Learning device and learning method, diagnosis device and fault diagnosis method for diagnosing whether equipment is abnormal based on time-domain vibration signals

Abstract

To provide a learning device and a learning method, and a device and a method for diagnosing abnormality of equipment based on a vibration signal in a time domain.SOLUTION: Acquiring, through a vibration sensor, a vibration signal in a time domain obtained by measuring a vibration generated when equipment is normally operated, acquiring a vibration signal for each frequency band filtered according to defined frequency bands by inputting the vibration signal to at least one filter, training a signal reconstruction model corresponding to each of the defined frequency bands using the vibration signal for each frequency band, and acquiring a reconstructed signal corresponding to each frequency band through the trained signal reconstruction model for each frequency band, A reconstruction error value indicating a difference between a vibration signal for each frequency band and a reconstruction signal corresponding to each frequency band may be determined, and a threshold value for abnormality detection corresponding to each defined frequency band may be determined based on the reconstruction error value determined for each defined frequency band.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present disclosure relates to a learning device and a learning method, and a diagnosis device and a fault diagnosis method for diagnosing abnormalities in equipment based on vibration signals in the time domain. [Background technology]

[0002] The equipment fault diagnosis technology using equipment vibration is a technology that diagnoses whether or not there is an abnormality in the equipment by analyzing vibration signals measured on the equipment. An unsupervised learning AI model can be used to diagnose equipment faults using equipment vibration. The unsupervised learning AI model can be trained using a large amount of vibration data collected during normal operation of the equipment. The vibration data can include frequency domain data or time domain data. Summary of the Invention [Means for solving the problem]

[0003] A learning device according to one embodiment includes a processor and a memory that stores instructions executable by the processor. When the instructions are executed by the processor, the instructions cause the learning device to acquire, via a vibration sensor, a time-domain vibration signal that measures vibrations generated when equipment is operated normally, input the vibration signal to at least one filter to acquire vibration signals for each frequency band filtered according to defined frequency bands, learn a signal reconstruction model corresponding to each of the defined frequency bands using the vibration signals for each frequency band, acquire a reconstructed signal corresponding to each frequency domain via the trained signal reconstruction model for each frequency band, determine a reconstruction error value indicating a difference between the vibration signal for each frequency band and the reconstructed signal corresponding to the frequency domain, and determine a threshold for anomaly detection corresponding to each of the defined frequency bands based on the reconstruction error value determined for each of the defined frequency bands.

[0004] The signal reconstruction model may include at least one of an autoencoder, a stacked autoencoder, a long short-term memory autoencoder (LSTM), and a convolutional autoencoder model.

[0005] When the instructions are executed by a processor, the instructions can cause the learning device to determine a reconstruction error value based on at least one of the average value of the differences between each frequency band vibration signal and the reconstructed signal corresponding to each frequency region, the average value of the squared differences, and the square root value of the average value of the squared differences.

[0006] When the instructions are executed by a processor, the instructions may cause a learning device to determine a maximum reconstruction error value for each frequency band or a threshold for the reconstruction error value based on the 3 Sigma rule.

[0007] According to an embodiment, a fault diagnosis apparatus for diagnosing the presence or absence of an anomaly in equipment using a signal reconstruction model includes a processor and a memory for storing instructions executable by the processor. When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to: acquire a time-domain vibration signal measuring vibration occurring in equipment via a vibration sensor; input the vibration signal to at least one filter to acquire a frequency-band-specific vibration signal filtered according to each defined frequency band; input each frequency-band-specific vibration signal to a signal reconstruction model corresponding to the frequency band to acquire a reconstructed signal corresponding to each defined frequency band from the signal reconstruction model; determine a reconstruction error value indicating a difference between the frequency-band-specific vibration signal and the reconstructed signal corresponding to the frequency band; and determine the presence or absence of an anomaly in the equipment based on the error value determined for each defined frequency band and a threshold value determined for each defined frequency band.

[0008] When the instructions are executed by a processor, the instructions may cause the fault diagnosis device to determine that there is an abnormality in the equipment if at least one of the reconstruction error values ​​determined for each frequency band is greater than a corresponding threshold value.

[0009] When the instructions are executed by a processor, the instructions can cause the fault diagnosis device to determine that the equipment is in a normal state if the reconstruction error value determined for each frequency band is less than or equal to the corresponding threshold value.

[0010] When the instructions are executed by a processor, the instructions can cause the fault diagnosis device to determine a reconstruction error value based on at least one of an average value of the differences between the vibration signals for each frequency band and the reconstructed signals corresponding to each frequency band, an average value of the squares of the differences, and a square root value of the average value of the squares of the differences.

[0011] When the instructions are executed by a processor, the instructions cause the fault diagnosis device to compare the reconstruction error value determined for each frequency band with the threshold determined for each frequency band, and when a frequency band in which the reconstruction error value is greater than the corresponding threshold is detected, the instruction can estimate an abnormality type based on the detected frequency band.

[0012] A learning method for training a signal reconstruction model using a learning device according to an embodiment may include an operation of acquiring, via a vibration sensor, a time-domain vibration signal measuring vibrations generated when equipment is operated normally; an operation of inputting the vibration signal to at least one filter to acquire vibration signals for each frequency band filtered according to defined frequency bands; an operation of training a signal reconstruction model corresponding to each defined frequency band using the vibration signals for each frequency band; an operation of acquiring reconstructed signals corresponding to each frequency domain through the trained signal reconstruction model for each frequency band; an operation of determining a reconstruction error value indicating a difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency domain; and an operation of determining a threshold value for anomaly detection corresponding to each defined frequency band based on the reconstruction error value determined for each defined frequency band.

