Analyzing device
Patent Information
- Application Number
- PCT/JP2025/008167
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-03-08
- Filing Date
- 2025-03-06
- Publication Date
- 2025-10-02
AI Technical Summary
Existing methods for determining abnormality in machinery by analyzing vibration data during operation lack accuracy in setting appropriate threshold values for anomaly detection, leading to potential inaccuracies in identifying machine abnormalities.
An analysis device that calculates signal strength for each frequency using vibration or sound signals, employs a trained model to determine abnormality scores, and sets thresholds based on these scores to accurately assess the severity of abnormalities in machinery.
Enables precise abnormality determination by objectively setting thresholds, reducing reliance on subjective methods and improving the accuracy of identifying machine conditions, with warnings for attention, maintenance, or replacement based on the severity of abnormalities.
Smart Images
Figure JP2025008167_02102025_PF_FP_ABST
Abstract
Description
analysis device
[0001] The present disclosure relates to an analysis device.
[0002] 2. Description of the Related Art Conventionally, a known method for inspecting machinery for abnormalities is to determine whether or not an abnormality exists in a machine by detecting a signal caused by abnormal vibrations during operation of the machine.
[0003] For example, Japanese Patent No. 6977364 (Patent Document 1) discloses an anomaly detection sensitivity setting device for a rotation mechanism. The anomaly detection sensitivity setting device includes a setting means for setting a plurality of threshold candidate values for determining whether vibration data during operation is normal or abnormal based on statistics of vibration data during normal operation, a verification means for detecting an anomaly in the vibration data during operation using the plurality of threshold candidate values and verifying the appropriateness of the anomaly detection sensitivity obtained for each threshold candidate value based on the results of the anomaly detection, and a determination means for determining a threshold value based on the appropriateness of the anomaly detection sensitivity.
[0004] Patent No. 6977364
[0005] In the technology disclosed in Patent Document 1, the appropriateness of the anomaly detection sensitivity obtained from multiple threshold candidate values is verified based on the anomaly detection results of vibration data (acceleration) during operation, and the threshold value is determined based on this appropriateness. However, it is believed that there is room for improvement in the method for determining the threshold value.
[0006] An object of one aspect of the present disclosure is to provide an analysis device that can accurately perform abnormality determination by simply and appropriately setting a threshold value used to determine whether or not an abnormality exists in a target device during operation.
[0007] An analysis device according to an embodiment includes a data calculation unit that calculates first data indicating signal strength for each frequency by analyzing signals based on vibrations or sounds of a normal target device in operation, and a score calculation unit that calculates a score indicating the degree of abnormality of the normal target device by inputting the first data into a trained model. The trained model has been trained so that, when data indicating signal strength for each frequency based on the device signal is input, it outputs a score indicating the degree of abnormality of the device as an estimation result. The data calculation unit further calculates second data indicating signal strength for each frequency by analyzing the signal of the target device in operation. The score calculation unit further calculates a score indicating the degree of abnormality of the target device by inputting the second data into the trained model. The analysis device further includes a threshold setting unit that sets at least one threshold based on multiple scores of the normal target device, and an abnormality determination unit that determines whether or not there is an abnormality in the target device based on the score of the target device and the at least one threshold.
[0008] Preferably, the at least one threshold value includes a first threshold value and a second threshold value greater than the first threshold value. When the score of the target device is less than the first threshold value, the abnormality determination unit determines that the target device is normal.
[0009] Preferably, the abnormality determination unit determines that the target equipment has a first abnormality when the score of the target equipment is greater than or equal to a first threshold and less than a second threshold, and determines that the target equipment has a second abnormality with a higher abnormality level than the first abnormality when the score of the target equipment is greater than or equal to the second threshold.
[0010] Preferably, when the predetermined score for the normal target device is equal to or greater than the first threshold, the threshold setting unit modifies the first threshold and the second threshold based on the predetermined score and the plurality of scores for the normal target device.
[0011] Preferably, the threshold setting unit sets the first threshold and the second threshold based on an average value and a standard deviation of the plurality of scores for normal target devices.
[0012] Preferably, the first threshold is a value obtained by adding a product of the standard deviation and a first coefficient to the average value, and the second threshold is a value obtained by adding a product of the standard deviation and a second coefficient greater than the first coefficient to the average value.
