Motor bearing wear state estimation device, bearing wear state estimation method, bearing wear state estimation program, and shield pump

Abstract

The invention provides a motor bearing wear state estimation device, a motor bearing wear state estimation method, a motor bearing wear state estimation program and a shield pump, and the motor bearing wear state estimation device can estimate the wear state without manual mechanical operation intervention even if the driving condition changes. This motor bearing wear state estimation device (5, 5A) estimates the wear state of a bearing (32, 33) using a plurality of detection coils (C1-C8). The device is provided with: a storage unit (56, 56A) that stores a first learning model (M11, M21) and a second learning model (M12, M22); an acquisition unit (550) that acquires each signal; and an estimation unit (551) that estimates the wear state by inputting the first differential signal, the second differential signal, and the amplitude signal to a first learning model and inputting the thrust differential signal and the amplitude signal to a second learning model. The amplitude signal is a first amplitude signal generated on the basis of the detection signals of the first radial detection coil and / or the second radial detection coil, or a second amplitude signal generated on the basis of the first composite signal and the second composite signal.

Description

Technical Field

[0001] The present invention relates to a motor bearing wear state estimation device, a bearing wear state estimation method, a bearing wear state estimation program and a canned motor pump. Background Art

[0002] A canned motor pump has a structure in which the pump and motor are integrated, and the liquid being processed is prevented from leaking. Generally speaking, the rotating structural parts of the canned motor pump (rotor, rotating shaft, bearings, and impeller) are sealed in a canned motor pump filled with the process liquid. Therefore, the internal structure of the canned motor pump cannot be visually monitored from the outside. Therefore, in order to efficiently operate a canned motor pump with such a structure, a device that monitors the wear state of the bearings (hereinafter referred to as a "monitoring device") is used (for example, see Patent Document 1).

[0003] The monitoring device disclosed in Patent Document 1 (a motor bearing wear monitoring device) measures the change in magnetic flux during rotor rotation by using detection coils installed at both ends of the stator in the long-side direction, thereby monitoring the displacement of the rotor (rotating shaft) in the radial and thrust directions caused by bearing wear. In this method, zero-point adjustment is required, that is, when the bearing is not worn, the output of the detection coil is adjusted so that the output of the detection coil shows zero displacement. The monitoring device detects the voltage induced in the detection coil due to the rotation of the motor. Therefore, the zero-point adjustment is performed while the motor is rotating according to the prescribed driving conditions used.

[0004] In canned motor pumps, when throttling, people use a method that tightens the valve of the pipeline connected to the discharge side of the canned motor pump. This method increases the resistance of the liquid flowing in the pipeline, resulting in energy loss. Therefore, in recent years, people have used a method that uses an inverter to make the motor drive conditions (e.g., drive frequency, drive voltage, etc.) variable (for example, see Patent Document 2). This method does not produce the energy loss of the method of tightening the valve, so adjusting the flow rate using this method has become the mainstream in recent years.

[0005] Prior art literature

[0006] Patent Literature:

[0007] Patent Document 1: Japanese Patent Application Laid-Open No. 10-080103

[0008] Patent Document 2: Japanese Patent Application Laid-Open No. 2007-162700 Summary of the Invention

[0009] Problems to be solved by the invention

[0010] However, as mentioned above, the monitoring device measures changes in magnetic flux as the rotor rotates. Therefore, when the drive conditions are changed by the inverter, the voltage induced in the detection coil changes, leading to technical issues such as erroneous detection due to a deviation in the correspondence between the measured value and the amount of wear. As a result, each time the drive conditions are changed, the correspondence must be manually adjusted (zero point adjustment). Thus, existing monitoring devices require manual mechanical operation to detect (estimate) the amount of wear (wear status) corresponding to changes in the drive conditions caused by the inverter.

[0011] The purpose of the present invention is to provide a motor bearing wear state estimation device, a bearing wear state estimation method, a bearing wear state estimation program and a shielded pump, which can estimate the wear state without human mechanical intervention even if the driving conditions change.

[0012] Solutions for solving problems

[0013] One embodiment of the present invention is a motor bearing wear state estimation device, which estimates the wear state of the bearing of the rotating shaft supporting the rotor based on the detection signals of multiple detection coils that detect the magnetic flux changes corresponding to the mechanical position changes of the rotor relative to the stator of the motor of the shielded pump, wherein the multiple detection coils are respectively installed on the stator, and the detection signal includes a fundamental component based on the driving frequency of the motor, and the multiple detection coils include: multiple radial detection coils that detect the magnetic flux changes of the rotating shaft in the radial direction; and multiple thrust detection coils that detect the magnetic flux changes of the rotating shaft in the thrust direction. The magnetic flux changes, the plurality of radial detection coils include: a first radial detection coil constituting one group; and a second radial detection coil constituting another group, the plurality of thrust detection coils include: a first thrust detection coil constituting one group; and a second thrust detection coil constituting another group, the motor bearing wear state estimation device has: a storage unit, which stores a learned first learning model and a learned second learning model, the learned first learning model is used as an input to represent the first differential signal between the detection signals of each of the first radial detection coils, and the first differential signal between the detection signals of each of the second radial detection coils. The machine learning is performed in a manner that the wear state in the radial direction is output when a second differential signal representing the difference between the detection signals of each of the first thrust detection coils and a second synthetic signal representing the amplitude of the fundamental wave component are input, and the learned second learning model is machine learning in a manner that the wear state in the thrust direction is output when a thrust differential signal representing the difference between the first synthetic signal synthesized by the detection signals of each of the first thrust detection coils and a second synthetic signal synthesized by the detection signals of each of the second thrust detection coils are input, and the amplitude signal is input; an acquisition unit that acquires the first differential signal The invention relates to a wear state estimation unit, comprising: a first differential signal, a second differential signal, a thrust differential signal and an amplitude signal obtained by the acquisition unit, inputting the first differential signal, the second differential signal and the amplitude signal obtained by the acquisition unit into the first learning model, and inputting the thrust differential signal and the amplitude signal obtained by the acquisition unit into the second learning model to estimate the wear state, wherein the amplitude signal is a first amplitude signal generated based on the detection signal of each of the first radial detection coil and / or the second radial detection coil, or a second amplitude signal generated based on the first composite signal and the second composite signal.

[0014] One embodiment of the present invention is a bearing wear state estimation method, which is performed by a motor bearing wear state estimation device, wherein the motor bearing wear state estimation device estimates the wear state of the bearing of the rotating shaft supporting the rotor based on the detection signals of multiple detection coils that detect the magnetic flux changes corresponding to the mechanical position changes of the rotor relative to the stator of the motor of the shielded pump, wherein the multiple detection coils are respectively installed on the stator, and the detection signals include a fundamental wave component based on the driving frequency of the motor, and the multiple detection coils include: multiple radial detection coils that detect the magnetic flux changes of the rotating shaft in the radial direction; and multiple thrust detection coils that detect The magnetic flux change of the rotating shaft in the thrust direction, the plurality of radial detection coils include: a first radial detection coil constituting one group; and a second radial detection coil constituting another group, the plurality of thrust detection coils include: a first thrust detection coil constituting one group; and a second thrust detection coil constituting another group, the motor bearing wear state estimation device includes a storage unit, the storage unit stores a learned first learning model and a learned second learning model, the learned first learning model is used as an input to represent the first differential signal between the detection signals of the respective first radial detection coils, and the first differential signal representing the difference between the detection signals of the respective second radial detection coils. The second learning model that has been learned performs machine learning in a manner that outputs the wear state in the radial direction when a second differential signal representing the difference between the detection signals of each of the first thrust detection coils and a second synthetic signal representing the amplitude of the fundamental wave component are input, and outputs the wear state in the thrust direction when a thrust differential signal representing the difference between the first synthetic signal synthesized by the detection signals of each of the first thrust detection coils and a second synthetic signal synthesized by the detection signals of each of the second thrust detection coils are input, and the amplitude signal is based on the first radial detection coil and / or the second thrust detection coil. The first amplitude signal is generated by the detection signal of each radial detection coil, or the second amplitude signal is generated based on the first composite signal and the second composite signal. The bearing wear state estimation method includes: the motor bearing wear state estimation device obtains the first differential signal, the second differential signal, the thrust differential signal and the amplitude signal; and the motor bearing wear state estimation device inputs the obtained first differential signal, the second differential signal and the amplitude signal into the first learning model, and inputs the obtained thrust differential signal and the amplitude signal into the second learning model to estimate the wear state.

