Normal range deviation diagnosis device, normal range deviation diagnosis system, normal range deviation diagnosis method, and program

Abstract

A normal range deviation diagnosis device (1) comprises: a data acquisition unit (11) that acquires data indicating the latest value of a feature amount related to an external device (3); a data waveform learning unit (13) that extracts a data waveform, which is a predetermined number of consecutive data points, from feature amount data, which is time series data in which data indicating the feature amount related to the external device (3) is accumulated, performs learning that aggregates similar data waveforms, and generates a predicted waveform which is an aggregated data waveform; a future data estimation unit (14) that uses the predicted waveform and the closest feature amount data to estimate future data that is the value of future feature amount data; a normal range definition unit (15) that defines a normal range for each value of the future data; and a normal range deviation determination unit (16) that determines whether the state of the external device (3) has deviated from the normal range on the basis of whether the latest value of the feature amount related to the external device (3) has deviated from the corresponding normal range.

Description

Normal range deviation diagnostic device, normal range deviation diagnostic system, normal range deviation diagnostic method, and program

[0001] The present disclosure relates to a normal range deviation diagnostic device, a normal range deviation diagnostic system, a normal range deviation diagnostic method, and a program.

[0002] In the manufacturing industry, there is a system that determines whether the state of a facility or device is outside of a normal range based on whether the measurement value of a sensor installed in the facility or device is outside of a normal range. For example, Patent Document 1 discloses a control device that acquires measurement values ​​(sensor signals and state quantities) of sensors installed in the facility or device, performs clustering processing on the state quantities to identify the state of the plant, and determines that an abnormality exists if any of the latest state quantities is outside of the normal range. The control device classifies state quantities accumulated in the past into clusters, identifies the cluster of the plant state to which a state quantity in a newly acquired sensor measurement value corresponds, and determines which state quantity is outside the normal range.

[0003] Japanese Patent Application Laid-Open No. 2023-109271

[0004] The technology described in Patent Document 1 has a problem in that if there is no past state quantity data sufficient for cluster classification, it is not possible to identify the state of the plant, and it is not possible to determine whether the state of an external device such as a facility or a machine has deviated from a normal range.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to make it possible to determine whether the state of an external device has deviated from the normal range even if there is not enough past data for cluster classification.

[0006] To achieve the above object, the normal range deviation diagnosis device according to the present disclosure includes a data acquisition unit, a data waveform learning unit, a future data estimation unit, a normal range definition unit, and a normal range deviation determination unit. The data acquisition unit acquires data indicating the latest values ​​of feature quantities related to the external device. The data waveform learning unit extracts a data waveform, which is a predetermined number of consecutive data points, from feature quantity data, which is time-series data that accumulates data indicating feature quantities related to the external device, and performs learning to aggregate similar data waveforms to generate a predicted waveform, which is the aggregated data waveform. The future data estimation unit estimates future data, which is the value of the future feature quantity data, using the predicted waveform and the most recent feature quantity data. The normal range definition unit defines a normal range for each value of the future data. The normal range deviation determination unit determines whether the state of the external device has deviated from the normal range based on whether the latest value of the feature quantity related to the external device deviates from the corresponding normal range.

[0007] According to the present disclosure, by extracting data waveforms from time-series data that accumulates data indicating feature quantities related to an external device, aggregating similar data waveforms to generate a predicted waveform, and defining a normal range for each value of future data estimated using the predicted waveform and the most recent feature quantity data, it becomes possible to determine whether the state of an external device has deviated from the normal range even if there is not enough past data for cluster classification.

[0008] FIG. 1 shows an example of the configuration of a normal range deviation diagnostic system according to embodiment 1. FIG. 2 shows an example of feature data according to embodiment 1. FIG. 3 shows an example of parameters according to embodiment 1. FIG. 4 shows an example of data waveform extraction according to embodiment 1. FIG. 5 shows an example of data waveform aggregation according to embodiment 1. FIG. 6 shows an example of extraction of a predicted waveform with the highest degree of matching according to embodiment 1. FIG. 7 shows an example of estimation of future measurement values ​​according to embodiment 1. FIG. 8 shows an example of definition of normal range according to embodiment 1. FIG. 9 shows an example of determination of whether or not there has been a deviation from the normal range according to embodiment 1. FIG. 10 shows an example of a determination of no deviation in determining whether or not there has been a deviation from the normal range according to embodiment 1. FIG. 11 shows an example of a determination of deviation in determining whether or not there has been a deviation from the normal range according to embodiment 1.

[0009] The following describes in detail the normal range deviation diagnostic device, normal range deviation diagnostic system, normal range deviation diagnostic method, and program according to the present embodiment with reference to the drawings. Note that identical or corresponding parts in the drawings are given the same reference numerals. In this embodiment, an example will be described in which it is determined whether the state of equipment or devices installed inside or outside a factory deviates from the normal range.

[0010] (Embodiment 1) The configuration of a normal range deviation diagnosis system 100 according to embodiment 1 will be described with reference to Fig. 1. The normal range deviation diagnosis system 100 includes a normal range deviation diagnosis device 1 that determines whether the state of an external device 3 deviates from a normal range, a sensor 2 that measures feature quantities related to the external device 3, and an external device 3 that is equipment or machinery installed inside or outside a factory. The external device 3 is, for example, a processing machine, a robot, a motor, or the like. In the figure, one external device 3 is shown as a representative example, but multiple external devices may be used. The normal range deviation diagnosis device 1 is capable of communicating with the sensor 2 and the external device 3 via a network. The network may be wired or wireless.

