there is no rubbing until a large rubbing occurs so that the raising and lowering of the servo hand SH is stopped. Conceivable. In order to capture such a change related to rubbing, what kind of statistic of the current value is monitored is determined by the monitoring condition determining device 50.

In such a use case, the sensor unit 21 measures a change over time (hereinafter, referred to as a current waveform) of the current value flowing through the elevating motor as time-series data closely related to rubbing. The abnormality detection device 30 outputs alarm information when the raising and lowering of the servo hand SH is stopped. The monitoring condition determination device 50 calculates the alarm generation distance for each current waveform, and calculates the correlation coefficient r with the statistic for each current waveform. The monitoring condition determination device 50 determines what kind of statistic to monitor in the current value (monitoring statistic type) based on the magnitude of the correlation coefficient r. As a result, it is possible to monitor a statistic that has a strong correlation with the alarm generation distance, and it is possible to accurately grasp the tendency of rubbing. Therefore, it is possible to prevent the servo hand SH from being stopped by taking measures such as correcting the bending of the pole before a large rubbing occurs.

As described above, the monitoring condition determination device 50 of the embodiment includes a time series data acquisition unit 510, an abnormality information acquisition unit 511, an abnormality occurrence distance calculation unit 520, and a monitoring statistic type determination unit 522. The time-series data acquisition unit 510 acquires a waveform (an example of "time-series data" indicating a change over time of variables) related to the work on the work object 10. The abnormality information acquisition unit 511 acquires alarm information (an example of "abnormality information") indicating a history of abnormalities related to work. The abnormality occurrence distance calculation unit 520 calculates an alarm occurrence distance (an example of "abnormality occurrence distance"), which is a distance between the waveform and the time when the alarm is generated, based on the waveform and the alarm information. The monitoring statistic type determination unit 522 determines the monitoring statistic type indicating the type of the statistic to be monitored in the waveform based on the relationship between the statistic in the waveform and the alarm generation distance. As a result, in the monitoring condition determination device 50 of the embodiment, the monitoring statistic type can be determined based on the alarm generation distance for each statistic, so that a peculiar change that occurs before an abnormality occurs in the time series data. It is possible to quantitatively determine the statistics that can indicate.

Further, in the monitoring condition determination device 50 of the embodiment, the monitoring statistic type determination unit 522 determines the statistic having the largest correlation coefficient r based on the correlation coefficient r between the statistic for each waveform and the alarm occurrence distance. , Determine the monitoring statistic type. As a result, the monitoring condition determination device 50 of the embodiment can determine a statistic having a strong correlation with the alarm generation distance for the monitoring statistic type, and has the same effect as the above-mentioned effect.
Further, in the monitoring condition determination device 50 of the embodiment, the monitoring statistic type determination unit 522 has a phase relationship based on the correlation coefficient r (an example of “correlation degree”) between the statistic for each waveform and the alarm generation distance. A statistic whose number r is larger than a predetermined threshold is determined as a monitoring statistic type. As a result, in the monitoring condition determination device 50 of the embodiment, the monitoring statistic type can be determined by a simple method of comparing the correlation coefficient and the threshold value.
Further, the monitoring condition determination device 50 of the embodiment further includes a monitoring threshold value determination unit 523 that determines one value selected from the values of the monitoring statistic type for each waveform as the monitoring threshold value. As a result, in the monitoring condition determination device 50 of the embodiment, the monitoring threshold value can be determined by a simple method of selecting from the values of the monitoring statistic type.
Further, in the monitoring condition determination device 50 of the embodiment, the monitoring threshold value determination unit 523 sets the values of the monitoring statistic type for each waveform into the order in which the values are rearranged according to the magnitude of the values and the correlation coefficient r. One value selected accordingly is determined as the monitoring threshold. As a result, the monitoring condition determination device 50 quantitatively determines the monitoring threshold value in descending order if the correlation coefficient is positive, counting in ascending order if the correlation coefficient is negative, and so on. be able to.

(Modified example of the embodiment)
In the above-described embodiment, the case where one waveform is acquired corresponding to one operation has been described as an example, but a plurality of operations may be performed in parallel in the manufacturing process. For example, when a plurality of raw materials are continuously supplied and agitated, a product is produced while adjusting the temperature, humidity, direction and speed of agitation.

When such a plurality of operations are performed in parallel, it is difficult to separate the start and end of the waveform because the start and end of each operation are not clear. In such a case, it becomes difficult to calculate the alarm generation distance with the time from the start to the end of one waveform as the unit of the alarm generation distance as described in the above-described embodiment. In such a case, for example, by using any of the following as a waveform delimiter, the period of the waveform is defined, and the alarm generation distance is calculated with the time corresponding to the period as the unit of the alarm generation distance.