[0013] Preferably, when the predetermined score for the normal target device is equal to or greater than the first threshold, the threshold setting unit modifies the first threshold and the second threshold by increasing the first coefficient and the second coefficient.
[0014] According to the present disclosure, by simply and appropriately setting a threshold value used to determine whether or not an abnormality exists in a target device during operation, it becomes possible to accurately determine an abnormality.
[0015] FIG. 1 is a diagram for explaining an overview of a system. FIG. 2 is a block diagram showing an example of the overall configuration of an analysis system. FIG. 3 is a block diagram showing an example of the hardware configuration of an analysis device. FIG. 4 is a flowchart showing an example of a threshold setting process. FIG. 5 is a diagram showing an example of a data set of signal strengths of each frequency band. FIG. 6 is a diagram for explaining a threshold setting method. FIG. 7 is a flowchart showing an example of an abnormality determination process. FIG. 8 is a block diagram showing an example of the functional configuration of an analysis device. FIG. 9 is a diagram for explaining a threshold setting method according to a modified example.
[0016] Hereinafter, the present embodiment will be described with reference to the drawings. In the following description, the same components are denoted by the same reference numerals. The names and functions of these components are also the same. Therefore, detailed description thereof will not be repeated.
[0017] <System Configuration> Fig. 1 is a diagram illustrating an overview of system 1000. Referring to Fig. 1, system 1000 is a system for determining an abnormality in a maintenance target device such as a pump (hereinafter simply referred to as "target device") by analyzing a signal of vibrations generated during operation of the target device. In the following description, the target device is a pump, but the present invention is not limited to this, and system 1000 can be applied to any target device that generates vibrations (or sounds) during operation. For example, system 1000 can also be applied to determining an abnormality in a motor, a part that vibrates due to vibrations from a vibrating body, or the like.
[0018] The system 1000 includes an analysis system 100, a plurality of vibration sensors 30, a terminal device 40, a network 50, and a plurality of pumps 70. The analysis system 100 performs an analysis of the pumps 70. The analysis system 100 includes an analysis device 10 and a sensor unit 20. The sensor unit 20 is electrically connected to the plurality of vibration sensors 30. In the system 1000, two sensor units 20 are connected to the analysis device 10, but a configuration in which three or more sensor units 20, or one sensor unit 20, is connected to the analysis device 10 may also be used. Each sensor unit 20 may be electrically connected to one vibration sensor 30. Each sensor unit 20 may be electrically connected to a plurality of vibration sensors 30 attached to the plurality of pumps 70, respectively.
[0019] The vibration sensor 30 is attached to the pump 70 and acquires a detection signal (vibration signal) detected due to vibrations and sounds of the pump. The analysis device 10 performs vibration analysis of the pump 70 based on the vibration signal input from the vibration sensor 30 via the sensor unit 20. The analysis device 10 is configured to be able to communicate with the terminal device 40 via the network 50. The analysis device 10 transmits vibration analysis results and the like to the terminal device 40.
[0020] The analysis device 10 typically has a structure conforming to a general-purpose computer architecture, and performs various processes described below by executing pre-installed programs using a processor. The analysis device 10 is, for example, a laptop PC (Personal Computer). However, the analysis device 10 may be any other device (for example, a desktop PC or a tablet terminal device) that is capable of executing the functions and processes described below.
[0021] The network 50 includes various networks such as the Internet, etc. The network 50 may employ a wired communication system or other wireless communication systems such as a wireless LAN (local area network).
[0022] Terminal device 40 is, for example, a portable tablet terminal device. However, terminal device 40 is not limited to this and may be realized as a smartphone, a desktop personal computer (PC), or the like. Note that analysis system 100 according to the present embodiment is configured as a separate device in which analysis device 10 and sensor unit 20 are separate, but may also be configured as an integrated device in which analysis device 10 and sensor unit 20 are integrated.
[0023] 2 is a block diagram showing an example of the overall configuration of analysis system 100. Referring to FIG. 2, analysis system 100 includes an analysis device 10 and a sensor unit 20.
