Abnormality diagnosis device and abnormality diagnosis method

Abstract

To provide an abnormality diagnostic device which can precisely detect an abnormality or an abnormal sign of a diagnosed object, minimizes a downtime of the diagnosed object, and decreases a maintenance cost of the diagnosed object.SOLUTION: An abnormality diagnostic device comprises: an MD value calculation unit 22 which calculates Mahalanobis distances between collected data from diagnosed objects and normal time data of the diagnosed objects and creates MD values; an abnormal wave form pattern storing unit 32 which classifies abnormalities about the diagnosed objects into a plurality of patterns and stores the patterns as abnormality wave form patterns; a wave form pattern comparison unit 23 which, when diagnosing diagnosed objects, compares wave forms of time series data of the generated MD value with the abnormal wave form patterns stored in the abnormal wave form storing unit 32; and a diagnosis unit 24 which, when the wave forms of the time series data of the MD value calculated from the collected data match the abnormality wave form patterns as a result of the comparison 23, generates a diagnosis result indicating that they match.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an abnormality diagnosis device and an abnormality diagnosis method using the MT method. [Background technology]

[0002] In recent years, an anomaly diagnosis technique using the Mahalanobis-Taguchi Method (MT method) has become well known. The MT method is a method for diagnosing whether a condition is "the same as usual" and is a method for determining an anomaly when the Mahalanobis distance of the diagnosis target is farther from the Mahalanobis space, which represents a preset, usual state, as the standard.

[0003] In the conventional MT method, when the calculated degree of abnormality (MD value) exceeds a preset threshold value, it is judged as "abnormal" or "there is a sign of abnormality." [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2013-041490 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, in actual plant facilities, there are many operating patterns depending on the load state, weather conditions, etc. Therefore, even if the MT method judges the equipment to be in a normal state, there is a certain degree of variability in the degree of abnormality (MD value), making it difficult to set an appropriate threshold value.

[0006] In addition, in the past, even if the waveform of the degree of anomaly (MD value) clearly showed abnormal behavior, it was not judged to be abnormal unless it exceeded the threshold, so there was a possibility that the abnormality would be overlooked. If the threshold was set low to avoid this, there was a possibility that a normal state would be erroneously detected as an abnormality.

[0007] The present invention has been made in consideration of the above-mentioned circumstances, and aims to provide an abnormality diagnosis device and an abnormality diagnosis method that can accurately detect abnormalities or signs of abnormalities in an object to be diagnosed, thereby minimizing downtime of the object to be diagnosed and reducing maintenance costs. [Means for solving the problem]

[0008] One aspect of the abnormality diagnosis device according to the present invention for achieving the above-mentioned object is an abnormality diagnosis device comprising: an MD value calculation unit that calculates a Mahalanobis distance between collected data from an object to be diagnosed and normal data of the object to be diagnosed to generate an MD value; an abnormal waveform pattern storage unit that classifies abnormalities related to the object to be diagnosed into a plurality of patterns and stores them as abnormal waveform patterns; a waveform pattern comparison unit that, when diagnosing the object to be diagnosed, compares the waveform of time-series data of the generated MD value with the abnormal waveform patterns stored in the abnormal waveform pattern storage unit; and a diagnosis unit that, when the waveform of the time-series data of the MD value calculated from the collected data matches the abnormal waveform pattern as a result of the comparison in the waveform pattern comparison unit, generates a diagnosis result to that effect.

[0009] Moreover, one aspect of the abnormality diagnosis method according to the present invention is an abnormality diagnosis method, which generates in advance a unit space from data of a diagnosis target when the diagnosis target is normal, stores in advance a plurality of patterns of abnormal waveform patterns of MD values ​​when the diagnosis target is abnormal, and when diagnosing the diagnosis target, compares the time-series waveform of MD values ​​generated from collected data from the diagnosis target and the unit space with the abnormal waveform pattern, and diagnoses the presence or absence of an abnormality in the diagnosis target based on a comparison result between the waveform of the time-series data of MD values ​​calculated from the collected data and the abnormal waveform pattern. Effect of the Invention