[0011] The sensor 2 and the external device 3 periodically, for example at a constant cycle, transmit data indicating the latest values ​​of the feature quantities related to the external device 3 to the normal range deviation diagnosis device 1. The latest values ​​of the feature quantities related to the external device 3 are, for example, values ​​measured immediately before the measurement values ​​of sound, vibration, temperature, etc. generated by the processing machine measured by the sensor 2, and the joint values ​​of the robot, the number of rotations of the motor, etc. measured by a measurement unit included in the external device 3. The measurement units included in the sensor 2 and the external device 3 are examples of measurement devices.

[0012] The normal range deviation diagnosis device 1 includes a data acquisition unit 11 that acquires data indicating the latest values ​​of feature quantities related to the external device 3 from the sensor 2 and the external device 3, and a data storage unit 12 that accumulates and stores the data indicating the feature quantities related to the external device 3 acquired by the data acquisition unit 11. The data indicating the feature quantities related to the external device 3 stored in the data storage unit 12 includes at least time information indicating either the measurement time or the time acquired by the data acquisition unit 11, and numerical information indicating the measurement value. Hereinafter, the time-series data accumulated with the data indicating the feature quantities related to the external device 3 stored in the data storage unit 12 will be referred to as feature quantity data.

[0013] The normal range deviation diagnosis device 1 also includes a data waveform learning unit 13 that extracts data waveforms from feature data and generates a predicted waveform by performing learning to aggregate similar data waveforms, and a future data estimation unit 14 that estimates future feature data values ​​using the predicted waveform and the most recent feature data. Hereinafter, the future feature data values ​​will be referred to as future data. The normal range deviation diagnosis device 1 also includes a normal range definition unit 15 that defines a normal range for the future data estimated by the future data estimation unit 14, and a normal range deviation determination unit 16 that compares data indicating the latest feature value for the external device 3 newly stored in the data storage unit 12 with the normal range defined by the normal range definition unit 15 and determines whether the state of the external device 3 has deviated from the normal range based on whether the latest value deviates from the normal range.

[0014] Every time the data acquisition unit 11 acquires data indicating the latest values ​​of feature quantities related to the external device 3 from the sensor 2 and the external device 3, it stores the data in the data storage unit 12. The data storage unit 12 stores feature quantity data, which is time-series data that accumulates data indicating feature quantities related to the external device 3.

[0015] The feature amount data stored in the data storage unit 12 will be described with reference to Fig. 2. In the example of Fig. 2, the feature amount data has a "measurement time" field indicating the measurement time when the feature amount was measured, and fields "measurement value A," "measurement value B," and "measurement value C" indicating the values ​​of the measured feature amount. Here, the number of measurement values ​​for one piece of time information is three, but this is not limited to this and may be one or more. Furthermore, the unit and number of digits of the measurement value are arbitrary. Hereinafter, the type of measurement value of the feature amount included in the feature amount data will be referred to as the data dimension.

[0016] Returning to Fig. 1, the data waveform learning unit 13 reads the feature data from the data storage unit 12, learns the data waveform, and generates a predicted waveform. The data waveform learning unit 13 extracts data waveforms, which are consecutive data waveform points (m), from the feature data of the maximum number of data points (N), which is the maximum number of data points stored in the data storage unit 12, and performs a data waveform learning process to aggregate similar data waveforms until the number of data waveforms becomes equal to or less than the maximum number of data waveforms (pt). The data waveform learning unit 13 stores the aggregated data waveforms that have become equal to or less than the maximum number of data waveforms (pt) in the data storage unit 12 as predicted waveforms.

[0017] As shown in FIG. 3, the data storage unit 12 stores, as parameters, the maximum number of data points (N), the number of data waveform points (m), the maximum number of data waveforms (pt), and the number of predicted points (p) for estimating future data. These parameters can be set by the user. The number of predicted points (p) is a parameter used by the future data estimation unit 14, which will be described later. The normal range deviation diagnosis device 1 may be connected to a user terminal used by the user, and the data storage unit 12 may store the values ​​of the parameters acquired from the user terminal.

[0018] In the example of FIG. 3 , the maximum number of data points (N) is 5,000. However, even if the number of data points of the feature data stored in the data storage unit 12 is less than 5,000 when the normal range deviation diagnosis device 1 is started up, the data waveform learning unit 13 may start the data waveform learning process when it becomes possible to extract data waveforms, which are the number of data waveform points (m) of consecutive data points, from the feature data of that number of data points and aggregate similar data waveforms until the number of data waveforms becomes equal to or less than the maximum number of data waveforms (pt).

[0019] Here, the data waveform learning method performed by the data waveform learning unit 13 will be described with reference to FIGS. 4 and 5. The data waveform learning unit 13 extracts data waveforms of m consecutive points (m<N) from the starting point of the N points of feature data stored in the data storage unit 12. This data waveform extraction process is repeated N-m+1 times by shifting the starting point one data point at a time from the latest value to the past, resulting in N-m+1 data waveforms. In the example of FIG. 4, N=6 and m=4, so 6-4+1=3 data waveforms are extracted. The data waveform learning unit 13 learns the data waveforms using machine learning and aggregates them so that the number of data waveforms is equal to or less than the maximum number of data waveforms (pt).

[0020] The machine learning algorithm used by the data waveform learning unit 13 may be a known algorithm such as supervised learning, unsupervised learning, or reinforcement learning. As an example, a case where "self-supervised learning," a type of unsupervised learning, is applied will be described. Unsupervised learning is a method of learning features in training data by providing the training data, which does not include results (labels), to a learning device. Self-supervised learning is a method of learning a teacher signal (labels are automatically generated) in the first stage, and then using the learned teacher signal for some supervised learning task in the second stage or later.