[0015] One aspect of the present invention is a bearing wear state estimation program that causes a computer to function as the motor bearing wear state estimation device according to the above aspect.

[0016] One embodiment of the present invention is a shielded pump, wherein the shielded pump comprises: a motor having a rotor, a stator for rotating the rotor, and a rotating shaft that rotates together with the rotor; a bearing for supporting the rotating shaft; a plurality of detection coils for detecting changes in magnetic flux corresponding to changes in the mechanical position of the rotor relative to the stator; and a motor bearing wear state estimation device according to the above embodiment, which estimates the wear state of the bearing based on detection signals output respectively from the plurality of detection coils.

[0017] Effects of the Invention

[0018] According to the present invention, a motor bearing wear state estimation device, a bearing wear state estimation method, a bearing wear state estimation program and a canned motor pump can be provided, which can estimate the wear state without manual mechanical intervention even if the driving conditions change. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a side view showing an embodiment of a canned motor pump according to the present invention.

[0020] Figure 2 It shows Figure 1 A schematic cross-sectional view of a longitudinal section of a motor portion of a canned motor pump.

[0021] Figure 3 It shows Figure 2 An enlarged schematic cross-sectional view of an enlarged portion A in the motor portion.

[0022] Figure 4 This is a functional block diagram showing an embodiment of a motor bearing wear state estimation device according to the present invention.

[0023] Figure 5 It shows Figure 4 A schematic perspective view of a stator core showing the arrangement of detection coils included in a motor bearing wear state estimation device.

[0024] Figure 6 It shows Figure 5 Enlarged stereoscopic image of part B in the middle.

[0025] Figure 7 It shows Figure 5 A schematic diagram of an example of a detection signal of a detection coil.

[0026] Figure 8 It is an explanation Figure 2 Schematic diagram of the relationship between the position of the rotor of the motor unit relative to the stator in the thrust direction and the synthetic signal obtained by synthesizing two detection signals.

[0027] Figure 9It shows Figure 4 Schematic diagram of an example of information stored in a storage unit included in a motor bearing wear state estimation device.

[0028] Figure 10 It shows Figure 4 Schematic diagram of the appearance of a display unit included in the motor bearing wear state estimation device.

[0029] Figure 11 This is a functional block diagram of the machine learning device in the present invention.

[0030] Figure 12 It means Figure 11 Schematic diagram of a machine learning method performed in a machine learning device.

[0031] Figure 13 It shows Figure 4 A flowchart of an example of the operation of the motor bearing wear state estimation device.

[0032] Figure 14 It shows Figure 13 The actions are included in a flowchart of an example of a radial wear state estimation process.

[0033] Figure 15 It shows Figure 13 A flowchart of an example of a thrust wear state estimation process included in the operation.

[0034] Figure 16 is a schematic diagram schematically showing changes in the judgment criteria for the thrust wear state before and after the drive frequency is changed. Figure 16 (a) shows the judgment criteria before the drive frequency is changed, Figure 16 (b) shows the determination criteria after the drive frequency is changed.

[0035] Figure 17 This is a functional block diagram showing another embodiment of the motor bearing wear state estimation device according to the present invention.

[0036] Figure 18 It shows Figure 17 Schematic diagram of an example of information stored in a storage unit included in a motor bearing wear state estimation device.

[0037] Figure 19 Yes Figure 16 A schematic diagram illustrating weighting performed by a signal generating unit of a motor bearing wear state estimation device, Figure 19 (a) shows the state before weighting, Figure 19 (b) shows the state after weighting. DETAILED DESCRIPTION

[0038] The present invention provides a device for estimating the wear state of motor bearings in a canned motor pump with the following functions: automatically acquiring a detection signal and a signal representing the amplitude (signal level, potential) of the detection signal; and automatically estimating the bearing wear state corresponding to the driving conditions using the acquired signal and a learned model. As a result, the present invention can estimate the wear state corresponding to the driving conditions without manual mechanical intervention, even when the driving conditions change, thereby enabling flow control in the canned motor pump using a frequency converter. Details of each term are described below.

[0039] The following describes the implementation methods of the shielded pump of the present invention (hereinafter referred to as "the present pump"), the motor bearing wear state estimation device of the present invention (hereinafter referred to as "the present device"), the bearing wear state estimation method of the present invention (hereinafter referred to as "the present method"), and the bearing wear state estimation program of the present invention (hereinafter referred to as "the present estimation program"). In the following description, appropriate references are made to the drawings. In each drawing, the same parts and elements are marked with the same figure marks, and repeated descriptions are omitted. In addition, for the sake of convenience of explanation, the dimensional ratios of the various elements are sometimes exaggerated and are not limited to the ratios shown in the drawings.

[0040] Shielded Pump

[0041] Structure of shielded pump

[0042] First, the structure of this pump will be described.

[0043] Figure 1 This is a side view showing an embodiment of the present pump.

[0044] For ease of explanation, the figure shows the upper half of the pump 1 in cross-section.

[0045] The pump 1 has a structure that prevents leakage of the treated liquid, and is particularly used for conveying high-temperature liquids or highly dangerous liquids (e.g., explosive, flammable, or toxic liquids). The pump 1 comprises a pump unit 2, a motor unit 3, an adapter 4, and a device 5.

[0046] In the structure of this pump 1, the structures of the pump unit 2, the motor unit 3 and the adapter 4 are common to those of a known canned motor pump. Therefore, in the following description, the structures of the pump unit 2, the motor unit 3 and the adapter 4 are only briefly described without further details.

[0047] In the following description, the “front end direction” refers to the direction (front) of the pump unit 2 relative to the motor unit 3 , and the “rear end direction” refers to the direction (rear) of the motor unit 3 relative to the pump unit 2 .

[0048] The pump unit 2 draws in and discharges the treatment liquid. The pump unit 2 includes a housing 20, an impeller 21, a pump chamber 22, a suction pipe 23, and a discharge pipe 24. The housing 20 defines: the pump chamber 22, which houses the impeller 21; the suction pipe 23, which serves as a path for the treatment liquid drawn into the pump chamber 22; and the discharge pipe 24, which serves as a path for the treatment liquid discharged from the pump chamber 22. The pump chamber 22 communicates with the suction pipe 23 and the discharge pipe 24.

[0049] The motor unit 3 is driven under predetermined driving conditions (e.g., driving voltage: 200 V, driving frequency: 60 Hz) to rotate the impeller 21 of the pump unit 2. The motor unit 3 includes a housing 30, a rotating shaft 31, two bearings 32 and 33, two thrust washers 34 and 35, a rotor 36, a stator 37, a shield sleeve 38, and terminal terminals 39. The motor unit 3 is an example of a motor in the present invention.

[0050] Figure 2 It is a schematic cross-sectional view showing a longitudinal section of the motor unit 3 .

[0051] Figure 3 It shows Figure 2 An enlarged schematic cross-sectional view of an enlarged portion A in the motor portion 3.

[0052] The housing 30 accommodates the stator 37 and the shield 38 in a liquid-tight manner.

[0053] The rotating shaft 31 rotates due to the rotation of the rotor 36, and transmits the rotational power to the impeller 21. The rotating shaft 31 is cylindrical in shape. The rotating shaft 31 is inserted through the rotor 36 and fixed. The front end of the rotating shaft 31 protrudes into the pump chamber 22 (see Figure 1 ) in the front end portion, the impeller 21 is mounted. The rotating shaft 31 includes cylindrical sleeves 31a and 31b for protecting the front and rear portions of the rotating shaft 31.

[0054] In the following description, the “thrust direction” refers to the axial direction of the rotating shaft 31 , the “radial direction” refers to the radial direction of the rotating shaft 31 , and the “circumferential direction” refers to the circumferential direction of the rotating shaft 31 .

