Abnormality cause estimation method, abnormality cause estimation device and recording medium

Abstract

The purpose of the present disclosure is to provide an abnormality cause estimation method that can improve the estimation accuracy of the abnormality cause. A method for estimating the abnormality cause includes: a step of calculating a priori probability as the probability of occurrence of each cause based on a cause table showing the occurrence frequency of the cause for each event; a step of calculating a posterior probability as the probability that a certain event is caused by a certain cause; and a step of multiplying the posterior probability by a weight coefficient related to the SN ratio gain value of the sensor measurement value, and calculating an index representing the probability of occurrence for each combination of the cause and the event.

Description

Technical Field

[0001] The present disclosure relates to an abnormality cause estimation method, an abnormality cause estimation device, and a recording medium. Background Art

[0002] Various devices are used in power generation equipment, boilers, gas turbines, chemical equipment, etc. It is desired to detect abnormalities (such as failures, signs of failures) in these devices or to estimate the causes thereof.

[0003] For example, Patent Document 1 discloses a method for diagnosing equipment faults (estimation method) that estimates the cause of a fault and the location of a fault using a weighted number of each cause based on a FT diagram (Fault Tree (FT) Diagram) including multiple causes and the know-how of a technician. In addition, an estimation method is also known that detects an abnormality of equipment by the Mahalanobis-Taguchi method, and estimates the occurrence and cause of the event by referring to the SN ratio (Signal to Noise Ratio) gain value of the sensor measurement value that affects the increase in the Mahalanobis distance.

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Publication No. 2006-99298 Summary of the invention

[0007] Problems to be solved by the invention

[0008] The above-mentioned conventional estimation method is a method of determining the accuracy of each cause, selecting possible causes from them, and estimating the cause of the abnormality. However, in reality, sometimes the abnormality of the equipment occurs due to the interaction of multiple causes, or multiple causes are mixed as the cause of the abnormality. Therefore, in the estimation of the cause of the abnormality achieved by the calculation result (absolute evaluation) of each cause, the estimation accuracy of the cause of the abnormality may be reduced.

[0009] In view of the above circumstances, an object of the present disclosure is to provide an abnormality cause estimation method and the like that can improve the estimation accuracy of the abnormality cause.

[0010] Technical Solution

[0011] The abnormality cause estimation method disclosed in the present invention includes:

[0012] A step of calculating a priori probability which is a probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event;

[0013] the step of calculating a posterior probability as the probability that an event was caused by a stated cause; and

[0014] A step of multiplying the posterior probability by a weight coefficient associated with an SN ratio gain value of a sensor measurement value, and calculating an index representing an occurrence probability for each combination of the cause and the event.

[0015] The abnormality cause estimation device disclosed in the present invention comprises:

[0016] a priori probability calculation unit configured to calculate the priori probability as the probability of occurrence of each cause based on a cause table showing the occurrence frequency of the cause for each event;

[0017] a posterior probability calculation unit configured to calculate a posterior probability as a probability that a certain event is caused by a certain cause; and

[0018] The index calculation unit multiplies the posterior probability by a weight coefficient associated with an SN ratio gain value of the sensor measurement value, and calculates an index indicating the probability of occurrence for each combination of the cause and the event.

[0019] The program of the present disclosure causes a computer to execute the following steps:

[0020] Calculating a priori probability indicating the probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event;

[0021] calculating the a posteriori probability as the probability that a stated event was caused by a stated cause; and

[0022] The posterior probability is multiplied by a weight coefficient associated with the SN ratio gain value of the sensor measurement value, and an index representing the probability of occurrence is calculated for each combination of the cause and the event.

[0023] Effects of the Invention

[0024] According to the present disclosure, it is possible to provide an abnormality cause estimation method and the like that can improve the estimation accuracy of the abnormality cause. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a block diagram schematically showing the hardware configuration of an abnormality cause estimation device according to one embodiment.

[0026] Figure 2 This is a block diagram schematically showing the functional structure of an abnormality cause estimation device according to one embodiment.

[0027] Figure 3 This is a flowchart for explaining an example of processing executed by the abnormality cause estimation device according to one embodiment.

[0028] Figure 4 1 is a diagram showing an example of the SN ratio gain value acquired by the abnormality cause estimation device according to one embodiment.

[0029] Figure 5 This is a diagram showing an example of a sensor table used by the abnormality cause estimation device according to one embodiment.

