Diagnostic apparatus, diagnostic method, and computer program product

Abstract

The present invention provides a diagnostic device, diagnostic method, and computer program product capable of accurately detecting abnormalities even when data is small or fluctuating. The diagnostic device comprises a Mahalanobis distance calculation unit that calculates the Mahalanobis distance of detection values, hereinafter referred to as the MD value; and an abnormality determination unit that determines the presence of an abnormality based on the MD value. The abnormality determination unit determines the presence of an abnormality such that the smaller the number of samples per unit space, the more likely it is to be determined that an abnormality is present, compared to when the number of samples is large.

Description

Technical Field

[0001] The present invention relates to a diagnostic device, a diagnostic method, and a program. This application claims priority based on Japanese patent application No. 2020-000427 filed on January 6, 2020, the contents of which are incorporated herein by reference. Background Art

[0002] The MT method (Maharanobis Taguchi System) using the Mahalanobis distance is widely used in abnormality detection systems for power generation equipment and remote monitoring systems (e.g., Patent Document 1). As described in Patent Document 1, the Mahalanobis distance assumes that normal data follows a normal distribution. However, real-world data often does not follow a normal distribution. Therefore, the abnormality detection preprocessing device described in Patent Document 1 determines whether distribution data related to two variables measured during normal conditions within a predetermined period follows a normal distribution. A regular amount of distribution data determined not to follow a normal distribution is fitted to a temporary nonlinear model. A correction term is calculated based on the difference between the temporary nonlinear model and the regression line to correct the distribution data. The temporary nonlinear model that yields the distribution data that best follows a normal distribution, corrected by the correction term, is selected as the abnormality detection model used for abnormality detection, and the correction term calculated based on the abnormality detection model is selected as the correction term used for abnormality detection.

[0003] Furthermore, the abnormality detection device described in Patent Document 1 calculates corrected judgment data based on correction terms selected by the abnormality detection preprocessing device. This corrected judgment data corrects the judgment data for the measurement data that is the subject of the normality / abnormality determination, and determines whether the corrected judgment data is abnormal based on the Mahalanobis distance. The abnormality detection preprocessing device and the abnormality detection device described in Patent Document 1 quantitatively evaluate the normal distribution of normal data and select correction terms for use in the abnormality detection model and abnormality detection based on the distribution data measured during normal conditions. This allows for the accurate detection of data that deviates from the measurement data obtained during normal conditions (i.e., abnormal data).

[0004] Previous technical literature

[0005] Patent Literature

[0006] Patent Document 1: Patent No. 6129508 Summary of the Invention

[0007] Technical issues to be solved by the invention

[0008] As described above, the abnormality detection preprocessing device described in Patent Document 1 selects correction items based on the results of determining whether the data distribution follows a normal distribution. Therefore, multiple data sets are required to determine whether the data follows a normal distribution. However, in gas turbine equipment, such as the example shown in Patent Document 1, which operates continuously (for example, for several months) once started, there is a problem: for data that can only be measured once per startup (for example, the time required for a predetermined state change during startup, the maximum, minimum, average, or total value of startup data, etc.), collecting multiple data sets requires a long time.

[0009] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a diagnostic device, a diagnostic method, and a program that can detect abnormalities with high accuracy even when there is little data or when the amount of data varies.

[0010] Means for solving technical problems

[0011] In order to solve the above-mentioned problems, the diagnostic device involved in the present invention includes: a Mahalanobis distance calculation unit, which calculates the Mahalanobis distance of the detection value (hereinafter referred to as the MD value); and an abnormality determination unit, which determines the presence or absence of an abnormality based on the MD value, and the abnormality determination unit determines the presence or absence of the abnormality in a manner that makes it easier to determine that there is no abnormality when the number of samples in the unit space is smaller than when the number of samples is larger.

[0012] Furthermore, the diagnostic method involved in the present invention has the following steps: a step of calculating the MD value of the detection value; and a step of determining whether the abnormality exists based on the MD value in a manner that the smaller the number of samples in the unit space is, the easier it is to determine that there is no abnormality than when the number of samples is large.

[0013] Furthermore, the program involved in the present invention enables the computer to execute the following steps: a step of calculating the MD value of the detection value; and a step of determining whether the abnormality exists based on the MD value in a manner that the smaller the number of samples in the unit space is, the easier it is to determine that there is no abnormality than when the number of samples is large.