[0055] Bearing 32 is positioned toward the front end of rotor 36 and rotatably supports rotating shaft 31. Bearing 33 is positioned toward the rear end of rotor 36 and rotatably supports rotating shaft 31. Bearings 32 and 33 are, for example, rolling bearings. Thrust washer 34 is mounted on rotating shaft 31 between bearing 32 and rotor 36 to restrict forward movement of rotating shaft 31. Thrust washer 35 is mounted on rotating shaft 31 between bearing 33 and rotor 36 to restrict rearward movement of rotating shaft 31.

[0056] A gap of length L1 is formed between the bearings 32 and 33 and the thrust washers 34 and 35. A gap of length L2 is formed between the bearings 32 and 33 and the sleeves 31a and 31b.

[0057] The rotor 36 rotates due to the rotating magnetic field generated by the stator 37. The rotor 36 has a cylindrical shape. It includes a plurality (28 in this embodiment) of rod-shaped rotor bars 36a embedded in the outer peripheral edge of the rotor 36 at equal intervals in the circumferential direction. When the bearings 32 and 33 are not worn, the rotor 36 is positioned in its initial position relative to the stator 37. In this embodiment, the "initial position" is a position where the center of the stator 37 coincides with the center of the rotor 36 in both the thrust and radial directions.

[0058] The stator 37 generates a rotating magnetic field for rotating the rotor 36. The stator 37 has a substantially cylindrical shape and includes a stator core 37a and a plurality of motor windings 37b.

[0059] The stator core 37a holds the motor winding 37b. The stator core 37a is cylindrical in shape. The stator core 37a includes a plurality of teeth 37c (see Figure 6 ; the same below).

[0060] The tooth portion 37c is formed with a slot 37d (see Figure 6 The teeth 37c are arranged at equal intervals on the inner circumference of the stator core 37a. The motor windings 37b are inserted into the slots 37d and connected to a power supply device (not shown), such as an inverter, via terminal 39.

[0061] The shielding sleeve 38 fluid-tightly houses the rotating shaft 31, bearings 32 and 33, thrust washers 34 and 35, and rotor 36. The shielding sleeve 38 is cylindrical in shape. A portion of the process liquid introduced through the suction pipe 23 is introduced into the shielding sleeve 38 to cool the bearings 32 and 33 and the motor 3, and is then discharged to the discharge pipe 24.

[0062] The drawings mainly referred to in this description are Figure 1 .

[0063] The adapter 4 is connected to the rear end of the pump unit 2 and the front end of the motor unit 3 to couple the pump unit 2 and the motor unit 3 .

[0064] This device 5 estimates the wear state of the bearings 32 and 33 supporting the rotating shaft 31 by detecting changes in magnetic flux corresponding to changes in the mechanical position of the rotor 36 relative to the stator 37. The specific structure of this device 5 will be described later.

[0065] Motor bearing wear state estimation device (1)

[0066] Structure of motor bearing wear state estimation device (1)

[0067] Next, the structure of the present device 5 will be described. In the following description, reference will be made to Figures 1 to 3 .

[0068] Figure 4 It is a functional block diagram showing an embodiment of the present device 5 .

[0069] The device 5 includes eight detection coils C1, C2, C3, C4, C5, C6, C7, and C8, a connection unit 50, filter circuits 51a, 51b, 51c, and 51d, signal processing circuits 52a, 52b, 52c, and 52d, a signal generating unit 53, an A / D converter 54, a control unit 55, a storage unit 56, a display unit 57, a D / A converter 58, and an offset processing unit 59. The A / D converter 54, the control unit 55, and the D / A converter 58 are configured, for example, by a microcomputer.

[0070] It should be noted that, in the present invention, the motor unit 3 may also include detection coils C1 to C8 .

[0071] Figure 5 It is a schematic perspective view of the stator core 37a showing the arrangement of the detection coils C1 to C8.

[0072] Figure 6 It shows Figure 5 Enlarged stereoscopic image of part B in the middle.

[0073] Detection coils C1-C8 detect changes in magnetic flux corresponding to changes in the position (displacement) of the rotor 36 relative to the stator 37, generate detection signals representing these changes, and output these detection signals. The rotor 36 displaces radially along with the rotating shaft 31 based on the radial wear (wear status) of the bearings 32 and 33, and displaces thrust-wise along with the rotating shaft 31 based on the thrust wear (wear status) of the bearings 32 and 33. In other words, the displacement of the rotor 36 can be considered the wear (wear status) of the bearings 32 and 33. Therefore, by using detection coils C1-C8 to obtain detection signals representing the displacement of the rotor 36, the present device 5 can estimate the wear status (wear amount) of the bearings 32 and 33. The detection coils C1-C8 are shaped like a flat bobbin. The detection coils C1-C8 are embedded in notches 37e formed in the teeth 37c at the front and rear ends of the stator 37.

[0074] Detection coils C1 to C4 are mounted at equal intervals (90° intervals) circumferentially at the front end of the teeth 37c of the stator 37. Detection coil C1 is positioned 180° opposite detection coil C3, and detection coil C2 is positioned 180° opposite detection coil C4. Detection coils C5 to C8 are mounted at equal intervals (90° intervals) circumferentially at the rear end of the teeth 37c of the stator 37. Detection coil C5 is positioned 180° opposite detection coil C7, and detection coil C6 is positioned 180° opposite detection coil C8.

[0075] Figure 7 Schematic diagram showing an example of a detection signal.

[0076] In this figure, the horizontal axis represents the rotation angle of rotor 36, and the vertical axis represents the output voltage (signal level) of the induced electromotive force of detection coils C1-C8. The detection signals from detection coils C1-C8 include a waveform corresponding to the changes in the main magnetic flux of motor unit 3 (hereinafter referred to as the "fundamental component") and a waveform corresponding to the changes in the magnetic flux caused by the induced current flowing through the rotor bars 36a of rotor 36 (hereinafter referred to as the "harmonic component"). The fundamental component is generated by the drive voltage of motor unit 3 and has the same frequency as the drive voltage. The harmonic component is generated by the induced current flowing through the rotor bars 36a and has a frequency that varies depending on the rotation of rotor 36 and the number of rotor bars 36a. For example, under the following conditions (drive frequency: 60 Hz, number of rotor bars 36a: 28), for every rotation of rotor 36, detection coils C1-C8 will detect the 28th order change in magnetic flux generated by the rotor bars 36a. Therefore, the frequency of the harmonic component is 60 Hz × 28 = 1.68 kHz. In this way, the fundamental wave component is determined based on the driving frequency, and the harmonic component is determined based on the rotation of the rotor 36, the driving frequency, and the number of rotor bars 36a.

[0077] Here, by utilizing frequency converter control to change driving frequency, the rotational speed of motor part 3 can be easily changed. At this time, if only driving frequency is reduced without reducing driving voltage, motor part 3 will burn out, so usually driving frequency and driving voltage will be changed simultaneously. The increase and decrease of this driving frequency is proportional to the increase and decrease of driving voltage. That is, driving voltage decreases in proportion to the reduction of driving frequency and increases in proportion to the increase of driving frequency. As mentioned above, the fundamental wave component of detection signal is generated by driving voltage, so the signal level (amplitude) of fundamental wave component is proportional to driving frequency and driving voltage. Therefore, the signal level (amplitude) of fundamental wave component increases and decreases in proportion to the increase and decrease of driving conditions (driving frequency, driving voltage).

[0078] The drawings mainly referred to in this description are Figure 4 and Figure 5.

[0079] Detection coils C1, C3, C5, and C7 can detect the radial displacement of rotor 36 (i.e., the radial wear of bearings 32 and 33) by detecting changes in magnetic flux corresponding to the radial displacement of rotor 36 caused by the widening of the gap (L2) between bearings 32 and 33 and sleeves 31a and 31b. Detection coils C1 and C3 form one set of radial detection coils, and detection coils C5 and C7 form another set of radial detection coils.

[0080] It should be noted that, in the present invention, the detection coils C5 and C7 may constitute one group of radial detection coils, while the detection coils C1 and C3 may constitute another group of radial detection coils.