[0013] The operation of determining the reconstruction error value may include an operation of determining the reconstruction error value based on at least one of the average value of the differences between each frequency band vibration signal and the reconstructed signal corresponding to each frequency region, the average value of the squared differences, and the square root value of the average value of the squared differences.

[0014] The operation of determining the threshold value may include an operation of determining a maximum value of the reconstruction error value for each frequency band or a threshold value for the reconstruction error value based on the 3 sigma rule.

[0015] A fault diagnosis method for diagnosing the presence or absence of an abnormality in equipment, which is performed by a fault diagnosis device according to one embodiment, may include an operation of acquiring a time-domain vibration signal measuring vibrations occurring in the equipment via a vibration sensor, an operation of acquiring vibration signals for each frequency band filtered according to the frequency band by inputting the vibration signal to at least one filter, an operation of inputting the vibration signals for each frequency band to a signal reconstruction model corresponding to the frequency band and acquiring a reconstructed signal corresponding to each defined frequency band from each signal reconstruction model, an operation of determining a reconstruction error value indicating a difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency band, and an operation of determining the presence or absence of an abnormality in the equipment based on the error value determined for each frequency band and the threshold value determined for each frequency band.

[0016] The operation of determining whether or not there is an abnormality may include an operation of determining that there is an abnormality in the equipment if at least one of the reconstruction error values ​​determined for each frequency band is greater than a corresponding threshold value.

[0017] The operation of determining whether or not there is an abnormality may include an operation of determining that the equipment is in a normal state if the reconstruction error value determined for each frequency band is equal to or less than a corresponding threshold value.

[0018] The operation of determining the reconstruction error value may include an operation of determining the reconstruction error value based on at least one of the average value of the differences between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency band, the average value of the squares of the differences, and the square root value of the average value of the squares of the differences.

[0019] The fault diagnosis method may further include an operation of comparing the reconstruction error value determined for each frequency band with a threshold determined for each frequency band, and when a frequency band in which the reconstruction error value is greater than the corresponding threshold is detected, estimating an abnormality type based on the detected frequency band. [Brief explanation of the drawings]

[0020] [Figure 1]1 is a diagram illustrating diagnosing whether or not equipment has a fault using a vibration signal in the time domain according to an embodiment; [Figure 2] FIG. 1 is a block diagram illustrating a configuration of a learning device according to an embodiment. [Figure 3] 1 is a block diagram for explaining the configuration of a fault diagnosis device according to an embodiment; [Figure 4] FIG. 10 is a diagram illustrating filtering of a time domain vibration signal according to one embodiment. [Figure 5] FIG. 10 is a diagram illustrating determining a reconstruction error value using a reconstructed signal according to one embodiment. [Figure 6] FIG. 10 is a diagram illustrating how a learning device according to an embodiment determines a threshold value. [Figure 7] 4 is a flowchart illustrating how a fault diagnosis device according to an embodiment diagnoses whether or not a fault exists in equipment. [Figure 8] 3 is a diagram illustrating how a fault diagnosis device according to an embodiment estimates whether or not an abnormality exists in equipment for each frequency band. FIG. [Figure 9] 10 is a flowchart illustrating the operation of a learning method according to an embodiment. [Figure 10] 4 is a flowchart illustrating the operation of a fault diagnosis method according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0021] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Therefore, the actual implementation is not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or alternatives included in the technical ideas described in the embodiments.

[0022] Although terms such as "first" or "second" may be used to describe various components, these terms should be construed only to distinguish one component from another. For example, a first component may be termed a second component, and similarly, a second component may be termed a first component.

[0023] When a component is described as being "coupled" to another component, it should be understood that the component may be directly coupled or connected to the other component, or there may be other components intervening therebetween.

[0024] "Singular expressions include plural expressions unless the context clearly dictates otherwise. In this document, each of the phrases "A or B," "at least one of A and B," "at least one of A or B," "A, B or C," "at least one of A, B, and C," etc., can include any or all possible combinations of the items listed together with the corresponding phrase. In this specification, terms such as "comprise" or "have" specify the presence of stated features, numbers, steps, operations, components, parts, or combinations thereof, and should be understood as not precluding the presence or possible addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0025] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted to have a meaning consistent with the meaning they have in the context of the relevant art, and should not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0026] The term "module" as used in this document refers to software or hardware components such as FPGAs or ASICs, and the "module" can function in either way. However, the term "module" is not limited to software or hardware. A "module" may reside on an addressable storage medium and be configured to implement one or more processes. For example, a "module" may include components such as software components, object-oriented software components, class components, or task components, as well as processes, functions, attributes, procedures, subroutines, program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The components and functionality provided within a "module" may be combined into fewer components and modules or further separated into additional components and modules. Furthermore, a component and a "module" may be implemented to implement one or more CPUs within a device or security multimedia card. A "module" may also include one or more processors.

[0027] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. In the description with reference to the accompanying drawings, the same components will be given the same reference numerals regardless of the reference numerals, and redundant description thereof will be omitted.

[0028] 1 is a diagram illustrating diagnosing whether or not an equipment fault exists using a time-domain vibration signal according to an embodiment. A fault diagnosis device (e.g., fault diagnosis device 300 in FIG. 3) can determine whether or not an equipment fault exists based on an AI (Artificial Intelligence) model (e.g., signal reconstruction models 221, 222, and 223 in FIG. 3), a reconstruction error value, and a threshold value.

[0029] Referring to FIG. 1, a fault diagnosis device may acquire a vibration signal via a sensor (e.g., a vibration signal detection sensor). The fault diagnosis device may determine whether or not an abnormality exists in the equipment using the vibration signal generated in the equipment. The signal used to determine whether or not an abnormality exists in the equipment may include a time-domain vibration signal. According to an embodiment, the fault diagnosis device may periodically acquire a vibration signal via a sensor. For example, the fault diagnosis device may periodically acquire a vibration signal of the equipment via a vibration sensor. Graph 110 is a graph of a periodic vibration signal acquired by the fault diagnosis device via the vibration sensor. Graph 110 is a time-domain vibration graph, which is a graph obtained by combining vibration signals of one or more frequency bands.