[0030] Figure 6 This is a diagram showing an example of calculation results of the abnormality cause estimation device according to one embodiment.

[0031] Figure 7 This is a diagram showing an example of calculation results (weight coefficients) of the abnormality cause estimation device according to one embodiment.

[0032] Figure 8 This is a diagram showing an example of the extraction result of the cause table by the abnormality cause estimation device according to one embodiment.

[0033] Fig. 9 This is a diagram showing an example of a calculation result of a priori probability by the abnormality cause estimation device according to one embodiment.

[0034] Fig.10 This is a diagram showing an example of a calculation result of likelihood by the abnormality cause estimation device according to one embodiment.

[0035] Fig.11 This is a diagram showing an example of calculation results of a posterior probability by the abnormality cause estimation device according to one embodiment.

[0036] Fig.12 This is a diagram showing an example of calculation results of the abnormality cause estimation device according to one embodiment.

[0037] Fig.13 This is a diagram showing an example of information output by the abnormality cause estimation device according to one embodiment. DETAILED DESCRIPTION

[0038] Several embodiments are described below with reference to the drawings. However, the dimensions, materials, shapes, and relative arrangements of the components described in the embodiments or shown in the drawings are not intended to limit the scope of the invention and are merely illustrative examples.

[0039] For example, expressions such as "in a certain direction", "along a certain direction", "parallel", "orthogonal", "center", "concentric" or "coaxial" that indicate relative or absolute configuration not only strictly indicate such a configuration, but also indicate a state of relative displacement at an angle or distance with a tolerance or a degree that can achieve the same function.

[0040] For example, expressions such as “same”, “equal”, and “homogeneous” that indicate that things are in an equal state not only strictly indicate an equal state, but also indicate a state in which there is a tolerance or a degree of difference that enables the same function to be obtained.

[0041] For example, expressions indicating shapes such as a quadrilateral shape and a cylindrical shape not only indicate shapes such as a quadrilateral shape and a cylindrical shape in a strict geometric sense, but also indicate shapes including concave and convex portions, chamfered portions, etc. within a range that can obtain the same effect.

[0042] On the other hand, expressions such as “having”, “including”, “having”, “including” or “having” a constituent element are not exclusive expressions that exclude the existence of other constituent elements.

[0043] (Configuration of abnormality cause estimation device)

[0044] The overall configuration of an abnormality cause estimation device 100 according to one embodiment will be described. Figure 1 This is a block diagram schematically showing the hardware configuration of an abnormality cause estimation device 100 according to one embodiment. The abnormality cause estimation device 100 is a device for estimating the cause of an abnormal event occurring in a device by acquiring sensor measurement values ​​from a sensor (not shown) that performs measurement related to the device, a server device (not shown) that stores the sensor measurement values, and the like.

[0045] For example Figure 1 As shown, the abnormality cause estimation device 100 is configured using a computer, which includes a processor 72 such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit); a RAM (Random Access Memory) 74; a ROM (Read Only Memory) 76; a HDD (Hard Disk Drive) 78; an input I / F (Input Interface) 80; and an output I / F (Output Interface) 82, which are connected to each other via a bus 84. The processor 72 of the abnormality cause estimation device 100 implements various functions described below by executing programs stored in a memory such as a ROM and a RAM.

[0046] Figure 2 1 is a block diagram schematically showing the functional structure of the abnormality cause estimation device 100 according to one embodiment. Figure 2As shown, the abnormality cause estimation device 100 functionally comprises: a gain acquisition unit 101, configured to acquire the SN ratio gain value; a weight coefficient acquisition unit 102, configured to acquire the weight coefficient; a priori probability calculation unit 103, configured to calculate the priori probability; a posterior probability calculation unit 104, configured to calculate the posterior probability; an index calculation unit 105, configured to calculate the index; and an output unit 106, configured to output various information (including estimation results).

[0047] (Processing flow)

[0048] Hereinafter, the flow of processing executed by the abnormality cause estimation device 100 according to one embodiment will be described. Figure 3 This is a flowchart for explaining an example of processing executed by the abnormality cause estimation device 100 according to one embodiment.