[0081] When the front side of the rotor 36 displaces radially from its initial position, the signal level of the harmonic components in the detection coils C1 and C3 that form a pair increases on the side closer to the rotor 36 (e.g., detection coil C1), while the signal level of the harmonic components decreases on the side farther from the rotor 36 (e.g., detection coil C3). At this time, the relative movement distance of the rotor 36 relative to the detection coils C1 and C3 is the same, so the increase and decrease in signal level are the same. Meanwhile, the signal level of the fundamental component does not increase or decrease. Therefore, when the detection signals of the detection coils C1 and C3 are combined to obtain (generate) their difference, the difference in the signal levels of the harmonic components in the combined signal (hereinafter referred to as "combined signal (C1C3)") increases as the amount of displacement increases. This difference allows the radial displacement of the front side of the rotor 36 to be detected. In other words, this difference indicates the amount of radial wear on the bearing 32, and the value of this difference is expressed as a voltage value. Similarly, in the composite signal of the detection signals of the respective detection coils C5 and C7 (hereinafter referred to as the "composite signal (C5C7)"), the difference in the signal level of the harmonic components increases as the displacement increases. This difference makes it possible to detect the displacement of the rear side of the rotor 36 in the radial direction. That is, this difference represents the amount of wear of the bearing 33 in the radial direction, and the value of this difference is expressed as a voltage value. Therefore, for example, when the rotor 36 is not displaced in the radial direction, the fundamental wave component and the harmonic components in each composite signal (C1C3, C5C7) cancel each other out, and the voltage value thereof is almost "0". On the other hand, when the rotor 36 is displaced in the radial direction, the difference in the harmonic components in each composite signal (C1C3, C5C7) increases as the displacement increases, and the voltage value thereof increases in accordance with this difference. Furthermore, since detection coils C1 and C3 are independent and separate from detection coils C5 and C7, by comparing the values ​​of the composite signal (C1 C3) and the composite signal (C5 C7), it is possible to detect uneven wear (a condition where one side is more worn than the other) on bearings 32 and 33. The signal representing the difference between the detection signals of detection coils C1 and C3 (i.e., composite signal (C1 C3)) is an example of the first differential signal in the present invention, and the signal representing the difference between the detection signals of detection coils C5 and C7 (i.e., composite signal (C5 C7)) is an example of the second differential signal in the present invention.

[0082] Detection coils C2, C4, C6, and C8 detect the displacement of rotor 36 in the thrust direction (i.e., the amount of wear on bearings 32 and 33 in the thrust direction) by detecting changes in magnetic flux. These changes correspond to the displacement of rotor 36 in the thrust direction caused by the widening of the gap (L1) between bearings 32 and 33 and thrust washers 34 and 35. Detection coils C2 and C4 form a set of thrust detection coils, connected so that their detection signals overlap. Therefore, the detection signals from detection coils C2 and C4 are combined in an overlapping manner to generate a composite signal (C2C4). Detection coils C6 and C8 form another set of thrust detection coils, connected so that their detection signals overlap. Therefore, the detection signals from detection coils C6 and C8 are combined in an overlapping manner to generate a composite signal (C6C8).

[0083] Figure 8 Schematic diagram for explaining the relationship between the position of the rotor 36 relative to the stator 37 in the thrust direction and each synthesized signal ( C2 C4 , C6 C8 ).

[0084] In this figure, the vertical axis represents the output voltage (signal level), and the horizontal axis represents the position of the rotor 36 in the thrust direction.

[0085] If the rotor 36 is displaced rearward from its initial position, the overlap between the detection coils C2 and C4 and the rotor 36 in the thrust direction decreases, but the overlap between the detection coils C6 and C8 and the rotor 36 remains unchanged. As a result, the signal level of the fundamental component of the composite signal (C2C4) decreases, but the signal level of the fundamental component of the composite signal (C6C8) remains almost unchanged. Similarly, if the rotor 36 is displaced forward from its initial position, the signal level of the fundamental component of the composite signal (C6C8) decreases, but the signal level of the fundamental component of the composite signal (C2C4) remains almost unchanged. Therefore, the displacement of the rotor 36 in the thrust direction can be detected by taking the difference between the composite signals (C2C4) and (C6C8). In other words, this difference represents the displacement of the rotor 36 in the thrust direction, that is, the amount of wear on the bearings 32 and 33 in the thrust direction, and the value of this difference is expressed as a voltage value. Therefore, for example, when the rotor 36 is not displaced in the thrust direction, the fundamental wave component and the harmonic components cancel each other out in the differential, and the voltage value is almost "0". On the other hand, when the rotor 36 is displaced in the thrust direction, the differential of the fundamental wave component increases according to the amount of displacement, and the voltage value increases according to the increase in the differential. Here, the reduction in signal level is affected by the magnetic flux density distribution of the detection coils C2, C4, C6, and C8 starting from the end of the rotor 36. Therefore, the change in voltage value with respect to the displacement of the rotor 36 is not a linear change, but rather a slightly convex curve. The signal representing the difference between the composite signal (C2C4) and the composite signal (C6C8) (the differential signal (Sd) described later) is an example of the thrust differential signal in the present invention.

[0086] The drawings mainly referred to in this description are Figure 4 .

[0087] The connection portion 50 is an interface for connecting the detection coils C1 to C8 .

[0088] Filter circuits 51a to 51d perform predetermined filtering on the detection signals from the corresponding detection coils C1, C3, C5, and C7. Filter circuits 51a to 51d are, for example, bandpass filters with a passband (e.g., 30 Hz to 2 kHz) that allows the fundamental and harmonic components to pass through. Filter circuit 51a is connected to detection coil C1, signal processing circuit 52a, and rectifier circuit 531, described later. Filter circuit 51b is connected to detection coil C3, signal processing circuit 52a, and rectifier circuit 532, described later. Filter circuit 51c is connected to detection coil C5, signal processing circuit 52b, and rectifier circuit 533, described later. Filter circuit 51d is connected to detection coil C7, signal processing circuit 52b, and rectifier circuit 534, described later.

[0089] Signal processing circuits 52a and 52b are comprised of, for example, a differential amplifier circuit, a rectifier circuit, and an integrator circuit. Signal processing circuit 52a generates a composite signal (C1C3) representing the difference between the detection signals of detection coils C1 and C3, forming one set. It then performs predetermined signal processing (rectification, AC-DC conversion) on the composite signal (C1C3), converting the composite signal (C1C3) from AC to DC. Signal processing circuit 52b generates a composite signal (C5C7) representing the difference between the detection signals of detection coils C5 and C7, forming another set. It then performs predetermined signal processing (rectification, AC-DC conversion) on the composite signal (C5C7), converting the composite signal (C5C7) from AC to DC.

[0090] Signal processing circuits 52c and 52d are composed of, for example, a filter circuit, a rectifier circuit, and an integrator circuit. The filter circuit is a low-pass filter with a cutoff frequency (e.g., 120 Hz) that allows the fundamental wave component to pass. Signal processing circuit 52c is connected to detection coils C2 and C4, while signal processing circuit 52d is connected to detection coils C6 and C8. Signal processing circuits 52c and 52d perform prescribed signal processing (filtering, rectification, AC-DC conversion) on the corresponding composite signals (C2C4, C6C8), converting the composite signals (C2C4, C6C8) from AC to DC.

[0091] The signal generator 53 generates a signal representing the amplitude of the fundamental wave component (hereinafter referred to as the "amplitude signal (Sa1)") based on the detection signals from each of the detection coils C1, C3, C5, and C7. The signal generator 53 is comprised of rectifier circuits 531, 532, 533, and 534, integration circuits 535, 536, 537, and 538, and an adder circuit 539. The integration circuits 535-538 are smoothing circuits that average the detection signals rectified by the corresponding rectifier circuits 531-534, converting the averaged detection signals into a DC value. The amplitude signal (Sa1) is an example of the first amplitude signal in the present invention.

[0092] Rectifier circuit 531 and integrator circuit 535 convert the detection signal from detection coil C1 from AC to DC. Rectifier circuit 532 and integrator circuit 536 convert the detection signal from detection coil C3 from AC to DC. Rectifier circuit 533 and integrator circuit 537 convert the detection signal from detection coil C5 from AC to DC. Rectifier circuit 534 and integrator circuit 538 convert the detection signal from detection coil C7 from AC to DC. At this point, the detection signals are averaged to form a signal representing the signal level (amplitude) of the fundamental wave component. Adder circuit 539 adds the DC signals from integrator circuits 535-538 to generate an amplitude signal (Sa1).