[0021] Specifically, the data waveform learning unit 13 uses one of the data waveforms as training data, learns the similarity between the data waveform and the training data, and aggregates data waveforms with high similarity into the training data. Subsequently, the data waveform learning unit 13 uses a data waveform with low similarity to the previous training data as the next training data, learns the similarity between the data waveform and the previous training data in the same manner, and aggregates the data waveform into the training data. This process is repeated until the aggregation of the data waveforms progresses and the number of data waveforms becomes equal to or less than the maximum number of data waveforms pt, at which point the learning is terminated. In the example of Figure 5, three data waveforms with high similarity are aggregated into one data waveform. The data waveform learning unit 13 stores the aggregated data waveforms that have become equal to or less than the maximum number of data waveforms pt as predicted waveforms in the data storage unit 12.

[0022] Returning to FIG. 1, the future data estimation unit 14 estimates future data for the number of predicted points (p) using the predicted waveform and the most recent feature amount data stored in the data storage unit 12.

[0023] The method of estimating future data executed by the future data estimation unit 14 will now be described with reference to FIGS. 6 and 7. The future data estimation unit 14 reads the predicted waveform from the data storage unit 12 and calculates the degree of matching between the feature data for (m-p) points in the past from the latest time, which is the most recent feature data, and the data for (m-p) points from the start point of the predicted waveform. The degree of matching is calculated by a method in which the smaller the absolute error between the feature data and the predicted data for the (m-p) points of data, the higher the degree of matching. For example, the average value of the absolute errors between the feature data and the predicted data for the (m-p) points of data is calculated, and the reciprocal of the average value is used as the degree of matching. In this case, when the absolute error between the feature data and the predicted data is 0, the degree of matching is considered to be the highest. The future data estimation unit 14 extracts the predicted waveform with the highest degree of matching. The future data estimation unit 14 performs fitting by overlapping data for (m-p) points from the start point of the extracted predicted waveform with data for (m-p) points from the latest time of the feature data, and calculates estimates for (p) predicted points from the latest time. The future data estimation unit 14 stores future data indicating the calculated estimates for (p) predicted points in the data storage unit 12.

[0024] In the example of FIG. 6 , m = 4 and p = 2, so the future data estimation unit 14 calculates the degree of matching between the data for two points in the past from the latest time and the data for two points from the starting points of the predicted waveforms W1, W2, and W3. The future data estimation unit 14 extracts the predicted waveform W1 with the highest degree of matching. As shown in FIG. 7 , the future data estimation unit 14 performs fitting by overlapping the data for two points from the starting point of the extracted predicted waveform W1 with the data for two points in the past from the latest time, and calculates values ​​corresponding to the third and fourth points of the predicted waveform W1 as estimated values ​​for two points in the future from the latest time. Hereinafter, the estimated value for the first point in the future from the latest time will be referred to as the one-step-ahead estimated value, and the estimated value for the second point will be referred to as the two-step-ahead estimated value.

[0025] Returning to FIG. 1 , the normal range definition unit 15 reads future data from the data storage unit 12 and defines a normal range for each value of the future data. The normal range may be defined as an estimated value ± a fixed value, or may be defined based on statistics (such as data range, standard deviation, and variance) obtained when the data waveform learning unit 13 learns the data waveform. For example, the width of the normal range may be set to the estimated value ± (1 / N of the data range: N≧2), or the normal range may be set to the estimated value ± standard deviation. By defining the normal range based on the statistics of past data for each data dimension, it is possible to set a normal range that matches the characteristics of each feature data.

[0026] Here, how the normal range defining unit 15 defines the normal ranges will be described with reference to Fig. 8. In the example of Fig. 8, the normal range defining unit 15 defines the normal ranges for the one-step-ahead estimated value and the two-step-ahead estimated value shown in Fig. 7 as estimated values ​​± fixed values.

[0027] 1 , the normal range deviation determination unit 16 determines whether the state of the external device 3 deviates from the normal range based on whether the latest value of the feature amount related to the external device 3 stored in the data storage unit 12 deviates from the normal range defined by the normal range definition unit 15. The normal range deviation determination unit 16 stores determination result information indicating the determination result in the data storage unit 12.

[0028] Here, a method for determining whether the state of the external device 3 has deviated from the normal range will be described with reference to Figures 9, 10A, and 10B. In the example of Figure 9, the normal range deviation determination unit 16 determines that the state of the external device 3 has deviated from the normal range when the latest value of the feature amount related to the external device 3 (the value at the latest time in the figure) does not fall within the normal range set for the one-step-ahead estimated value. In the examples of Figures 10A and 10B, the normal range deviation determination unit 16 determines that the state of the external device 3 has deviated from the normal range when the latest value of the feature amount related to the external device 3 has deviated from the normal range two times in a row.

[0029] 10A , if the value of the feature quantity at time T1 falls within the normal range set for the one-step-ahead estimated value and the value of the feature quantity at time T2 does not fall within the normal range set for the two-step-ahead estimated value, the normal-range deviation determination unit 16 determines that the state of the external device 3 does not fall outside the normal range and is normal. As shown in FIG. 10B , if the value of the feature quantity at time T1 does not fall within the normal range set for the one-step-ahead estimated value and the value of the feature quantity at time T2 does not fall within the normal range set for the two-step-ahead estimated value, the normal-range deviation determination unit 16 determines that the state of the external device 3 falls outside the normal range.