[0049] like Figure 3 As shown, the gain acquisition unit 101 acquires the SN ratio gain value of the sensor measurement value (step S1). It should be noted that the gain acquisition unit 101 is configured to monitor the Mahalanobis distance based on the sensor measurement value, and acquire the SN ratio gain value when an abnormal event is detected based on the Mahalanobis distance. In this case, when an abnormal event is detected, the abnormal cause is estimated, so that the calculation process accompanying the abnormal cause estimation can be suppressed.

[0050] Figure 4 FIG. 1 is a diagram showing an example of the SN ratio gain value acquired by the abnormality cause estimation device 100 according to one embodiment. Figure 4 As shown in FIG. 1 , the gain acquisition unit 101 acquires the SN ratio gain value of each signal from a plurality of sensor measurement values ​​(signals A to F) that can be acquired in time series. Figure 4 , the case where the SN ratio gain value at one time is acquired is shown, but the gain acquisition unit 101 may acquire the SN ratio gain value at each time at a plurality of times.

[0051] like Figure 3 As shown, the weight coefficient acquisition unit 102 acquires the weight coefficient related to the SN ratio gain value of the sensor measurement value (step S2). The weight coefficient may be the SN ratio gain value, or may be a value obtained by performing threshold processing on the SN ratio gain value as in the following example.

[0052] Here, an example of a case where a weight coefficient is acquired by threshold processing will be described. Figure 5 1 is a diagram showing an example of a sensor table used by the abnormality cause estimation device 100 according to one embodiment. Figure 6 It is a diagram showing an example of calculation results of the abnormality cause estimation device 100 according to one embodiment. Figure 7It is a diagram showing an example of calculation results (weight coefficients) of the abnormality cause estimation device 100 according to one embodiment.

[0053] The weight coefficient acquisition unit 102 can use the weight coefficient to calculate the weight coefficient. Figure 5 The sensor table shown in the figure. The sensor table is information indicating which signal is associated with which occurrence event. In the sensor table, for example, for signal A, it is related to occurrence events 2, 4, and 8, so the value is 1, while it is not related to other occurrence events, so the value is 0. The correlation is represented by these two values ​​1 and 0. It should be noted that the weight coefficient acquisition unit 102 is not limited to the configuration using the sensor table. The weight coefficient acquisition unit 102 can also be configured to extract the occurrence events related to each signal without using the sensor table.

[0054] The occurrence event is, for example, gas turbine shaft vibration, abnormal increase in exhaust temperature, etc. For example, if the cause of the increase in exhaust temperature is insufficient cooling air, sensor failure, etc. There are cases where the occurrence event and the cause are clearly one-to-one corresponding, but there are also events that occur due to complex causes, so such screening is meaningful.

[0055] The weight coefficient acquisition unit 102 can Figure 5 Multiply the sensor table values ​​shown by Figure 4 The SN ratio gain values ​​shown are used to obtain Figure 6 The calculation results are shown. Figure 6 In the calculation results shown, for signal A, the occurrence of events 1, 3, 5 to 7 is the product of 0 and 5.1, so the value is 0; the occurrence of events 2, 4, and 8 is the product of 1 and 5.1, so the value is 5.1. For signals B to F, Figure 4 The SN ratio gain values ​​shown are Figure 5 The same calculation is performed for the product of 1 or 0 of the sensor table shown.

[0056] The weight coefficient acquisition unit 102 can obtain Figure 6 The calculation result shown in the figure obtains the A value. The A value is the maximum value of the SN ratio gain value in each occurrence event. For example, Figure 6 In the occurrence event 1 shown in FIG. 1 , the calculated value corresponding to signal B is 2.2, the calculated value corresponding to signal F is 1.8, and the calculated values ​​corresponding to the other signals are 0. In this case, Figure 7 As shown, the A value of occurrence event 1 is 2.2. The A values ​​of other occurrence events are obtained in the same manner.

[0057] The weight coefficient acquisition unit 102 can be as follows Figure 7 As shown, the threshold of the SN ratio gain value is used to obtain the B value corresponding to the A value. Figure 7In the example shown, the threshold of the SN ratio gain value is 3. The SN ratio gain value is set to a value that serves as a criterion for estimating that an effect has been exerted on the increase of the Mahalanobis distance. The threshold of the SN ratio gain value may be a value manually input by the user based on his / her knowledge, or a value automatically calculated by a statistical method. Figure 7 In the example shown, the same threshold value 3 is set for all events. However, a different threshold value may be set for each event.

