WO2020039565A1 - Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante - Google Patents

Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante Download PDF

Info

Publication number
WO2020039565A1
WO2020039565A1 PCT/JP2018/031217 JP2018031217W WO2020039565A1 WO 2020039565 A1 WO2020039565 A1 WO 2020039565A1 JP 2018031217 W JP2018031217 W JP 2018031217W WO 2020039565 A1 WO2020039565 A1 WO 2020039565A1
Authority
WO
WIPO (PCT)
Prior art keywords
bearing
machine
detection signal
evaluation value
frequency
Prior art date
Application number
PCT/JP2018/031217
Other languages
English (en)
Japanese (ja)
Inventor
佐藤　哲哉
顕 山下
彰教 杉垣
増田　新
Original Assignee
村田機械株式会社
国立大学法人京都工芸繊維大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 村田機械株式会社, 国立大学法人京都工芸繊維大学 filed Critical 村田機械株式会社
Priority to PCT/JP2018/031217 priority Critical patent/WO2020039565A1/fr
Publication of WO2020039565A1 publication Critical patent/WO2020039565A1/fr

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M13/00—Testing of machine parts
    • G01M13/04—Bearings

Definitions

  • the present invention relates to a method for diagnosing an abnormality of a bearing used in a rotating machine.
  • a motor is used as a drive source of a movable part of various logistics equipment such as a stacker crane and a ceiling carrier, and a machine tool such as a lathe and a laser beam machine, and a drive shaft of the motor is supported by a bearing.
  • Industrial equipment such as logistics equipment and machine tools operate according to the logistics plan and production plan specified by the user.However, if a bearing failure that the user does not expect occurs, the logistics plan and the production plan will be greatly disrupted. become severely damaged. For this reason, it is desirable to detect (predict) the occurrence of a failure in the bearing in advance, prepare a replacement bearing in advance, and replace the bearing at a pre-planned timing immediately before the occurrence of the failure. By doing so, it is less likely that the user's distribution plan and production plan will be out of order.
  • Patent Document 1 discloses an abnormality in a bearing by performing envelope processing and frequency analysis on a waveform of vibration generated from a bearing, and comparing a calculated value (normalized) obtained by dividing a peak value of an envelope spectrum by an overall value (normalized) with a reference value. Disclosed is a method for diagnosing an abnormality of a bearing that determines that there is a failure.
  • Patent Literature 1 a person observes a test result to determine a reference value for abnormality determination. It is necessary to select an appropriate reference value for the bearing (for example, whether it is a stacker crane or a lathe) according to the target, location, operation situation, and individual difference of the bearing. However, it is not realistic to perform an operation test every time using the same or similar machine to calculate the reference value. Therefore, an object of the present invention is to make a device such as a computer automatically determine a transition between a normal state and an abnormal state of a bearing in a method of diagnosing a bearing abnormality without depending on human experience and intuition.
  • a device such as a computer automatically determine a transition between a normal state and an abnormal state of a bearing in a method of diagnosing a bearing abnormality without depending on human experience and intuition.
  • a method is a method for diagnosing abnormality of a bearing used in a rotating machine, and includes the following steps. ⁇ Discrete Fourier transform processing step to obtain frequency spectrum by executing discrete Fourier transform processing on detection signal obtained from rotating machine ⁇ Determine learned discriminant by applying machine learning to detection signal Machine learning step ⁇ Evaluation value calculation step of calculating the evaluation value on the time axis by inputting the frequency spectrum to the learned discriminant ⁇ The bearing is in a normal state based on the time when the evaluation value drops sharply on the time axis Abnormality determination step for determining that the state has transitioned to the abnormal state Note that a sharp decrease in the evaluation value on the time axis means that the amount of decrease in the evaluation value per unit time is large.
  • the normal state is, for example, stages 1 and 2 in a 5-stage model
  • the abnormal state is, for example, stages 3 to 5.
  • the five-stage model is a model of the physical phenomena of bearing anomalies. When damage occurs to the bearing, small damage repeats "occurrence” ⁇ "growth” ⁇ “dull” regardless of the cause, and eventually ends. Destruction will occur.
  • the five-stage model consists of a stage 1 where the surface irregularities can be "made", a stage 2 in a stable state, a stage 3 that becomes dull after the first damage occurs, and a stage 3 that becomes dull after new damage starts from the previous damage Is repeated and stage 5 of ultimate destruction.
  • the discriminant can be accurately determined using the detection signal obtained from the rotating machine (bearing) as learning data, the difference between the normal and abnormal states can be determined regardless of human experience and intuition. A discriminant that can be clearly discriminated can be obtained. As a result, the transition between the normal and abnormal states of the bearing is automatically determined with high accuracy. This enables predictive maintenance and planned maintenance.
  • the method may further include a notification step of notifying that the state has transitioned to the abnormal state.
  • the notification means displaying information or notifying the state of the machine by voice or light. According to this method, the notification step allows the operator to quickly know that the bearing has transitioned to the abnormal state.
  • the notification step may notify information that prompts the user to prepare a replacement bearing.
  • the operator is prompted to prepare the replacement bearing as the next operation.
  • the machine learning algorithm may use only normal data as data necessary for learning.
  • One Class SVM is used as a machine learning algorithm. With this method, machine learning can be realized using only normal data.
  • the machine learning algorithm may use both normal data and abnormal data as data necessary for learning.
  • Boosting one of methods for creating a more accurate classifier by combining weak learners
  • SVM Small Vector Machine
  • the machine learning step may separate the detection signal into a starting (running in) stage and a stable state, and use the detection signal of the stable state as data for machine learning.
  • the starting stage is a stage 1 of a five-stage model
  • the stable state stage is a stage 2 of a five-stage model.