[0030] The method of determining whether the state of the external device 3 has deviated from the normal range may be to determine that the state has deviated from the normal range when the state does not fall within the normal range once, as shown in FIG. 9 , or to determine that the state has deviated from the normal range when the state has deviated from the normal range twice consecutively, as shown in FIGS. 10A and 10B . Alternatively, the external device 3 may be determined to have deviated from the normal range when the state has deviated from the normal range three or more consecutive times. In this case, the prediction score (p) is 3 or more. Alternatively, a condition may be set such that the state is determined to have deviated from the normal range when a threshold or more times out of a predetermined number of times have not fallen within the normal range. In this case, the prediction score (p) is the predetermined number of times. By evaluating the number of times the state has not fallen within the normal range and determining that the state has deviated, the accuracy of the deviation determination can be improved.

[0031] Next, the normal range deviation diagnostic process executed by the normal range deviation diagnostic device 1 will be described with reference to FIG. 11 . The normal range deviation diagnostic process shown in FIG. 11 starts, for example, when the normal range deviation diagnostic device 1 is powered on. If the data acquisition unit 11 of the normal range deviation diagnostic device 1 does not acquire data indicating the latest values ​​of the feature quantities related to the external device 3 from the sensor 2 and the external device 3 (step S11; NO), the process proceeds to step S21. If the data acquisition unit 11 acquires data indicating the latest values ​​of the feature quantities related to the external device 3 from the sensor 2 and the external device 3 (step S11; YES), the data indicating the latest values ​​of the feature quantities related to the external device 3 is stored in the data storage unit 12 (step S12). The data storage unit 12 stores time-series feature quantity data as shown in FIG. 2 . In the example of FIG. 2 , the feature quantity data has a field for “measurement time” indicating the measurement time at which the feature quantity was measured, and fields for “measurement value A,” “measurement value B,” and “measurement value C” indicating the values ​​of the measured feature quantities.

[0032] 11 , if the number of data points of the feature vector data stored in the data storage unit 12 has not reached the maximum number of data points (step S13; NO), the process proceeds to step S21. If the number of data points of the feature vector data stored in the data storage unit 12 has reached the maximum number of data points (step S13; YES), the data waveform learning unit 13 reads the feature vector data from the data storage unit 12 and performs a data waveform learning process to learn the data waveform (step S14). The data waveform learning unit 13 stores the data waveform aggregated by the data waveform learning process as a predicted waveform in the data storage unit 12 (step S15). Note that even if the number of data points of the feature vector data stored in the data storage unit 12 does not reach the maximum number of data points, if the data waveform learning process is configured to start when it becomes possible to perform the data waveform learning process, the process determines in step S13 whether the data waveform learning process is possible.

[0033] The future data estimation unit 14 reads the predicted waveform stored in the data storage unit 12 and performs a future data estimation process to estimate future data using the predicted waveform and the most recent feature data (step S16). The future data estimation unit 14 stores the estimated future data in the data storage unit 12 (step S17). The normal range definition unit 15 defines normal ranges for the future data stored in the data storage unit 12 (step S18). In the example of Fig. 8, normal ranges are defined for one-step ahead estimated values ​​and two-step ahead estimated values, which are future data.

[0034] 11 , the normal range deviation determination unit 16 determines whether the state of the external device 3 deviates from the normal range based on whether the latest value of the feature amount related to the external device 3, acquired by the data acquisition unit 11 and stored in the data storage unit 12 in step S11, deviates from the normal range defined by the normal range definition unit 15 (step S19). The normal range deviation determination unit 16 stores determination result information indicating the determination result in the data storage unit 12 (step S20). If the power supply of the normal range deviation diagnosis device 1 is not turned off (step S21; NO), the process returns to step S11, and steps S11 to S21 are repeated. If the power supply is turned off (step S21; YES), the process ends.

[0035] The data waveform learning process executed in step S14 of the flowchart shown in FIG. 11 will be described with reference to FIG. 12. The data waveform learning unit 13 of the normal range deviation diagnosis device 1 sets i = 0 (step S31) and extracts m consecutive data waveforms (m < N) from the N points of feature data stored in the data storage unit 12, starting from the most recent value - i (step S32). The data waveform learning unit 13 sets i = i + 1 (step S33) and determines whether m + i ≦ N (step S34). If m + i ≦ N (step S34; YES), the process returns to step S32, and steps S32 to S34 are repeated. If m + i ≦ N is not true (step S34; NO), the data waveform learning unit 13 uses machine learning to learn the data waveforms and aggregates them until the number of data waveforms becomes equal to or less than the maximum number of data waveforms (pt) (step S35), and then ends the process.

[0036] The future data estimation process executed in step S16 of the flowchart shown in FIG. 11 will be described with reference to FIG. 13. The future data estimation unit 14 of the normal range deviation diagnosis device 1 calculates the degree of matching between the most recent feature data (m-p) points in the past from the latest time and the data (m-p) points from the start point of the predicted waveform (step S41). The future data estimation unit 14 extracts the predicted waveform with the highest degree of matching (step S42). The future data estimation unit 14 performs fitting by overlapping the extracted data (m-p) points in the past from the start point of the predicted waveform with the most recent feature data (m-p) points in the past from the latest time, and calculates estimates for the number of predicted points (p) in the future from the latest time (step S43), after which the process ends.

[0037] 11 may be performed for each data dimension, or may be performed for multiple data dimensions showing similar data trends. Steps S14 to S18 do not have to be performed every cycle. For example, after the initial execution, they may be performed every predetermined number of cycles, such as once every five cycles or once every ten cycles.