[0058] The B value is the value obtained by filtering the A value through the threshold of the SN ratio gain value. In general, the B value is a value that sets the A value below the threshold of the SN ratio gain value to 0 and sets the A value that exceeds the threshold of the SN ratio gain value directly. For example, for event 1, the A value is 2.2, which is below 3, so the B value is 0. For event 2, the A value is 5.1, which is above 3, so the B value is 5.1.

[0059] The weight coefficient acquisition unit 102 can be as follows Figure 7 As shown, the B value is used to obtain the C value of each occurrence event. The C value is a weight coefficient. The C value as a weight coefficient can be a coefficient set based on the excess amount of the SN ratio gain value relative to the threshold. For example, in the example shown in the figure, the C value is calculated by multiplying the value obtained by subtracting the threshold of the SN ratio gain value from the B value by 2. However, the calculation formula of the C value is not limited to this and can be appropriately changed.

[0060] In the above, an example of obtaining a weight coefficient by threshold processing is described. However, the step of obtaining a weight coefficient by threshold processing is not limited to the above example. For example, the weight coefficient obtaining unit 102 may be used in Figure 4 The SN ratio gain values ​​shown are pre-thresholded and multiplied by Figure 5 In addition, the weight coefficient acquisition unit 102 is not configured as Figure 7 In the example shown, the B value is calculated from the A value and then the C value is calculated, but the C value is calculated directly from the A value and the C value with a negative calculation result is set to 0. Even if the steps are changed in this way, the same result can be obtained.

[0061] If such a weight coefficient is used in the calculation of the index described later, not only the event and cause of the SN ratio gain value being greater than the reference value can be calculated, but also the magnitude relationship of the SN ratio gain value can be reflected in the weight coefficient and used in the calculation. Therefore, the magnitude relationship of the SN ratio gain value can be made significant.

[0062] like Figure 3 As shown, the priori probability calculation unit 103 calculates the priori probability which is the probability of occurrence of each cause (step S3). The priori probability calculation unit 103 calculates the priori probability based on the cause table which shows the occurrence frequency of the cause for each event.

[0063] The information included in the cause table also includes information such as rarely occurring events, special events, and events with unknown causes. When such information is used to estimate the cause of the abnormality, there is a problem of reduced accuracy. Therefore, the prior probability calculation unit 103 may be configured to extract and use information related to events and causes associated with sensor measurement values ​​with SN ratio gain values ​​greater than the reference value from the information included in the cause table. In this case, the information related to events and causes associated with sensor measurement values ​​with SN ratio gain values ​​greater than the reference value from the information included in the cause table can be filtered out and the prior probability and posterior probability can be calculated, thereby achieving high accuracy. It should be noted that the prior probability calculation unit 103 may also be configured to directly calculate the prior probability based on the cause table without performing such extraction.

[0064] Figure 8 FIG. 1 is a diagram showing an example of the extraction result of the cause table based on the abnormality cause estimation device 100 according to one embodiment. Figure 4 As shown, the SN ratio gain can be seen in the six sensors (signals), and the events related to the six signals (signals A to F) are as follows: Figure 5 The eight occurrence events 1 to 8 are shown. These eight occurrence events and their respective causes are extracted from the cause table to form Figure 8 In this way, the prior probability calculation unit 103 extracts several occurrence events and their respective causes from the cause table that associates occurrence events with causes. Figure 8 For example, for event 1, the number of occurrences due to cause 7 is 2 times, the number of occurrences due to cause 8 is 3 times, and the number of occurrences due to causes 1 to 6 is 0 times. Figure 8 The extraction result shown shows the number of occurrences of each occurrence event by cause, and also shows the subtotal value for each cause (for example, the number of occurrences of cause 7 is 27 times) and the total value obtained by adding up the subtotal values.

[0065] Fig. 9 This is a diagram showing an example of a calculation result of a priori probability by the abnormality cause estimation device 100 according to one embodiment. Fig. 9 R1~R8 shown in the figure are Figure 8 The causes 1 to 8 shown in the figure correspond to each other. P represents the probability. For example, P(R1) represents the probability of cause 1 occurring (the prior probability of cause 1). The prior probability is calculated by Figure 8 The prior probability calculation unit 103 performs such calculation for each cause.

[0066] like Figure 3 As shown, the posterior probability calculation unit 104 calculates the posterior probability as the probability that a certain event is caused by a certain cause (step S4). The posterior probability calculation unit 104 can calculate the posterior probability P(Rj|Fi) using the prior probability of each cause and the likelihood P(Fi|Rj) as the probability that each event Fi occurs due to each cause Rj. In this case, the posterior probability can be easily calculated using the prior probability and the likelihood.