  • the detection signal obtained from the rotating machine may be at least one selected from the group of vibration amplitude, vibration acceleration, F / B (feedback) torque of the motor of the rotating machine, sound, and the like. With this method, it is possible to diagnose whether the rotating machine is normal or abnormal using at least one of the plurality of detection signals.
  • the sensor that detects the detection signal from the rotating machine may be retrofittable to the rotating machine in operation.
  • this method for example, by retrofitting a sensor to a machine whose operation period has elapsed, it is possible to predict an abnormality in a bearing of the machine.
  • the abnormality determination step may include the following steps. ⁇ Based on the state information of the rotating machine, a detection signal determination step of determining a detection signal used to determine that the transition from the normal state to the abnormal state, the state information of the rotating machine, for example, the rotation direction of the bearing, This is the rotation speed of the bearing. In this method, since the detection signal is determined based on the state information of the rotating machine, the most appropriate detection signal of the plurality of sensors can be used.
  • the method may further include the following steps.
  • the Propagation stage is Stage 4 of the 5-stage model
  • the Damage growth stage is Stage 5 of the 5-stage model. According to this method, it can be seen that the bearing has transitioned from the defect propagation stage to the damage growth stage.
  • the method may further include the following steps.
  • a frequency spectrogram and / or an envelope frequency spectrogram are obtained based on the frequency spectrum and / or the envelope frequency spectrum.
  • Frequency spectrogram obtaining step ⁇ The frequency spectrogram and / or the envelope frequency spectrogram indicate that the frequency corresponding to the frequency resulting from the failure of the bearing has appeared.
  • Confirmation step confirms in advance that it is possible to transition determination from the normal state to an abnormal state should be noted that signals obtained from the accelerated test machine, for example, a vibration acceleration. According to this method, it is possible to confirm in advance in an acceleration test that it is possible to determine the transition of the bearing from a normal state to an abnormal state during normal operation.
  • the rotating machine may be any one of a physical distribution machine, a machine tool, a textile machine, and a rehabilitation device.
  • the distribution transport machine includes a stacker crane, a conveyor, an overhead transport vehicle, an AGV, an RGV, and an autonomous traveling robot.
  • Machine tools include lathes, press brakes, punch presses, laser machines, deburring machines, loaders, and unloaders.
  • Textile machines include automatic winders, spinning machines, draw false twisters, and filament winders. According to this method, the above effects can be obtained in a physical distribution machine, a machine tool, a textile machine, and a rehabilitation device.
  • a transition between a normal state and an abnormal state of a bearing can be automatically determined by a computer without depending on human experience and intuition.
  • FIG. 1 is a schematic configuration diagram of a bearing abnormality diagnosis device according to a first embodiment.
  • FIG. 3 is a block diagram showing an example of a configuration of a diagnostic unit.
  • FIG. 7 is a block diagram showing another example of the configuration of the diagnostic unit.
  • 5 is a graph showing a 5-stage model. Time change of measured value of vibration acceleration and frequency spectrogram of vibration acceleration. 9 is a flowchart of a machine learning operation. 7 is a graph showing a time change of an evaluation value based on a change in hyper parameter of One ⁇ Class ⁇ SVM.
  • 5 is a graph showing a time change of a frequency spectrogram, a vibration acceleration, and an evaluation value of machine learning.
  • FIG. 6 is a graph showing a change in an evaluation value calculated by a learned discriminant using One ⁇ Class ⁇ SVM for vibration acceleration.
  • 7 is a graph showing a change in an evaluation value calculated by a learned discriminant by the SVM for the vibration acceleration.
  • 9 is a graph showing a change in an evaluation value calculated by a learned discriminant by Boosting for a vibration acceleration.
  • 9 is a flowchart of an evaluation value calculation and determination operation by the analysis unit in the stage 1 or 2.
  • 7 is a flowchart of evaluation value calculation and determination operation by an analysis unit in stage 4.
  • FIG. 2 is a schematic configuration diagram when an abnormality diagnosis device is attached to an acceleration tester. Vibration acceleration frequency spectrogram and envelope frequency spectrogram.
  • FIG. 1 is a schematic configuration diagram of a bearing abnormality diagnosis device according to the first embodiment.
  • FIG. 2A is a block diagram illustrating an example of a configuration of a diagnosis unit.
  • FIG. 2B is a block diagram illustrating another example of the configuration of the diagnosis unit.
  • the device 1 diagnoses, for example, an abnormality of a bearing 7 that supports the main shaft 5 used in the machine tool 3. Therefore, the abnormality of the bearing 7 can be diagnosed accurately and inexpensively at an early timing when the machine tool 3 is not damaged.
  • the main shaft 5 is a rotary shaft device that grips and rotates a tool or a work.
  • the machine tool 3 further has a motor 9 and a servo amplifier (not shown).
  • the main shaft 5 extends from the motor 9 and is rotatably supported by bearings 7.
  • the bearing 7 is a ball bearing, and includes a plurality of rolling elements rotatably arranged between the inner ring 7a fitted to the main shaft 5, the outer ring 7b fitted to the housing 13, and the inner ring 7a and the outer ring 7b. It has a moving body 7c and a retainer (not shown) for rotatably holding the rolling body 7c.
  • the type of the bearing is not limited, and a roller bearing may be used.
  • FIG. 3 is a graph showing a five-stage model.
  • FIG. 3 shows a change in the degree of damage (Dynamic impact of wear severity) and a change in the surface topology (Evolution of surface topology) over time.
  • the five-stage model includes a stage 1 (starting, running in) in which irregularities such as minute burrs due to surface processing are polished, and a stage 2 (stable, steady state) in which mechanical contact is smooth and operation is stable.
  • stage 3 defect initiation
  • stage 4 in which dulling is repeated after new damage has occurred from the previous damage
  • Defect propagation Defect propagation
