Apparatus, method, and computer readable medium

Abstract

Provided are an apparatus, a method, and a computer-readable medium, the apparatus including: an acquisition unit that acquires measurement data indicating a state of an object; and a supply unit that supplies the measurement data acquired by the acquisition unit to a plurality of classification models that complete learning by learning data for each of different periods including the measurement data for a period in which the state of the object is normal. And classifies the measurement data as either normal or abnormal according to the input of the measurement data; and a determination unit that determines, on the basis of a plurality of classification results output from the plurality of classification models, whether the state of the object is normal or abnormal.

Description

Technical Field

[0001] The present invention relates to apparatus, methods and computer-readable media. Background Art

[0002] Patent Document 1 states that “… a threshold value may be used to detect whether or not the device 10 has an abnormality” (paragraph 0055) and the like. Prior art literature Patent Document 1: Japanese Patent Publication No. 2020-95425 Summary of the invention

[0003] In a first aspect of the present invention, a device is provided, comprising: an acquisition unit, which acquires measurement data representing a state of an object; a supply unit, which supplies the measurement data acquired by the acquisition unit to a plurality of classification models, the plurality of classification models respectively completing learning by using learning data of each different period including measurement data of a period in which the state of the object is normal, the plurality of classification models classifying the measurement data as either normal or abnormal according to the input of the measurement data; and a judgment unit, which judges the state of the object as either normal or abnormal based on a plurality of classification results output from the plurality of classification models.

[0004] The above-mentioned device may further include a learning processing unit that generates a new classification model included in the multiple classification models by performing learning processing using learning data including measurement data of the period acquired by the acquisition unit during each period when the state of the object is normal.

[0005] In any of the above-mentioned devices, there may also be a selection unit, which selects the multiple classification models from more than two classification models, and the two or more classification models respectively complete learning by using learning data of each different period including measurement data of a period in which the state of the object is normal, and classifies the measurement data into either normal or abnormal based on the input of the measurement data.

[0006] In the above-mentioned device including the selection unit, the selection unit may select at least one classification model learned by learning data including measurement data of a latest period as at least one of the plurality of classification models.

[0007] In any of the above-mentioned devices including the selection unit, the selection unit may select at least one classification model designated by an operator among the two or more classification models as at least one of the plurality of classification models.

[0008] In any of the above-mentioned devices, the determination unit may take a logical product of the plurality of classification results to determine whether the state of the object is normal or abnormal.

[0009] In any of the above-mentioned devices, the determination unit may determine the state of the object as either normal or abnormal by taking a majority decision of the plurality of classification results.

[0010] In the above-mentioned device, the judgment unit may assign a greater weight to the classification result of the classification model among the multiple classification models that has completed learning by using learning data that includes measurement data of a more recent period, and take the weighted majority of the multiple classification results to determine the state of the object as either normal or abnormal.

[0011] Any of the above-mentioned devices may further include a determination unit that determines the severity of the abnormal state of the subject based on the number of times the state of the subject is continuously determined to be abnormal.

[0012] In the above-mentioned device, there may also be a setting unit, which sets any one of the measurement data of each period acquired by the acquisition unit and supplied from the supply unit to the multiple classification models, the measurement data of the period in which the severity determined by the determination unit is lower than the benchmark severity, and the measurement data acquired by the acquisition unit and supplied from the supply unit to the multiple classification models, except for each measurement data that caused continuous abnormality judgments until the severity determined by the determination unit reached the benchmark severity, and each measurement data that continued to cause continuous abnormality judgments after the measurement data, as measurement data included in the new learning data.

[0013] In a second aspect of the present invention, a method is provided, comprising: an acquisition stage, in which measurement data representing a state of an object is acquired; a supply stage, in which the measurement data acquired by the acquisition stage is supplied to a plurality of classification models, the plurality of classification models respectively completing learning by including learning data of each different period of measurement data during which the state of the object is normal, and classifying the measurement data as either normal or abnormal based on the input of the measurement data; and a judgment stage, in which the state of the object is judged as either normal or abnormal based on a plurality of classification results output from the plurality of classification models.

[0014] In a third aspect of the present invention, a computer-readable medium is provided, which stores a program, and a computer functions as an acquisition unit, a supply unit and a judgment unit by executing the program, wherein the acquisition unit acquires measurement data representing the state of an object, the supply unit supplies the measurement data acquired by the acquisition unit to multiple classification models, the multiple classification models respectively complete learning by including learning data of each different period of measurement data during which the state of the object is normal, and classify the measurement data into either normal or abnormal based on the input of the measurement data, and the judgment unit judges the state of the object as either normal or abnormal based on multiple classification results output from the multiple classification models.

[0015] In addition, the above summary of the invention does not list all the essential features of the present invention. In addition, sub-combinations of these feature groups may also constitute inventions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A system 1 according to an embodiment is shown. Figure 2 Indicates the action of device 3. Figure 3 It represents the relationship between the judgment result, the number of abnormal judgments x and the severity. Figure 4 An example of a computer 2200 is shown in which various aspects of the present invention may be implemented in whole or in part. Description of Reference Numerals 1 system, 2 equipment, 3 device, 20 sensor, 31 acquisition unit, 32 storage unit, 33 supply unit, 34 judgment unit, 35 determination unit, 36 output unit, 37 setting unit, 38 learning processing unit, 39 selection unit, 321 measurement data file, 322 classification model, 2200 computer, 2201 DVD-ROM, 2210 main controller, 2212 CPU, 2214 RAM, 2216 graphic controller, 2218 display device, 2220 input / output controller, 2222 communication interface, 2224 hard disk drive, 2226 DVD-ROM drive, 2230 ROM, 2240 input / output chip, 2242 keyboard. DETAILED DESCRIPTION

[0017] The present invention will be described below by way of the embodiments of the invention, but the following embodiments do not limit the invention according to the claims. In addition, not all combinations of features described in the embodiments are essential to the solution means of the invention.