[0038] According to the normal range deviation diagnosis device 1 of embodiment 1, a data waveform is extracted from feature amount data, which is time-series data indicating feature amounts related to the external device 3, similar data waveforms are aggregated to generate a predicted waveform, and a normal range is defined for each value of future data estimated using the predicted waveform and the most recent feature amount data, thereby making it possible to determine whether the state of the external device 3 has deviated from the normal range even if there is not enough past data for cluster classification. Furthermore, the technology described in Patent Document 1 requires cluster classification of state amount data accumulated in the past and labeling of various states, which results in a large computational load, but the normal range deviation diagnosis device 1 of embodiment 1 requires a smaller amount of data to be learned than the technology described in Patent Document 1 and does not require labeling, thereby reducing the computational load.

[0039] (Embodiment 2) In embodiment 2, the parameters stored in data storage unit 12 are automatically adjusted to achieve conditions that provide the best accuracy in prediction results. The configuration of normal range deviation diagnostic system 100 according to embodiment 2 will be described with reference to FIG. 14. As shown in FIG. 14, normal range deviation diagnostic device 1 of normal range deviation diagnostic system 100 includes data acquisition unit 11, data storage unit 12, data waveform learning unit 13, future data estimating unit 14, normal range defining unit 15, normal range deviation determining unit 16, and also parameter adjusting unit 17 that adjusts parameters.

[0040] The parameter adjustment unit 17 reads the parameters to be adjusted and the parameter variation ranges. The parameters to be adjusted and the parameter variation ranges may be obtained from an external device or system (not shown), may be input by a user, may be obtained from a user terminal, or may be stored in a setting file accessible by the parameter adjustment unit 17. Alternatively, an API (Application Programming Interface) for directly calling the parameter adjustment unit 17 may be created and the parameters may be passed as arguments to this API.

[0041] The parameter adjustment unit 17 sets the value of the parameter to be adjusted to the minimum value of the variation range, and causes the data waveform learning unit 13 and the future data estimation unit 14 to execute the data waveform learning process and the future data estimation process, respectively. The learning data used by the data waveform learning unit 13 to learn the data waveform is not based on the current time, but is instead extracted from data older than the latest time of the feature vector data stored in the data storage unit 12, with a number of data points greater than the number (m) of data waveform points. The future data estimation unit 14 calculates an estimate of point P, which is a time in the future from the latest point in the learning data. The error between this estimate of point P and the actual measurement value included in the feature vector data stored in the data storage unit 12 is evaluated. The parameter adjustment unit 17 adds new data from a time one point in the future of the latest point in the learning data to the learning data, repeating the above process until there are no more latest points to add. This process is hereinafter referred to as error evaluation processing.

[0042] The parameter adjustment unit 17 repeatedly performs error evaluation processing by increasing the parameter value by one unit within the variation range, and obtains a combination of the parameter value and the error evaluation result. When the parameter value reaches the maximum value within the variation range, the parameter adjustment unit 17 determines the optimal value of the parameter. The optimal value is determined based on the error evaluation result. For example, the average error rate for each parameter value, the average absolute value of the error rate, the error variance, the standard deviation, etc. may be used as the selection criterion for the optimal value. Furthermore, the normal range obtained by applying the processing of the normal range definition unit 15 may be used to determine the number of times and the amount of deviation of the actual measurement value from the normal range as the error. The parameter adjustment unit 17 determines the optimal value of the parameter based on a predetermined or user-specified error evaluation method and optimal value selection criterion. The parameter adjustment unit 17 rewrites the parameter stored in the data storage unit 12 to the optimal value. Other functions of the normal range deviation diagnosis device 1 are the same as those in the first embodiment.

[0043] The parameter adjustment process executed by the parameter adjustment unit 17 will now be described with reference to Fig. 15. The parameter adjustment process shown in Fig. 15 may be executed automatically each time one cycle of the normal range deviation diagnosis process is executed, or may be executed in response to a command from a user, or may be configured to be executed when the prediction accuracy drops below a threshold by adding a function for automatically evaluating the prediction accuracy after one cycle of the normal range deviation diagnosis process is executed.

[0044] The parameter adjustment unit 17 reads the parameter to be adjusted and the parameter's variation range (step S51). The parameter adjustment unit 17 sets j to the minimum value of the variation range (step S52) and causes the data waveform learning unit 13 and the future data estimation unit 14 to execute the data waveform learning process and the future data estimation process, respectively (step S53). The parameter adjustment unit 17 evaluates the error between the estimated value of point P calculated by the future data estimation unit 14 and the actual measurement value included in the feature data stored in the data storage unit 12 (step S54). The parameter adjustment unit 17 sets j to j + 1 (step S55) and determines whether j is equal to or less than the maximum value of the variation range (step S56). If j is equal to or less than the maximum value of the variation range (step S56; YES), the process returns to step S53, and steps S53 to S56 are repeated. If j is not equal to or less than the maximum value of the variation range (step S56; NO), the parameter adjustment unit 17 determines the optimum value of the parameter based on the error evaluation method and the optimum value selection criteria specified by the user (step S57), and ends the process. The parameter adjustment unit 17 rewrites the parameter stored in the data storage unit 12 to the optimum value.

[0045] Although the above description has been given using an example in which only one parameter is adjusted in the parameter adjustment process, two or more parameters may be evaluated simultaneously. In this case, the process of changing the parameter value by one unit within the variation range may be performed in multiple stages.

[0046] According to the normal range deviation diagnostic device 1 of the second embodiment, the parameter adjustment process can be performed to constantly maintain the parameters at optimal values. In addition, the time and effort required for the user to set the parameters can be reduced.