[0030] The fault diagnosis device may filter a time-domain vibration signal for each frequency band. The fault diagnosis device may input a vibration signal to one or more filters and obtain a time-domain vibration signal for each defined frequency band from the filters. For example, the fault diagnosis device may input the vibration signal of graph 110 to a filter to obtain a time-domain vibration signal in a first frequency band and a time-domain vibration signal in a second frequency band. Graph 120 is a graph showing a time-domain vibration signal in a first frequency band, and graph 130 is a graph showing a time-domain vibration signal in a second frequency band. Various filters may be used to filter the vibration signal, which will be described in more detail with reference to FIG. 4.

[0031] According to an embodiment, the fault diagnosis apparatus may input the time-domain vibration signals filtered for each frequency band to the corresponding signal reconstruction model. For example, the fault diagnosis apparatus may input the time-domain vibration signals in the first frequency band of the graph 120 to the first signal reconstruction model 121, and input the time-domain vibration signals in the second frequency band to the second signal reconstruction model 131.

[0032] According to an embodiment, the signal reconstruction models 121 and 131 may be artificial intelligence models trained through unsupervised learning. The signal reconstruction models 121 and 131 may be artificial intelligence models trained through unsupervised learning using a learning device (e.g., the learning device 200 of FIG. 2). The signal reconstruction models 121 and 131 are described in more detail in FIG. 2. The fault diagnosis device may receive a time-domain vibration signal as input and obtain a reconstructed signal corresponding to the input vibration signal from each signal reconstruction model. For example, the fault diagnosis device may obtain a reconstructed signal in a first frequency band from the first signal reconstruction model 121 and a reconstructed signal in a second frequency band from the second signal reconstruction model 131. Graph 122 illustrates the reconstructed signal in the first frequency band from the first signal reconstruction model 121, and graph 132 illustrates the reconstructed signal in the second frequency band from the second signal reconstruction model 131. Generation of the reconstructed signals by the fault diagnosis device will be described in more detail in FIG. 5.

[0033] The fault diagnosis device can determine whether or not the equipment is abnormal by comparing the reconstruction error value with a threshold value for each frequency band. The fault diagnosis device can determine whether or not the equipment is abnormal based on the reconstruction error value for each frequency and the threshold value for each frequency. For example, the fault diagnosis device can determine that the equipment is abnormal in the first frequency band if the first reconstruction error value in the first frequency band is greater than the first threshold value (if “YES” in Operation 123), and can determine that the equipment is normal in the first frequency band if the first reconstruction error value is equal to or less than the first threshold value (if “NO” in Operation 123). The fault diagnosis device can determine that the equipment is abnormal in the second frequency band if the second reconstruction error value in the second frequency band is greater than the second threshold value (if “YES” in Operation 133), and can determine that the equipment is normal in the second frequency band if the second reconstruction error value is equal to or less than the second threshold value (if “NO” in Operation 133).

[0034] FIG. 2 is a block diagram illustrating the configuration of a learning device according to an embodiment.

[0035] 2, the learning device 200 may include a vibration signal acquisition unit 210, a signal filtering 215, a signal reconstruction unit 220, a reconstruction error value determination unit 230, and a threshold determination unit 240. The learning device 200 may include a processor (not shown) for operating the vibration signal acquisition unit 210, the signal filtering 215, the signal reconstruction unit 220, the reconstruction error value determination unit 230, and the threshold determination unit 240, and a memory that stores instructions executable by the processor (not shown).

[0036] The vibration signal acquisition unit 210 may acquire a time-domain vibration signal that measures vibrations generated during normal operation of the equipment through a vibration sensor. According to an embodiment, the vibration sensor may include a micro-electromechanical system (MEMS), integrated electronics piezo-electric (IEPE), or other vibration sensors. The measurement time range of the vibration signal measured through the vibration sensor may vary depending on the specifications of the sensor and the environment in which the vibration signal is collected. The vibration signal measured through the vibration sensor may be visualized in the time domain through a graph.

[0037] The signal filtering unit 215 can acquire the vibration signals for each frequency band from the filters. The signal filtering unit 215 can acquire the vibration signals for each frequency band by inputting the vibration signal to at least one filter. For example, the signal filtering unit 215 can acquire the vibration signals for each frequency band by inputting the vibration signal to a plurality of band pass filters (BPFs).

[0038] The signal reconstruction unit 220 can train signal reconstruction models 221, 222, and 223 corresponding to each of the defined frequency bands using the vibration signals for each frequency band. According to an embodiment, the signal reconstruction unit 220 can train (e.g., unsupervised learning) the signal reconstruction models using vibration signals measured in equipment under normal conditions as training data. The signal reconstruction unit 220 can include a plurality of signal reconstruction models (e.g., a first signal reconstruction model 221, a second signal reconstruction model 222, and a third signal reconstruction model 223). The first signal reconstruction model 221 can correspond to the first signal reconstruction model 121 of FIG. 1, and the second signal reconstruction model 222 can correspond to the second signal reconstruction model 131 of FIG. 1.

[0039] According to an embodiment, the signal reconstruction unit 220 may perform unsupervised learning of the signal reconstruction models 221, 222, and 223. For example, the signal reconstruction unit 220 may perform unsupervised learning of the signal reconstruction models 221, 222, and 223 using a time-domain vibration signal measured in equipment under normal conditions. The signal reconstruction models 221, 222, and 223 that the signal reconstruction unit 220 performs unsupervised learning of may include at least one of an autoencoder, a stacked autoencoder, a long short-term memory autoencoder (LSTM) autoencoder, and a convolutional autoencoder model.