[0018] <1. System 1> Figure 1 A system 1 according to this embodiment is shown. The system 1 includes equipment 2 and a device 3 .

[0019] <1.1. Equipment 2> Device 2 can be an example of an object, and the state can be monitored by device 3. Device 2 can be a facility or device with one or more devices (not shown). For example, device 2 can be a factory, or a composite device that combines multiple devices. As factories, in addition to chemical or biological industrial factories, there can also be factories that manage and control wellheads and their surroundings such as gas fields or oil fields, factories that manage and control power generation such as hydropower, thermal power, and atomic energy, factories that manage and control environmental power generation such as solar power or wind power, and factories that manage and control water supply and drainage or dams. The equipment possessed by device 2 can be an appliance, a machine, a device, or a so-called field device. For example, the equipment can be a sensor device such as a pressure gauge, a flow meter, a temperature sensor, a valve device such as a flow control valve or an on-off valve, or an actuator device such as a fan or a motor.

[0020] One or more sensors 20 may be provided in the device 2. Each sensor 20 may measure a physical quantity indicating the state of the device 2. The measured physical quantity may be, for example, pressure, temperature, pH, speed, flow rate, etc., or may be the output of the product of the device 2, the proportion of impurities mixed into the product, etc. Each sensor 20 may be of a different type from each other, and at least a portion of two or more sensors 20 may be of the same type. Each sensor 20 may supply the measured value of the physical quantity, etc. to the device 3. The communication between the sensor 20 and the device 3 may be carried out, for example, through the ISA (International Society of Automation) wireless communication protocol, and as an example, may be carried out through ISA100, HART (Highway Addressable Remote Transducer) (registered trademark), BRAIN (registered trademark), FOUNDATION Fieldbus, PROFIBUS, etc.

[0021] <1.2. Device 3> The device 3 includes an acquisition unit 31 , a storage unit 32 , a supply unit 33 , a determination unit 34 , a determination unit 35 , an output unit 36 ​​, a setting unit 37 , a learning processing unit 38 , and a selection unit 39 .

[0022] <1.2.1. Acquisition unit 31> The acquisition unit 31 acquires measurement data indicating the state of the device 2. The acquisition unit 31 may acquire measurement data including one measurement value measured by a single sensor 20, or may acquire measurement data including multiple measurement values ​​measured by multiple sensors 20 at the same (or substantially the same) timing. A timestamp indicating the measurement timing of the sensor 20 or the acquisition timing of the acquisition unit 31 may be added to the measurement data. The acquisition unit 31 may supply the acquired measurement data to the storage unit 32 and store it in a measurement data file 321 described later. The acquisition unit 31 may supply the acquired measurement data to the supply unit 33.

[0023] <1.2.2. Storage unit 32> The storage unit 32 stores various information. For example, the storage unit 32 may store a measurement data file 321 and a plurality of classification models 322 .

[0024] <1.2.2(1). Measurement data file 321> The measurement data file 321 stores the measurement data supplied from the acquisition unit 31. In addition, each measurement data in the measurement data file 321 can indicate which period of a plurality of periods that are different for each reference time width (one week as an example in this embodiment) the measurement data belongs to by using the timestamp of the measurement data. For example, according to the timestamp of January 1, 2023, it can be indicated that the measurement data belongs to the period of the first week of 2023, and according to the timestamp of January 8, 2023, it can be indicated that the measurement data belongs to the period of the second week of 2023.

[0025] <1.2.2(2). Classification Model 322> Each classification model 322 classifies the measurement data into normal or abnormal based on the input measurement data. Each classification model 322 may output a classification result binarized by a value indicating normality and a value indicating abnormality. Each classification model 322 may output the classification result to the determination unit 34.

[0026] Each classification model 322 can complete learning separately using learning data of each different period, and the learning data includes measurement data of the period when the state of device 2 is normal. The measurement data of the period when the state of device 2 is normal can be measurement data including measurement values ​​measured during the period when the state of device 2 is normal. The period when the state of device 2 is normal can be a period when the operator or at least one of the judgment units 34 described later judges that the state of device 2 is normal. The learning data of each period can include measurement data measured at two or more timings (as an example, measurement data in a time series order). In the present embodiment, as an example, the learning data of each period can include measurement data within a period of the same time width (also referred to as a reference time width). As an example, the reference time width can be one week. Different periods can be periods where at least one of the start time or the end time is different. In the present embodiment, as an example, it can be a period that does not include a common time point.

[0027] Each classification model 322 can classify the measurement data of the classification object as normal according to whether the measurement data satisfies a condition, or classify the measurement data as abnormal according to whether the condition is not satisfied, and the condition is derived from the characteristics of the measurement data contained in the learning data. The characteristics of the measurement data contained in the learning data may be a value representing the distribution of various measurement values, and as an example, may be at least one of the maximum value, minimum value, mean value, median value, mode, standard deviation, or variance, but may also be a value representing a waveform of a timing measurement value. As an example, the condition derived from the characteristics of the measurement data may be that the measurement value of the measurement data of the classification object converges within the range of the minimum value to the maximum value, or may be that the measurement value converges within the range of (mean value - 3σ) to (mean value + 3σ) (where σ represents the standard deviation).

[0028] Each classification model 322 can be learned by unsupervised learning using only learning data, or can be generated by the same learning algorithm, wherein the learning data includes measurement data that is pre-classified as normal measurement data. In the present embodiment, as an example, each classification model 322 can be generated by a learning algorithm of a class SVM (Support Vector Machine). Each classification model 322 can be learned using different learning data.

[0029] At least one of the classification models 322 stored in the storage unit 32 may be generated by a learning process performed by the learning processing unit 38 described later. At least one of the classification models 322 stored in the storage unit 32 may be a classification model 322 (also referred to as a classification model 322 of a use object) used for judgment in the judgment unit 34. The details will be described later, and the classification model 322 of a use object may be selected by the selection unit 39.