[0014] Effects of the Invention

[0015] According to the diagnostic device, diagnostic method and program of the present invention, an abnormality can be detected with good accuracy even when there is little data or the amount of data varies. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a diagram showing a configuration example of a diagnostic device according to the first embodiment of the present invention.

[0017] Figure 2This is a flowchart showing an example of the operation of the diagnostic device according to the first embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram for explaining a diagnostic device according to a second embodiment of the present invention.

[0019] Figure 4 This is a flowchart showing an example of the operation of the diagnostic device according to the second embodiment of the present invention.

[0020] Figure 5 This is a system flowchart showing an operation example of the diagnostic device according to the second embodiment of the present invention.

[0021] Figure 6 This is a schematic diagram for explaining a diagnostic device according to a third embodiment of the present invention.

[0022] Figure 7 This is a flowchart showing an example of the operation of the diagnostic device according to the third embodiment of the present invention.

[0023] Figure 8 This is a system flowchart showing an operation example of the diagnostic device according to the third embodiment of the present invention.

[0024] Figure 9 This is a system flowchart showing an operation example of the diagnostic device according to the fourth embodiment of the present invention.

[0025] Figure 10 This is a schematic block diagram showing the structure of a computer according to at least one embodiment. DETAILED DESCRIPTION

[0026] <First embodiment>

[0027] (Structure of Diagnostic Device)

[0028] Below, reference Figure 1 and Figure 2 A diagnostic device according to an embodiment of the present invention will be described. In the figures, identical or corresponding components are denoted by the same reference numerals, and their descriptions are omitted as appropriate. While this embodiment assumes that the diagnostic device 10 is installed in a monitoring center that monitors a gas turbine and is used to detect abnormalities in the gas turbine, the present invention is not limited to this configuration.

[0029] Figure 1 It is a diagram showing a configuration example of a diagnostic device 10 according to the first embodiment of the present invention. Figure 1The illustrated diagnostic device 10 monitors, for example, a gas turbine facility 31 installed at a power plant and diagnoses any abnormalities. The gas turbine facility 31 and the diagnostic device 10 are connected via a network 32 to enable information transmission and reception. For example, the diagnostic device 10 receives gas turbine operating data, alerts, and inquiries from the gas turbine facility 31 at predetermined times. The diagnostic device 10 stores the acquired information in a storage unit 13 (described in detail later). While this embodiment describes a case where the diagnostic device 10 is remotely located from the gas turbine facility 31 via the network 32, the location of the diagnostic device 10 is not limited to this.

[0030] The diagnostic device 10 can be configured using, for example, a computer such as a server or personal computer and its peripheral devices. The diagnostic device 10 comprises a Mahalanobis distance calculation unit 11, an abnormality determination unit 12, and a storage unit 13. The diagnostic device 10 is a functional structure composed of a combination of hardware such as a computer and its peripheral devices, and software such as a program executed by the computer.

[0031] The storage unit 13 stores measurement data (such as operating data) acquired over a predetermined period from measuring instruments installed at various locations within the gas turbine facility 31. This measurement data includes, for example, measurement data acquired over a predetermined period during non-steady-state conditions, such as during gas turbine startup, when normal data is particularly unstable; and measurement data that serves as a target for determining whether the gas turbine facility is normal or abnormal. In this embodiment, the various locations within the gas turbine facility include, for example, combustors and compressors, and the measurement data includes information such as temperature, voltage, current, rotational speed, and pressure values acquired from these locations; the time required for predetermined state changes during startup; and the maximum, minimum, average, or total value of data acquired during startup. The storage unit 13 also stores values related to unit space in the MT method, such as the number of samples (number of data points), the average value, the standard deviation, the inverse of the correlation matrix, and threshold values for determining abnormality.

[0032] The Mahalanobis distance calculation unit 11 calculates the MD value of the measurement data (detection values). The Mahalanobis distance calculation unit 11 calculates the k-dimensional MD value using the following formula. Here, k is the number of items, i and j range from 1 to k, αij represents the i and j components of the inverse correlation matrix, and mi, mj, σi, and σj represent the mean and standard deviation in the unit space, respectively. The unit space is a reference data set consisting of multiple MD values based on normal measurement data, and is calculated based on multiple samples of the k-item measurement data set. The unit space is updated, for example, by the Mahalanobis distance calculation unit 11 based on new measurement data.