[0093] As described above, the increase (decrease) in the signal level of the harmonic components of the detection signal from detection coil C1 is the same as the decrease (increase) in the signal level of the harmonic components of the detection signal from detection coil C3. Similarly, the increase (decrease) in the signal level of the harmonic components of the detection signal from detection coil C5 is the same as the decrease (increase) in the signal level of the harmonic components of the detection signal from detection coil C7. Therefore, when the DC signals of these detection signals are added together to generate the amplitude signal (Sa1), the increase and decrease in the signal level of the harmonic components representing the radial displacement of rotor 36 in the amplitude signal (Sa1) cancel each other out. In other words, the amplitude signal (Sa1) functions as information indicating the amplitude of the fundamental wave component and does not include information about the radial displacement of rotor 36. In this embodiment, the amplitude signal (Sa1) is the sum of the four detection signals, and its signal level is approximately four times that of the detection signals. In other words, the signal level (amplitude) of the amplitude signal (Sa1) is proportional to the signal level (amplitude) of the fundamental wave component. As described above, the signal level of the fundamental wave component varies in proportion to changes in the driving conditions. Therefore, the present device 5 can indirectly detect changes in the driving conditions (i.e., the current driving frequency) based on changes in the signal level of the amplitude signal (Sa1) by acquiring the amplitude signal (Sa1).

[0094] The A / D converter 54 is connected to the signal processing circuits 52 a to 52 d , the signal generating unit 53 , and the offset processing unit 59 , converts analog signals inputted thereto into digital signals, and outputs the digital signals to the control unit 55 .

[0095] The control unit 55 controls the overall operation of the device 5. For example, the control unit 55 includes a processor such as a CPU (Central Processing Unit) 55a, volatile memory such as a RAM (Random Access Memory) 55b that functions as a work area for the CPU 55a, and non-volatile memory such as a ROM (Read Only Memory) 55c that stores various information, including the estimation program and other control programs. The control unit 55 includes an acquisition unit 550, an estimation unit 551, and a display control unit 552.

[0096] This estimation program runs in the control unit 55, cooperating with the hardware resources of the present device 5 to implement the present method described below. Furthermore, by causing the processor (CPU 55a) constituting the control unit 55 to execute this estimation program, the program enables the processor to function as the acquisition unit 550, estimation unit 551, and display control unit 552, thereby enabling the processor to execute this method. Similarly, by causing a computer to execute this estimation program, the program enables the computer to function as the present device 5.

[0097] It should be noted that, in the present invention, the estimation program may also be stored in the storage unit 56. Furthermore, the estimation program may also be stored in an installable file format or an executable file format in a non-transitory storage medium (e.g., a CD (Compact Disk), a DVD (Digital Versatile Disc), a USB (Universal Serial Bus) memory, etc.) and provided to the present device 5 via a dedicated read medium.

[0098] The acquisition unit 550 acquires a signal representing the difference between the detection signals of detection coils C1 and C3 (composite signal (C1 C3)), a signal representing the difference between the detection signals of detection coils C5 and C7 (composite signal (C5 C7)), a signal representing the difference between the composite signal (C2 C4) and the composite signal (C6 C8) (difference signal (Sd) described later), and an amplitude signal (Sa1). The specific operation of the acquisition unit 550 will be described later.

[0099] Estimation unit 551 inputs the composite signal (C1, C3, C5, C7) and the amplitude signal (Sa1) acquired by acquisition unit 550 into a first learning model M11 to estimate the radial wear state, described later. It also inputs the differential signal (Sd) and the amplitude signal (Sa1) acquired by acquisition unit 550 into a second learning model M12 to estimate the thrust wear state, described later. The detailed operation of estimation unit 551, first learning model M11, and second learning model M12 will be described later.

[0100] The display control unit 552 controls the display of the wear state on the display unit 57 based on the wear state estimated by the estimation unit 551 .

[0101] The storage unit 56 stores information necessary for the operation of the device 5 (e.g., the first learning model M11, the second learning model M12, offset information, etc.). The storage unit 56 is, for example, a nonvolatile memory such as an EEPROM (Electrically Erasable Programmable Read-Only Memory) or flash memory.

[0102] Figure 9 It is a schematic diagram showing an example of information (first learning model M11 and second learning model M12) stored in the storage unit 56.

[0103] The "first learning model M11" is a machine learning algorithm (i.e., a learning model) that has been trained to output the radial wear state of bearings 32 and 33 (hereinafter referred to as the "radial wear state") when the composite signals (C1, C3, C5, and C7) and the amplitude signal (Sa1) are input. The first learning model M11 is pre-generated, for example, by the machine learning device 6 (described later) and stored in the storage unit 56.

[0104] The second learning model M12 is a model that has been machine-learned to output the wear state of bearings 32 and 33 in the thrust direction (hereinafter referred to as the "thrust wear state") when the differential signal (Sd) and the amplitude signal (Sa1) are input. The second learning model M12 is pre-generated, for example, by the machine learning device 6 (described later) and stored in the storage unit 56.

[0105] "Offset information" is information (e.g., a voltage value) indicating the offset voltage added to or subtracted from the composite signal (C2C4) during the offset processing. The offset information is measured or set in advance under predetermined reference driving conditions (e.g., driving frequency: 60 Hz, driving voltage: 200 V) before the pump 1 is shipped, and is stored in the storage unit 56.

[0106] "Offset processing" refers to the addition or subtraction of an offset voltage from the composite signal (C2C4) under reference driving conditions, ensuring that the difference between the composite signal (C2C4) and the composite signal (C6C8) accurately represents the displacement of rotor 36 in the thrust direction (the amount of wear on bearings 32 and 33 in the thrust direction). This offset processing virtually aligns the magnetic center position of rotor 36 relative to stator 37 in the thrust direction with the mechanical center position (the two center positions coincide).

[0107] Figure 10 Schematic diagram showing the appearance of the display unit 57 .

[0108] The display unit 57 shows the wear status and rotational direction of the bearings 32 and 33. For example, the display unit 57 is composed of multiple LEDs (light emitting diodes) that display the radial wear status, thrust wear status, and rotational direction. The display unit 57 displays the radial and thrust wear status in three levels: green, yellow, and red. The rotational direction is indicated by lighting (forward rotation) and off (reverse rotation).

[0109] The drawings mainly referred to in this description are Figure 4 .

[0110] The D / A converter 58 converts the offset information input from the control section 55 from a digital signal into an analog signal, and sends the converted analog signal to the offset processing section 59 .

[0111] The offset processing unit 59 performs offset processing based on the offset information. The offset processing unit 59 is comprised of, for example, an offset voltage generation circuit, an arithmetic circuit, and a differential absolute value conversion circuit (none of which are shown). The offset processing unit 59 performs offset processing on the composite signal (C2C4) by generating an offset voltage based on the offset information in the offset voltage generation circuit and adding or subtracting the offset voltage from the composite signal (C2C4) in the arithmetic circuit. Furthermore, the offset processing unit 59 calculates (generates) the difference between the offset-processed composite signal (C2C4) and the composite signal (C6C8) in the arithmetic circuit, and converts the difference into an absolute value in the differential absolute value conversion circuit. The signal representing this absolute value (hereinafter referred to as the "difference signal (Sd)") is converted into a digital signal by the A / D converter 54 and input to the control unit 55.

[0112] Machine Learning Device

[0113] Next, the machine learning device 6 that generates the first learning model M11 and the second learning model M12 will be described.

[0114] The structure of the machine learning device

[0115] Figure 11 This is a functional block diagram of the machine learning device 6 in the present invention.

[0116] For the convenience of explanation, external devices described later are also shown in this figure.

[0117] The machine learning device 6 is constituted by, for example, a personal computer and includes a connection unit 61 , a communication unit 62 , a control unit 63 , a storage unit 64 , an input unit 65 , and a display unit 66 .

[0118] The connection portion 61 is, for example, a well-known interface including terminals for connection to an external device.