[0067] Fig.10 1 is a diagram showing an example of calculation results of likelihood P(Fi|Rj) by the abnormality cause estimation device 100 according to one embodiment. Fig.10 Fi shown shows any occurrence event i among occurrence events 1 to 8. For example, P(Fi|R1) shows the likelihood as the probability that event Fi occurs due to cause R1. For example, the likelihood that event F2 (occurrence event 2) occurs due to cause R1 is 0.30303.

[0068] By will be Figure 8 The likelihood is calculated by dividing the number of times an event Fi occurs due to a certain cause Ri by the subtotal value of each cause in the extraction result of the cause table shown in the figure. For example, the likelihood P(F1|R7) is the value of dividing the number of times event F1 occurs due to cause R7 by 2 divided by the subtotal value 27, as shown in Fig.10 As shown, it is 0.074074. Fig.10 As shown, the posterior probability calculation unit 104 may calculate such likelihood for each combination of each event and each cause.

[0069] The posterior probability calculation unit 104 may calculate the posterior probability using such likelihood. Fig.11 This is a diagram showing an example of a calculation result of a posterior probability P(Rj|Fi) of an abnormality cause estimation device 100 according to an embodiment. The posterior probability P(Rj|Fi) is calculated as the probability that a certain event Fi is caused by a certain cause Rj by dividing the product of a certain likelihood P(Fi|Rj) and a certain prior probability P(Rj) by the sum of the products of each likelihood and each prior probability corresponding to the cause Rj.

[0070] The posterior probability P(Rj|Fi) when there are n causes is calculated by the formula of posterior probability P(Rj|Fi)=P(Fi|Rj)·P(Rj) / (P(Fi|R1)·P(R1)+P(Fi|R2)·P(R2)+···+P(Fi|Rn)·P(Rn)). The posterior probability is calculated by the formula that also takes into account the case where a certain event is caused by multiple causes (R1 to Rn).

[0071] It should be noted that, for example, Fig.11In the case of finding the posterior probability P(R1|F2) that event 2 is caused by cause 1, the posterior probability P(R1|F2) = P(F2|R1)·P(R1) / (P(F2|R1)·P(R1)+P(F2|R2)·P(R2)+P(F2|R3)·P(R3)+P(F2|R4)·P(R4)+P(F2|R5)·P(R5)+P(F2|R6)·P(R6)+P(F2|R7)·P(R7)+P(F2|R8)·P(R8)) = 0.357143. It should be noted that in Fig.11 In the figure, the posterior probability is subtotaled for each cause or each event. If these are totaled, the total value becomes 8, which is the same as the number of events that occurred, as shown in the lower right corner of the figure.

[0072] like Figure 3 As shown, the index calculation unit 105 calculates the index representing the probability of occurrence (step S5). The index calculation unit 105 multiplies the posterior probability calculated by the posterior probability calculation unit 104 by the weight coefficient (C value) obtained by the weight coefficient acquisition unit 102, and calculates the index representing the probability of occurrence for each combination of the cause and the occurrence event. The index representing the probability of occurrence can be a normalized value of the previous stage, or it can be a normalized probability of occurrence.

[0073] First, a specific example of the former will be described. Fig.12 This is a diagram showing an example of calculation results of the abnormality cause estimation device 100 according to one embodiment. For each combination of a cause and an occurrence event, a value obtained by multiplying the posterior probability calculated by the posterior probability calculation unit 104 by the weight coefficient (C value) acquired by the weight coefficient acquisition unit 102, that is, an index representing the occurrence probability, is shown.

[0074] For example, the index representing the probability of event 1 occurring due to cause 2 is the product of the posterior probability 0 and the C value 0, so it is 0. The index representing the probability of event 2 occurring due to cause 2 is the product of the posterior probability 0.53714 and the C value 4.2, so it is 2.25. Figure 7 The C value corresponding to each occurrence event is multiplied by Fig.11 The index representing each occurrence probability is calculated using the values ​​of the posterior probability table shown.

[0075] It should be noted that in Fig.12 In the figure, the index representing the probability of occurrence is subtotaled for each cause or each event. If these are added up, the total value becomes 13.2 as shown in the lower right corner of the figure. This total value is affected by the weight coefficient and thus becomes a value different from the number of events. In this example, it is not 8 but 13.2.