[0040] According to an embodiment, the signal reconstruction unit 220 can obtain reconstructed signals corresponding to each frequency domain from the signal reconstruction models 221, 222, and 223. For example, when the signal reconstruction unit 220 inputs a time-domain vibration signal measured in equipment under normal conditions to the signal reconstruction models 221, 222, and 223, the signal reconstruction unit 220 can obtain reconstructed signals similar to the time-domain vibration signals for each frequency band from the signal reconstruction models 221, 222, and 223.

[0041] The reconstruction error value determiner 230 may determine a reconstruction error value for the reconstructed signal. The reconstruction error value indicates a difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency region. According to an embodiment, the reconstruction error value determiner 230 may determine each frequency band error value based on the difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency region. For example, the reconstruction error value determiner 230 may determine the reconstruction error value based on at least one of the average value of the difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency region, the average value of the squared difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency region, and the square root value of the average value of the squared difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency region. The reconstruction error value will be described in more detail with reference to FIG. 5.

[0042] The threshold determination unit 240 may determine a threshold for detecting an anomaly for each defined frequency band. The threshold determination unit 240 may determine the threshold based on the reconstruction error value determined for each defined frequency band. For example, the threshold determination unit 240 may determine the threshold based on the maximum reconstruction error value for each frequency band or the 3-Sigma rule. The threshold determination by the threshold determination unit 240 will be described in more detail with reference to FIG. 6.

[0043] FIG. 3 is a block diagram for explaining the configuration of the fault diagnosis device according to one embodiment.

[0044] 3, the fault diagnosis device 300 may include a vibration signal acquisition unit 210, a signal filtering unit 215, a signal reconstruction unit 220, a reconstruction error value determination unit 230, and an abnormality presence / absence determination unit 310. The fault diagnosis device 300 may include a processor (not shown) for operating the vibration signal acquisition unit 210, the signal filtering unit 215, the signal reconstruction unit 220, the reconstruction error value determination unit 230, and the abnormality presence / absence determination unit 310, and a memory for storing instructions executable by the processor (not shown).

[0045] The vibration signal acquisition unit 210, the signal filtering unit 215, the signal reconstruction unit 220, and the reconstruction error value determination unit 230 have been explained in FIG. 2, so a duplicate explanation will be omitted.

[0046] The abnormality determination unit 310 may determine whether or not there is an abnormality in the equipment based on the error value determined for each defined frequency band and the threshold value determined for each defined frequency band.

[0047] According to one embodiment, the abnormality determination unit 310 may determine whether or not there is an abnormality in the entire equipment based on the presence or absence of an abnormality for each frequency. For example, the abnormality determination unit 310 may determine whether or not there is an abnormality in the entire equipment by performing a logical operation (e.g., an OR logical operation (normal (1), abnormal (0))) based on the abnormality presence / absence result in the first frequency band and the abnormality presence / absence result in the second frequency band. If the abnormality presence / absence determination value in the first frequency band is 1 (normal) and the abnormality presence / absence result value in the second frequency band is 1 (normal), the abnormality determination unit 310 may determine that there is no abnormality in the entire equipment since the OR logical operation value is 1.

[0048] According to an embodiment, the abnormality determination unit 310 can determine whether or not an abnormality exists in the entire equipment based on N (a natural number equal to or greater than 2) filters and N signal reconstruction models. The combination of the result values ​​for each frequency based on the defined N frequency regions can be calculated as follows: N If all of the result values ​​in the N frequency domains are 1 (for example, the OR logical operation value is 1), the abnormality presence / absence determining unit 310 can determine that there is no abnormality in the entire equipment.

[0049] FIG. 4 is a diagram illustrating filtering of a vibration signal in the time domain according to one embodiment.

[0050] Referring to FIG. 4, a signal filtering unit (e.g., signal filtering unit 215 of FIG. 2) of a fault diagnosis device (e.g., fault diagnosis device 300 of FIG. 3) can filter a time-domain vibration signal for each defined frequency band. The signal filtering unit of the fault diagnosis device can include at least one filter for filtering the vibration signal. The fault diagnosis device can acquire a time-domain vibration signal for each defined frequency band from a band-specific filter (e.g., first filter 420, second filter 430). Signal filtering (or frequency filtering) is a method of passing or suppressing signals in a specific frequency band using a filter. Filters can include a band-pass filter (BPF), a high-pass filter (HPF), a low-pass filter (LPF), a band-stop filter (BSF), etc. A band-pass filter (BPF) passes only a specific frequency band, and a high-pass filter (HPF) passes only signals above a specific frequency. A low-pass filter (LPF) is a filter that passes only signals below a specific frequency, and a band-elimination filter (BSF) is a filter that blocks only signals in a specific frequency band.

[0051] Graph 410 is a graph showing a time-domain vibration signal acquired via a vibration sensor. According to an embodiment, the signal filtering unit may acquire vibration signals for each defined frequency band using a band-pass filter (BPF). For example, the signal filtering unit may acquire a vibration signal for a first frequency band using a first filter 420 (e.g., a first band-pass filter (BPF)) that passes only frequency signals in a first frequency band (e.g., frequencies equal to or less than 124 Hz), and may acquire a vibration signal for a second frequency band using a second filter 430 (a second band-pass filter (BPF)) that passes only frequency signals in a second frequency band (e.g., frequencies greater than 124 Hz). Graph 421 is a graph showing a vibration signal for the first frequency band in the time domain, and graph 431 is a graph showing a vibration signal for the second frequency band in the time domain.

[0052] The number of filters and the range of the frequency band passing through each filter may vary depending on the facility operating environment and the frequency region of interest, and are not limited to the above embodiment.