[0047] (Embodiment 3) In embodiment 3, a process of removing high-frequency components from feature data written in a data storage unit is performed. The configuration of a normal range deviation diagnostic system 100 according to embodiment 3 will be described with reference to FIG. 16. As shown in FIG. 16, a normal range deviation diagnostic device 1 of normal range deviation diagnostic system 100 includes a data acquisition unit 11, a data storage unit 12, a data waveform learning unit 13, a future data estimation unit 14, a normal range definition unit 15, and a normal range deviation determination unit 16, as well as a high-frequency component removal unit 18 that performs a process of removing high-frequency components from the feature data.

[0048] The high frequency component removal unit 18 performs a process of removing high frequency components from the feature data stored in the data storage unit 12. A general low pass filter process can be used for the process of removing high frequency components. FIG. 17A shows an example of feature data stored in the data storage unit 12 before the process of removing high frequency components is performed. By performing the process of removing high frequency components, the data waveform is smoothed and noise is removed, as shown in FIG. 17B. Other functions of the normal range deviation diagnosis device 1 are the same as those in the first embodiment.

[0049] The normal range deviation diagnostic processing executed by the normal range deviation diagnostic device 1 according to the third embodiment will now be described with reference to Fig. 18. The normal range deviation diagnostic processing shown in Fig. 18 is a flowchart in which a step of removing high frequency components from the feature amount data stored in the data storage unit 12 is added before the data waveform learning processing in the flowchart of the normal range deviation diagnostic processing according to the first embodiment shown in Fig. 11. Steps S61 to S63 and steps S66 to S72 are similar to steps S11 to S13 and steps S15 to S21, respectively, and therefore will not be described here.

[0050] If the number of data points of the feature data stored in the data storage unit 12 has reached the maximum number of data points (step S63; YES), the high frequency component removal unit 18 performs a process of removing high frequency components from the feature data stored in the data storage unit 12 (step S64). The data waveform learning unit 13 reads the feature data from which the high frequency components have been removed from the data storage unit 12, and performs a data waveform learning process of learning the data waveform (step S65).

[0051] According to the normal range deviation diagnosis device 1 of embodiment 3, by performing a process to remove high frequency components from the feature data before performing the data waveform learning process, it is possible to determine whether feature data including fine noise components is deviating from the normal range without being affected by noise.

[0052] The hardware configuration of the normal range deviation diagnostic device 1 will be described with reference to Fig. 19. As shown in Fig. 19, the normal range deviation diagnostic device 1 includes a temporary storage unit 101, a storage unit 102, a calculation unit 103, an input unit 104, a transmission / reception unit 105, and a display unit 106. The temporary storage unit 101, the storage unit 102, the input unit 104, the transmission / reception unit 105, and the display unit 106 are all connected to the calculation unit 103 via a BUS.

[0053] The calculation unit 103 is, for example, a CPU (Central Processing Unit), and executes the processes of the data waveform learning unit 13, the future data estimation unit 14, the normal range definition unit 15, the normal range deviation determination unit 16, the parameter adjustment unit 17, and the high frequency component removal unit 18 in accordance with the control program stored in the storage unit 102.

[0054] The temporary storage unit 101 is, for example, a random-access memory (RAM). The temporary storage unit 101 loads the control program stored in the storage unit 102 and is used as a work area for the calculation unit 103.

[0055] The storage unit 102 is a non-volatile memory such as a flash memory, a hard disk, a DVD-RAM (Digital Versatile Disc - Random Access Memory), or a DVD-RW (Digital Versatile Disc - Rewritable). The storage unit 102 pre-stores a program for causing the calculation unit 103 to perform processing of the normal range deviation diagnosis device 1, and also supplies data stored by this program to the calculation unit 103 in accordance with instructions from the calculation unit 103, and stores the data supplied from the calculation unit 103. The data storage unit 12 is configured in the storage unit 102.

[0056] The input unit 104 is an interface device that connects input devices such as a keyboard, a pointing device, and a voice input device to the BUS. Information input by the user is supplied to the calculation unit 103 via the input unit 104. In a configuration in which the user inputs parameters to be stored in the data storage unit 12, the user inputs the parameter values ​​via the input unit 104. In a configuration in which the user inputs parameters to be adjusted and the parameter variation ranges, the input unit 104 functions as the parameter adjustment unit 17.

[0057] The transmitting / receiving unit 105 is a network termination device or a wireless communication device that connects to the network, and a serial interface or a LAN (Local Area Network) interface that connects to them. The transmitting / receiving unit 105 functions as the data acquiring unit 11. Information input by the user is supplied to the calculating unit 103 via the input unit 104. In a configuration in which the parameters stored in the data storage unit 12 are acquired from a user terminal, the parameter values ​​are acquired from the user terminal via the transmitting / receiving unit 105. In a configuration in which the parameters to be adjusted and the parameter variation ranges are acquired from the user terminal, the transmitting / receiving unit 105 functions as the parameter adjusting unit 17.

[0058] The display unit 106 is a display device such as an LCD (Liquid Crystal Display), an organic EL (Electroluminescence) display, etc. In a configuration in which the user inputs parameters stored in the data storage unit 12, the display unit 106 displays an input screen for inputting parameter values. In a configuration in which the user inputs parameters to be adjusted and their variation ranges, the display unit 106 displays an input screen for inputting the parameters to be adjusted and their variation ranges.