  • Stage 5 Defect propagation
  • the stages 1 and 2 are in a state where the bearings are normal
  • the stages 3 to 5 are a state where the bearings are abnormal.
  • FIG. 4 is an example of a time change of a measured value of the vibration acceleration and a frequency spectrogram of the vibration acceleration.
  • the time change of the frequency spectrum of the vibration acceleration in FIG. 4 was obtained by performing a discrete Fourier transform for each time on the measured value of the vibration acceleration in FIG.
  • the numbers (1) to (5) above the frequency spectrogram of the vibration acceleration indicate each stage of the 5-stage model.
  • the amplitudes of all frequency components of the vibration acceleration appear over a remarkably wide band.
  • the appearance of all the frequency components of the vibration acceleration over a wide band is high.
  • the device 1 includes a diagnosis unit 15, a vibration sensor 17, a microphone 19, and a feedback current input unit 21.
  • the vibration sensor 17, the microphone 19, and the feedback current input unit 21 are sensors for detecting the state of the bearing 7 and the axis of the rotating machine, and output a detection signal to the diagnosis unit 15. That is, the detection signal obtained from the machine tool 3 is one of the vibration acceleration, the F / B torque of the motor 9, the sound, or a combination thereof.
  • the vibration sensor 17 is attached to the bearing 7 via the housing 13, and detects vibration of the bearing 7. Specifically, the vibration sensor 17 is provided on the surface of the outer race 7b via the housing 13.
  • the vibration sensor 17 for example, an acceleration sensor, an AE (Acoustic Emission) sensor, an ultrasonic sensor, a shock pulse sensor, or the like is used. Further, by detecting acceleration, speed, strain, stress, displacement, minute cracks, and the like, a device capable of equivalently detecting vibration and converting it into an electric signal can be used as appropriate.
  • the microphone 19 is arranged at a position where the operation sound near the bearing 7 is picked up.
  • the feedback current input unit 21 measures a feedback current output from a servo amplifier (not shown) that supplies power for driving the motor 9. Since the feedback current supplied to the motor 9 corresponds to the feedback torque generated from the motor 9 (hereinafter, referred to as F / B torque), it can be said that the feedback current input unit 21 measures the F / B torque. .
  • the F / B torque is data that can be obtained without adding a sensor for measuring the torque to the axis of the rotating machine or the like, and thus can be applied to, for example, a case where the sensor cannot be attached to the axis of the rotating machine. .
  • the feedback current input unit 21 may output the command value to the servo amplifier to the diagnostic unit 15 as a measured value of the feedback current.
  • the vibration sensor 17, the microphone 19, and the feedback current input unit 21 can be retrofitted to the machine tool 3 in operation (existing). Therefore, for example, an abnormality of the bearing 7 of the machine tool 3 can be predicted by retrofitting the sensor to the existing machine tool 3.
  • the diagnosis unit 15 includes a processor (for example, CPU), a storage device (for example, ROM, RAM, HDD, SSD, etc.), various interfaces (for example, A / D converter, D / A converter, and communication). Interface, etc.).
  • the diagnostic unit 15 performs various control operations by executing a program stored in a storage unit (corresponding to a part or all of the storage area of the storage device).
  • the diagnosis unit 15 may be configured by a single processor, or may be configured by a plurality of independent processors for each control.
  • each element of the diagnosis unit 15 may be realized as a program that can be executed by a computer system constituting the control unit. In addition, a part of the function of each element of the control unit may be configured by a custom IC.
  • the diagnostic unit 15 is connected to a sensor for detecting a state of an object of the machine tool 3, a sensor and a switch for detecting a state of each device, and an information input device.
  • the diagnosis unit 15 has, specifically, a data conversion unit 25 and an analysis unit 29.
  • the analysis unit 29 has a machine learning unit 41 and a determination unit 43.
  • the data converter 25 converts a detection signal from the sensor into a frequency spectrum.
  • the analysis unit 29 calculates an evaluation value by machine learning based on the frequency spectrum, and diagnoses an abnormality of the bearing 7 from the evaluation value.
  • the function of the diagnosis unit 15 may be realized by one computer system as shown in FIG. 2A, or may be realized by a plurality of computer systems.
  • the function of the diagnosis unit 15 is realized by a plurality of computer systems, for example, as illustrated in FIG. 2B, the data conversion unit 25 and the machine learning unit 41 are realized by one computer system (first computer system 15a).
  • the data conversion unit 25 and the determination unit 43 may be realized by another computer system (the second computer system 15b).
  • the diagnostic unit 15 will be described in more detail with reference to FIGS. 2A and 2B.
  • the data conversion unit 25 has an AD conversion unit 31.
  • the AD converter 31 converts a detection signal, which is an analog signal, into a digital signal.
  • the data conversion unit 25 has a feature amount extraction unit 35.
  • the feature amount extraction unit 35 performs a discrete Fourier transform process on the input digital signal or envelope. Specifically, the feature amount extraction unit 35 performs a discrete Fourier transform process on the detection signal obtained from the machine tool 3 to convert the detection signal into a frequency and an intensity of a signal vibrating at the frequency.
  • the data conversion unit 25 may include an envelope processing unit 33 between the AD conversion unit 31 and the feature amount extraction unit 35.
  • the envelope processing unit 33 obtains an envelope (envelope waveform) of the digital signal (vibration occurring in the bearing 7) generated by the AD conversion unit 31.
  • the AD converter 31 may be omitted.
  • the data conversion unit 25 associates the time at which the one piece of data was acquired with the frequency spectrum generated for one particular piece of data (digital signal / envelope).
  • the data conversion unit 25 has a spectrogram creation unit 37.
  • the spectrogram creating unit 37 creates a frequency spectrogram or an envelope frequency spectrogram based on the frequency spectrum of the vibration information (the spectrum obtained by Fourier-transforming the digital signal as it is) or the envelope frequency spectrum (the spectrum obtained by performing the Fourier transform on the data after the envelope processing). .