[0053] FIG. 5 is a diagram illustrating determining a reconstruction error value using a reconstructed signal according to one embodiment.

[0054] 5, a fault diagnosis apparatus (e.g., fault diagnosis apparatus 300 of FIG. 3) may determine a reconstruction error value based on a time-domain vibration signal and a reconstructed signal. The generated reconstructed signal of graph 530 is a resultant value output by inputting the vibration signal of the first frequency band of graph 510 into first signal reconstruction model 520. The reconstructed signal output from first signal reconstruction model 520 may be a reconstructed signal similar to the input vibration signal to the extent that the input vibration signal is similar to the vibration signal used for training first signal reconstruction model 520. First signal reconstruction model 520 may correspond to first signal reconstruction model 121 of FIG. 1 and / or first signal reconstruction model 221 of FIG. 2. Graph 540 is a graph showing a vibration signal of the first frequency band and a reconstructed signal corresponding to the vibration signal of the first frequency band.

[0055] An error value determiner of the fault diagnosis device according to an embodiment may determine a reconstruction error value based on a difference between the vibration signal of a first frequency band and the reconstructed signal corresponding to the vibration signal of the first frequency band in graph 540. For example, the error value determiner of the fault diagnosis device may determine the reconstruction error value using at least one of an average value of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, an average value of the square of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, and a square root of the average value of the square of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region.

[0056] The mean absolute error (MAE) of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region is expressed by the following Equation 1.

[0057] [Formula 1]

[0058] JPEG0007825245000001.jpg1744

[0059] The mean squared error (MSE) of the square of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region is expressed by the following Equation 2.

[0060] [Formula 2]

[0061] JPEG0007825245000002.jpg1847

[0062] The root mean squared error (RMSE) of the square of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region is expressed by the following Equation 3.

[0063] [Formula 3]

[0064] JPEG0007825245000003.jpg2846

[0065] JPEG0007825245000004.jpg14169

[0066] FIG. 6 is a diagram for explaining how a learning device according to an embodiment determines a threshold value.

[0067] 6, a reconstruction error value determiner (e.g., reconstruction error value determiner 230 of FIG. 3) of a fault diagnosis apparatus (e.g., fault diagnosis apparatus 300 of FIG. 3) may determine a threshold value using a reconstructed signal. For example, the reconstruction error value determiner may determine a threshold value using a maximum value of the reconstructed signal or based on the 3-Sigma rule. The method for determining the threshold value may be determined differently depending on the equipment operation environment. According to an embodiment, when there are N frequency domains into which the frequency is divided, the reconstruction error value determiner may determine a threshold value in each frequency domain for each of the N reconstructed signals.

[0068] According to an embodiment, a threshold value determiner (e.g., threshold value determiner 240 of FIG. 2) may determine a threshold value for each frequency band (e.g., a first frequency band (a frequency band equal to or less than 124 Hz) and a second frequency band (a frequency band exceeding 124 Hz)). For example, graph 610 is a graph showing reconstruction error values ​​in a first frequency band. The threshold value determiner may determine, as a first threshold, a maximum value 611 of the reconstruction error values ​​for the first frequency band or a value 612 that is three times the standard deviation of the reconstruction error values ​​from the average value of the reconstruction error values. Graph 620 shows reconstruction error values ​​in a second frequency band. The threshold value determiner may determine, as a second threshold, a maximum value 621 of the reconstruction error values ​​for the second frequency band or a value 622 that is three times the standard deviation of the reconstruction error values ​​from the average value of the reconstruction error values.

[0069] FIG. 7 is a flowchart illustrating how the fault diagnosis device according to one embodiment diagnoses whether or not there is a fault in the equipment.

[0070] 7, a fault diagnosis device (e.g., fault diagnosis device 300 of FIG. 3) can determine whether or not an abnormality exists in the entire equipment based on the error value and threshold value for each frequency. The error value and threshold value for each frequency have been described in FIGS. 5 and 6, so a repeated description will be omitted.

[0071] According to an embodiment, a fault diagnosis device may determine that an equipment malfunctions if at least one of the reconstruction errors determined for each frequency band is greater than a corresponding threshold. For example, if a first reconstruction error value is greater than a threshold for a first frequency band ("YES" in operation 710) or if a second reconstruction error value is greater than a threshold for a second frequency band ("YES" in operation 720), the fault diagnosis device may determine that the equipment malfunctions. If the first reconstruction error value is equal to or less than the threshold for the first frequency band ("NO" in operation 710) and the second reconstruction error value is equal to or less than the threshold for the second frequency band ("NO" in operation 720), the fault diagnosis device may determine that the equipment is normal.

[0072] According to an embodiment, the fault diagnosis device can determine whether or not there is an abnormality in the entire equipment as well as whether or not there is a fault in the equipment for each frequency band. A fault related to a first frequency band may be an imbalance, and a fault related to a second frequency band may be a misalignment. If the first reconstruction error value is equal to or less than a threshold value for the first frequency band and the second reconstruction error value is greater than a threshold value for the second frequency band, the fault diagnosis device can determine that the equipment is misaligned.

[0073] In one embodiment, when there are N frequency bands, the fault diagnosis device can determine that the equipment is normal only if the error value of each frequency band is equal to or less than the threshold value corresponding to each frequency band.

[0074] FIG. 8 is a diagram illustrating how the fault diagnosis device according to an embodiment estimates an abnormality type of equipment for each frequency band.

[0075] 8, a fault diagnosis device (e.g., fault diagnosis device 300 of FIG. 3) can compare a reconstruction error value determined for each frequency band with a threshold value determined for each frequency band. When a frequency band in which a reconstruction error value is greater than a corresponding threshold value is detected, the fault diagnosis device can estimate an abnormality type (e.g., abnormality type 1x) of the equipment based on the detected frequency band.