[0033] [Formula 1]

[0034]

[0035] Furthermore, the Mahalanobis distance method uses a single parameter (the Mahalanobis distance) to represent a characteristic quantity (multivariate) of a group, and evaluates the quality of a particular measurement data based on its distance from the baseline data of a healthy group (normal measurement data). If the measurement data is poor, the distance from the healthy group increases, while if the measurement data is good, the distance from the healthy group decreases.

[0036] The abnormality determination unit 12 determines the presence of an abnormality based on the MD value calculated by the Mahalanobis distance calculation unit 11. In this case, the abnormality determination unit 12 determines the presence of an abnormality so that the smaller the number of samples per unit space, the easier it is to determine that there is no abnormality than when the number of samples is large. For example, the abnormality determination unit 12 sets a threshold (also called an MD threshold) for the Mahalanobis distance. If the Mahalanobis distance is below the threshold, the system is determined to be normal; if the distance is above the threshold, the system is determined to be abnormal.

[0037] (Operation of the diagnostic device)

[0038] Next, refer to Figure 2 ,right Figure 1 The basic operation of the diagnostic device 10 shown will be described. Figure 2 This is a flowchart showing an example of the operation of the diagnostic device according to the first embodiment of the present invention.

[0039] Figure 2 The illustrated process is initiated, for example, upon a predetermined operator input. Furthermore, the storage unit 13 stores measurement data related to the operation of the gas turbine, measured by the gas turbine equipment 31. In the diagnostic device 10, the Mahalanobis distance calculation unit 11 first calculates the MD value of the measurement data (detection value) (step S11). Next, the abnormality determination unit 12 determines the presence of an abnormality, such that the smaller the number of samples per unit space, the more likely it is to be determined as abnormal (step S12).

[0040] In step S12, the abnormality determination unit 12 determines the presence of an abnormality by, for example, increasing the abnormality determination threshold relative to the MD value as the number of samples in the unit space decreases. As the threshold increases, the number of times an abnormality is determined to be present when the MD value is large decreases. Alternatively, the abnormality determination unit 12 may correct the MD value calculated by the Mahalanobis distance calculation unit 11 by a coefficient that decreases the value as the number of samples in the unit space decreases, thereby decreasing the MD value. As the MD value decreases, the number of times an abnormality is determined to be present decreases, even if the abnormality determination threshold remains unchanged. Alternatively, the abnormality determination unit 12 may combine threshold correction with MD value correction.

[0041] If the number of samples is small, the distribution of the MD value that forms the basis of the unit space does not follow the normal distribution. In this case, it is considered that, for example, when the threshold value is set assuming a normal distribution, the MD value exceeds the threshold value even though it is normal. In contrast, by setting the number of samples in the unit space to be smaller than that in the case of a large number of samples, it is easier to determine that there is no abnormality. In this embodiment, it is possible to reduce the number of cases where the normal value is mistakenly detected as abnormal. That is, the occurrence of false detection can be suppressed and the accuracy of abnormality detection can be improved. In addition, the abnormality determination unit 12 can prevent the occurrence of false detections where the abnormality is determined to be normal even though it is abnormal, for example, by combining the abnormality detection method described in Patent Document 1.

[0042] The determination result of whether there is an abnormality based on the abnormality determination unit 12 can be stored in the storage unit 13, or output from a display device, printing device, sound output device, etc. not shown in the figure of the diagnostic device 10, or sent to an external terminal via a communication device not shown in the figure of the diagnostic device 10.

[0043] <Second embodiment>

[0044] refer to Figures 3 to 5 , a second embodiment of the present invention is described. Figure 3 This is a schematic diagram for explaining a diagnostic device according to a second embodiment of the present invention. Figure 4 This is a flowchart showing an example of the operation of the diagnostic device according to the second embodiment of the present invention. Figure 5 This is a system flowchart showing an operation example of the diagnostic device according to the second embodiment of the present invention.

[0045] The basic structure of the diagnostic device of the second embodiment is Figure 1 The second embodiment is the same as the diagnostic device 10 of the first embodiment shown. Figure 1 A part of the operation of the abnormality determination unit 12 is shown in more detail.