[0010] According to one aspect of the present invention, it is possible to accurately detect abnormalities or signs of abnormalities in an object to be diagnosed, thereby minimizing downtime of the object to be diagnosed and reducing maintenance costs. [Brief description of the drawings]

[0011] [Figure 1] 1 is a block diagram showing the configuration of a first embodiment of an abnormality diagnosis device according to the present invention; [Diagram 2] 4 is a flowchart showing a processing procedure according to the first embodiment. [Diagram 3] FIG. 1 is a waveform diagram showing an abnormal waveform pattern 1 of an MD value. [Figure 4] FIG. 13 is a waveform diagram showing an abnormal waveform pattern 2 of the MD value. [Diagram 5] FIG. 13 is a waveform diagram showing an abnormal waveform pattern 3 of the MD value. [Figure 6] FIG. 13 is a waveform diagram showing an abnormal waveform pattern 4 of the MD value. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Configuration of the embodiment FIG. 1 is a block diagram showing the configuration of an embodiment of an abnormality diagnostic device according to the present invention.

[0013] As shown in the figure, the abnormality diagnostic device 1 is made up of a computer, and includes an input device 10, a central processing unit (CPU) 20, and a storage device 30. It also includes an alarm output device 40 that outputs an alarm to the outside.

[0014] The input device 10 includes a data acquisition unit 11 for inputting collected data collected in a plant to be diagnosed, and also includes input devices such as a keyboard, a mouse, operation buttons, and a touch panel.

[0015] The calculation processing device 20 includes a unit space calculation section 21, an MD value calculation section 22, a waveform pattern comparison section 23, and a diagnosis section 24. The storage device 30 includes a unit space data storage section 31, an abnormal waveform pattern storage section 32, and a history data storage section 33.

[0016] The unit space calculation unit 21 generates a unit space that is a reference for data diagnosis in the normal state of the plant based on the acquired data acquired by the data acquisition unit 11. The generated unit space is stored in the unit space data storage unit 31.

[0017] The MD value calculation unit 22 calculates, during diagnosis of the plant, the Mahalanobis distance between the data acquired from the plant and the unit space as the MD value.

[0018] The waveform pattern comparison unit 23 compares the time series data of the MD value with an abnormal waveform pattern to detect the presence or absence of an abnormality in the acquired data.

[0019] The diagnosing unit 24 diagnoses an abnormality in the plant based on the collation result of the waveform pattern collation unit 23.

[0020] The unit space data storage unit 31 stores the unit space calculated by the unit space calculation unit 21, which is data on the plant in a normal state.

[0021] The abnormal waveform pattern storage unit 32 classifies abnormalities in the plant to be diagnosed into a plurality of patterns and stores them as abnormal waveform patterns. Specifically, the abnormal waveform patterns 1 to 4 shown in Figs. 3 to 6 are stored.

[0022] The history data storage unit 33 stores the diagnosis results as history data. This history data is used for updating or correcting the abnormal waveform pattern.

[0023] Processing Procedure of the Embodiment Next, the processing procedure of the embodiment will be described with reference to the flowchart of FIG.

[0024] A data group (collected data) collected in a plant to be diagnosed is input to the data acquisition unit 11. Prior to diagnosis, the unit space calculation unit 21 calculates a unit space that serves as a reference for diagnosis and stores it in the unit space data storage unit 31 as unit space data.

[0025] During abnormality diagnosis, data is acquired from a plant in operation via the data acquisition unit 11 (step S1). The MD value calculation unit 22 calculates the MD value from the acquired data (step S2).

[0026] Specifically, first, the average value and standard deviation are calculated for the acquired data. Next, the data is standardized using the value of the acquired data and the average value and standard deviation calculated for the acquired data to calculate a standardized value for each variable. Next, the MD value calculation unit 22 calculates a correlation matrix for the acquired data using the standardized value calculated for the acquired data, and also calculates an inverse matrix of the correlation matrix. Finally, the Mahalanobis distance is calculated using the calculated inverse matrix. The Mahalanobis distance (MD value) calculated in this way makes it possible to identify the difference between the current state and the usual state (normal state) of the acquired data from the diagnosis target.