(1) The waveform is divided by the periodic operation of the work. For example, the waveform cycle is defined based on the opening / closing cycle of the door that supplies the raw material, the rotation cycle of the stirring parts, and the like.
(2) Divide the waveform according to the time. For example, the waveform is defined for each predetermined unit time such as 1 minute.
(3) The waveform is divided in units of variables (for example, volume) that change according to the work. For example, in the case of continuously adding a liquid raw material, the waveform is defined every time the total volume increases by a unit volume (for example, 1 liter).
Which of the above (1) to (3) to specify the waveform may be arbitrarily selected according to the situation of the manufacturing process. For example, it may be selected depending on the reason that the calculation in the latter stage is easy, the waveform is easy to measure, or the measurement is easy.

All or part of the monitoring condition determination device 50 in the above-described embodiment may be realized by a computer. In that case, the program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by the computer system and executed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM, or a storage device such as a hard disk built in a computer system. Further, a "computer-readable recording medium" is a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line, and dynamically holds the program for a short period of time. In that case, a program may be held for a certain period of time, such as a volatile memory inside a computer system serving as a server or a client. Further, the above program may be for realizing a part of the above-mentioned functions, and may be further realized for realizing the above-mentioned functions in combination with a program already recorded in the computer system. It may be realized by using a programmable logic device such as FPGA (Field Programmable Gate Array).

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs and the like within a range that does not deviate from the gist of the present invention.

1 Monitoring system 10 Work object 20 Work device 21 Sensor unit 30 Abnormality detection device 50 Monitoring condition determination device 51 Acquisition unit 510 Time series data acquisition unit 511 Abnormality information acquisition unit 512 Update instruction acquisition unit 52 Calculation unit 520 Abnormality occurrence distance calculation unit 521 Statistic calculation unit 522 Monitoring statistic type determination unit 523 Monitoring threshold determination unit 53 Update control unit 54 Output unit 55 Storage unit 550 Abnormality occurrence distance information 551 Statistics information 552 Correlation coefficient information 552 Monitoring statistic type information 554 Monitoring threshold Information 60 Monitoring device 70 Update device

Claims (7)

A time-series data acquisition unit that acquires time-series data indicating changes over time in variables related to work in the work object,
An abnormality information acquisition unit that acquires abnormality information indicating the history of abnormalities related to the work, and
Based on the time-series data and the abnormality information, an abnormality occurrence distance calculation unit that calculates an abnormality occurrence distance, which is a distance between the time-series data and the time when the abnormality occurred,
Based on the relationship between the statistic for each time-series data and the abnormality occurrence distance, a monitoring statistic type determination unit that determines a monitoring statistic type indicating the type of statistic to be monitored in the time-series data, and a monitoring statistic type determination unit.
A monitoring condition determining device comprising. The monitoring statistic type determination unit determines the statistic having the highest correlation degree as the monitoring statistic type based on the degree of correlation between the statistic for each time series data and the abnormality occurrence distance.
The monitoring condition determination device according to claim 1. The monitoring statistic type determination unit determines a statistic whose degree of correlation is greater than a predetermined threshold as the monitoring statistic type, based on the degree of correlation between the statistic for each time series data and the abnormality occurrence distance. To do,
The monitoring condition determining device according to claim 1 or 2, wherein the monitoring condition is determined. A monitoring threshold value determining unit for determining one value selected from the values of the monitoring statistic type for each time series data as a monitoring threshold value, which is a monitoring threshold value, is further provided.
The monitoring condition determination device according to claim 2 or 3, wherein the monitoring condition determination device is characterized. The monitoring threshold value determination unit selects one value selected according to the order in which the values of the monitoring statistic type for each time series data are rearranged according to the magnitude of the value and the degree of correlation. Determine the monitoring threshold,
The monitoring condition determining apparatus according to claim 4. The time-series data acquisition unit acquires time-series data indicating changes in variables related to work in the work object over time.
The abnormality information acquisition unit acquires the abnormality information indicating the history of the abnormality related to the work, and obtains the abnormality information.
The abnormality occurrence distance calculation unit calculates the abnormality occurrence distance, which is the distance between the time series data and the time when the abnormality occurs, based on the time series data and the abnormality information.
The monitoring statistic type determination unit determines the monitoring statistic type indicating the type of the statistic to be monitored in the time series data based on the relationship between the statistic for each time series data and the abnormality occurrence distance.
A method for determining monitoring conditions. Computer,
Time-series data acquisition means for acquiring time-series data indicating changes over time in variables related to work in a work object,
Abnormality information acquisition means for acquiring abnormal information indicating a history of abnormalities related to the work
An abnormality occurrence distance calculation means for calculating an abnormality occurrence distance, which is a distance between the time series data and the time when the abnormality occurred, based on the time series data and the abnormality information.
A monitoring statistic type determining means for determining a monitoring statistic type indicating the type of statistic to be monitored in the time series data based on the relationship between the statistic for each time series data and the abnormality occurrence distance.
A program to function as.

Images