  • the spectrogram is displayed by the display unit 39.
  • the display unit 39 is various displays.
  • the data conversion unit 25 has a confirmation unit 40 (described later).
  • the analysis unit 29 has a machine learning unit 41.
  • the machine learning unit 41 determines a coefficient used in the learned determination formula based on the frequency spectrum of the specific region obtained by the feature amount extraction unit 35 (machine learning step).
  • FIG. 5 is a flowchart of the machine learning operation.
  • the control flowchart described below is an example, and each step can be omitted or replaced as needed. Further, a plurality of steps may be executed simultaneously, or some or all steps may be executed in an overlapping manner. Further, each block in the control flowchart is not limited to a single control operation, and can be replaced with a plurality of control operations represented by a plurality of blocks.
  • the operation of each device is a result of a command from the control unit to each device, and these are expressed by each step of the software application.
  • training data (data used for learning) is acquired using the device 1.
  • the training data can be all the detection signals acquired during the period from when the bearing 7 reaches the stages 1 to 5 of the five-stage model.
  • step S2 the acquired training data is labeled. Specifically, the training data is classified into “abnormal data” and “non-abnormal data”. For example, the training data is classified into “abnormal data” and “non-abnormal data” at a location where the data value of the training data suddenly changes.
  • step S2 and step S3 described below may be interchanged.
  • step S3 the feature amount extraction unit 35 performs a discrete Fourier transform process on the training data (discrete Fourier transform process step). Thereby, the frequency spectrum of the training data is extracted.
  • the feature amount extraction unit 35 smoothes the frequency spectrum of the training data. Specifically, for example, smoothing is performed by convolving the frequency spectrum of the training data with a Gaussian filter.
  • step S4 the machine learning unit 41 determines a discriminant for discriminating between normal data and abnormal data using the frequency spectrum of the normal data by a machine learning algorithm.
  • the following equation 1 is used as the discriminant.
  • f (x) is a discriminant.
  • K (xi, x) outputs a value based on the difference between the frequency spectrum of the normal data and the frequency spectrum of the target detection signal.
  • X in the discriminant f (x) is the intensity of the frequency spectrum of the normal data when the machine learning is performed.
  • x is the intensity of the frequency spectrum of the detection signal to be determined as normal or abnormal.
  • N represents the number of dimensions of the data, and in the present embodiment, corresponds to the number of frequency components included in the frequency spectrum of normal data.
  • ⁇ i, xi, ⁇ , and ⁇ are parameters characterizing the discriminant. Among these parameters, optimal values of ⁇ i, xi, and ⁇ are determined by a machine learning algorithm. ⁇ is called a “hyperparameter” and determines a value that is considered optimal before machine learning as follows.
  • FIG. 6 is a graph showing a change in the evaluation value based on a change in the hyperparameter of One Class SVM.
  • FIG. 6 shows that the discriminant is determined by machine learning after fixing the hyperparameter ⁇ to a predetermined value (a plurality of values of large and small types), and the frequency spectrum of the detection signal (the frequency spectrum of the training data used for machine learning) , Or may be the frequency spectrum of an actually measured detection signal).
  • the evaluation value is saturated early in the first evaluation value E where ⁇ is a small value.
  • the second evaluation value F where ⁇ is an intermediate value
  • the second evaluation value F is saturated later than the first evaluation value E. No saturation is observed with the third evaluation value G where ⁇ is a large value.
  • the hyper parameter ⁇ is the variance of the Gaussian kernel, and can be said to indicate the “sensitivity” of the discriminant for the input data. It is important to set an appropriate value for the hyperparameter ⁇ based on the magnitude of the data value, the variation of the data, the difference (distance) between the normal data and the abnormal data, and the like. More specifically, the hyperparameter ⁇ is a value such that the evaluation value does not saturate as the difference between the abnormal data and the normal data increases, as shown in the third evaluation value G in FIG. Is preferred.
  • the above discriminant includes hyperparameters other than the hyperparameter ⁇ , and the hyperparameter can also set its optimum value in the same manner as ⁇ .
  • a discriminant is determined by machine learning.
  • the machine learning algorithm uses One Class SVM that uses only normal data among the training data.
  • the machine learning algorithm sequentially inputs the frequency spectrum of the normal data (inputs the normal data to “x” in the discriminant of Equation 1 above), calculates the output value of the discriminant for each normal data, The optimum values of these parameters are determined by changing ⁇ i, xi, and ⁇ so that the calculated output value approaches 0.
  • the machine learning algorithm uses Boosting or SVM (Support ⁇ Vector ⁇ Machine) that uses both normal data and abnormal data as training data.
  • Boosting or SVM (Support ⁇ Vector ⁇ Machine) that uses both normal data and abnormal data as training data.
  • SVM Serial ⁇ Vector ⁇ Machine
  • higher-precision machine learning can be realized by using abnormal data indicating “abnormal” as teacher data.
  • a learned discriminant was obtained by a plurality of types of machine learning regarding the vibration acceleration. Thereafter, the learned discriminant was evaluated by actually calculating an evaluation value (described later) using the learned discriminant.
  • an evaluation value described later
  • a discriminant different from the one represented by the above equation may be calculated, but for any discriminant, whether or not the input data is normal is used as an evaluation value. It remains the same.
  • FIG. 7 is a graph showing a change in the evaluation value calculated by the learned discriminant using One ⁇ Class ⁇ SVM for the vibration acceleration.
  • FIG. 8 is a graph showing a change in the evaluation value calculated by the learned discriminant by the SVM for the vibration acceleration.
  • FIG. 9 is a graph showing a change in the evaluation value calculated by the learned discriminant by Boosting for the vibration acceleration.