[0027] Next, if there is a waveform pattern exceeding a preset threshold value among the waveform data obtained from the time series data of the calculated MD value, the waveform pattern comparison unit 23 compares the waveform pattern with abnormal waveform patterns stored in advance (step S3). If there is no corresponding abnormal waveform pattern as a result of the comparison (step S4 NO), the process returns to step S1 and continues the data acquisition process.

[0028] If there is a corresponding waveform pattern (step S4: YES), the diagnosis unit 24 judges which abnormal waveform pattern the waveform pattern corresponds to (step S5). If the waveform pattern is abnormal waveform pattern 1, a warning to that effect is output (steps S5: YES, S6). If the waveform pattern is not abnormal waveform pattern 1, it judges whether the waveform pattern corresponds to abnormal waveform pattern 2 (step S7).

[0029] If the waveform pattern is abnormal waveform pattern 2, a warning to that effect is output (steps S7 YES, S8). If the waveform pattern is not abnormal waveform pattern 2, it is determined whether or not the waveform pattern corresponds to abnormal waveform pattern 3 (step S9).

[0030] If the waveform pattern is the abnormal waveform pattern 3, a warning to that effect is output (Steps S9 YES, S10). If the waveform pattern is not the abnormal waveform pattern 3, it is determined whether the waveform pattern corresponds to the abnormal waveform pattern 4 (Step S11).

[0031] If the waveform pattern is the abnormal waveform pattern 4, a warning to that effect is output (Steps S9 YES, S10).

[0032] In this way, it is collated which abnormal waveform pattern the waveform pattern corresponds to. When a matching abnormal waveform pattern is detected, a warning to that effect is output. If the waveform pattern is not any of the abnormal waveform patterns 1 to 4, there may be a false detection, so a report to that effect is made.

[0033] In the above processing, if there is an abnormality or a sign of an abnormality, the warning output device 40 immediately notifies related persons such as maintenance personnel by email or the like. Also, the abnormality detection history is stored in the history data storage unit 33.

[0034] Next, the abnormal waveform patterns used in the above-described processing procedure will be specifically described. In the present embodiment, the following four patterns are assumed as the abnormal waveform patterns. Note that the time axes of the following Patterns 1 to 3 are, for example, several seconds to several hours, or several days, while Pattern 4 is in units of several months to several years.

[0035] <Abnormal waveform pattern 1 of MD value> In the abnormal waveform pattern 1 shown in FIG. 3, in the case where, due to a false detection of the sensor or the like, the MD value exceeds the threshold Th1 although it is not originally abnormal, it is not detected as abnormal. That is, when the MD value exceeds the threshold Th1 and then continues to be in a state below the threshold Th1, it is not determined as abnormal.

[0036] However, when viewed over a slightly longer span, if it shows a behavior of exceeding the threshold Th1 periodically or a specified number of times, it is regarded as abnormal.

[0037] For example, it applies to the detection of abnormalities in the conveyor belt. When damage or foreign matter adheres to a part of the belt, it may appear as an abnormality such as abnormal vibration. When a sensor such as a vibration meter is attached somewhere on the conveyor, the MD value periodically exceeds the threshold Th1 only when the location of the abnormality passes near the sensor. If foreign matter adheres and exceeds the threshold Th1 but immediately detaches and returns to the original state, it will change as a curve shown by the dashed line in Figure 3.

[0038] <Abnormal waveform pattern 2 of MD value> The abnormal waveform pattern 2 shown in Figure 4 is the case where the MD value exceeds the threshold Th1 in a relatively short period and maintains that state for a certain period.

[0039] For example, as an example, when a part such as a bolt fixing the motor is damaged, it is assumed that the vibration behavior of the equipment will change. Also, as another example, when foreign matter jams in the drive part of a device driven by a motor, the load increases, and the motor current value fluctuates, the MD value may change as a curve shown in Figure 4.

[0040] <Abnormal waveform pattern 3 of MD value> The abnormal waveform pattern 3 shown in Figure 5 is the case where the MD value shows a value clearly larger than the threshold. Even if there is only a momentary change in the MD value, if the MD value shows an overly large numerical value, it is judged as an abnormality.

[0041] For example, as shown in Figure 5, if the MD value shows a value that further exceeds the second threshold Th2 (= Th1 × α, for example, α > 10) which greatly exceeds the first threshold Th1 even for a moment, a warning is issued assuming that some abnormality has occurred in the plant equipment.