[0119] An "external device" is a device connected to the machine learning device 6 that generates learning data acquired by the machine learning device 6. Examples of the external device include a test device 7 that simulates the pump 1, a signal generator 8 that simulates a signal obtained by the pump 1, and a work terminal device 9 used by an operator who generates a learning model using the machine learning device 6.

[0120] The communication unit 62 is connected to, for example, a communication network line (for example, a wireless communication line such as the Internet or an intranet), and transmits and receives information to and from other computers or the like.

[0121] The control unit 63 controls the overall operation of the machine learning device 6. For example, the control unit 63 includes a processor such as a CPU (Central Processing Unit) 63a, volatile memory such as a RAM (Random Access Memory) 63b that functions as a work area for the CPU 63a, and nonvolatile memory such as a ROM (Read Only Memory) 63c that stores various information, including machine learning programs and other control programs. The control unit 63 includes a data acquisition unit 630, a learning dataset generation unit 631, and a learning unit 632.

[0122] The machine learning program runs in the control unit 63, and the machine learning program cooperates with the hardware resources of the machine learning device 6 to implement the machine learning method described below. Furthermore, by causing the processor (CPU 63a) constituting the control unit 63 to execute the machine learning program, the machine learning program enables the processor to function as the data acquisition unit 630, the learning data set generation unit 631, and the learning unit 632, thereby enabling the processor to execute the machine learning method. Similarly, by causing a computer to execute the machine learning program, the machine learning program enables the computer to function as the machine learning device 6.

[0123] It should be noted that, in the present invention, the machine learning program may also be stored in the storage unit 64. Furthermore, the machine learning program may also be stored in an installable file format or an executable file format in a non-transitory storage medium (e.g., a CD (Compact Disk), a DVD (Digital Versatile Disc), a USB (Universal Serial Bus) memory, etc.) and provided to the machine learning device 6 via a dedicated read medium.

[0124] The data acquisition unit 630 acquires learning data from an external device via the connection unit 61 .

[0125] “Learning data” includes information serving as input data to a learning model and information serving as output data (teacher data) to a learning model.

[0126] "Input data" is an explanatory variable in machine learning. In this embodiment, it is a composite signal (C1 C3: first differential signal), a composite signal (C5 C7: second differential signal), a differential signal (Sd: thrust differential signal), and an amplitude signal (Sa1: first amplitude signal) in a specified state (specified driving conditions, specified wear state, specified period). Specifically, the input data is, for example, a voltage value obtained from each signal. As described above, the composite signals (C1 C3, C5 C7) represent the displacement (wear) of the bearings 32 and 33 in the radial direction, that is, the radial wear state. The differential signal (Sd) represents the displacement (wear) of the bearings 32 and 33 in the thrust direction, that is, the thrust wear state. The signal level (voltage value) of the amplitude signal (Sa1) is proportional to the driving conditions (driving frequency, driving voltage). The voltage values ​​of the composite signals (C1 C3, C5 C7) and the differential signal (Sd) increase or decrease according to the increase or decrease of the driving conditions. That is, the voltage values ​​of the composite signals (C1, C3, C5, C7) and the differential signal (Sd) increase or decrease according to the increase or decrease of the amplitude signal (Sa1). Therefore, there is a correlation between the composite signals (C1, C3, C5, C7) and the differential signal (Sd) and the amplitude signal (Sa1). Furthermore, if the displacement of bearings 32 and 33 increases (the wear condition worsens), the voltage values ​​of the composite signals (C1, C3, C5, C7) and the differential signal (Sd) increase. In other words, if the voltage values ​​of the composite signals (C1, C3, C5, C7) and the differential signal (Sd) increase, the wear condition of bearings 32 and 33 worsens (the amount of wear increases). Therefore, there is a correlation between the composite signals (C1, C3, C5, C7) and the differential signal (Sd) and the wear condition of bearings 32 and 33. Furthermore, if the driving frequency (i.e., the amplitude signal (Sa1)) is changed, the displacement (wear amount) represented by each composite signal (C1, C3, C5, C7) and the differential signal (Sd), i.e., the wear state, changes (see also for details). Figure 16 ). Therefore, there is a correlation between the amplitude signal (Sa1) and the wear state of the bearings 32 and 33. In this way, there is a correlation between the input data and the output data. The input data are, for example, the signals (voltage values) obtained during a specified period when the test device 7 is operated under specified driving conditions and specified wear conditions. Alternatively, the input data are, for example, the simulated signals (voltage values) generated by the signal generating device 8 during a specified period based on actual data (for example, log data) obtained in the past in a pump of the same model as the present pump 1. The data acquisition unit 630 acquires the composite signal (C1 C3), the composite signal (C5C7), the differential signal (Sd) and the amplitude signal (Sa1) from the test device 7 and / or the signal generating device 8 as input data.

[0127] Here, the "prescribed driving condition" is, for example, a driving frequency between 40 Hz and 60 Hz. The prescribed driving condition is set in steps of several Hz (for example, 5 Hz). The "prescribed wear state" is formed, for example, by artificially moving the position of the rotor 36 (rotating shaft 31) so that the composite signal (C1, C3, C5, C7) and the differential signal (Sd) indicate the wear state of "normal", "caution", and "abnormal". The "prescribed period" is the time (for example, several seconds to several tens of seconds) required to obtain a signal that can be used as learning data under the prescribed driving conditions and the prescribed wear state.

[0128] "Output data" is a criterion variable in machine learning. In this embodiment, it represents information indicating the wear status of bearings 32 and 33. In this embodiment, "wear status" includes three states: "Normal," indicating a wear status that does not require replacement of bearings 32 and 33 (mild); "Caution," indicating a wear status that requires replacement of bearings 32 and 33 (moderate); and "Abnormal," indicating a wear status that requires replacement of bearings 32 and 33 (severe). Wear status includes radial wear and thrust wear. The operator determines the wear status for each input data item as the wear status corresponding to the input data. The operator inputs the determination result into the work terminal device 9 as output data corresponding to the input data and stores it in the work terminal device 9. The output data is associated with information that can be used by the operator to determine the wear status (e.g., an ID for each input data item or information indicating the period during which the input data was generated). The data acquisition unit 630 acquires the wear status from the work terminal device 9 as output data. At this time, a numerical value such as "0", "1", or "2" is assigned to each wear state.

[0129] It should be noted that in the present invention, the radial wear state and the thrust wear state can be determined for each input data at the same time, or the radial wear state corresponding to the composite signal (C1 C3, C5C7) and the amplitude signal (Sa1) in the input data, and the thrust wear state corresponding to the differential signal (Sd) and the amplitude signal (Sa1) in the input data can be determined one by one.

[0130] Furthermore, in the present invention, output data is not limited to three states. For example, the output data may include only two states: "Normal" and "Abnormal." Furthermore, the output data may include multiple "Normal" and / or "Caution" states, divided into several stages (e.g., three stages) depending on the progression of wear.

[0131] The learning dataset generation unit 631 generates one or more learning datasets based on the learning data acquired by the data acquisition unit 630. The learning dataset generation unit 631 generates the learning datasets by, for example, associating input data from a predetermined period with output data corresponding to the input data. The generated learning datasets are stored in the storage unit 64.

[0132] The learning unit 632 performs machine learning on the learning model (constructs a learning model) using the learning dataset generated by the learning dataset generation unit 631. The machine learning performed by the learning unit 632 is performed using a well-known machine learning algorithm (e.g., a neural network having an input layer, multiple intermediate layers, and an output layer).

[0133] It should be noted that in the present invention, the machine learning algorithm used by learning unit 632 is not limited to a neural network; any algorithm can be used as long as it can estimate the wear state by generating a learning model through machine learning using a learning dataset. For example, learning unit 632 may also use a random forest, decision tree, support vector machine, or the like.

[0134] The storage unit 64 stores information necessary for the operation of the machine learning device 6 (learning data, a learning data set used to train the learning model, and completed learning models (first learning model M11, second learning model M12, etc.)). The storage unit 64 is, for example, a storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive).

[0135] The input unit 65 is, for example, a device (eg, a keyboard, a mouse, etc.) that receives input operations from a user of the machine learning device 6 .

[0136] The display unit 66 is, for example, a device (for example, a monitor, a liquid crystal display, etc.) that displays information to an operator or the like.