[0030] <1.2.3. Supply unit 33> The supply unit 33 supplies the measurement data acquired by the acquisition unit 31 to one or more classification models 322. The supply unit 33 can supply the measurement data to each classification model 322 of the target. Thus, the classification result can be output to the determination unit 34 from each classification model 322 of the target to which the measurement data is supplied.

[0031] <1.2.4. Judgment Unit 34> The determination unit 34 determines whether the state of the device 2 is normal or abnormal based on the plurality of classification results output from the plurality of classification models 322. The determination unit 34 may obtain the classification results from each classification model 322 of the object of use and make the determination.

[0032] The judgment unit 34 may take the logical product of multiple classification results and judge the state of the device 2 as either normal or abnormal. In the present embodiment, as an example, the judgment unit 34 may set the classification result of the measurement data being abnormal to true (True), and the classification result of the measurement data being normal to false (False), and take the logical product of multiple classification results. In other words, the state of the device 2 may be set to abnormal based on the measurement data being abnormal in all classification results, and the state of the device 2 may be set to normal based on the measurement data being normal in at least one classification result. The judgment unit 34 may supply the judgment result to the determination unit 35.

[0033] <1.2.5. Determination unit 35> The determination unit 35 determines the severity of the abnormal state of the device 2 based on the number of times the device 2 is continuously determined to be abnormal (also referred to as the number of consecutive abnormal determinations). Continuous determinations of abnormality may be determinations of abnormality without intervening normal determinations.

[0034] The more consecutive abnormality judgments there are, the higher the severity can be determined by the determination unit 35. For example, when the consecutive abnormality judgment times are set to "x", the first reference times are set to "a" (where a is a natural number greater than 2), and the second reference times are set to "b" (where b is a natural number of a<b), the determination unit 35 can determine the severity as a low level according to x<a, determine the severity as a medium level (also called a caution level) according to a≤x<b, and determine the severity as a high level (also called a warning level) according to b≤x. As an example, the first reference times a can be 5 times, and the second reference times b can be 10 times.

[0035] The determination section 35 may determine the severity every time the determination result is supplied from the determination section 34. The determination section 35 may supply the determined severity together with the determination result of the determination section 34 to the output section 36 and the setting section 37.

[0036] <1.2.6. Output unit 36> The output unit 36 ​​outputs the judgment result of the judgment unit 34 and the determination result of the determination unit 35. The output unit 36 ​​can display the judgment result and the determination result on a display device (not shown).

[0037] <1.2.7. Setting unit 37> The setting unit 37 sets the measurement data of the period in which the severity determined by the determination unit 35 is lower than the reference severity among the measurement data of each period acquired by the acquisition unit 31 and supplied from the supply unit 33 to the classification model 322 of the object of use as the measurement data included in the new learning data (also referred to as the measurement data for learning). The measurement data of the period in which the severity determined by the determination unit 35 is lower than the reference severity among the measurement data of each period supplied to the classification model 322 of the object of use may be the measurement data of the period in which the severity determined by the determination unit 35 is maintained lower than the reference severity among the measurement data of each period. The severity lower than the reference severity may be that the device 2 is normal or that the device 2 is abnormal but the severity of the abnormality is lower than the reference severity. In the present embodiment, as an example, the reference severity may be a high level, i.e., a warning level.

[0038] In addition, during a period when the severity is lower than the reference severity, even if the state of the device 2 is determined to be abnormal, since the abnormal state occurs only temporarily and once, the device 2 is considered to be normal, but the measurement data is classified as abnormal due to noise or drift of the measurement value, etc. The setting unit 37 can set the measurement data during the period when the device 2 is normal as the measurement data for learning.

[0039] The setting unit 37 may attach a label indicating that the measurement data is for learning (also referred to as a learning label) to the measurement data of each period in the measurement data file 321, the measurement data of each period having a severity lower than the reference severity. In the present embodiment, as an example, the setting unit 37 may determine whether the determined severity is lower than the high level each time the determination result of the severity is supplied from the determination unit 35, and attach a learning label to the measurement data of the period in which the severity is maintained lower than the high level. The setting unit 37 may supply a notification of the addition of the learning label to the measurement data to the learning processing unit 38.

[0040] <1.2.8. Learning Processing Unit 38> The learning processing unit 38 generates a new classification model 322 included in the plurality of classification models 322 stored in the storage unit 32 by performing a learning process using the learning data including the measurement data of the period acquired by the acquisition unit 31 during each period in which the state of the device 2 is normal. Generating a new classification model 322 during each period in which the state of the device 2 is normal may be generating a new classification model 322 for each period in which the state of the device 2 is normal.

[0041] The learning processing unit 38 can generate a new classification model 322 by using the learning data including the measurement data of the most recent period among the measurement data to which the learning labels are attached in the measurement data file 321, based on the notification received from the setting unit 37 indicating that the learning labels are attached to the measurement data. Therefore, when the learning labels are attached to the measurement data of a plurality of consecutive periods, a new classification model 322 can be generated each time each period passes.

[0042] The learning processing unit 38 can generate the classification model 322 by unsupervised learning using only the learning data. In the present embodiment, as an example, the classification model 322 can be generated by a learning algorithm of a class SVM, wherein the learning data includes measurement data that has been pre-classified by adding learning labels. The learning processing unit 38 can provide a notification indicating the generation of the classification model 322 to the selection unit 39.

[0043] <1.2.9. Selection section 39> The selection unit 39 selects a classification model 322 to be used from among the classification models 322 stored in the storage unit 32. When there are two or more classification models 322 stored in the storage unit 32, the selection unit 39 may select a plurality of classification models 322 to be used from the two or more classification models 322. When there are more than a reference number (four as an example) of classification models 322 stored in the storage unit 32, the selection unit 39 may select the reference number of classification models 322 as the classification model 322 to be used.

[0044] The selection unit 39 can select at least one classification model 322 that has been learned using the learning data including the measurement data of the latest period as at least one of the classification models 322 of the object to be used. The selection unit 39 can select the classification model 322 of the object to be used, respectively, based on the notification indicating that the classification model 322 is newly generated, received from the learning processing unit 38. The selection unit 39 can supply the measurement data from the supply unit 33 to the classification model 322 of the object to be used by supplying the identification information of the classification model 322 of the object to be used to the supply unit 33.