[0024] The vibration sensor 30 connected to the sensor unit 20 is a sensor capable of detecting vibration signals, and is configured, for example, by an acceleration sensor using an organic piezoelectric element. Note that the vibration sensor 30 may be any sensor capable of detecting vibration signals, and may be configured as an acceleration sensor of another type (for example, a servo type) or may be configured as various other sensors.
[0025] If the signal obtained by the vibration sensor 30 is a charge signal, a charge converter is provided between the vibration sensor 30 and the analysis system 100. In this case, the charge converter converts the charge signal from the vibration sensor 30 into a voltage signal and outputs it to the analysis system 100. Note that if the vibration sensor 30 has the function of converting a charge signal into a voltage signal, the charge converter is not necessary.
[0026] The sensor unit 20 converts the vibration signal acquired from the vibration sensor 30 (or the charge converter) into a signal that can be processed by the analysis device 10. Specifically, the sensor unit 20 includes a filter 21, an amplifier 22, and an A / D converter 23.
[0027] The filter 21 is an analog filter that removes noise components from the vibration signal output from the vibration sensor 30. The filter 21 is configured by a low-pass filter, a high-pass filter, or the like.
[0028] The amplifier 22 amplifies the analog signal output from the filter 21 by a predetermined factor and outputs the amplified signal to the A / D converter 23 .
[0029] The A / D converter 23 converts the signal input from the amplifier 22 from an analog signal to a digital signal at a predetermined sampling frequency, and outputs the digitally converted signal to the analysis device 10.
[0030] Fig. 3 is a block diagram showing an example of the hardware configuration of analysis device 10. Referring to Fig. 3, analysis device 10 includes processor 101, memory 103, display 105, input device 107, signal input interface (I / F) 109, and communication interface (I / F) 111. These components are connected to each other so as to be able to communicate data with each other.
[0031] The processor 101 is typically an arithmetic processing unit such as a CPU (Central Processing Unit), an MPU (Multi Processing Unit), etc. The processor 101 controls the operation of each unit of the analysis device 10 by reading and executing a program stored in the memory 103. More specifically, the processor 101 realizes each function of the analysis device 10 by executing the program.
[0032] The memory 103 is realized by a random access memory (RAM), a read-only memory (ROM), a flash memory, a hard disk, etc. The memory 103 stores programs executed by the processor 101, etc.
[0033] The display 105 is, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, etc. The display 105 may be configured integrally with the analysis device 10, or may be configured separately from the analysis device 10.
[0034] The input device 107 accepts operation inputs to the analysis device 10. The input device 107 is realized by, for example, a keyboard, buttons, a mouse, etc. The input device 107 may also be realized as a touch panel.
[0035] The signal input interface 109 mediates data transmission between the processor 101 and the sensor unit 20. The signal input interface (I / F) 109 accepts input of a vibration signal from the vibration sensor 30 via the sensor unit 20. Specifically, the signal input interface 109 accepts input of a digital signal from the A / D converter 23.
[0036] The communication interface 111 mediates data transmission between the processor 101 and the terminal device 40, etc. As a communication method, for example, a wireless communication method such as Bluetooth (registered trademark) or a wireless LAN (Local Area Network) is used. Note that a wired communication method such as a USB (Universal Serial Bus) may also be used as the communication method.
[0037] <Threshold Setting Process> An overview of the threshold setting process according to this embodiment will be described. In the threshold setting process, for example, the vibration state of the pump 70 at the beginning of operation is measured. Since the pump 70 at the beginning of operation is in a brand new state, the vibration state of the pump 70 in a normal state is measured as a reference. However, instead of the pump 70, a pump of the same type as the pump 70 in a normal state may be separately prepared, and the vibration state of that pump may be measured as a reference.
[0038] 4 is a flowchart showing an example of the threshold setting process. Typically, the following steps are realized by processor 101 of analysis device 10 executing a program stored in memory 103.
[0039] 4, the processor 101 acquires a vibration signal output from the vibration sensor 30 via the sensor unit 20 (step S10). Specifically, the processor 101 acquires a vibration signal (a vibration signal indicating the vibration state of the pump 70) corresponding to a target device (e.g., the pump 70) in a normal state from the vibration sensor 30. In the present embodiment, the target device in a normal state is also referred to as a "normal target device."