[0046] Figure 3 This is a diagram schematically showing the probability distribution of the Mahalanobis distance MD using the normal distribution and n (or Student's t-distribution). Figure 3 In the figure, the shading indicates the normal distribution cumulative probability (the inverse cumulative distribution function of the normal distribution) of MD ≥ 3 and the t distribution cumulative probability (the inverse cumulative distribution function of the t distribution) of MD ≥ 5.

[0047] A mathematically correct probability density function follows a t-distribution when the number of samples is small, and approaches a normal distribution when the number of samples is large. When the sample size is infinite, the t-distribution coincides with the normal distribution. Furthermore, when the number of sensors is small (e.g., one), the same as the t-test occurs. Even for the same MD, the cumulative probability changes depending on the assumed probability density function. In particular, the t-distribution has a wider range than the normal distribution. When diagnosing "3 < MD is abnormal," while "3 < MD is relatively common with the t-distribution even when the number of data is small," there are still many false positives that are considered abnormal.

[0048] Therefore, in the second embodiment, the abnormality determination unit 12 corrects the MD value calculated by the Mahalanobis distance calculation unit 11 to an MD' value (a second MD value) based on the cumulative probability of the t distribution with the degree of freedom corresponding to the number of samples (hereinafter referred to as the t distribution cumulative probability) and the cumulative probability of the normal distribution corresponding to the t distribution cumulative probability, and determines the presence or absence of an abnormality based on the result of comparing the MD' value with a specified threshold value.

[0049] like Figure 4 As shown in the second embodiment, first, the Mahalanobis distance calculation unit 11 calculates the MD value of the measurement data (detection value) (step S21). In this case, it is assumed that the MD value is "5".

[0050] Next, the abnormality determination unit 12 obtains the t-distribution cumulative probability up to the MD value (“5”) in the t-distribution based on the MD value calculated by the Mahalanobis distance calculation unit 11 using the MT method (step S22 ).

[0051] Next, the abnormality determination unit 12 obtains the MD' value that gives the same cumulative probability (normal distribution cumulative probability) as the obtained t-distribution cumulative probability in the normal distribution (step S23). In this case, the MD' value is assumed to be "3".

[0052] Next, the abnormality determination unit 12 compares the MD′ value with a threshold value to perform abnormality diagnosis (step S24 ).

[0053] Furthermore, the abnormality determination unit 12 can obtain the MD′ value, which is the corrected MD value, using the following equation, for example.

[0054] [Formula 2]

[0055]

[0056] Here, tcdf is the cumulative distribution function of the t-distribution, erfinv is the inverse function of the error function, ν is the degree of freedom (NA), N is the number of samples per unit space (the number of data), and A is the number of sensors.

[0057] In addition, if Figure 5 As shown, the abnormality determination unit 12 calculates the t-distribution cumulative probability (103) based on the MD value (101) calculated by the MD Mahalanobis distance calculation unit 11 using the MT method and the number of samples (number of specimens) (102). Furthermore, the abnormality determination unit 12 calculates the MD' value (104) in the normal distribution that has the same cumulative probability. Then, the abnormality determination unit 12 compares the MD' value (104) with the MD threshold value (105).

[0058] As described above, according to the second embodiment, in the same manner as the first embodiment, it is possible to reduce the number of cases where a normal condition is mistakenly detected as abnormal. Furthermore, by using the above formula, the same calculation formula can be used to handle a large or small number of samples.

[0059] <Third embodiment>

[0060] refer to Figures 6 to 8 , a third embodiment of the present invention is described. Figure 6 This is a schematic diagram for explaining a diagnostic device according to a third embodiment of the present invention. Figure 7 This is a flowchart showing an example of the operation of the diagnostic device according to the third embodiment of the present invention. Figure 8 This is a system flowchart showing an operation example of the diagnostic device according to the third embodiment of the present invention.

[0061] The basic structure of the diagnostic device of the third embodiment is Figure 1 The third embodiment is the same as the diagnostic device 10 of the first embodiment shown. Figure 1 A part of the operation of the abnormality determination unit 12 is shown in more detail.