[0045] According to the above device 3, the state of the device 2 is judged to be normal or abnormal based on the multiple classification results of normality or abnormality of the measurement data output by supplying the measurement data to the classification model 322. Therefore, unlike the case where an index value indicating the degree of normality or abnormality of the measurement data is obtained and the index value is compared with a threshold value to make a judgment, it is possible to eliminate the trouble of the operator setting the threshold value through repeated attempts, and it is possible to prevent the judgment result from being different depending on the size of the threshold value. In addition, by using a plurality of classification models 322 that have been learned using learning data for each period that is different from each other, including measurement data for a period in which the state of the device 2 is normal, a high-precision classification result can be obtained compared to the case where a single classification model is used that has been learned using learning data for each measurement data for a plurality of periods in which the state of the device 2 is normal. That is, even if the state of the device 2 is normal, the trend of the measurement value may change, for example, in the case of drift of the measurement value, in the case of a change in the state of the device 2, and in the case of a change in the operation method of the device 2. In addition, even if the state of the device 2 is normal, the measurement value may contain various noises. In this way, in the case where the trend of the measurement value changes or in the case where various noises are contained in the measurement value, if a single classification model that has been learned using learning data for each measurement data for a plurality of periods is used, even if the measurement data of the classification object deviates from the normal range in each period, the measurement data of the classification object may be classified as normal measurement data if the plurality of periods as a whole are contained in the normal range. In contrast, if a plurality of classification models 322 that have been learned using learning data for each period that is different from each other are used, when the measurement data to be classified deviates from the normal range in each period, the measurement data can be correctly classified as abnormal measurement data by each classification model 322. Furthermore, by making judgments using a plurality of such classification results, a judgment result with high accuracy can be obtained.

[0046] In addition, by taking the logical product of multiple classification results based on multiple classification models 322, it is determined whether the state of the device 2 is normal or abnormal, so compared with the case of taking the logical sum for judgment, a high-precision judgment result can be obtained.

[0047] In addition, during each period when the state of the device 2 is normal, a new classification model 322 is generated by a learning process using learning data including the measurement data of the period. Therefore, since the classification models 322 corresponding to the changes in the trend of the measurement data can be sequentially generated and used for judgment, a highly accurate judgment result can be obtained.

[0048] Furthermore, a plurality of classification models 322 to be used are selected from the two or more classification models 322 stored in the storage unit 32. Therefore, an appropriate classification model 322 can be appropriately selected and used for determination.

[0049] Furthermore, a classification model 322 that has been learned using learning data including measurement data of the latest period is selected as at least one of the classification models 322 to be used, so that the classification model 322 corresponding to the most recent state of the device 2 is used for judgment.

[0050] Furthermore, the severity of the abnormal state of the device 2 is determined based on the number of times the state of the device 2 is continuously determined to be abnormal, so the severity of the abnormal state can be acquired.

[0051] Furthermore, among the measurement data of each period supplied to the plurality of classification models 322, the measurement data of the period having a severity lower than the reference severity is set as the measurement data included in the new learning data. Therefore, since the measurement data of the period when the device 2 is originally normal can be set for learning, by generating the classification model 322 using the new learning data including such measurement data, the classification model 322 capable of correctly classifying the originally normal measurement data as normal can be obtained.

[0052] <2. Operation of device 3> Figure 2 The operation of the device 3 is shown. The device 3 supports the monitoring of the equipment 2 by performing the processing of steps S1 to S55. In this embodiment, as an example, a case where the classification model 322 is not stored in the storage unit 32 at the start time of this operation is described.

[0053] In step S1, the acquisition unit 31 acquires measurement data during a period when the state of the device 2 is normal (as an example, a period of a reference time width). In this step, the period during which the state of the device 2 is normal may be a period during which the operator determines that the state of the device 2 is normal. As an example, when the operator determines that the state of the device 2 is normal during the reference time width, it may be this period. When the operator determines that the state of the device 2 is normal during a period longer than the reference time width, it may be a period of the reference time width included in this period. The acquired measurement data may include measurement values ​​during a period during which the operator determines that the state of the device 2 is normal. The acquired measurement data may be used for learning the classification model 322, and a learning label may be pre-attached. The acquisition unit 31 may acquire measurement values ​​from one or more sensors 20 as measurement data. The acquisition unit 31 may perform so-called preprocessing on the acquired measurement data. As an example, the acquisition unit 31 may calculate a moving average of the measurement values ​​from the most recent plurality of measurement data obtained as the measurement values ​​of the obtained measurement data, or may remove deviation values ​​detected by statistical analysis using past measurement data from the measurement values ​​of the obtained measurement data.

[0054] In step S3, the learning processing unit 38 generates a new classification model 322 by a learning process using the learning data including the measurement data acquired in step S1. The learning processing unit 38 may store the generated new classification model 322 in the storage unit 32.

[0055] In step S5, the selection unit 39 selects the classification model 322 generated in step S3 as the classification model 322 to be used. Thereby, in step S15 described later, the measurement data can be supplied to the selected classification model 322.

[0056] In step S11, the supply unit 33 starts measuring the reference time width, thereby starting a new period.

[0057] In step S13, the acquisition unit 31 acquires measurement data indicating the state of the device 2. The acquisition unit 31 may acquire measurement values ​​as measurement data from one or more sensors 20. The acquisition unit 31 may pre-process the acquired measurement data in the same manner as in step S1.

[0058] In step S15, the supply unit 33 supplies the measurement data acquired in step S13 to at least one classification model 322. Thus, each classification model 322 supplied with the measurement data can output a classification result of classifying the measurement data as normal or abnormal.