[0040] The processor 101 performs octave analysis on the vibration signals accumulated for a predetermined time (e.g., several tens to several hundreds of milliseconds) (step S12). In this embodiment, 1 / 3 octave analysis is used. Therefore, each vibration signal is separated into 48 bands, for example, from 0.4 Hz to 20 kHz, using a 1 / 3 band-pass filter, and the signal strength (vibration strength) is averaged for each band (i.e., frequency band). In the following description, the signal strength averaged over a frequency band is also simply referred to as the "signal strength of the frequency band."
[0041] The processor 101 stores, for each frequency band, the signal strength of the frequency band corresponding to the pump 70 in a normal state as signal strength data R in the memory 103 (step S14).
[0042] FIG. 5 is a diagram showing an example of a data set of signal strength for each frequency band. Referring to FIG. 5, data set 310 includes signal strength L for each frequency band f1 to fn (where n is a natural number, n<m) for each time T1 to Tm (where m is a natural number). If each vibration signal is separated into 48 bands, n=48. For example, data set 310 includes signal strengths L1_1 to L1_n for each frequency band f1 to fn at time T1, and signal strengths Lm_1 to Lm_n for each frequency band f1 to fn at time Tm. The period from time T1 to Tx corresponds to the initial period after the start of operation. In the following description, "signal strength data R for time T" is defined as data including "signal strength L for each frequency band f1 to fn at time T."
[0043] For example, if measurements are taken twice a day, time T1 corresponds to a certain time in the morning of day 1, time T2 corresponds to a certain time in the afternoon of day 1, time T3 corresponds to a certain time in the morning of day 2, and time T4 corresponds to a certain time in the afternoon of day 2. The same applies to the other times T5 to Tm.
[0044] 4, the processor 101 calculates a score for the pump 70 in a normal state by inputting the signal strength data R for each of times T1 to Tx into a trained model G1 for estimating a score indicating the degree of abnormality of the pump 70 (step S16). For example, the processor 101 calculates the score corresponding to time T1 by inputting the signal strength data R for time T1 (i.e., the signal strengths L1_1 to L1_n for each of frequency bands f1 to fn) into the trained model G1. Similarly, the processor 101 calculates the scores corresponding to each of times T2 to Tx.
[0045] The trained model G1 is a model obtained by performing a known learning process (e.g., supervised learning) using a training dataset. Typically, the training dataset includes signal strength data of an apparatus (e.g., pump 70) measured at a certain time and training data including a result indicating the degree of abnormality of the apparatus (e.g., normal, first abnormality with a low level of abnormality, second abnormality with a high level of abnormality, etc.). The trained model G1 has undergone a learning process using many training datasets so that, when signal strength data of an apparatus is input, a score indicating the degree of abnormality of the apparatus is output as an estimated result. More specifically, the trained model G1 is a model trained to output a score when signal strength data of an apparatus is input, such that the score becomes higher as the degree of abnormality of the apparatus becomes greater (higher).
[0046] The processor 101 sets at least one threshold based on a plurality of scores of the normal target device (for example, the pump 70 in a normal state) (step S18).
[0047] 6 is a diagram illustrating a threshold setting method. Referring to FIG. 6, an information table 501 includes a plurality of scores for normal target devices. An information table 502 includes a plurality of thresholds calculated (set) based on the plurality of scores.
[0048] As shown in information table 501, the score for No. 1 is "7.2." This indicates that when signal strength data of a normal target device at time T1 is input to learned model G1, the score output from learned model G1 is "7.2." Similarly, for example, a score of "6.9" for No. 3 is obtained based on signal strength data of a normal target device at time T3, and a score of "6.4" for No. 30 is obtained based on signal strength data of a normal target device at time T30.
[0049] The processor 101 calculates, for example, the average value M and standard deviation σ of 30 scores No. 1 to No. 30. The processor 101 calculates thresholds Th1, Th2, and Th3 for determining the abnormality level of the target device based on the average value M and the standard deviation σ. Typically, the threshold value Th1 is a threshold for determining whether or not "attention" should be directed to the target device. The threshold value Th2 is a threshold for determining whether or not "maintenance" should be performed on the target device. The threshold value Th3 is a threshold for determining whether or not the target device is in a "dangerous state" that requires replacement or the like.