[0076] Next, a specific example of the latter will be described. Fig.13 It is a diagram showing an example of information output by the abnormality cause estimation device 100 according to one embodiment. Fig.13 The probability of occurrence shown shows that by Fig.12 The values ​​shown are divided by the total value of 13.2 to normalize the probability of occurrence. Fig.13 In the table of the probability of occurrence, if the subtotals of the probability of occurrence of each cause are added up, then it becomes 1, i.e., 100%, as shown in the lower right of the figure. In this way, the index representing the probability of occurrence is the probability of occurrence of each cause when a certain event occurs, and the probability of occurrence of each cause can also be calculated by dividing the subtotal by the total, wherein the subtotal is the subtotal of the multiplication of the posterior probability and the weight coefficient for each cause, and the total is the value of adding up these subtotals in all the causes. In this case, the probability of occurrence of each cause can be grasped.

[0077] like Figure 3 As shown in FIG. 1 , the output unit 106 outputs the estimation result (step S6). Fig.13 As shown, the output unit 106 can mark the probability of occurrence of each cause in order of magnitude and output the cause with the highest level. In this case, it is easy to understand which cause caused the abnormal event. For example, it can be known that the probability of the event occurring due to cause 2 is 24.7%, which is the highest level. It should be noted that the output unit 106 can output only one cause with the highest level, or it can output multiple causes selected in sequence from the highest level. In addition, as Fig.12 or Fig.13 As shown, the output unit 106 may output an index representing the probability of occurrence as an estimation result.

[0078] Above, refer to Figure 3 The process flow of the process performed by the abnormality cause estimation device 100 according to one embodiment has been described. Figure 3 Part of the processing shown may also be performed by the user. Figure 3 The order of processing shown. For example, steps S3 and S4 may be executed before steps S1 and S2. The prior probability, likelihood, and posterior probability may be calculated based on the cause table instead of the sensor measurement value. Therefore, the prior probability, likelihood, and posterior probability may be calculated in advance, and then the SN ratio gain value and weight coefficient may be obtained to calculate an index representing the probability of occurrence.

[0079] In addition, the abnormal cause inference device 100 may also use a table representing pre-acquired posterior probabilities to calculate an indicator representing the probability of occurrence instead of using a cause table or calculating prior probabilities, likelihoods, and posterior probabilities. In this case, steps S3 and S4 may be omitted. Preferably, the cause table and tables representing prior probabilities, likelihoods, and posterior probabilities are updated in a manner that reflects the latest information. Such an update process may be performed automatically by the abnormal cause inference device 100 or by manual input by the user. In the case where the update frequency is high, it is preferred as follows. Figure 3 As shown, the prior probability, likelihood and posterior probability are calculated each time.

[0080] The abnormal cause estimation device 100 can also be configured to select a cause table used in the calculation from a plurality of cause tables for each process. For example, in a device or a system equipped with the device, in a transitional action state such as when starting or stopping, or in a special action state such as when the measured values ​​of temperature, pressure, and vibration are above the usual 2σ (reference value of the dispersion value), sometimes an activity different from the usual action state is indicated. In this case, the event and the cause also differ depending on the process of the operating state. Therefore, for example, sometimes a cause table is prepared for each process such as when starting, when operating, and when stopping, and by selecting and using the cause table corresponding to the process when the abnormal event occurs, the estimation accuracy can be improved.

[0081] The present invention is not limited to the above-described embodiment, and includes modified forms of the above-described embodiment and forms in which these forms are appropriately combined.

[0082] (Summarize)

[0083] The contents described in the above-mentioned embodiments can be understood, for example, as follows.

[0084] (1) The abnormality cause estimation method disclosed in the present invention includes:

[0085] A step of calculating a priori probability which is a probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event;

[0086] the step of calculating a posterior probability as the probability that an event was caused by a stated cause; and

[0087] A step of multiplying the posterior probability by a weight coefficient associated with an SN ratio gain value of a sensor measurement value, and calculating an index representing an occurrence probability for each combination of the cause and the event.