[0137] It should be noted that, in the present invention, the input unit 65 and the display unit 66 may be formed of, for example, a touch panel display.

[0138] Actions of machine learning devices

[0139] Next, the operation (machine learning method) of the machine learning device 6 will be described.

[0140] Figure 12 It is a schematic diagram illustrating the machine learning method executed in the machine learning device 6.

[0141] First, the data acquisition unit 630 acquires input data in a predetermined state (predetermined period, predetermined wear state) from an external device (e.g., the testing device 7). Next, the data acquisition unit 630 acquires output data corresponding to the acquired input data from the work terminal device 9. The acquired input data is associated with the corresponding output data and stored in the storage unit 64.

[0142] Next, the learning dataset generation unit 631 generates a learning dataset based on the input data and output data acquired by the data acquisition unit 630. Specifically, the learning dataset generation unit 631 generates a set of learning datasets corresponding to the radial wear state (hereinafter referred to as "the first learning dataset DS1") by associating the synthetic signal (C1 C3, C5C7) and the amplitude signal (Sa1) in a specified state (specified operating conditions, specified wear state, specified period) with the radial wear state. In addition, the learning dataset generation unit 631 generates a set of learning datasets corresponding to the thrust wear state (hereinafter referred to as "the second learning dataset DS2") by associating the differential signal (Sd) and the amplitude signal (Sa1) in a specified state (specified period, specified wear state) with the thrust wear state. That is, the learning dataset includes the first learning dataset DS1 and the second learning dataset DS2. The generated learning datasets are stored in the storage unit 64.

[0143] Here, the number of generated learning data sets is appropriately set according to the estimation accuracy of the generated learning models (first learning model M11 and second learning model M12 ).

[0144] Next, the learning unit 632 generates (constructs) a learning model by repeatedly subjecting the learning dataset to machine learning using a known machine learning algorithm. Specifically, the learning unit 632 generates a first learning model M11 by subjecting the machine learning algorithm to machine learning using multiple first learning datasets DS1. Furthermore, the learning unit 632 generates a second learning model M12 by subjecting multiple second learning datasets DS2 to machine learning using a known machine learning algorithm. The specific machine learning method is well known, so a detailed description thereof is omitted. The generated first learning model M11 and second learning model M12 are stored in the storage unit 64.

[0145] The first learning model M11 thus generated is capable of outputting a radial wear state corresponding to the drive condition by receiving the composite signal (C1, C3, C5, C7) and the amplitude signal (Sa1) as input data. Specifically, the first learning model M11 performs machine learning so that, when the composite signal (C1, C3, C5, C7) and the amplitude signal (Sa1) are input data, the model outputs a radial wear state corresponding to the drive condition (derived from the amplitude signal (Sa1)). Similarly, the second learning model M12 performs machine learning so that, when the differential signal (Sd) and the amplitude signal (Sa1) are input data, the model outputs a thrust wear state corresponding to the drive condition (derived from the amplitude signal (Sa1)).

[0146] Operation of shielded pump (motor bearing wear state estimation device (1))

[0147] Next, the operation of the present pump 1 (i.e., the present method) will be described below, centering on the operation of the present device 5. Figures 1 to 12 .

[0148] Figure 13 This is a flowchart showing an example of the operation of the present device 5 .

[0149] While pump 1 is operating, power under specified drive conditions is supplied to motor unit 3, causing rotor 36, rotary shaft 31, and impeller 21 to rotate at a specified rotational speed. During this operation, device 5 periodically (e.g., every several seconds) repeatedly executes radial wear state estimation processing (ST1) and thrust wear state estimation processing (ST2).

[0150] Radial wear state estimation processing

[0151] Figure 14 This is a flowchart showing an example of the radial wear state estimation process (ST1).

[0152] The "Radial Wear State Estimation Process (ST1)" is a process in which the device 5 acquires input data (synthesized signals (C1, C3, C5, C7) and amplitude signal (Sa1)) and estimates the radial wear state while the pump 1 is operating. The radial wear state estimation process (ST1) is an example of this method.

[0153] First, the acquisition unit 550 acquires the composite signals (C1, C3, C5, C7) and the amplitude signal (Sa1) as input data (ST11: acquisition step). The acquired composite signal (C1, C3) is generated by the signal processing circuit 52a and converted into a digital signal by the A / D converter 54. It represents the difference between the detection signals of the detection coils C1 and C3. Similarly, the acquired composite signal (C5, C7) is generated by the signal processing circuit 52b and converted into a digital signal by the A / D converter 54. It represents the difference between the detection signals of the detection coils C5 and C7. Furthermore, the amplitude signal (Sa1) is generated by the signal generation unit 53 by generating DC signals from the detection signals of the detection coils C1, C3, C5, and C7, respectively, and summing all of these DC signals. As described above, the amplitude signal (Sa1) functions as information indicating only the amplitude of the fundamental wave component and does not include information indicating radial displacement. The acquired amplitude signal (Sa1) is the signal converted into a digital signal by the A / D converter 54.

[0154] Next, the estimation unit 551 inputs the acquired input data into the first learning model M11 and estimates the radial wear state based on the output of the first learning model M11 (ST12: estimation step). Specifically, for example, if the output of the first learning model M11 is "normal," the estimation unit 551 estimates the radial wear state to be "normal."

[0155] Next, the display control unit 552 determines the display mode of the display unit 57 based on the estimated radial wear state, and controls the display of the display unit 57 so that the display of the display unit 57 changes to the determined display mode (ST13: display step). In other words, for example, when the estimated radial wear state is "normal", the display control unit 552 lights the "green" LED of the display unit 57.

[0156] Thrust wear state estimation processing

[0157] Figure 15 This is a flowchart showing an example of the thrust wear state estimation process (ST2).

[0158] The thrust wear state estimation process (ST2) is a process in which the device 5 acquires input data (differential signal (Sd) and amplitude signal (Sa1)) and estimates the thrust wear state during the operation of the pump 1. The thrust wear state estimation process (ST2) is an example of this method.

[0159] First, the acquisition unit 550 acquires the differential signal (Sd) and the amplitude signal (Sa1) as input data (ST21: acquisition step). As described above, the acquired differential signal (Sd) is generated by the offset processing unit 59 and converted into a digital signal by the A / D converter 54. It represents the absolute value of the difference between the composite signal (C2C4) and the composite signal (C6C8).

[0160] Next, the estimation unit 551 inputs the acquired input data into the second learning model M12 and estimates the thrust wear state based on the output of the second learning model M12 (ST22: estimation step). Specifically, for example, if the output of the second learning model M12 is "Caution," the estimation unit 551 estimates the thrust wear state to be "Caution."

[0161] Next, the display control unit 552 determines the display mode of the display unit 57 based on the estimated thrust wear state, and controls the display of the display unit 57 so that the display of the display unit 57 changes to the determined display mode (ST23: display step). That is, for example, when the estimated thrust wear state is "Caution", the display control unit 552 lights the "yellow" LED of the display unit 57.

[0162] Figure 16 is a schematic diagram schematically showing changes in the judgment criteria for the thrust wear state before and after the drive frequency is changed. Figure 16 (a) shows the judgment criteria before the drive frequency is changed, Figure 16 (b) shows the determination criteria after the drive frequency is changed.

[0163] In this figure, the vertical axis represents the voltage value of the differential signal (Sd), and the horizontal axis represents the amount of wear of the bearings 32 and 33 in the thrust direction (the displacement of the rotor 36 in the thrust direction). In (a) of the figure, the "V"-shaped solid line represents the correspondence between the voltage value and the wear amount in the drive frequency before the change. In (b) of the figure, the "V"-shaped solid line represents the correspondence in the drive frequency after the change, and the double-dotted line represents the correspondence in the drive frequency before the change. In addition, the figure paints the thrust wear state corresponding to the drive frequency into three colors. Furthermore, the figure shows the change in the judgment criterion when the drive frequency is lowered from the state shown in (a) of the figure to the state shown in (b) of the figure.