  • the analysis unit 29 has a judgment unit 43.
  • the determination unit 43 inputs a frequency spectrum of a detection signal (referred to as a frequency spectrum for diagnosis) to be determined as a normal / abnormal to a discriminant determined by the machine learning unit 41 (referred to as a learned discriminant).
  • a learned discriminant determined by the machine learning unit 41
  • an evaluation value indicating the normal / abnormal of the data is calculated.
  • the feature amount extraction unit 35 performs a discrete Fourier transform process on a diagnostic detection signal acquired at a predetermined time interval to acquire a diagnostic frequency spectrum (discrete Fourier transform). Conversion processing step).
  • the determination unit 43 obtains the evaluation value on the time axis by inputting the frequency spectrum for diagnosis at each time into the learned discriminant and calculating the evaluation value at the time when each detection signal is obtained. (Evaluation value calculation step).
  • FIG. 10 is a graph showing a time change of the evaluation value of the frequency spectrogram, the vibration acceleration, and the machine learning.
  • A is the vibration acceleration
  • B is the evaluation value.
  • Each is displayed as a plurality of points, but the overlapping portion is displayed as an area.
  • the evaluation value sharply changes from a value close to 0 to a negative value. Has changed. That is, the evaluation value rapidly changes from a value indicating normal to a value indicating abnormal.
  • the discriminant is determined by using the detection signal obtained from the bearing 7 as learning data, so that the discriminant is determined based on human experience and intuition. It is possible to obtain a discriminant capable of clearly discriminating between a normal state and an abnormal state without relying on it. As a result, the transition between the normal and abnormal states of the bearing is automatically determined with high accuracy. This enables predictive maintenance and planned maintenance.
  • the determining unit 43 determines that the bearing 7 has transitioned from the normal state to the abnormal state based on the time when the evaluation value sharply decreases on the time axis (abnormality determining step). Note that a sharp decrease in the evaluation value on the time axis means that the decrease amount of the evaluation value per unit time is large.
  • the determination result (for example, the abnormality diagnosis result) is output to the notification unit 45. At this time, the state transition may be determined more accurately by considering a downward spike-like change observed prior to a sharp decrease in the evaluation value as a sign of the stage transition.
  • the frequency spectrum based on the detection signals of the plurality of sensors may be input to the analysis unit 29, respectively.
  • the analysis unit 29 determines a detection signal used to determine that the state has transitioned from the normal state to the abnormal state based on the state information of the machine tool 3 (detection signal determination step).
  • the state information of the machine tool 3 is, for example, the rotation direction of the bearing 7 and the number of rotations of the bearing 7. In this method, since the detection signal is determined based on the state information of the machine tool 3, the most appropriate detection signal of the plurality of sensors can be used.
  • the determination unit 43 determines that the bearing 7 is in an abnormal state and the stage 4 (defect propagation stage) to the stage 5 (damage propagation stage) at time t2 in FIGS. (Growth stage).
  • the notification unit 45 may be a display such as a liquid crystal monitor, an alarm such as a light or a buzzer, or a combination thereof.
  • the notification unit 45 notifies that the bearing 7 has transitioned to the abnormal state.
  • the notification is a display of information or a notification by voice or light. Thereby, the operator can quickly know that the bearing 7 has transitioned to the abnormal state.
  • the notification is preferably information that prompts the user to prepare a replacement bearing. Thereby, the worker is prompted to prepare the replacement bearing as the next work.
  • the notification unit 45 may notify that the bearing 7 has transitioned from the stage 4 (Defect propagation stage) to the stage 5 (Damage growth stage) in an abnormal state. Further, the above notification may be realized by displaying information on the display unit 39.
  • FIG. 11 is a flowchart of the evaluation value calculation and determination operation by the analysis unit in the stage 1 or 2.
  • the feature amount extraction unit 35 performs a discrete Fourier transform process on the diagnostic detection signal (or an envelope obtained by performing an envelope process on the detection signal) to perform a diagnostic frequency spectrum or an evaluation frequency spectrum.
  • the judgment unit 43 inputs the diagnostic frequency spectrum or the diagnostic envelope frequency spectrum (when the training data is converted into the envelope frequency spectrum and executes the machine learning) into the learned discriminant to input the evaluation value. Calculation (evaluation value calculation step).
  • step S6 the determination unit 43 determines whether the evaluation value is less than the threshold (abnormality determination step). If it is less than the threshold, the process proceeds to step S7, and if it is more than the threshold, the process returns to step S5.
  • the fact that the evaluation value is less than the threshold value means that the input frequency spectrum is significantly different from the normal data. That is, it means that the state of the bearing 7 has transitioned from the normal state (stage 1 or 2) to the abnormal state (stage 3) (time t1 in FIGS. 7 to 9). As described above, the calculation and determination of the evaluation value are repeatedly executed until the state transits to the abnormal state.
  • step S7 the determination unit 43 transmits a determination result (transition to an abnormal state) to the notification unit 45 and / or the display unit 39.
  • a determination result transition to an abnormal state
  • the notification unit 45 and / or the display unit 39 for example, one or more of “transition to an abnormal stage”, “preparation of a replacement bearing”, and “remaining operable period” are performed. Is notified (notification step). Note that, even after the transition to the abnormal state, the calculation and determination of the evaluation value may be repeatedly performed.
  • FIG. 12 is a flowchart of the evaluation value calculation and determination operation by the analysis unit in stage 4.