[0076] According to an embodiment, the fault diagnosis apparatus may compare the reconstruction error value determined for each frequency band with a threshold value determined for each frequency band. When a frequency band in which the reconstruction error value is greater than the threshold value is detected as a result of comparing the reconstruction error value with the threshold value, the fault diagnosis apparatus may estimate an abnormality type based on the detected frequency band. When the reconstruction error value is equal to or less than the threshold value in the frequency band of graph 810 and a vibration frequency signal is not detected in other frequency bands, the fault diagnosis apparatus may determine that the equipment is in a normal state. When the reconstruction error value is greater than the threshold value in frequency band 821 of graph 820, the fault diagnosis apparatus may estimate an abnormality type (e.g., 1X type) of the equipment associated with frequency band 821. When the reconstruction error value is greater than the threshold value in each of frequency bands 831, 832, and 833 of graph 830, the fault diagnosis apparatus may estimate an abnormality type (e.g., 1X type, 2X type, or 3X type) of the equipment associated with each of frequency bands 831, 832, and 833. If each reconstruction error value corresponding to frequency band 841 to frequency band 842 of graph 840 is greater than each threshold value, the fault diagnosis device can estimate the abnormality type (e.g., 1X type, 2X type, 3X type, 4X type, 5X type, 6X type, 7X type, 8X type, 9X type) of the equipment associated with each frequency band.

[0077] The fault diagnosis device can estimate the type of abnormality occurring in the equipment by utilizing the occurrence of abnormal amplitude in a specific frequency range. Since the vibration signals occurring in the equipment show high amplitude in a specific frequency range depending on the type of various defects, the fault diagnosis device can perform effective equipment fault diagnosis.

[0078] Fault diagnosis systems can improve their anomaly detection performance (True Positive Rate, TPR) by determining whether equipment is faulty by utilizing abnormal amplitudes that occur in specific frequency ranges. For example, fault diagnosis systems can estimate equipment abnormalities such as unbalance, misalignment, looseness, and bearing faults using a rotating machinery simulator and an IEPE (Integrated Electronics Piezo-Electric) type vibration sensor.

[0079] FIG. 9 is a flowchart illustrating the operation of the learning method according to one embodiment.

[0080] The operations of the training method for training a signal reconstruction model can be performed by a training device (eg, training device 200 of FIG. 2).

[0081] In operation 910, the learning device can acquire a vibration signal via a vibration sensor. The learning device can acquire a time-domain vibration signal that measures vibrations that occur when the equipment is operating normally via the vibration sensor. For example, the learning device can acquire a vibration signal that occurs in the equipment via a MEMS (Micro-Electro Mechanical System) sensor.

[0082] In operation 920, the learning device can obtain the vibration signals for each frequency band through a filter. The learning device can obtain the vibration signals for each frequency band filtered according to the defined frequency bands by inputting the vibration signal to at least one filter. For example, the learning device can obtain the vibration signals for each frequency band from multiple band-pass filters to which the vibration signal is input.

[0083] In operation 930, the learning device can train a signal reconstruction model corresponding to each of the frequency bands. The learning device can train a signal reconstruction model corresponding to each of the defined frequency bands using the vibration signal for each frequency band. For example, the learning device can train the signal reconstruction model in an unsupervised manner using the vibration signal for each frequency band of the equipment in a normal state. The signal reconstruction model according to one embodiment can include at least one of an autoencoder, a stacked autoencoder, a long short-term memory autoencoder (LSTM), and a convolutional autoencoder model.

[0084] In operation 940, the learning device can obtain a reconstructed signal corresponding to each frequency region through the trained signal reconstruction model for each frequency band.

[0085] In operation 950, the learning device can determine a reconstruction error value corresponding to each frequency region. The learning device can determine a reconstruction error value indicating a difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region. For example, the learning device can determine the reconstruction error value based on any one of the average value of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, the average value of the squared difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, and the square root value of the average value of the squared difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region.

[0086] In operation 960, the learning device may determine a threshold value corresponding to each frequency region. The learning device may determine a threshold value for anomaly detection corresponding to each defined frequency band based on the reconstruction error value determined for each defined frequency band. According to an embodiment, the learning device may determine a threshold value for the reconstruction error value based on the maximum value of the reconstruction error value for each frequency band or the 3-Sigma rule. The 3-Sigma rule determines a value that is three times the standard deviation of the reconstruction error value from the average value of the reconstruction error value as the threshold.

[0087] FIG. 10 is a flowchart for explaining the operation of the fault diagnosis method according to one embodiment.

[0088] The operation of the fault diagnosis method for diagnosing whether or not there is an abnormality in the equipment can be performed by a fault diagnosis device (for example, the fault diagnosis device 300 of FIG. 3).

[0089] In operation 1010, the fault diagnosis device may acquire a vibration signal via a vibration sensor. The vibration signal acquired via the vibration sensor may include a time-domain vibration signal. For example, the fault diagnosis device may acquire the time-domain vibration signal via an Integrated Electronics Piezo-Electric (IEPE) vibration sensor.

[0090] In Operation 1020, the fault diagnosis device can acquire vibration signals for each frequency band through a filter. The fault diagnosis device can acquire vibration signals for each frequency band filtered according to the frequency band by inputting the time-domain vibration signal to at least one filter. For example, the fault diagnosis device can acquire vibration signals for a band greater than 124 Hz and vibration signals for a band less than 124 Hz by inputting the time-domain vibration signal to a high-pass filter (HPF) (cutoff frequency equal to or less than 124 Hz) and a low-pass filter (LPF) (cutoff frequency exceeding 124 Hz).