[0059] The supply unit 33 can supply measurement data to each classification model 322 of the use object selected by the selection unit 39. As an example, when only a single classification model 322 is selected as the use object through the above-mentioned step S5, the supply unit 33 can supply measurement data to the single classification model 322. When multiple classification models 322 are selected as the use object through the later-described step S55, the supply unit 33 can supply measurement data to each selected classification model 322. When step S15 is performed for the second time or later, the processing of step S15 can be performed every reference interval (for example, 30 minutes).

[0060] In step S17 , the determination unit 34 acquires the classification result of the measurement data from each classification model 322 to be used.

[0061] In step S19 , the determination unit 34 determines whether the state of the device 2 is normal or abnormal based on the classification results acquired in step S17 .

[0062] In the case where only a single classification result is obtained from the single classification model 322 in the above step S17, the judgment unit 34 can judge the state of the device 2 as normal or abnormal based on the single classification result. As an example, the judgment unit 34 can judge the state of the device 2 as normal based on the classification result that the measurement data is normal, and judge the state of the device 2 as abnormal based on the classification result that the measurement data is abnormal.

[0063] In the case where multiple classification results are obtained from multiple classification models 322 in the above step S17, the judgment unit 34 can judge the state of the device 2 as either normal or abnormal based on the multiple classification results. The judgment unit 34 can take the logical product of the multiple classification results and judge the state of the device 2 as either normal or abnormal. In the case where it is judged in step S19 that the device 2 is normal (step S19; normal), the processing can be transferred to step S41. In the case where it is judged in step S19 that the device 2 is abnormal (step S19; abnormal), the processing can be transferred to step S21.

[0064] In step S21, the determination unit 35 increments the number of consecutive abnormality determinations x. When step S21 is first performed, the number of consecutive abnormality determinations may be incremented from an initial value of 0 to 1.

[0065] In step S23, the determination unit 35 determines whether the number of consecutive abnormality determinations x is less than the first reference number a, that is, whether x < a. If it is determined that x < a (step S23; yes), the process can be transferred to step S25. If it is determined that x < a, that is, a ≤ x (step S23; no), the process can be transferred to step S27.

[0066] In step S25, the determination section 35 determines that the severity of the abnormal state of the device 2 is at a low level. If the process of step S25 is completed, the process may be transferred to step S45 described later.

[0067] In step S27, the determination unit 35 determines whether the number of consecutive abnormality determinations x is greater than or equal to the first reference number a and less than the second reference number b, that is, whether a≤x<b. If it is determined that a≤x<b (step S27; yes), the process can be transferred to step S29. If it is determined that a≤x<b is not, that is, b≤x (step S27; no), the process can be transferred to step S31.

[0068] In step S29, the determination unit 35 determines the severity of the abnormal state of the device 2 as a medium level, that is, a caution level. If the process of step S29 is completed, the process may be transferred to step S45 described later.

[0069] In step S31, the determination unit 35 determines the severity of the abnormal state of the device 2 to be a high level, that is, a warning level. If the process of step S31 is completed, the process may be transferred to step S45 described later.

[0070] In step S41 , the determination unit 35 resets the number of consecutive abnormality determinations x to zero. In step S43 , the determination section 35 determines the severity of the abnormal state of the device 2 to be zero.

[0071] In step S45, the output unit 36 ​​outputs the determination result of the determination unit 34 in step S19 and the determination result of the determination unit 35 in step S25, S29, S31 or S43. The output unit 36 ​​may display the determination result and the determination result on a display device (not shown).

[0072] In step S47, the supply unit 33 determines whether the timing of the reference time width is completed, that is, whether the reference time width period (in this embodiment, one week as an example) has passed since the above step S11. If it is determined that the reference time width period has passed (step S47; yes), the process can be transferred to step S49. If it is determined that the reference time width period has not passed (step S47; no), the process can be transferred to step S13.

[0073] In step S49, the setting unit 37 determines whether the severity level has been maintained below the reference severity level (in this embodiment, a high level as an example) during the recent period. If it is determined that the severity level has not been maintained below the high level (step S49; No), the process may be transferred to the above-mentioned step S11. If it is determined that the severity level has been maintained below the high level (step S49; Yes), the process may be transferred to step S51.

[0074] In step S51, the setting unit 37 sets the measurement data of the most recent period as the measurement data included in the new learning data. In the present embodiment, as an example, the setting unit 37 may set the measurement data of the period during which the judgment that the state of the device 2 is normal made in step S19 as the measurement data included in the new learning data. On this basis, the setting unit 37 may set the measurement data of the period during which the severity is maintained below the high level by determining the severity as a low level, a medium level or 0 in steps S25, S29, and S43 as the measurement data included in the new learning data. The setting unit 37 may add a learning label to the measurement data of the most recent period stored in the measurement data file 321.

[0075] In step S53, the learning processing unit 38 generates a new classification model 322 by a learning process using learning data including the most recent measurement data to which the learning labels are added. The learning processing unit 38 may additionally store the generated new classification model 322 in the storage unit 32.

[0076] In step S55, the selection unit 39 selects a classification model 322 to be used from the plurality of classification models 322 stored in the storage unit 32. Thus, measurement data can be supplied to each classification model 322 to be used selected in step S15.

[0077] When only the classification models 322 with a number less than the reference number are stored in the storage unit 32, the selection unit 39 may select each classification model 322 as the classification model 322 to be used. When more classification models 322 than the reference number are stored in the storage unit 32, the selection unit 39 may select the classification models 322 with the reference number from among these classification models 322 as the classification model 322 to be used. For example, the selection unit 39 may select at least one latest classification model 322 generated by step S53 from among the classification models 322 in the storage unit 32 as the classification model 322 to be used. Thus, until the number of classification models 322 in the storage unit 32 reaches the reference number, the measurement data may be classified by all the classification models 322, and after the number of classification models 322 in the storage unit 32 exceeds the reference number, the measurement data may be classified by the classification models 322 with the reference number selected. The selection unit 39 may sequentially delete the classification models 322 generated previously according to the capacity of the storage unit 32. If the processing of step S55 is completed, the processing can be transferred to the above-mentioned step S11.