  • the feature amount extraction unit 35 performs a discrete Fourier transform process on the diagnostic detection signal (or an envelope obtained by performing envelope processing on the detection signal), and performs a diagnostic frequency spectrum or a diagnostic frequency spectrum.
  • the judgment unit 43 inputs the diagnostic frequency spectrum or the diagnostic envelope frequency spectrum (when the training data is converted into the envelope frequency spectrum and executes the machine learning) into the learned discriminant to input the evaluation value. Calculation (evaluation value calculation step).
  • step S9 the determination unit 43 determines whether the variation of the evaluation value in the predetermined range is equal to or larger than a threshold (abnormality determination step, damage growth determination step). If the variation in the evaluation value is equal to or greater than the threshold, the process proceeds to step S10, and if less than the threshold, the process returns to step S8.
  • the fact that the variation in the evaluation value is equal to or larger than the threshold value means that the input frequency spectrum greatly fluctuates. That is, it means that the state of the bearing 7 has transitioned from the defect propagation stage (stage 4) to the damage growth stage (stage 5) in the abnormal state (the time in FIGS. 8 to 10). t2).
  • step S ⁇ b> 10 the determination unit 43 transmits the determination result (transition to the damage growth stage) to the notification unit 45 and / or the display unit 39.
  • the calculation / determination of the evaluation value may be repeatedly executed even after the transition to the damage growth (Damage growth) stage.
  • the determination unit 43 uses the learned discriminant to determine the normal state of the apparatus. It is necessary to confirm both whether or not the measurement data obtained from the (bearing 7) can be determined to be normal and whether or not the measurement data obtained from the device (the bearing 7) in an abnormal state can be determined to be abnormal.
  • the bearing 7 is used for an industrial machine having a service life of more than ten years, such as a stacker crane (an example of a logistics transport machine) or a lathe (an example of a machine tool), under normal operating conditions. It takes several years to several tens of years until the bearing 7 becomes abnormal. Further, since the occurrence of the abnormality is a stochastic event, the abnormality may occur earlier, but the probability is low.
  • the inventors of the present application performed an accelerated test in consideration of the following concerns (i) and (ii) and verified whether or not the concerns could be resolved. (I) Whether the bearing 7 in which the abnormality has occurred in the acceleration test and the bearing 7 in which the abnormality has occurred under the normal use condition show the same behavior. (Ii) Whether an abnormality can occur in the bearing 7 by the acceleration test.
  • FIG. 13 is a schematic configuration diagram when the abnormality diagnosis device is attached to an acceleration tester.
  • the acceleration test machine 101 is, for example, a device simulating the machine tool 3.
  • the acceleration test machine 101 has the same motor 9 as in the above embodiment.
  • a bearing 7A of the same type as the bearing 7 of the embodiment is installed on the main shaft 5 of the motor 9 (bearing installation step).
  • the vibration sensor 17 is arranged near the outer ring 7b of the bearing 7A, and specifically, is close to or in contact with the outer ring 7b.
  • the acceleration test machine 101 has a load overload section 103.
  • the load overload portion 103 is a device that applies a radial load F to the bearing 7A.
  • the feature amount extraction unit 35 performs a discrete Fourier transform process on the vibration acceleration obtained from the vibration sensor 17 in the acceleration test machine 101 to obtain a frequency spectrum (frequency spectrum obtaining step).
  • the envelope processing unit 33 performs an envelope process on the detection signal, and the feature amount extraction unit 35 performs a discrete Fourier transform process to obtain an envelope frequency spectrum. May be.
  • the spectrogram creating section 37 acquires a frequency spectrogram and / or an envelope frequency spectrogram based on the frequency spectrum and / or the envelope frequency spectrum (frequency spectrogram acquiring step).
  • the confirmation unit 40 changes the normal state of the bearing 7A from the normal state to the abnormal state during the normal operation due to the fact that in the frequency spectrogram and / or the envelope frequency spectrogram, a frequency corresponding to the frequency resulting from the failure of the bearing 7A is developed. It is confirmed in advance that the transition determination of (1) is possible (confirmation step).
  • FIG. 14 shows a frequency spectrogram of the vibration acceleration and an envelope frequency spectrogram.
  • the inventors have experimentally found that in the frequency spectrogram and / or the envelope frequency spectrogram, a frequency f1 that matches a frequency (referred to as a characteristic frequency) due to the failure of the bearing 7A is expressed.
  • a characteristic frequency is 85.3 Hz for the defect of the outer ring.
  • the bearing abnormality diagnosis method of the present invention the normal state and the abnormal state of the bearing can be determined by evaluating the evaluation value of the bearing during normal operation using the parameters obtained in the acceleration test.
  • the data supplied to the analysis unit 29 is a frequency spectrum, but may be data obtained by performing processing (for example, smoothing in the frequency direction) on the frequency spectrum.
  • the rotating machine may be any one of a logistics transport machine, a machine tool, a textile machine, and a rehabilitation machine.
  • the distribution transport machine includes a stacker crane, a conveyor, an overhead transport vehicle, an AGV, an RGV, and an autonomous traveling robot.
  • Machine tools include lathes, press brakes, punch presses, laser machines, deburring machines, loaders, and unloaders.
  • Textile machines include automatic winders, spinning machines, draw false twisters, and filament winders.
  • the present invention can be widely applied to a method for diagnosing an abnormality of a bearing used for a rotating machine.
  • Abnormality diagnosis device 3 Machine tool 5: Main shaft 7: Bearing 7A: Bearing 7a: Inner ring 7b: Outer ring 7c: Rolling element 9: Motor 15: Diagnosis unit 17: Vibration sensor 19: Microphone 21: Feedback current input unit 25: Data conversion unit 29: Analysis unit 31: AD conversion unit 33: Envelope processing unit 35: Feature extraction unit 37: Spectrogram creation unit 39: Display unit 40: Confirmation unit 41: Machine learning unit 43: Judgment unit 45: Notification unit