[0059] The processing of the data acquisition unit 11, data storage unit 12, data waveform learning unit 13, future data estimation unit 14, normal range definition unit 15, normal range deviation determination unit 16, parameter adjustment unit 17, and high frequency component removal unit 18 of the normal range deviation diagnosis device 1 shown in Figures 1, 14, and 16 is executed by a control program using the temporary storage unit 101, calculation unit 103, storage unit 102, input unit 104, transmission / reception unit 105, display unit 106, etc. as resources.

[0060] Furthermore, the above hardware configuration and flowchart are merely examples and can be changed and modified as desired.

[0061] The core components of the normal range deviation diagnostic device 1, such as the calculation unit 103, temporary storage unit 101, storage unit 102, input unit 104, transmission / reception unit 105, and display unit 106, can be realized using an ordinary computer system rather than a dedicated system. For example, the normal range deviation diagnostic device 1 that executes the above-described processes may be configured by storing and distributing a computer-readable recording medium such as a flexible disk, a CD-ROM (Compact Disc - Read Only Memory), or a DVD-ROM (Digital Versatile Disc - Read Only Memory), and installing the computer program on a computer. Alternatively, the normal range deviation diagnostic device 1 may be configured by storing the computer program in a storage device of a server device on a communication network, such as the Internet, and downloading it into an ordinary computer system.

[0062] In addition, when the functions of the normal range deviation diagnosis device 1 are realized by sharing the functions between an OS (Operating System) and an application program, or by collaboration between an OS and an application program, only the application program portion may be stored on a recording medium or storage device.

[0063] It is also possible to superimpose a computer program on a carrier wave and provide it via a communication network. For example, the computer program may be posted on a bulletin board system (BBS) on the communication network and provided via the communication network. The computer program may then be started and executed under the control of an OS in the same way as other application programs, thereby enabling the above-mentioned processing to be performed.

[0064] In the above first to third embodiments, examples have been described in which it is determined whether the state of a facility or device installed inside or outside a factory deviates from a normal range, but this is not limiting. The normal range deviation diagnosis device 1 can determine whether the state of an external device capable of measuring some feature value deviates from a normal range.

[0065] In the above first to third embodiments, the normal range deviation diagnosis device 1 is connected to the sensor 2 and the external device 3, but this is not limited thereto. For example, if the data indicating the feature quantities related to the external device 3 are only the measurement values ​​of the sound, vibration, temperature, etc. generated by the processing machine measured by the sensor 2, the normal range deviation diagnosis device 1 may be connected only to the sensor 2. If the data indicating the feature quantities related to the external device 3 are only the measurement values ​​of the robot joint values, motor rotation speed, etc. measured by a measurement unit included in the external device 3, the normal range deviation diagnosis device 1 may be connected only to the external device 3. Alternatively, the data acquisition unit 11 of the normal range deviation diagnosis device 1 may acquire data indicating the latest values ​​of the feature quantities related to the external device 3 from an external device or system.

[0066] In the above-described first to third embodiments, the data indicating the feature quantities related to the external device 3 stored in the data storage unit 12 includes time information as information indicating a time series, but this is not limited to this. For example, in a configuration in which the sensor 2 and the external device 3 periodically transmit data indicating the latest values ​​of the feature quantities related to the external device 3 to the normal range deviation diagnosis device 1 at regular intervals, the information indicating the time series may be information indicating the order in which the data was acquired.

[0067] In the first to third embodiments described above, the normal range deviation determination unit 16 stores determination result information indicating the determination result in the data storage unit 12. However, this is not limited to this. For example, when the normal range deviation determination unit 16 determines that the state of the external device 3 deviates from the normal range, the normal range deviation determination unit 16 may be configured to notify a user that the state of the external device 3 deviates from the normal range, or to notify a control device that controls the external device 3 that the state of the external device 3 deviates from the normal range. Alternatively, when the normal range deviation diagnosis device 1 is applied to the control device that controls the external device 3, the normal range deviation determination unit 16 may be configured to notify a control unit that controls the external device 3 that the state of the external device 3 deviates from the normal range. An example of a control device that controls the external device 3 is a programmable logic controller (PLC). When the control device or control unit that controls the external device 3 is notified that the state of the external device 3 deviates from the normal range, the control device or control unit that controls the external device 3 performs control to return the state of the external device 3 to the normal range. When the characteristic amount of the external device 3 is a parameter controllable by the control device or control unit, stable operation of the external device 3 can be achieved by maintaining the value within the normal range.

[0068] In the third embodiment, the data waveform learning unit 13 performs the data waveform learning process on the feature data after high-frequency components have been removed. However, this is not limited to this. The data waveform learning unit 13 may also perform the data waveform learning process on both the feature data before high-frequency components have been removed and the feature data after high-frequency components have been removed to generate two types of predicted waveforms. In this case, the future data estimation unit 14 estimates two types of future data using the two types of predicted waveforms and the most recent feature data. The normal range definition unit 15 defines two types of normal ranges for each value of the two types of future data. The normal range deviation determination unit 16 determines whether the state of the external device 3 has deviated from the normal range based on whether the latest value of the feature related to the external device 3 deviates from the corresponding two types of normal ranges. Using both raw data with large fluctuations and low-frequency components with small fluctuations can reduce the probability of erroneous determination.

[0069] Although the preferred embodiments have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the claims.

[0070] It should be noted that various embodiments and modifications of the present disclosure are possible without departing from the broad spirit and scope of the present disclosure. Furthermore, the above-described first embodiment is intended to explain the present disclosure and does not limit the scope of the present disclosure. That is, the scope of the present disclosure is defined by the claims, not the embodiments. Various modifications made within the scope of the claims and the meaning of equivalent disclosures are considered to be within the scope of the present disclosure.