[0088] According to the above method, an index representing the probability of occurrence is calculated for each combination of cause and event based on the posterior probability. The posterior probability is calculated by a formula that also takes into account the situation where a certain event is caused by multiple causes. Therefore, instead of inferring the abnormal cause by the calculation result (absolute evaluation) of each cause, the abnormal cause can be inferred by the calculation (relative evaluation) that takes into account the interaction of multiple causes. Therefore, the accuracy of inferring the abnormal cause can be improved.

[0089] (2) In some embodiments, according to the method described in (1) above,

[0090] In the step of calculating the prior probability, information on the event and the cause associated with the sensor measurement value in which the SN ratio gain value is greater than a reference value is extracted and used from the information included in the cause table.

[0091] The information included in the cause table also includes information on rarely occurring events, special events, and events with unknown causes. When such information is used to estimate the cause of an abnormality, there is a problem of reduced accuracy. In this regard, according to the above method, information related to events and causes associated with sensor measurement values ​​with SN ratio gain values ​​greater than the reference value can be screened out from the information included in the cause table and the prior probability and the posterior probability can be calculated, thereby achieving high accuracy.

[0092] (3) In some embodiments, according to the method described in (1) or (2) above,

[0093] In the step of calculating the posterior probability, the posterior probability is calculated using the prior probability of each cause and the likelihood which is the probability of each event occurring due to each cause.

[0094] According to the above method, the posterior probability can be easily calculated using the prior probability and the likelihood.

[0095] (4) In some embodiments, according to any one of the methods (1) to (3) above,

[0096] The weight coefficient is a coefficient set based on the amount by which the SN ratio gain value exceeds a threshold value.

[0097] According to the above method, not only the events and causes of the SN ratio gain value being greater than the reference value are calculated, but also the magnitude relationship of the SN ratio gain value is reflected in the weight coefficient and used in the calculation. Therefore, the magnitude relationship of the SN ratio gain value can be made significant.

[0098] (5) In some embodiments, according to any one of the methods (1) to (4) above,

[0099] The index representing the probability of occurrence is the probability of occurrence of each cause when a certain event occurs.

[0100] The probability of occurrence of each cause can also be calculated by dividing the subtotal value by the total value, wherein the subtotal value is the subtotal of the product of the posterior probability and the weight coefficient for each cause, and the total value is the sum of the subtotal values ​​among all the causes.

[0101] According to the above method, the occurrence probability of each cause can be grasped.

[0102] (6) In some embodiments, the method according to (5) above includes:

[0103] The occurrence probability of each cause is ranked in order of magnitude, and the step of outputting the cause at the highest level is performed.

[0104] According to the above method, it is possible to easily understand the cause of the abnormal event.

[0105] (7) In some embodiments, the method according to any one of (1) to (6) above includes:

[0106] a step of monitoring a Mahalanobis distance based on the sensor measurement value; and

[0107] A step of acquiring the SN ratio gain value when the event of abnormality is detected based on the Mahalanobis distance.

[0108] According to the above method, when an abnormal event is detected, the abnormality cause is estimated, so that the calculation process accompanying the abnormality cause estimation can be suppressed.

[0109] (8) In some embodiments, the method according to any one of (1) to (7) above further includes:

[0110] A step of selecting the cause table used in calculation from among the plurality of cause tables for each process.

[0111] According to the above method, by selecting and using the cause table corresponding to the process when the abnormal event occurred, it is possible to improve the estimation accuracy.

[0112] (9) The abnormality cause estimation device (100) of the present disclosure comprises:

[0113] A priori probability calculation unit (103) is configured to calculate a priori probability as a probability of occurrence of each cause based on a cause table showing the occurrence frequency of the cause for each event;

[0114] a posterior probability calculation unit (104) configured to calculate a posterior probability as a probability that a certain event is caused by a certain cause; and

[0115] An index calculation unit (105) multiplies the posterior probability by a weight coefficient associated with an SN ratio gain value of a sensor measurement value, and calculates an index representing the probability of occurrence for each combination of the cause and the event.

[0116] According to the above configuration, the abnormality cause estimation device (100) calculates an index representing the probability of occurrence for each combination of cause and event based on the posterior probability. The posterior probability is calculated by a formula that also takes into account the situation where a certain event is caused by multiple causes. Therefore, instead of estimating the abnormality cause by the calculation result (absolute evaluation) of each cause, the abnormality cause can be estimated by the calculation (relative evaluation) that takes into account the interaction of multiple causes. Therefore, the estimation accuracy of the abnormality cause can be improved.