[0078] <3. Action example> Figure 3The relationship between the judgment result, the number of consecutive abnormality judgments x, and the severity of the abnormality is shown. In this figure, the horizontal axis represents time, and the dotted lines in the vertical direction represent the division of time periods.

[0079] As shown in this figure, the number of consecutive abnormal judgments x increases according to the continuous abnormal judgment result of the state of the device 2, and is reset according to the normal judgment result of the state of the device 2. In addition, when the number of consecutive abnormal judgments x is 0, the severity is 0, when the number of consecutive abnormal judgments x is 1 to 4, the severity is low, when the number of consecutive abnormal judgments x is 5 to 9, the severity is medium (i.e., attention level), and when the number of consecutive abnormal judgments x is 10, the severity is high (i.e., warning level).

[0080] <4. Modifications> In addition, in the above-mentioned embodiment, it is explained that the selection unit 39 selects at least one classification model 322 that has been learned by learning data including measurement data of the latest period as at least one of the classification models 322 to be used, but on this basis or instead of this, at least one classification model 322 designated by the operator among the two or more classification models 322 stored in the storage unit 32 can be selected as at least one of the classification models 322 to be used. Thus, an arbitrary classification model 322 can be selected by the operator and used for judgment. Therefore, in the case where the current device 2 is normal, the classification model 322 that determines the current measurement data as normal can be selected and used for judgment. In addition, in the test using normal measurement data and abnormal measurement data performed in advance, the classification model indicating high classification accuracy can be selected and used for judgment. The selection unit 39 can fix the classification model 322 as the classification model 322 to be used according to the classification model 322 designated by the operator, and can automatically select the classification model 322 in the process of step S55.

[0081] In addition, the judgment unit 34 takes the logical product of multiple classification results from multiple classification models 322 of the object of use and judges the state of the device 2 as either normal or abnormal, but it is also possible to take the majority decision of multiple classification results and judge the state of the device 2 as either normal or abnormal. On this basis, the judgment unit 34 may also give a greater weight to the classification result of the classification model 322 that has been learned by learning data including measurement data of a newer period among the multiple classification models 322 of the object of use, take the weighted majority decision of multiple classification results, and judge the state of the device 2 as either normal or abnormal. The weighting of the classification results and taking the weighted majority decision may be to increase the number of each classification result according to the weight and take the majority decision. As an example, the classification result of the classification model 322 learned by the learning data of the latest period is set as the first classification result, the classification result of the classification model 322 learned by the learning data of the second newest period is set as the second classification result, and the classification result of the classification model 322 learned by the learning data of the third newest period is set as the third classification result. In addition, the weights of the first to third classification results are set to w1 to w3 (where w1>w2>w3), respectively. In this case, the judgment unit 34 can take the first classification result as the classification result of 1×w1, the second classification result as the classification result of 1×w2, and the third classification result as the classification result of 1×w3, thereby making a majority decision. As a result, the newer the classification model 322, that is, the classification model 322 that is closer to the state of the most recent device 2, the greater the influence on the judgment result, and the judgment result corresponding to the transition of the state of the device 2 can be obtained.

[0082] In addition, it is explained that the judgment unit 34 makes judgments based on multiple classification results output from multiple classification models 322 generated by a common learning algorithm, but in addition to making judgments based on multiple classification results output from these multiple classification models 322, it can also make judgments based on classification results output from classification models generated by other publicly known learning algorithms.

[0083] Furthermore, although the learning processing unit 38 generates the classification model 322 based on the notification received from the setting unit 37 indicating that a learning label is added to the measurement data, the classification model 322 may be generated according to a predetermined schedule or according to an operator's instruction. The learning processing unit 38 may generate the classification model 322 using the learning data of each period in which the state of the device 2 is normal and the learning data that is not used in the learning process.

[0084] In addition, the apparatus 3 is described as having a storage unit 32, a determination unit 35, an output unit 36, a setting unit 37, a learning processing unit 38, and a selection unit 39, but it may not have any of them. In the case where the apparatus 3 does not have the storage unit 32, the measurement data file 321 and the classification model 322 may be stored in a storage device connected to the outside of the apparatus 3. In the case where the apparatus 3 does not have the determination unit 35, the output unit 36 ​​may output the determination result of the determination unit 34 instead of the determination result of the determination unit 35, and the setting unit 37 may set the measurement data of the period determined by the determination unit 34 as the measurement data included in the new learning data. In the case where the apparatus 3 does not have the output unit 36, the determination unit 34 and the determination unit 35 may store the determination result and the determination result in correspondence with the corresponding measurement data in the measurement data file 321. In the case where the apparatus 3 does not have the setting unit 37, the learning processing unit 38 may generate the classification model 322 using the measurement data in the period selected by the operator among the measurement data of each period. When the device 3 does not include the learning processing unit 38, the determination unit 34 may perform determination using only the classification model 322 stored by default in the storage unit 32. When the device 3 does not include the selection unit 39, the determination unit 34 may perform determination using each classification model 322 in the storage unit 32.

[0085] In addition, it is described that each learning data includes measurement data within the same reference time width, but it can also include measurement data within different time widths. In this case, the operator can arbitrarily set the start time and end time of each period. In addition, in the processing of the above-mentioned step S1, the acquisition unit 31 can acquire measurement data during an arbitrary time width when the state of the device 2 is normal. As an example, it can acquire measurement data during an arbitrary time width when the operator determines that the state of the device 2 is normal.