[0091] In Operation 1030, the fault diagnosis apparatus may obtain a reconstructed signal from each signal reconstruction model. For example, the fault diagnosis apparatus may input the vibration signal for each frequency band to a signal reconstruction model corresponding to the frequency band, and obtain a reconstructed signal corresponding to each defined frequency band from each signal reconstruction model. According to an embodiment, the signal reconstruction model may include at least one of an autoencoder, a stacked autoencoder, an LSTM autoencoder, and a convolutional autoencoder model.

[0092] In Operation 1040, the fault diagnosis apparatus may determine a reconstruction error value for each frequency band. For example, the fault diagnosis apparatus may determine a reconstruction error value indicating a difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency band. According to an embodiment, the fault diagnosis apparatus may determine the reconstruction error value based on at least one of an average value of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, an average value of the squares of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region, and a square root value of the average value of the squares of the difference between the vibration signal for each frequency band and the reconstructed signal corresponding to each frequency region.

[0093] The fault diagnosis device can determine whether or not the equipment is abnormal based on the error value determined for each frequency band and the threshold value determined for each frequency band. If at least one of the reconstruction error values ​​determined for each frequency band is equal to or less than the threshold value determined for each frequency band ("NO" in Operation 1050), the fault diagnosis device can determine in Operation 1060 that the equipment is in a normal state.

[0094] If at least one of the reconstruction error values ​​determined for each frequency band is greater than the threshold determined for that frequency band ("YES" in Operation 1050), the fault diagnosis device may determine in Operation 1070 that the equipment is in an abnormal state. If the fault diagnosis device determines that the equipment is in an abnormal state, the fault diagnosis device may estimate an abnormality type based on the detected frequency bands in Operation 1080. The fault diagnosis device compares the reconstruction error value determined for each frequency band with the threshold determined for that frequency band, and if a frequency band in which the reconstruction error value is greater than the corresponding threshold is detected, the fault diagnosis device may estimate an abnormality type based on the detected frequency band. A fault type that can occur in a first frequency band may be a 1X type, and a fault type that can occur in a second frequency band may be a 2X type. If the reconstruction error value in the first frequency band is greater than the corresponding threshold, the fault diagnosis device may determine that the equipment has a 1X type fault.

[0095] The fault diagnosis device can obtain vibration signals for each frequency band by filtering the time-domain vibration signal through a band-pass filter (BPF), a high-pass filter (HPF), a low-pass filter (LPF), etc. The fault diagnosis device can input the vibration signals for each frequency band to a signal reconstruction model for each frequency band to estimate whether an abnormality has occurred corresponding to each frequency band. The fault diagnosis device can determine whether an abnormality exists in the entire equipment based on the type of abnormality of the equipment estimated based on the vibration signals for each frequency band.

[0096] The above-described embodiments may be implemented using hardware components, software components, and / or a combination of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as a processor, controller, arithmetic logic unit (ALU), digital signal processor, microcomputer, field programmable gate array (FPGA), programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and software applications that run on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a processing device may be described as being a single device, but those skilled in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, the processing device may include multiple processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

[0097] Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing device or, independently or collectively, instruct the processing device to operate as desired. The software and / or data may be permanently or temporarily embodied in some type of machine, component, physical device, virtual device, computer storage medium, or device to be interpreted by the processing device or to provide instructions or data to the processing device. The software may also be distributed across computer systems connected by a network, stored or executed in a distributed manner. The software and data may be stored on a computer-readable storage medium.

[0098] Methods according to the embodiments may be embodied in the form of program instructions that can be executed by various computer means and may be recorded on a computer-readable medium. The computer-readable medium may store program instructions, data files, data structures, and the like, alone or in combination. The program instructions stored on the medium may be those specially designed and constructed for the embodiments, or may be those well known and available to those skilled in the art of computer software. Examples of computer-readable storage media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include not only machine language code produced by a compiler, but also high-level language code that can be executed by a computer using an interpreter, for example.

[0099] The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0100] Although the embodiments have been described above using limited drawings, those skilled in the art can apply various technical modifications and variations thereto. For example, the described techniques may be performed in a different order than described, and / or the components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different manner than described, or replaced with other components or equivalents, and still achieve suitable results.

[0101] Accordingly, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

Claims

1. A learning device, comprising: a processor; a memory storing instructions executable by said processor; When the instructions are executed by the processor, the instructions cause the learning device to: Via a vibration sensor, a time domain vibration signal is acquired that measures the vibration generated when the equipment is operating normally, inputting the vibration signal to at least one filter to obtain vibration signals for each frequency band filtered according to defined frequency bands; learning a signal reconstruction model corresponding to each of the defined frequency bands using the vibration signals for each frequency band; Obtaining a reconstructed signal corresponding to each frequency region through the trained signal reconstruction model for each frequency band; determining a reconstruction error value indicating a difference between each of the frequency band vibration signals and a reconstructed signal corresponding to each of the frequency regions; determining a threshold value for detecting an anomaly corresponding to each of the defined frequency bands based on the reconstruction error value determined for each of the defined frequency bands; A learning device, wherein the defined frequency bands are bands that are set in advance to correspond to each of the types of abnormalities that may occur in the equipment.

2. The signal reconstruction model is 2. The learning device of claim 1, comprising at least one of an autoencoder, a stacked autoencoder, a long short-term memory autoencoder (LSTM) autoencoder, and a convolutional autoencoder model.

3. When the instructions are executed by the processor, the instructions cause the learning device to:

2. The learning device according to claim 1, wherein the reconstruction error value is determined based on at least one of an average value of the differences between the vibration signals for each frequency band and the reconstructed signals corresponding to each frequency region, an average value of the squares of the differences, and a square root value of the average value of the squares of the differences.

4. When the instructions are executed by the processor, the instructions cause the learning device to: The learning device according to claim 1 , wherein the threshold value for the reconstruction error value is determined for each frequency band based on a maximum value of the reconstruction error value or a 3 SIGMA rule.