[0062] Figure 6 and Figure 3 Similarly, the probability distribution of the Mahalanobis distance MD is schematically represented by the normal distribution and the t distribution. Figure 6 In the figure, the normal distribution cumulative probability (inverse cumulative distribution function of the normal distribution) for MD ≥ 3 and the t distribution cumulative probability for MD ≥ 5 are indicated by shading. Figure 6 , the value of the threshold MDc is "3".

[0063] In the third embodiment, the abnormality determination unit 12 calculates the cumulative probability up to a threshold value MDc specified in the normal distribution (normal distribution cumulative probability), calculates the corresponding value MDt corresponding to the threshold value MDc in the t distribution with the degree of freedom corresponding to the number of samples that is equal to the normal distribution cumulative probability, and determines the presence or absence of an abnormality based on the result of comparing the MD' value (second MD value) obtained by correcting the MD value based on the threshold value MDc and the corresponding value MDt with the threshold value MDc. Figure 6 In the example shown, the corresponding value MDt corresponding to the threshold MDc of “3” is “5.” In this case, for example, the value obtained by multiplying the MD value by 3 / 5 (MD′ value) is compared with the threshold MDc.

[0064] like Figure 7 As shown, in the third embodiment, first, the Mahalanobis distance calculation unit 11 calculates the MD value of the measurement data (detection value) (step S31). Then, the abnormality determination unit 12 calculates the cumulative probability up to the threshold value (MDc) in the normal distribution (step S32). Then, the abnormality determination unit 12 calculates the corresponding value (MDt) in the t distribution that has the same cumulative probability as the calculated normal distribution cumulative probability (step S33). Figure 6 In the t-distribution, MDc = 3 and MDt = 5. Therefore, the probability of occurrence of a phenomenon with MD = 3 in the normal distribution is equivalent to MD = 5 in the t-distribution. The corresponding value of the t-distribution changes depending on the number of samples.

[0065] Next, the abnormality determination unit 12 multiplies the MD value calculated by the Mahalanobis distance calculation unit 11 using the MT method by MDc / MDt to obtain an MD′ value (second MD value) (step S34). Figure 6 In the example shown, MD′ value=MD value×3 / 5. Next, the abnormality determination unit 12 compares the MD′ value with the MD threshold value (MDc) to perform abnormality diagnosis (step S35).

[0066] Furthermore, the abnormality determination unit 12 can obtain the MD′ value, which is the corrected MD value, using the following equation, for example.

[0067] [Formula 3]

[0068]

[0069] Here, MDc is the threshold of the MT method, tinv is the inverse cumulative distribution function of the t distribution, erf is the error function, ν is the degree of freedom (NA), N is the number of samples per unit space (the number of data), and A is the number of sensors.

[0070] In addition, if Figure 8As shown, the abnormality determination unit 12 of the third embodiment can compare the MD (MD value) (201) calculated by the MD Mahalanobis distance calculation unit 11 using the MT method with an MD threshold value selected from the MD threshold value for normal distribution (203) or the MD threshold value for t distribution (204) based on the number of samples (number of specimens) (205). In this case, the MD threshold value for t distribution (204) is smaller than the MD threshold value for normal distribution (203) and is selected when the number of samples is small. That is, in the third embodiment, when the number of samples is small, the predetermined threshold value to be compared with the MD value is increased, and when the MD value is greater than the threshold value, the abnormality determination unit 12 determines that an abnormality exists.

[0071] As described above, according to the third embodiment, in the same manner as the first embodiment, it is possible to reduce the number of cases where a normal condition is mistakenly detected as an abnormality. Furthermore, by using the above formula, the calculation process can be simplified (the function parameter can be set to the constant threshold value MDc rather than the MD value), thereby reducing the processing load compared to the second embodiment.

[0072] <Fourth embodiment>

[0073] refer to Figure 9 , a fourth embodiment of the present invention is described. Figure 9 This is a system flow chart showing an example of the operation of the diagnostic device according to the fourth embodiment of the present invention. Figure 1 The fourth embodiment is the same as the diagnostic device 10 of the first embodiment shown. Figure 1 A part of the operation of the abnormality determination unit 12 is shown in more detail.