[0117] (10) The program of the present disclosure causes a computer to execute the following steps:

[0118] Calculating a priori probability as the probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event;

[0119] calculating the a posteriori probability as the probability that a stated event was caused by a stated cause; and

[0120] The posterior probability is multiplied by a weight coefficient associated with the SN ratio gain value of the sensor measurement value, and an index representing the probability of occurrence is calculated for each combination of the cause and the event.

[0121] According to the above procedure, an index representing the probability of occurrence is calculated for each combination of cause and event based on the posterior probability. The posterior probability is calculated by a formula that also takes into account the situation where a certain event is caused by multiple causes. Therefore, instead of inferring the abnormal cause by the calculation result (absolute evaluation) of each cause, the abnormal cause can be inferred by the calculation (relative evaluation) that takes into account the interaction of multiple causes. Therefore, the accuracy of inferring the abnormal cause can be improved.

[0122] Description of Reference Numerals

[0123] 72 Processors

[0124] 74 RAM

[0125] 76 ROM

[0126] 78 HDD

[0127] 80 Input I / F

[0128] 82 Output I / F

[0129] 84 Bus

[0130] 100 Abnormality cause estimation device

[0131] 101 Gain acquisition unit

[0132] 102 Weight coefficient acquisition unit

[0133] 103 Prior Probability Calculation Department

[0134] 104 Posterior Probability Calculation Unit

[0135] 105 Index Calculation Department

[0136] 106 Output

Claims

1. A method for inferring the cause of an abnormality, comprising: A step of calculating a priori probability which is a probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event; A step of calculating a posterior probability as the probability that a certain event is caused by a certain cause using the prior probability of each cause and the likelihood as the probability that each event occurs due to each cause; as well as The step of multiplying the posterior probability by a weight coefficient for each of the events related to an SN ratio gain value of a sensor measurement value, and calculating an index representing the probability of occurrence for each combination of the cause and the event.

2. The abnormality cause estimation method according to claim 1, wherein: In the step of calculating the prior probability, information on the event and the cause associated with the sensor measurement value in which the SN ratio gain value is greater than a reference value is extracted and used from the information included in the cause table.

3. The abnormality cause estimation method according to claim 1 or 2, wherein: The weight coefficient is a coefficient set based on the amount by which the SN ratio gain value exceeds a threshold value.

4. The abnormality cause estimation method according to claim 1 or 2, wherein: The index representing the probability of occurrence is the probability of occurrence of each cause when a certain event occurs. The occurrence probability of each cause is calculated by dividing a subtotal value obtained by subtotaling the product of the posterior probability and the weight coefficient for each cause by a total value obtained by summing up the subtotal values ​​for all the causes.

5. The abnormality cause estimation method according to claim 4, wherein: include: The occurrence probability of each cause is ranked in order of magnitude, and the step of outputting the cause at the highest level is performed.

6. The abnormality cause estimation method according to claim 1 or 2, wherein: include: a step of monitoring a Mahalanobis distance based on the sensor measurement value; as well as A step of acquiring the SN ratio gain value when the event of abnormality is detected based on the Mahalanobis distance.

7. The abnormality cause estimation method according to claim 1 or 2, wherein: Also includes: A step of selecting the cause table used in calculation from among the plurality of cause tables for each process.

8. An abnormality cause estimation device, comprising: a priori probability calculation unit configured to calculate the priori probability as the probability of occurrence of each cause based on a cause table showing the occurrence frequency of the cause for each event; a posterior probability calculation unit configured to calculate a posterior probability as a probability that a certain event is caused by a certain cause using the prior probability of each cause and the likelihood as a probability that each event occurs due to each cause; and The index calculation unit multiplies the posterior probability by a weight coefficient for each of the events related to an SN ratio gain value of the sensor measurement value, and calculates an index indicating the probability of occurrence for each combination of the cause and the event.

9. A computer-readable recording medium storing a program, the program causing a computer to execute the following steps: Calculating a priori probability as the probability of occurrence of each cause based on a cause table showing the frequency of occurrence of the cause for each event; calculating a posterior probability as the probability that a certain event is caused by a certain cause using the prior probability of each cause and the likelihood as the probability that each event occurs due to each cause; and The posterior probability is multiplied by a weight coefficient for each of the events related to the SN ratio gain value of the sensor measurement value, and an index representing the probability of occurrence is calculated for each combination of the cause and the event.

Images