[0050] The thresholds Th1 to Th3 are values obtained by adding the product of the standard deviation σ and a predetermined coefficient to the average value M. For example, the threshold Th1 is the sum of the average value M and three times the standard deviation σ (i.e., Th1 = M + 3σ). The threshold Th2 is the sum of the average value M and 3.5 times the standard deviation σ (i.e., Th2 = M + 3.5σ). The threshold Th3 is the sum of the average value M and four times the standard deviation σ (i.e., Th3 = M + 4σ).
[0051] 7 is a flowchart showing an example of the abnormality determination process. Typically, the following steps are realized by the processor 101 of the analysis device 10 executing a program stored in the memory 103.
[0052] 7, the processor 101 acquires a vibration signal output from the vibration sensor 30 via the sensor unit 20 (step S50). Specifically, the processor 101 acquires, from the vibration sensor 30, a vibration signal corresponding to a target device (e.g., the pump 70) during a normal period (e.g., an operating period after the end of an initial period after the start of operation).
[0053] The processor 101 performs octave analysis on the vibration signals accumulated for a predetermined period of time (step S52). The processor 101 stores, in the memory 103, the signal strength of each frequency band corresponding to the pump 70 (step S54). Specifically, the signal strength of each frequency band in the pump 70 at a certain time Ts is stored in the form of a data set 310. Here, the series of signal strengths Ls_1 to Ls_n of frequency bands f1 to fn in the pump 70 at time Ts is also referred to as signal strength data Ps. Note that the period from time T1 to Tx corresponds to the initial period of operation, and therefore the normal period corresponds to the period from time Tx+1 to time Tm. Therefore, time Ts is any of the periods from time Tx+1 to time Tm.
[0054] The processor 101 inputs the signal strength data Ps for the time Ts into the trained model G1 to calculate the score Ks of the pump 70 (step S56). The processor 101 determines whether or not there is an abnormality in the pump 70 based on the score Ks and the thresholds Th1 to Th3 (step S58).
[0055] If the score Ks is less than the threshold value Th1, the processor 101 determines that the pump 70 is normal. If the score Ks is equal to or greater than the threshold value Th1 and less than the threshold value Th2, the processor 101 determines that the pump 70 has an abnormality E1 with a low abnormality level. If the score Ks is equal to or greater than the threshold value Th2 and less than the threshold value Th3, the processor 101 determines that the pump 70 has an abnormality E2 with a slightly high abnormality level. If the score Ks is equal to or greater than the threshold value Th3, the processor 101 determines that the pump 70 has an abnormality E3 with a high abnormality level.
[0056] The processor 101 outputs the abnormality determination result for the pump 70 (step S60). Specifically, the processor 101 displays on the display 105 a result indicating whether the pump 70 is normal or in one of the abnormal states E1 to E3. For example, if the pump 70 is in abnormal state E1, the processor 101 displays a warning (e.g., "Caution") indicating that the pump 70 requires attention. If the pump 70 is in abnormal state E2, the processor 101 displays a warning (e.g., "Maintenance Recommended") indicating that maintenance of the pump 70 is recommended. If the pump 70 is in abnormal state E3, the processor 101 displays a warning (e.g., "Danger") urging replacement of the pump 70. Note that these warnings may be output as audio via a speaker.
[0057] <Functional Configuration> Fig. 8 is a block diagram showing an example of the functional configuration of the analysis device 10. Referring to Fig. 8, the analysis device 10 mainly includes a signal input unit 202, a data calculation unit 204, a learned model storage unit 206, a score calculation unit 208, a threshold setting unit 210, an abnormality determination unit 212, and an output control unit 214. Each of these functions is realized, for example, by the processor 101 of the analysis device 10 executing a program stored in the memory 103. Note that some or all of these functions may be configured to be realized by hardware.
[0058] The signal input unit 202 receives an input of a signal (i.e., a vibration signal) based on the vibration of the pump 70, which is detected by the vibration sensor 30 attached to the pump 70 during operation. Specifically, the signal input unit 202 receives the vibration signal detected by the vibration sensor 30 via the sensor unit 20.