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

L'invention concerne un procédé de diagnostic d'anomalie pour des paliers utilisés dans une machine tournante, lequel procédé comprend : une étape de traitement à transformée de Fourier discrète (étape S1, étape S2) pour acquérir un spectre de fréquence, par l'exécution d'un traitement à transformée de Fourier discrète sur un signal de détection obtenu à partir d'une machine-outil (3) ; une étape d'apprentissage automatique (étape S4) pour déterminer un discriminant appris en appliquant un apprentissage automatique au signal de détection ; une étape de calcul de valeur d'évaluation (étape S5) pour acquérir une valeur d'évaluation sur l'axe de temps, en injectant un spectre de fréquence au discriminant appris de fréquence ; et une étape de détermination d'anomalie (étape S6, étape S7) pour déterminer que le palier est passé de l'état normal à l'état anormal, sur la base du temps où la valeur d'évaluation change brutalement sur l'axe de temps.
PCT/JP2018/031217 2018-08-23 2018-08-23 Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante WO2020039565A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/JP2018/031217 WO2020039565A1 (fr) 2018-08-23 2018-08-23 Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2018/031217 WO2020039565A1 (fr) 2018-08-23 2018-08-23 Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante

Publications (1)

Publication Number Publication Date
WO2020039565A1 true WO2020039565A1 (fr) 2020-02-27

Family

ID=69592903

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2018/031217 WO2020039565A1 (fr) 2018-08-23 2018-08-23 Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante

Country Status (1)

Country Link
WO (1) WO2020039565A1 (fr)

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112302966A (zh) * 2020-11-02 2021-02-02 湘潭大学 一种离心泵运行状态判定方法及判定系统
CN112484981A (zh) * 2020-11-27 2021-03-12 上海电气电站设备有限公司 一种基于卷积自编码器的发电机异常状态监测方法
CN112798279A (zh) * 2020-12-30 2021-05-14 杭州朗阳科技有限公司 一种新的诊断电机轴承故障的检测方法
CN113358359A (zh) * 2021-05-25 2021-09-07 中国人民解放军92493部队计量测试研究所 一种柴油机工作性能状态检测方法
JP2021146368A (ja) * 2020-03-18 2021-09-27 株式会社リコー 診断装置、診断方法及び診断プログラム
CN113466021A (zh) * 2020-03-31 2021-10-01 丰田自动车株式会社 加压检查方法以及加压检查装置
CN114061922A (zh) * 2020-07-30 2022-02-18 宝山钢铁股份有限公司 基于振动数据的圆盘剪异常状况预警方法
JP2022168363A (ja) * 2021-04-26 2022-11-08 三菱重工業株式会社 風車翼の診断方法
EP4102206A3 (fr) * 2021-06-10 2022-12-21 Robert Bosch GmbH Procédé et dispositif d'estimation de la durée de vie d'un composant soumis à des contraintes tribologiques et produit programme informatique
CN116659860A (zh) * 2022-10-24 2023-08-29 中国人民解放军93208部队 一种服役环境下航空发动机主轴承故障演化监测新方法
CN117664576A (zh) * 2023-11-29 2024-03-08 东北大学 一种基于改进优化器深度学习模型的轴承健康评估方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005188979A (ja) * 2003-12-24 2005-07-14 Toribo Tex Kk 転がり軸受の劣化評価方法
JP2013140135A (ja) * 2011-12-09 2013-07-18 Tokyo Electron Ltd 周期的駆動系の異常検知装置、周期的駆動系を有する処理装置、周期的駆動系の異常検知方法、およびコンピュータプログラム
WO2015068446A1 (fr) * 2013-11-08 2015-05-14 三菱電機株式会社 Dispositif de diagnostic de son anormal
JP2015514220A (ja) * 2012-04-12 2015-05-18 シーメンス アクチエンゲゼルシヤフトＳｉｅｍｅｎｓ Ａｋｔｉｅｎｇｅｓｅｌｌｓｃｈａｆｔ アコースティック・エミッションセンサを備えたセンサ素子
JP2017102765A (ja) * 2015-12-03 2017-06-08 株式会社明電舎 異常予兆検出システムおよび異常予兆検出方法
JP2017142153A (ja) * 2016-02-10 2017-08-17 セイコーエプソン株式会社 寿命予測方法、寿命予測装置、および寿命予測システム
JP2017173321A (ja) * 2016-03-17 2017-09-28 Ntn株式会社 状態監視システムおよび風力発電装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005188979A (ja) * 2003-12-24 2005-07-14 Toribo Tex Kk 転がり軸受の劣化評価方法
JP2013140135A (ja) * 2011-12-09 2013-07-18 Tokyo Electron Ltd 周期的駆動系の異常検知装置、周期的駆動系を有する処理装置、周期的駆動系の異常検知方法、およびコンピュータプログラム
JP2015514220A (ja) * 2012-04-12 2015-05-18 シーメンス アクチエンゲゼルシヤフトＳｉｅｍｅｎｓ Ａｋｔｉｅｎｇｅｓｅｌｌｓｃｈａｆｔ アコースティック・エミッションセンサを備えたセンサ素子
WO2015068446A1 (fr) * 2013-11-08 2015-05-14 三菱電機株式会社 Dispositif de diagnostic de son anormal
JP2017102765A (ja) * 2015-12-03 2017-06-08 株式会社明電舎 異常予兆検出システムおよび異常予兆検出方法
JP2017142153A (ja) * 2016-02-10 2017-08-17 セイコーエプソン株式会社 寿命予測方法、寿命予測装置、および寿命予測システム
JP2017173321A (ja) * 2016-03-17 2017-09-28 Ntn株式会社 状態監視システムおよび風力発電装置