5. A fault diagnosis device that diagnoses the presence or absence of an abnormality in equipment using a signal reconstruction model, a processor; a memory storing instructions executable by said processor; When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to: Via a vibration sensor, a time domain vibration signal is acquired that measures the vibration occurring in the equipment, inputting the vibration signal to at least one filter to obtain vibration signals for each frequency band filtered according to each defined frequency band; inputting the vibration signals for each frequency band into a signal reconstruction model corresponding to the frequency band, and obtaining a reconstructed signal corresponding to each of the defined frequency bands from each signal reconstruction model; determining a reconstruction error value indicating a difference between the vibration signal for each frequency band and a reconstructed signal corresponding to each frequency band; determining whether or not there is an abnormality in the equipment based on the error value determined for each of the defined frequency bands and the threshold value determined for each of the defined frequency bands; and when a frequency band in which the reconstruction error value is greater than a corresponding threshold value is detected as a result of comparing the reconstruction error value determined for each frequency band with a threshold value determined for each frequency band, the fault diagnosis device estimates an abnormality type based on the detected frequency band.

6. When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to: The fault diagnosis device according to claim 5 , wherein if at least one of the reconstruction error values ​​determined for each frequency band is greater than a corresponding threshold, it is determined that there is an abnormality in the equipment.

7. When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to: The fault diagnosis device of claim 5 , wherein the device determines that the equipment is in a normal state when the reconstruction error value determined for each frequency band is equal to or less than a corresponding threshold value.

8. When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to: The signal reconstruction model is The fault diagnosis apparatus of claim 5 , comprising at least one of an autoencoder, a stacked autoencoder, an LSTM autoencoder, and a convolutional autoencoder model.

9. When the instructions are executed by the processor, the instructions cause the fault diagnosis apparatus to:

6. The fault diagnosis device according to claim 5, wherein the reconstruction error value is determined based on at least one of an average value of differences between the vibration signals for each frequency band and the reconstructed signals corresponding to the respective frequency bands, an average value of the squares of the differences, and a square root value of the average value of the squares of the differences.

10. A learning method for learning a signal reconstruction model by a learning device, comprising: An operation of acquiring a time domain vibration signal that measures vibrations generated when the equipment is operated normally via a vibration sensor; inputting the vibration signal into at least one filter to obtain vibration signals for each frequency band filtered according to defined frequency bands; an operation of learning a signal reconstruction model corresponding to each of the defined frequency bands using the vibration signals for each frequency band; An operation of obtaining a reconstructed signal corresponding to each frequency region through the learned signal reconstruction model for each frequency band; determining a reconstruction error value indicating a difference between each frequency band vibration signal and a reconstructed signal corresponding to each frequency band; and determining a threshold value for anomaly detection corresponding to each of the defined frequency bands based on the reconstruction error value determined for each of the defined frequency bands; A learning method, wherein the defined frequency bands are bands that are set in advance to correspond to each of the types of abnormalities that may occur in the equipment.

11. The signal reconstruction model is 11. The method of claim 10, comprising at least one of an autoencoder, a stacked autoencoder, a long short-term memory autoencoder, and a convolutional autoencoder model.

12. The operation of determining the reconstruction error value comprises:

11. The learning method according to claim 10, further comprising: determining the reconstruction error value based on at least one of an average value of differences between the vibration signals for each frequency band and the reconstructed signals corresponding to each frequency region, an average value of the squares of the differences, and a square root value of the average value of the squares of the differences.

13. The act of determining the threshold value includes: The learning method of claim 10 , further comprising determining a threshold value for the reconstruction error value based on a maximum value of the reconstruction error value or a 3 SIGMA rule for each frequency band.

14. A fault diagnosis method for diagnosing whether or not there is an abnormality in equipment, which is performed by a fault diagnosis device, comprising: An operation of acquiring a time domain vibration signal that measures vibration occurring in the equipment via a vibration sensor; inputting the vibration signal to at least one filter to obtain vibration signals for each frequency band filtered according to the frequency band; an operation of inputting the vibration signals for each frequency band into a signal reconstruction model corresponding to the frequency band, and obtaining a reconstructed signal corresponding to each defined frequency band from each signal reconstruction model; determining a reconstruction error value indicating a difference between the vibration signal for each frequency band and a reconstructed signal corresponding to each frequency band; and determining whether or not there is an abnormality in the equipment based on the error value determined for each frequency band and the threshold value determined for each frequency band; the operation of determining whether or not an abnormality exists further includes an operation of estimating an abnormality type based on the detected frequency band when a frequency band in which the reconstruction error value is greater than a corresponding threshold is detected as a result of comparing the reconstruction error value determined for each frequency band with a threshold determined for each frequency band.

15. The operation of determining whether or not there is an abnormality is The fault diagnosis method of claim 14 , further comprising: determining that there is an abnormality in the equipment when at least one of the reconstruction error values ​​determined for each frequency band is greater than a corresponding threshold value.

16. The operation of determining whether or not there is an abnormality is The fault diagnosis method of claim 14 , further comprising determining that the equipment is in a normal state if the reconstruction error value determined for each frequency band is equal to or less than a corresponding threshold value.

17. The signal reconstruction model is The fault diagnosis method of claim 14 , comprising at least one of an autoencoder, a stacked autoencoder, an LSTM autoencoder, and a convolutional autoencoder model.

18. The operation of determining the reconstruction error value comprises:

15. The fault diagnosis method according to claim 14, further comprising: determining the reconstruction error value based on at least one of an average value of differences between vibration signals for each frequency band and reconstructed signals corresponding to the respective frequency bands, an average value of squares of the differences, and a square root value of the average value of squares of the differences.

Images