[0086] In addition, it is explained that the setting unit 37 sets the measurement data during the period when the severity determined by the determination unit 35 is maintained below the baseline severity (a high level as an example) as measurement data for learning, that is, the period during which the severity is above the baseline severity is excluded and the measurement data for the remaining period is set as measurement data for learning, but the measurement data for learning can also be set in other ways. For example, when the number of consecutive abnormality judgments is greater than the second reference number "b" (i.e., the reference number for determining the severity as a high level), the setting unit 37 may exclude the measurement data that causes abnormality judgments after the "b"th time from the measurement data acquired by the acquisition unit 31 and supplied from the supply unit 33 to the plurality of classification models 322, and set the remaining measurement data as measurement data for learning. Alternatively, the setting unit 37 may exclude the measurement data that causes consecutive abnormality judgments until the severity determined by the determination unit 35 becomes a high level (i.e., the measurement data that causes the first abnormality judgment to the "b"th abnormality judgment) and the measurement data that causes consecutive abnormality judgments after the measurement data (i.e., the measurement data that causes consecutive abnormality judgments after the "b+1"th time) from the measurement data acquired by the acquisition unit 31 and supplied from the supply unit 33 to the plurality of classification models 322 as measurement data for learning. In these cases, the measurement data for learning may include measurement data within periods of different time widths.

[0087] Furthermore, although the object of monitoring by the apparatus 3 is described as the device 2 , it may be a device installed in the device 2 (for example, an actuator device).

[0088] In addition, it is explained Figure 2 At the start time of the action, no classification model 322 is stored in the storage unit 32, but one or more classification models 322 may be pre-stored. In this case, the action may be started from step S11 in a state where the selection unit 39 pre-selects each classification model 322 in the storage unit 32 as the classification model 322 to be used. In addition, in step S55, the selection unit 39 may select the classification model 322 generated by the learning processing unit 38, or may select the classification model 322 pre-stored in the storage unit 32.

[0089] In addition, it is explained Figure 2 In the case where the new classification model 322 generated by the learning processing unit 38 in the operation is stored in the storage unit 32, the new classification model 322 generated by the learning processing unit 38 in the external device through the same learning process as the learning processing unit 38 can also be stored in the storage unit 32. In this case, based on the classification model 322 stored in the storage unit 32, the selection unit 39 can select any classification model 322 as the classification model 322 to be used.

[0090] In addition, various embodiments of the present invention may be described with reference to flow charts and block diagrams, where a module may represent (1) a stage of a process for performing an operation or (2) a portion of a device having the function of performing an operation. Specific stages and portions may be implemented by dedicated circuits, programmable circuits supplied together with computer-readable instructions stored on a computer-readable medium, and / or processors supplied together with computer-readable instructions stored on a computer-readable medium. Dedicated circuits may include digital and / or analog hardware circuits, and may also include integrated circuits (ICs) and / or discrete circuits. Programmable circuits may include reconfigurable hardware circuits, which include logical AND, logical OR, logical XOR, logical NAND, logical NOR and other logical operations, flip-flops, registers, field programmable gate arrays (FPGAs), programmable logic arrays (PLAs) and other memory elements, etc.

[0091] Computer-readable media may include any tangible device capable of storing instructions executed by an appropriate device, and as a result, a computer-readable medium having instructions stored therein includes a product containing instructions that can be executed in order to make a means for performing the operations specified by the flowchart or block diagram. Examples of computer-readable media may include: electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable media may include: floppy (registered trademark) disks, magnetic disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), electrically erasable programmable read-only memories (EEPROM), static random access memories (SRAM), compact disk read-only memories (CD-ROM), digital versatile disks (DVD), Blu-ray (RTM) disks, memory sticks, integrated circuit cards, etc.

[0092] Computer readable instructions include any of source code and object code described by any combination of one or more programming languages ​​including assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or object-oriented programming languages ​​such as Smalltalk (registered trademark), JAVA (registered trademark), C++, and existing procedural programming languages ​​such as the "C" programming language or similar programming languages.

[0093] The computer readable instructions may be provided to a processor or programmable circuit of a general-purpose computer, a special-purpose computer or other programmable data processing device via a local or local area network (LAN), a wide area network (WAN) such as the Internet, and the computer readable instructions are executed to produce a means for performing the operations specified by the flowchart or block diagram. Examples of processors include: computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0094] Figure 4 The computer 2200 is an example of a computer that can implement various modes of the present invention in whole or in part. The computer 2200 can perform operations associated with the device of the embodiment of the present invention or functions of one or more parts of the device, or perform the operations or the one or more parts, and / or the computer 2200 can perform processes of the embodiment of the present invention or stages of the processes, through the program installed in the computer 2200. In order to make the computer 2200 perform specific operations associated with some or all of the modules in the flowcharts and block diagrams described in this specification, the CPU 2212 can execute such a program.

[0095] The computer 2200 of this embodiment includes a CPU 2212, a RAM 2214, a graphic controller 2216, and a display device 2218, which are connected to each other through a main controller 2210. The computer 2200 also includes an input / output unit such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the main controller 2210 via an input / output controller 2220. The computer also includes a conventional input / output unit such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0096] The CPU 2212 controls each unit by operating according to the program stored in the ROM 2230 and the RAM 2214. The graphic controller 2216 acquires image data generated by the CPU 2212 in a frame buffer or the like provided in the RAM 2214 or in itself, and displays the image data on the display device 2218.

[0097] The communication interface 2222 can communicate with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads programs or data from the DVD-ROM 2201 and provides the programs or data to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0098] The ROM 2230 stores therein a boot program or the like executed by the computer 2200 upon activation and / or a program that depends on the hardware of the computer 2200. The input / output chip 2240 may also connect various input / output units to the input / output controller 2220 via a parallel port, a serial port, a keyboard port, a mouse port, and the like.

[0099] The program is provided by a computer-readable medium such as DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium and installed in the hard disk drive 2224, RAM 2214 or ROM 2230, which is also an example of a computer-readable medium, and is executed by the CPU 2212. The information processing described in these programs is read to the computer 2200, thereby bringing about cooperation between the program and the various types of hardware resources described above. The device or method can be constituted by realizing the operation or processing of information with the use of the computer 2200.

[0100] For example, when communication is performed between the computer 2200 and an external device, the CPU 2212 can execute a communication program loaded in the RAM 2214, and instruct the communication interface 2222 to perform communication processing based on the processing described in the communication program. Under the control of the CPU 2212, the communication interface 2222 reads the transmission data stored in the transmission buffer processing area provided in the RAM 2214, the hard disk drive 2224, the DVD-ROM 2201, or a recording medium such as an IC card, and transmits the read transmission data to the network, or writes the reception data received from the network to the reception buffer processing area provided on the recording medium, etc.