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7435103B2 (ja) 2020-03-18 2024-02-21 株式会社リコー 診断装置、診断方法及び診断プログラム
JP2021146368A (ja) * 2020-03-18 2021-09-27 株式会社リコー 診断装置、診断方法及び診断プログラム
CN113466021A (zh) * 2020-03-31 2021-10-01 丰田自动车株式会社 加压检查方法以及加压检查装置
CN114061922A (zh) * 2020-07-30 2022-02-18 宝山钢铁股份有限公司 基于振动数据的圆盘剪异常状况预警方法
CN112302966A (zh) * 2020-11-02 2021-02-02 湘潭大学 一种离心泵运行状态判定方法及判定系统
CN112302966B (zh) * 2020-11-02 2022-06-14 湘潭大学 一种离心泵运行状态判定方法及判定系统
CN112484981A (zh) * 2020-11-27 2021-03-12 上海电气电站设备有限公司 一种基于卷积自编码器的发电机异常状态监测方法
CN112798279A (zh) * 2020-12-30 2021-05-14 杭州朗阳科技有限公司 一种新的诊断电机轴承故障的检测方法
JP2022168363A (ja) * 2021-04-26 2022-11-08 三菱重工業株式会社 風車翼の診断方法
JP2022168865A (ja) * 2021-04-26 2022-11-08 三菱重工業株式会社 風車翼の診断方法
JP7245866B2 (ja) 2021-04-26 2023-03-24 三菱重工業株式会社 風車翼の診断方法
JP7438288B2 (ja) 2021-04-26 2024-02-26 三菱重工業株式会社 風車翼の診断方法
CN113358359A (zh) * 2021-05-25 2021-09-07 中国人民解放军92493部队计量测试研究所 一种柴油机工作性能状态检测方法
EP4102206A3 (fr) * 2021-06-10 2022-12-21 Robert Bosch GmbH Procédé et dispositif d'estimation de la durée de vie d'un composant soumis à des contraintes tribologiques et produit programme informatique
CN116659860A (zh) * 2022-10-24 2023-08-29 中国人民解放军93208部队 一种服役环境下航空发动机主轴承故障演化监测新方法
CN116659860B (zh) * 2022-10-24 2024-03-22 中国人民解放军93208部队 一种服役环境下航空发动机主轴承故障演化监测新方法
CN117664576A (zh) * 2023-11-29 2024-03-08 东北大学 一种基于改进优化器深度学习模型的轴承健康评估方法

Similar Documents

Publication Publication Date Title
WO2020039565A1 (fr) Procédé de diagnostic d'anomalie pour paliers utilisés dans une machine tournante
US11521105B2 (en) Machine learning device and machine learning method for learning fault prediction of main shaft or motor which drives main shaft, and fault prediction device and fault prediction system including machine learning device
JP6810097B2 (ja) 異常検出器
JP6508017B2 (ja) 機械設備の評価方法
US20130148817A1 (en) Abnormality detection apparatus for periodic driving system, processing apparatus including periodic driving system, abnormality detection method for periodic driving system, and computer program
EP3186598B1 (fr) Contrôle d'un dispositif comportant une partie mobile
JP6571704B2 (ja) 診断用データ取得システム、診断用システム、及び診断用データ取得プログラム
JP2018156151A (ja) 異常検知装置及び機械学習装置
US11544554B2 (en) Additional learning method for deterioration diagnosis system
JP6714806B2 (ja) 状態監視装置及び状態監視方法
JP6733838B1 (ja) 転がり軸受の異常診断方法及び異常診断装置
JP2018179735A (ja) 回転部品の異常診断方法及び異常診断装置
KR20190081933A (ko) 제조 설비의 이상 감지 및 진단 방법
JP6603260B2 (ja) 数値制御装置
Medjaher et al. Feature extraction and evaluation for health assessment and failure prognostics
JP4417318B2 (ja) 設備診断装置
KR20140072331A (ko) 이상진단 사전감시 방법
JP2016200451A (ja) 信号処理方法、信号処理装置、及び切削加工用異常検出装置
KR102433483B1 (ko) 진동 센서를 통한 설비 예지 보전 시스템
JP6497919B2 (ja) 回転体およびその軸受を含む設備の診断方法と診断システム
CN114830052A (zh) 用于确定机器的动态特性的方法和系统
JP2013205048A (ja) 回転機械の健全性診断方法
Kohlheb et al. Embedded autoencoder-based condition monitoring of rotating machinery
Prosvirin et al. Fault prediction of rolling element bearings using one class least squares SVM
JP2020184203A (ja) 衝突発生時の最大荷重推定装置および機械ダメージ診断装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18930972

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18930972

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: JP