[0042] <Abnormal waveform pattern 4 of MD value> Abnormal waveform pattern 4 shown in Fig. 6 is a waveform of the change in MD value due to aging. It is assumed that the change will be exponential over a long period of time. The time axis in Fig. 6 is set with reference to the general lifespan of the device, judging from descriptions in catalogs, etc., and is, for example, a period of several months to several years.

[0043] As described above, in one embodiment of the present invention, the MT method is used to diagnose abnormalities in plant equipment. Time-series data of the degree of abnormality (MD value: Mahalanobis distance value) calculated by the MT method is classified into multiple patterns according to the shape of the data waveform, and the presence or absence of an abnormality or a sign of an abnormality is judged. This improves the accuracy and precision of abnormality diagnosis compared to the conventional abnormality judgment based on a threshold value. Furthermore, by classifying abnormalities in advance, the time required to identify the cause can be shortened.

[0044] In the above embodiment, four abnormal waveform patterns 1 to 4 are assumed. However, these abnormal waveform patterns 1 to 4 are only examples, and various other abnormal waveform patterns can be assumed depending on the device or facility to be diagnosed and its installation environment, etc.

[0045] Although the present invention has been described in detail using the embodiments, the present invention is not limited to the embodiments described in the specification and drawings. The scope of the present invention is determined by the description of the claims and the equivalent scope of the claims. [Explanation of symbols]

[0046] 1... Abnormality diagnosis device, 10... Input device, 11... Data acquisition unit, 20... Arithmetic processing device, 21... Unit space calculation unit, 22... MD value calculation unit, 23... Waveform pattern comparison unit, 24... Diagnosis unit, 30... Storage device, 31... Unit space data storage unit, 32... Abnormal waveform pattern storage unit, 33... History data storage unit, 40... Alarm output device (alarm output unit)

Claims

1. An abnormality diagnosis device for detecting an abnormality in a conveyor belt, comprising: an MD value calculation unit that calculates a Mahalanobis distance between collected data from a diagnostic object and normal data of the diagnostic object to generate an MD value; an abnormal waveform pattern storage unit that stores an abnormal waveform pattern that periodically repeats a behavior in which the MD value exceeds a threshold value indicating an abnormality and then continues to be below the threshold value at a timing corresponding to the rotational movement of the belt of the conveyor; a waveform pattern comparison unit that compares the waveform of the time series data of the generated MD value with an abnormal waveform pattern stored in the abnormal waveform pattern storage unit when diagnosing the diagnosis target; and a diagnosis unit that generates a diagnosis result to that effect when the waveform of the time series data of the MD value calculated from the collected data matches the abnormal waveform pattern as a result of the comparison in the waveform pattern comparison unit.

2. The abnormal waveform pattern further includes at least one of a second abnormal waveform pattern in which the state exceeding the threshold is maintained for a certain period of time, a third abnormal waveform pattern in which the MD value is clearly greater than the threshold value, and a fourth abnormal waveform pattern in which the MD value increases over time.

3. The abnormality diagnosis device according to claim 1 , further comprising a history data storage unit that stores the diagnosis results as history data.

4. 4. The abnormality diagnosis device according to claim 1, further comprising an alarm output unit that outputs an alarm based on a diagnosis result from the diagnosis unit.

5. An abnormality diagnosis device for detecting abnormalities in a conveyor belt, A unit space is generated in advance from the normal data of the diagnostic target, As an abnormal waveform pattern of the MD value when the diagnosis target is abnormal, an abnormal waveform pattern in which the MD value exceeds a threshold value indicating an abnormality and then continues to be below the threshold value is stored in advance, and the abnormal waveform pattern is periodically repeated at a timing corresponding to the rotational movement of the conveyor belt, When diagnosing the diagnosis object, a time series waveform of MD values ​​generated from the collected data from the diagnosis object and the unit space is compared with the abnormal waveform pattern; The abnormality diagnosis method diagnoses whether or not the object to be diagnosed has an abnormality based on a comparison result between the waveform of time-series data of MD values ​​calculated from the collected data and the abnormal waveform pattern.

Images