[0101] Furthermore, the CPU 2212 can read all or a necessary part of a file or database stored in an external recording medium such as the hard disk drive 2224, the DVD-ROM drive 2226 (DVD-ROM 2201), or an IC card into the RAM 2214, and perform various types of processing on the data on the RAM 2214. Then, the CPU 2212 writes the processed data back to the external recording medium.

[0102] Various types of information such as various types of programs, data, tables, and databases can be stored in the recording medium and receive information processing. CPU2212 performs various types of processing described in various places of this disclosure on the data read from RAM2214 and writes the results back to RAM2214. The various types of processing include various types of operations specified by the instruction sequence of the program, information processing, conditional judgment, conditional branching, unconditional branching, information retrieval / replacement, etc. In addition, CPU2212 can retrieve information in files, databases, etc. in the recording medium. For example, in the case where multiple entries having attribute values ​​of the first attribute associated with the attribute values ​​of the second attribute are stored in the recording medium, CPU2212 can retrieve an entry consistent with the condition specifying the attribute value of the first attribute from the multiple entries, and read the attribute value of the second attribute stored in the entry, thereby obtaining the attribute value of the second attribute associated with the first attribute that meets the predetermined condition.

[0103] The program or software module described above may be stored in a computer-readable medium on or near the computer 2200. In addition, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable medium, thereby providing the program to the computer 2200 via the network.

[0104] The present invention has been described above using the embodiments, but the technical scope of the present invention is not limited to the scope described in the above embodiments. It is obvious to those skilled in the art that various changes or improvements can be made to the above embodiments. According to the description of the claims, the method of making such changes or improvements can also be included in the technical scope of the present invention.

[0105] The execution order of each process such as actions, processes, steps and stages in the apparatus, system, program and method shown in the claims, specifications and drawings is not specifically indicated as "earlier", "before", etc. In addition, it should be noted that as long as the output of the previous process is not used in the subsequent process, it can be implemented in any order. Even if the action flow in the claims, specifications and drawings is described using "first," "next," etc. for the sake of convenience, it does not mean that it must be implemented in this order.

Claims

1. A device, characterized in that have: an acquisition unit that acquires measurement data indicating a state of an object; a supply unit that supplies the measurement data acquired by the acquisition unit to a plurality of classification models, the plurality of classification models respectively completing learning by using learning data of each period different from each other including measurement data of a period in which the state of the object is normal, and the plurality of classification models classifying the measurement data as either normal or abnormal based on input of the measurement data; as well as The determination unit determines whether the state of the object is normal or abnormal based on the plurality of classification results output from the plurality of classification models.

2. The device according to claim 1, characterized in that The device further includes a learning processing unit that generates a new classification model included in the plurality of classification models by performing a learning process using learning data including measurement data of the period acquired by the acquisition unit, for each period in which the state of the object is normal.

3. The device according to claim 1, characterized in that The device also includes a selection unit, which selects the multiple classification models from more than two classification models, and the two or more classification models respectively complete learning by using learning data of each different period including measurement data of a period in which the state of the object is normal, and classifies the measurement data into either normal or abnormal based on the input of the measurement data.

4. The device according to claim 3, characterized in that The selection unit selects at least one classification model learned by learning data including measurement data of a latest period as at least one of the plurality of classification models.

5. The device according to claim 3, characterized in that The selection unit selects at least one classification model designated by an operator among the two or more classification models as at least one of the plurality of classification models.

6. The device according to claim 1, characterized in that The determination unit calculates a logical product of the plurality of classification results and determines whether the state of the object is normal or abnormal.

7. The device according to claim 1, characterized in that The determination unit determines the state of the object as either normal or abnormal by taking a majority decision of the plurality of classification results.

8. The device according to claim 7, characterized in that The judgment unit gives a greater weight to the classification result of the classification model among the multiple classification models that has completed learning by using learning data that includes measurement data from a more recent period, and takes a weighted majority decision of the multiple classification results to judge the state of the object as either normal or abnormal.

9. The device according to claim 1, characterized in that The apparatus further includes a determination unit that determines the severity of the abnormal state of the subject based on the number of times the state of the subject is continuously determined to be abnormal.

10. The device according to claim 9, characterized in that The device also includes a setting unit, which sets any one of the measurement data of each period acquired by the acquisition unit and supplied from the supply unit to the multiple classification models, the measurement data of the period in which the severity determined by the determination unit is lower than the benchmark severity, and the measurement data acquired by the acquisition unit and supplied from the supply unit to the multiple classification models, except for each measurement data that caused continuous abnormality judgments until the severity determined by the determination unit reached the benchmark severity and each measurement data that continued to cause continuous abnormality judgments after the measurement data, as measurement data included in the new learning data.

11. A method, characterized in that have: The acquisition phase acquires measurement data representing the state of the object; A supplying stage, supplying the measurement data acquired in the acquiring stage to a plurality of classification models, wherein the plurality of classification models complete learning by using learning data of each different period including measurement data of a period in which the state of the object is normal, and classifying the measurement data into either normal or abnormal according to the input of the measurement data; as well as In the judgment stage, the state of the object is judged as either normal or abnormal based on the multiple classification results output from the multiple classification models.

12. A computer-readable medium, characterized in that A program is stored, and the computer functions as an acquisition unit, a supply unit, and a determination unit by executing the program. The acquisition unit acquires measurement data indicating a state of the object, The supply unit supplies the measurement data acquired by the acquisition unit to a plurality of classification models, the plurality of classification models respectively complete learning by using learning data of each period different from each other including the measurement data of a period in which the state of the object is normal, and classify the measurement data into either normal or abnormal according to the input of the measurement data, The determination unit determines whether the state of the object is normal or abnormal based on a plurality of classification results output from the plurality of classification models.

Images