[0059] Data calculation unit 204 calculates data indicating signal strength for each frequency by analyzing the vibration signal received by signal input unit 202. In one aspect, data calculation unit 204 calculates signal strength data R by analyzing the vibration signal of a normal target device in operation. In another aspect, data calculation unit 204 calculates signal strength data Ps by analyzing the vibration signal of a target device in operation during operation.
[0060] For example, the data calculation unit 204 calculates the signal strength of each frequency band (for example, the signal strength L of each frequency band f1 to fm) by performing octave analysis (for example, 1 / 3 octave analysis) on the vibration signal corresponding to the pump 70. Note that the data calculation unit 204 may be configured to calculate the signal strength of each frequency band by fast Fourier transform (FFT).
[0061] The trained model storage unit 206 stores the trained model G1. For example, the trained model G1 is generated by pre-machine learning, through supervised learning, of the relationship between data indicating the signal strength for each frequency based on the vibration signal of an equipment (e.g., the pump 70) and a score indicating the degree of abnormality of the equipment. However, the learning algorithm used to generate the trained model G1 is not limited to supervised learning. For example, reinforcement learning, unsupervised learning, semi-supervised learning, deep learning, etc. may be applied as the learning algorithm. Furthermore, other known methods (e.g., genetic programming, functional logic programming, support vector machines) may be applied as the learning algorithm.
[0062] The score calculation unit 208 reads out the learned model G1 stored in the learned model storage unit 206 and calculates the score of the normal target device by inputting the signal strength data R to the learned model G1. The score calculation unit 208 also calculates the score of the target device during operation by inputting the signal strength data Ps to the learned model G1.
[0063] The threshold setting unit 210 sets at least one threshold based on multiple scores for normal target devices. Specifically, the threshold setting unit 210 sets thresholds Th1 to Th3 based on the average value M and standard deviation σ of the multiple scores for normal target devices. For example, the threshold setting unit 210 sets thresholds Th1 to Th3 using the threshold setting method described in FIG. 6. As another example, the threshold setting unit 210 may be configured to assume a normal distribution for the scores for normal target devices and set the upper limit of the 95% confidence interval as the threshold. Note that the threshold setting unit 210 may be configured to set the threshold at any timing, such as when a threshold setting instruction is received from a user (e.g., an instruction via the input device 107).
[0064] The abnormality determination unit 212 determines whether or not the target device has an abnormality based on the score of the target device during operation and at least one threshold value (e.g., threshold values Th1 to Th3). In one aspect, the abnormality determination unit 212 determines that the target device is normal when the score of the target device is less than threshold value Th1. In another aspect, the abnormality determination unit 212 determines that the target device has an abnormality E1 when the score of the target device is equal to or greater than threshold value Th1 and less than threshold value Th2, and determines that the target device has an abnormality E2 (or an abnormality E3) that is a higher level of abnormality than abnormality E1 when the score of the target device is equal to or greater than threshold value Th2.
[0065] The output control unit 214 outputs the abnormality determination result of the target device. Typically, the output control unit 214 displays the abnormality determination result on the display 105.
[0066] The threshold setting unit 210 may be configured to appropriately modify each of the thresholds Th1 to Th3. For example, if the target device is normal but the predetermined score calculated for that target device (i.e., normal target device) is equal to or greater than the currently set threshold Th1, it is possible that the settings of each of the thresholds Th1 to Th3 are too strict.
[0067] Therefore, if the predetermined score for a normal target device is equal to or greater than the threshold Th1, the threshold setting unit 210 may modify the current thresholds Th1 to Th3 based on the predetermined score and multiple scores (e.g., scores No. 1 to No. 30 in FIG. 6 ) used to set the current threshold Th1. The threshold setting unit 210 calculates the average value M1 and standard deviation σ1 of the predetermined score and the 31 scores No. 1 to No. 30, and calculates new thresholds Th1, Th2, and Th3 based on the average value M1 and standard deviation σ1. The new threshold Th1 is the sum of the average value M1 and three times the standard deviation σ1. The new threshold Th2 is the sum of the average value M1 and 3.5 times the standard deviation σ1. The new threshold Th3 is the sum of the average value M1 and four times the standard deviation σ1.