[0074] like Figure 9 As shown, the abnormality determination unit 12 of the fourth embodiment calculates the t-distribution cumulative probability (303) based on the MD value (301) calculated by the MD Mahalanobis distance calculation unit 11 using the MT method and the number of samples (number of specimens) (302), and compares the t-distribution cumulative probability (303) with a cumulative probability threshold (304) that is the cumulative probability of a normal distribution corresponding to a preset threshold, thereby performing abnormality determination. That is, in the diagnostic device 10 of the fourth embodiment, the Mahalanobis distance calculation unit 11 calculates the MD value of the measurement data (detection value). Then, the abnormality determination unit 12 compares the cumulative probability up to the MD value obtained based on the number of samples in the unit space with a predetermined cumulative probability threshold, and determines the presence or absence of an abnormality based on the comparison result, thereby determining the presence or absence of an abnormality based on the MD value.

[0075] (Other Implementation Methods)

[0076] While the embodiments of the present invention have been described in detail with reference to the accompanying drawings, the specific configuration is not limited to the embodiments and includes design changes that do not depart from the spirit of the present invention.

[0077] Computer Structure

[0078] Figure 10 This is a schematic block diagram showing the structure of a computer according to at least one embodiment.

[0079] The computer 90 includes a processor 91 , a main memory 92 , a storage 93 , and an interface 94 .

[0080] The diagnostic device 10 is installed in a computer 90. The operations of the aforementioned processing units are stored in a memory 93 as a program. A processor 91 reads the program from memory 93, expands it into a main memory 92, and executes the aforementioned processing according to the program. Furthermore, the processor 91 reserves storage areas corresponding to the aforementioned storage units in the main memory 92 according to the program.

[0081] The program can be used to implement a portion of the functions that computer 90 performs. For example, the program can be combined with other programs already stored in memory or installed in other devices to perform the functions. Furthermore, in other embodiments, the computer may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or in place of the above-described structure. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions implemented by the processor can be implemented by this integrated circuit.

[0082] Examples of the memory 93 include an HDD (Hard Disk Drive), an SSD (Solid State Drive), a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disc Read Only Memory), a DVD-ROM (Digital Versatile Disc Read Only Memory), and a semiconductor memory. The memory 93 may be an internal medium directly connected to the bus of the computer 90, or an external medium connected to the computer 90 via an interface 94 or a communication line. Furthermore, when the program is transferred to the computer 90 via a communication line, the computer 90 that has received the transfer may also expand the program into the main memory 92 and execute the above-described processing. In at least one embodiment, the memory 93 is a non-transitory, tangible storage medium.

[0083] <Note>

[0084] The diagnostic device 10 according to each embodiment can be understood as follows, for example.

[0085] (1) A diagnostic device 10 according to a first embodiment includes: a Mahalanobis distance calculation unit 11 that calculates an MD value of measurement data (detection value); and an abnormality determination unit 12 that determines the presence or absence of an abnormality based on the MD value. The abnormality determination unit 12 determines the presence or absence of an abnormality in such a manner that the smaller the number of samples per unit space, the easier it is to determine that there is no abnormality than in a case where the number of samples is large.

[0086] (2) Diagnostic device 10 of the second embodiment In the diagnostic device 10 of (1), the abnormality determination unit 12 corrects the MD value to the MD' value (second MD value) based on the cumulative probability of the t distribution with degrees of freedom corresponding to the number of samples and the cumulative probability of the normal distribution corresponding to the cumulative probability of the t distribution, and determines the presence or absence of an abnormality based on the result of comparing the MD' value (second MD value) with a predetermined threshold value.

[0087] (3) Diagnostic device of the third embodiment In the diagnostic device 10 of (1), the abnormality determination unit 12 calculates the cumulative probability up to a threshold value MDc specified in the normal distribution, and calculates the corresponding value MDt corresponding to the threshold value in the t-distribution with the degree of freedom corresponding to the number of samples that becomes a cumulative probability equal to the cumulative probability, and determines the presence or absence of an abnormality based on the result of comparing the MD' value (second MD value) obtained by correcting the MD value based on the threshold value Mdc and the corresponding value MDt with the threshold value MDc.

[0088] (4) Diagnostic Device of the Fourth Embodiment In the diagnostic device 10 of (1), when the number of samples is small, the predetermined threshold value for comparison with the MD value is increased, and when the MD value is larger than the threshold value, the abnormality determination unit 12 determines that an abnormality exists.