[0071] 1 Normal range deviation diagnosis device, 2 Sensor, 3 External device, 11 Data acquisition unit, 12 Data storage unit, 13 Data waveform learning unit, 14 Future data estimation unit, 15 Normal range definition unit, 16 Normal range deviation determination unit, 17 Parameter adjustment unit, 18 High frequency component removal unit, 100 Normal range deviation diagnosis system, 101 Temporary storage unit, 102 Storage unit, 103 Calculation unit, 104 Input unit, 105 Transmission / reception unit, 106 Display unit.

Claims

1. A normal range deviation diagnosis device comprising: a data acquisition unit that acquires data indicating the latest values ​​of feature quantities related to an external device; a data waveform learning unit that extracts data waveforms, which are a predetermined number of consecutive data points, from feature quantity data, which is time-series data that accumulates data indicating feature quantities related to the external device, and performs learning to aggregate similar data waveforms to generate a predicted waveform, which is the aggregated data waveform; a future data estimation unit that uses the predicted waveform and the most recent feature quantity data to estimate future data, which are future values ​​of the feature quantity data; a normal range definition unit that defines a normal range for each value of the future data; and a normal range deviation determination unit that determines whether the state of the external device has deviated from the normal range based on whether the latest value of the feature quantity related to the external device deviates from the corresponding normal range.

2. The normal range deviation diagnostic device according to claim 1, wherein the normal range deviation determination unit determines that the state of the external device has deviated from the normal range when the latest values ​​of the feature quantities related to the external device deviate from all of the normal ranges, or when the latest values ​​deviate from the normal range a threshold or more times out of a predetermined number of times.

3. The normal range deviation diagnosis device according to claim 1 or 2, further comprising: a parameter adjustment unit that determines optimal values ​​for the parameters based on evaluation results of the error evaluation process by varying values ​​of at least one of the parameters of the maximum number of data points, the number of data waveform points, the maximum number of data waveforms, and the predicted waveform; and a parameter adjustment unit that determines optimal values ​​for the parameters based on evaluation results of the error evaluation process by varying values ​​of at least one of the parameters of the maximum number of data points, the number of data waveform points, the maximum number of data waveforms, and the predicted waveform.

4. A normal range deviation diagnostic device according to any one of claims 1 to 3, further comprising a high frequency component removal unit that performs processing to remove high frequency components from the feature data, wherein the data waveform learning unit extracts data waveforms of a predetermined number of data points from the feature data from which the high frequency components have been removed, and performs learning to aggregate similar data waveforms to generate the predicted waveform.

5. The normal range deviation diagnostic device according to claim 4, wherein the data waveform learning unit extracts a predetermined number of data points of data waveforms from each of the feature data before high frequency components are removed and the feature data after high frequency components are removed, and performs learning to aggregate similar data waveforms to generate two types of predicted waveforms; the future data estimation unit estimates two types of future data using the two types of predicted waveforms and the most recent feature data; the normal range definition unit defines two types of normal ranges for each value of the two types of future data; and the normal range deviation determination unit determines whether the state of the external device has deviated from the normal range based on whether the latest value of the feature related to the external device has deviated from the corresponding two types of normal ranges.

6. A normal range deviation diagnostic device according to any one of claims 1 to 5, wherein the normal range definition unit defines the normal range based on statistics obtained when the data waveform learning unit learns the data waveform.

7. A normal range deviation diagnostic device according to any one of claims 1 to 6, further comprising a control unit that controls the external device, wherein the control unit performs control to return the state of the external device to within the normal range when the normal range deviation determination unit determines that the state of the external device has deviated from the normal range.

8. A normal range deviation diagnosis system comprising: a normal range deviation diagnosis device according to any one of claims 1 to 7; and a measurement device that measures a characteristic quantity of the external device, wherein the data acquisition unit acquires data indicating the latest value of the characteristic quantity related to the external device from the measurement device.

9. A normal range deviation diagnostic system comprising: a normal range deviation diagnostic device according to any one of claims 1 to 6; and a control device that controls the external device, wherein the control device controls the external device to return the state of the external device to within the normal range when the normal range deviation determination unit determines that the state of the external device has deviated from the normal range.

10. A normal range deviation diagnostic method executed by a normal range deviation diagnostic device, comprising: a step of acquiring data indicating the latest values ​​of feature quantities related to an external device; a step of extracting data waveforms of a predetermined number of data points from feature quantity data, which is time-series data that accumulates data indicating feature quantities related to the external device, and performing learning to aggregate similar data waveforms to generate a predicted waveform, which is the aggregated data waveform; a step of estimating future data, which is future values ​​of the feature quantity data, using the predicted waveform and the most recent feature quantity data; a step of defining a normal range for each value of the future data; and a step of determining whether the state of the external device has deviated from the normal range, based on whether the latest value of the feature quantity related to the external device has deviated from the corresponding normal range.

11. A program that causes a computer to function as: a data waveform learning unit that extracts data waveforms of a predetermined number of data points from feature data, which is time-series data that accumulates data indicating feature quantities related to an external device, and performs learning to aggregate similar data waveforms to generate a predicted waveform, which is the aggregated data waveform; a future data estimation unit that uses the predicted waveform and the most recent feature data to estimate future data, which is the future value of the feature data; a normal range definition unit that defines a normal range for each value of the future data; and a normal range deviation determination unit that determines whether the state of the external device has deviated from the normal range based on whether the latest value of the feature quantity related to the external device has deviated from the corresponding normal range.

Images