[0068] Note that when the predetermined score for a normal target device is equal to or greater than the threshold value Th1, the threshold value setting unit 210 does not necessarily have to modify the current threshold values Th1 to Th3. For example, the threshold value setting unit 210 may be configured to modify the current threshold values Th1 to Th3 at any timing as needed, for example, when the predetermined score is equal to or greater than the threshold value Th1 multiple times in succession, when a command to modify the threshold value is received from the user, or the like.
[0069] In another aspect, when the predetermined score for a normal target device is equal to or greater than threshold Th1, threshold setting unit 210 may modify thresholds Th1 to Th3 by increasing the coefficient of standard deviation σ. For example, threshold setting unit 210 may set the sum of mean value M and 3.5 times the standard deviation σ as new threshold Th1, set the sum of mean value M and 4 times the standard deviation σ as new threshold Th2, and set the sum of mean value M and 4.5 times the standard deviation σ as new threshold Th3.
[0070] <Modifications> In the above, a configuration has been described in which a vibration sensor attached to a target device (e.g., the pump 70) detects a vibration signal based on the vibration of the target device, and signal strength data for each frequency is calculated based on the vibration signal. However, depending on the device, it may be difficult to directly attach a vibration sensor. In such cases, a configuration may be used in which a sound sensor (e.g., a microphone) that generates a sound signal based on air vibrations detects a sound signal based on the sound of the target device, and signal strength data for each frequency is calculated based on the sound signal.
[0071] Specifically, the signal input unit 202 receives an input of a sound signal detected by a sound sensor provided near the pump 70 during operation. The data calculation unit 204 calculates data indicating the signal strength for each frequency by analyzing (e.g., octave analysis) the sound signal received by the signal input unit 202. For example, the signal strength data Ra is calculated by analyzing a sound signal of a normal target device during operation. In another aspect, the data calculation unit 204 calculates the signal strength data Psa by analyzing a vibration signal of the target device during operation. Typically, these signal strength data are stored in a format similar to the data set 310 shown in FIG. 5 .
[0072] The trained model storage unit 206 stores a trained model G2 for sound signals. The trained model G2 is trained in advance by supervised learning to determine the relationship between data indicating the signal strength for each frequency based on the sound signal of an appliance (e.g., the pump 70) and a score indicating the degree of abnormality of the appliance.
[0073] The score calculation unit 208 calculates the score of the normal target device by inputting the signal strength data Ra to the trained model G2, and calculates the score of the target device during operation by inputting the signal strength data Psa to the trained model G2. The threshold setting unit 210 sets at least one threshold based on the multiple scores of the normal target device.
[0074] 9 is a diagram illustrating a threshold setting method according to a modified example. Referring to FIG. 9, an information table 511 includes a plurality of scores for normal target devices. An information table 512 includes a plurality of thresholds calculated based on the plurality of scores.
[0075] As shown in information table 511, the score for No. 1 is "13.8." This indicates that when the signal strength data of the normal target device at time T1 is input to trained model G2, a score of "13.8" is output from trained model G2. Similarly, for example, a score of "7.9" is obtained for No. 3 and a score of "10.0" is obtained for No. 30.
[0076] The threshold setting unit 210 calculates, for example, the average value Ma and standard deviation σa of 30 scores No. 1 to No. 30. The threshold setting unit 210 calculates thresholds Th1a, Th2a, and Th3a for determining the abnormality level of the target device based on the average value Ma and the standard deviation σa. The calculation method for each threshold is the same as the method described in FIG. 6. Specifically, "Th1a = Ma + 3σa", "Th2a = Ma + 3.5σa", and "Th3a = Ma + 4σa". The determination method used by the abnormality determination unit 212 is the same as above. In the above, the thresholds Th1 to Th3 can be replaced with thresholds Th1a to Th3a, respectively.
[0077] <Advantages> According to this embodiment, the threshold for determining an abnormality can be appropriately set based on the score of the target device when it is normal, thereby enabling more accurate abnormality determination. Furthermore, because the threshold is set based on an objective index, namely the score of the target device when it is normal, the abnormality determination is not dependent on each individual, and an objective abnormality determination is possible. Furthermore, because the process for calculating the threshold from the score is simple, calculations for abnormality determination can be performed using a simple computer.