[0089] (5) The diagnostic device of the fourth embodiment comprises: a Mahalanobis distance calculation unit 11, which calculates the Mahalanobis distance of the measurement data (detection value); and an abnormality determination unit 12, which determines whether there is an abnormality based on the MD value. The abnormality determination unit 12 compares the cumulative probability (303) up to the MD value obtained based on the number of samples in the unit space with a predetermined cumulative probability threshold (304), and determines whether there is an abnormality based on the comparison result.

[0090] According to the above-described aspects, when the number of samples (data used for calculation in a unit space) is small, it is possible to reduce erroneous detections in which normal values are determined to be abnormal values, thereby improving detection accuracy.

[0091] Explanation of symbols

[0092] 10- Diagnostic device, 11- Mahalanobis distance calculation unit, 12- Abnormality determination unit.

Claims

1. A diagnostic device comprising: A Mahalanobis distance calculation unit that calculates the Mahalanobis distance of the detection value, i.e., the MD value; and an abnormality determination unit that determines whether an abnormality exists based on the MD value, The abnormality determination unit corrects the MD value to a second MD value based on the cumulative probability of the t distribution with the degree of freedom corresponding to the number of samples in the unit space and the cumulative probability of the normal distribution corresponding to the cumulative probability of the t distribution, and determines the presence or absence of the abnormality based on the result of comparing the second MD value with a predetermined threshold value. The abnormality determination unit obtains the t-distribution cumulative probability up to the MD value in the t-distribution, and then obtains the second MD value in the normal distribution that has a cumulative probability equal to the obtained t-distribution cumulative probability.

2. A diagnostic device comprising: A Mahalanobis distance calculation unit that calculates the Mahalanobis distance of the detection value, i.e., the MD value; and an abnormality determination unit that determines whether an abnormality exists based on the MD value, The abnormality determination unit calculates the cumulative probability up to a specified threshold value MDc in the normal distribution, and calculates the corresponding value MDt corresponding to the threshold value MDc in the t-distribution with a cumulative probability equal to the cumulative probability and a degree of freedom corresponding to the number of samples in the unit space, and determines the presence or absence of the abnormality based on a result of comparing the second MD value obtained by multiplying the MD value by MDc / MDt with the threshold value MDc.

3. A diagnostic device comprising: A Mahalanobis distance calculation unit that calculates the Mahalanobis distance of the detection value, i.e., the MD value; and an abnormality determination unit that determines whether an abnormality exists based on the MD value, The abnormality determination unit compares the cumulative probability up to the MD value in the t distribution with the degree of freedom corresponding to the number of samples in the unit space with the cumulative probability threshold in the normal distribution corresponding to a predetermined threshold, and determines the presence or absence of the abnormality based on the comparison result.

4. A diagnostic method comprising the following steps: The diagnostic device calculates the Mahalanobis distance of the detection value, i.e., the MD value; and The diagnostic device determines whether or not there is an abnormality based on the MD value, In the step of determining the presence or absence of the abnormality, the diagnostic device corrects the MD value to a second MD value based on the cumulative probability of the t distribution with degrees of freedom corresponding to the number of samples in the unit space and the cumulative probability of the normal distribution corresponding to the cumulative probability of the t distribution, and determines the presence or absence of the abnormality based on a result obtained by comparing the second MD value with a predetermined threshold value. In the step of determining the presence or absence of the abnormality, after obtaining the t distribution cumulative probability up to the MD value in the t distribution, the second MD value having the same cumulative probability as the obtained t distribution cumulative probability is obtained in the normal distribution.

5. A computer program product comprising a program for causing a computer to perform the following steps: The step of calculating the Mahalanobis distance of the detection value, i.e., the MD value; and The step of determining whether there is an abnormality based on the MD value, The program causes the computer to correct the MD value to a second MD value based on the cumulative probability of the t distribution with degrees of freedom corresponding to the number of samples in the unit space and the cumulative probability of the normal distribution corresponding to the cumulative probability of the t distribution in the step of determining the presence or absence of the abnormality, and to determine the presence or absence of the abnormality based on a result obtained by comparing the second MD value with a predetermined threshold value. In the step of determining the presence or absence of the abnormality, after obtaining the t distribution cumulative probability up to the MD value in the t distribution, the second MD value having the same cumulative probability as the obtained t distribution cumulative probability is obtained in the normal distribution.

Images