[0078] Other Embodiments (1) In the above-described embodiment, a configuration has been described in which signal intensity data obtained by octave analysis of a signal from a target device (e.g., a vibration signal or a sound signal) is input to a trained model. However, this is not limited to this. For example, a configuration may be adopted in which a signal from a target device is analyzed and converted into a mel spectrogram in image format, and the mel spectrogram is input to a trained model as signal intensity data for each frequency. In this case, the trained model is trained to output a score for a device when signal intensity data (mel spectrogram) of the device is input.
[0079] (2) In the above-described embodiment, a program may be provided that causes a computer to function and execute the control described in the above flowchart. Such a program may be recorded on a non-transitory computer-readable recording medium such as a flexible disk, secondary storage device, main storage device, or memory card attached to the computer and provided as a program product. Alternatively, the program may be recorded on a recording medium such as a hard disk built into the computer and provided. The program may also be provided by downloading via a network.
[0080] (3) The configurations exemplified as the above-described embodiments are merely examples of the configurations of the present invention, and may be combined with other known technologies, or may be modified, such as by omitting some parts, without departing from the spirit of the present invention. Furthermore, the above-described embodiments may be implemented by appropriately adopting the processes and configurations described in other embodiments.
[0081] The embodiments disclosed herein should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the above description, and is intended to include all modifications within the meaning and scope of the claims.
[0082] 10 Analysis device, 20 Sensor unit, 21 Filter, 22 Amplifier, 23 A / D converter, 30 Vibration sensor, 40 Terminal device, 50 Network, 70 Pump, 100 Analysis system, 101 Processor, 103 Memory, 105 Display, 107 Input device, 109 Signal input interface, 111 Communication interface, 202 Signal input unit, 204 Data calculation unit, 206 Learned model storage unit, 208 Score calculation unit, 210 Threshold setting unit, 212 Abnormality determination unit, 214 Output control unit.
Claims
1. An analysis device comprising: a data calculation unit that calculates first data indicating signal strength for each frequency by analyzing signals based on vibrations or sounds of normal target equipment in operation; and a score calculation unit that calculates a score indicating the degree of abnormality of the normal target equipment by inputting the first data into a trained model, wherein the trained model has undergone a training process such that when data indicating signal strength for each frequency based on the signals of the equipment is input, a score indicating the degree of abnormality of the equipment is output as an estimated result; the data calculation unit further calculates second data indicating signal strength for each frequency by analyzing the signals of the target equipment in operation; and the score calculation unit further calculates a score indicating the degree of abnormality of the target equipment by inputting the second data into the trained model; a threshold setting unit that sets at least one threshold based on the multiple scores of the normal target equipment; and an abnormality determination unit that determines whether or not there is an abnormality in the target equipment based on the scores of the target equipment and the at least one threshold.
2. The analysis device of claim 1, wherein the at least one threshold value includes a first threshold value and a second threshold value greater than the first threshold value, and the abnormality determination unit determines that the target device is normal if the score of the target device is less than the first threshold value.
3. The analysis device according to claim 2, wherein the abnormality determination unit determines that the target device has a first abnormality when the score of the target device is equal to or greater than the first threshold value and less than the second threshold value, and determines that the target device has a second abnormality with a higher abnormality level than the first abnormality when the score of the target device is equal to or greater than the second threshold value.
4. An analysis device as described in claim 2 or 3, wherein when the predetermined score for the normal target device is equal to or greater than the first threshold, the threshold setting unit modifies the first threshold and the second threshold based on the predetermined score and the multiple scores for the normal target device.
5. The analysis device according to claim 2 or 3, wherein the threshold setting unit sets the first threshold and the second threshold based on the average value and standard deviation of the scores of the normal target devices.
6. The analysis device described in claim 5, wherein the first threshold is a value obtained by adding the product of the standard deviation and a first coefficient to the average value, and the second threshold is a value obtained by adding the product of the standard deviation and a second coefficient greater than the first coefficient to the average value.
7. The analysis device of claim 6, wherein when the predetermined score for the normal target device is equal to or greater than the first threshold, the threshold setting unit modifies the first threshold and the second threshold by increasing the first coefficient and the second coefficient.