[0164] The thrust wear state is determined based on the corresponding relationship of the "V" shape. Figure 16In the example shown, the voltage value corresponding to the specified wear amount "T1%" and "T2%" in the corresponding relationship becomes the judgment standard for the thrust wear state. When the wear amount is less than "T1%", the thrust wear state is judged as "normal", when the wear amount is greater than "T1%" and less than "T2%", the thrust wear state is judged as "caution", and when the wear amount is greater than "T2%", the thrust wear state is judged as "abnormal". Figure 16 As shown in FIG, the wear amount (T1%, T2%) as the threshold value of the wear state remains unchanged before and after the drive frequency is changed, but the voltage value as the threshold value of the wear state increases and decreases with the increase and decrease of the drive frequency. Therefore, if the judgment standard is not changed according to the change of the drive frequency, the present device 5 cannot estimate the correct thrust wear state. That is, for example, Figure 16 In the example shown, when the voltage value is "V1" at the changed driving frequency, the judgment criterion ( Figure 16 In (b)), the wear amount is "a2%", and the thrust wear status becomes "Caution". However, the judgment criteria before the change ( Figure 16 In (a)), the wear amount is "a1%," and the thrust wear state is "normal." In the present invention, the first learning model M11 and the second learning model M12 perform machine learning on the judgment criteria to avoid such misjudgments. As a result, even if the drive conditions (drive frequency, drive voltage) are changed, the present device 5 can estimate the wear state corresponding to the drive conditions without manual mechanical intervention. It should be noted that the same applies to radial wear.

[0165] Summary (1)

[0166] According to the embodiment described above, the plurality of detection coils C1 to C8 each output a detection signal representing a change in magnetic flux corresponding to a change in the mechanical position of the rotor 36 relative to the stator 37. The plurality of detection coils C1 to C8 include detection coils C1, C3, C5, and C7, which detect changes in magnetic flux in the radial direction, and detection coils C2, C4, C6, and C8, which detect changes in magnetic flux in the thrust direction. The detection signal includes a fundamental component based on the drive frequency of the motor unit 3. The present device 5 includes an acquisition unit 550, an estimation unit 551, and a storage unit 56. The storage unit 56 stores a first learning model M11 and a second learning model M12. The first learning model M11 is machine-learned to output a radial wear state when fed with a composite signal (C1 to C3), a composite signal (C5 to C7), and an amplitude signal (Sa1). The second learning model M12 is machine-learned to output a thrust wear state when fed with a differential signal (Sd) and an amplitude signal (Sa1). The amplitude signal (Sa1) is generated based on the detection signals from each of the detection coils C1, C3, C5, and C7. With this configuration, the present device 5 uses the amplitude signal (Sa1), which contains only the amplitude (signal level) of the fundamental wave component that increases or decreases proportionally with the drive conditions, as well as the composite signals (C1, C3, C5, and C7) representing the wear amount (wear state) of bearings 32 and 33, and the differential signal (Sd) as input data for the learned model. This allows the present device 5 to estimate the wear state corresponding to the drive conditions. Therefore, even if the drive conditions change, the present device 5 can estimate the wear state corresponding to the drive conditions without requiring manual mechanical intervention.

[0167] Furthermore, according to the embodiment described above, the device 5 includes a signal generation unit 53 that generates an amplitude signal (Sa1) based on the detection signals of each of the detection coils C1, C3, C5, and C7. This configuration allows the acquisition unit 550 to acquire the amplitude signal (Sa1) as input data for the learning model at any time. Therefore, even if the driving conditions change, the device 5 can estimate the wear state corresponding to the driving conditions without requiring manual intervention.

[0168] Furthermore, according to the embodiment described above, the signal generation unit 53 adds the detection signals (DC signals) from the detection coils C1, C3, C5, and C7. This structure cancels out all radial displacement information from the front and rear sides in the amplitude signal (Sa1). Consequently, the estimation unit 551 can estimate the radial wear and thrust wear conditions corresponding to the driving conditions without being affected by the displacement information. Consequently, even if the driving conditions change, the device 5 can estimate the wear conditions corresponding to the driving conditions without requiring any manual mechanical intervention.

[0169] It should be noted that in the above-described embodiment, the signal generating unit 53 may generate an average value of the four detection signals (DC signals) as the amplitude signal instead of generating the sum of the four detection signals (DC signals).

[0170] Furthermore, in the embodiment described above, the signal generation unit 53 may generate the amplitude signal (Sa1) by summing only the detection signals of the detection coils C1 and C3 that form a pair, or by summing only the detection signals of the detection coils C5 and C7 that form a pair. In this configuration, the increase and decrease in harmonic components included in the detection signals of the detection coils C1 and C3 that form a pair are canceled out, and the increase and decrease in harmonic components included in the detection signals of the detection coils C5 and C7 that form another pair are also canceled out. Therefore, the amplitude signal (Sa1) does not include information about the displacement of the rotor 36, and functions as information indicating only the amplitude of the fundamental wave component.

[0171] Furthermore, in the embodiment described above, the control unit 55 can also function as a signal generating unit. In this configuration, the circuit group constituting the signal generating unit 53 is unnecessary, and the circuit configuration can be simplified.

[0172] Furthermore, in the embodiment described above, the present device 5 may also have the function of serving as a machine learning device 6. That is, for example, the control unit 55 may also have a data acquisition unit 630, a learning data set generation unit 631, and a learning unit 632. In this case, for example, the data acquisition unit 630 may also operate the present pump 1 by artificially displacing the rotor 36 (rotating shaft 31) under various driving conditions before the present pump 1 is shipped, thereby acquiring input data. In addition, the data acquisition unit 630 may also acquire, from the operation terminal device 9, a wear state corresponding to the input data determined by the operator and input into the operation terminal device 9. Furthermore, the present device 5 may be connected to an external device, and the data acquisition unit 630 may acquire learning data from the external device.

[0173] Furthermore, in the embodiment described above, the machine learning device 6 may also generate a single integrated learning model that functions as the first learning model M11 and the second learning model M12 by performing machine learning on the learning model using the composite signal (C1, C3, C5, C7), the differential signal (Sd), and the amplitude signal (Sa1) as input data and the radial wear state and thrust wear state as output data. In this case, the storage unit 56 may store the integrated learning model in place of the first learning model M11 and the second learning model M12, and the estimation unit 551 may use the integrated learning model to estimate the wear state.

[0174] Furthermore, in the embodiment described above, the pump 1 may not include the device 5. Specifically, for example, the detection coils C1 to C8 may be provided in the motor unit 3, and the device 5 may be provided separately from the pump unit 2, the motor unit 3, and the adapter 4. In this case, the device 5 may be connected to the detection coils C1 to C8 via a cable, or the device 5 and the motor unit 3 may be configured to have a communication function, so that the device 5 is connected to the motor unit 3 via a wireless communication link and is capable of receiving detection signals.

[0175] Motor bearing wear state estimation device (2)

[0176] Next, another embodiment of the motor bearing wear state estimation device of the present invention (hereinafter referred to as the "second embodiment") will be described, focusing on the differences from the previously described embodiment (hereinafter referred to as the "first embodiment"). The second embodiment differs from the first embodiment in the structure and method for acquiring the amplitude signal. In the following description, elements common to the first embodiment are denoted by the same reference numerals, and their description will be omitted.

[0177] Structure of motor bearing wear state estimation device (2)

[0178] Figure 17 This is a functional block diagram showing another embodiment (second embodiment) of the present device.

[0179] This device 5A includes eight detection coils C1 to C8, a connection unit 50, signal processing circuits 52c, 52d, 52e, and 52f, an A / D converter 54A, a control unit 55A, a storage unit 56A, a display unit 57, a D / A converter 58, and an offset processing unit 59. The A / D converter 54A, the control unit 55A, and the D / A converter 58 are constituted by, for example, a microcomputer.

[0180] The detection coils C1 and C3 constitute a set of radial detection coils, which are connected in such a way that their respective detection signals cancel each other. The detection coils C5 and C7 constitute another set of radial detection coils, which are connected in such a way that their respective detection signals cancel each other. Therefore, the composite signal (C1 C3) of the detection coils C1 and C3 represents the difference between the detection signals of the detection coils C1 and C3. Through this difference, the displacement of the front side of the rotor 36 in the radial direction is detected. That is, the difference represents the amount of wear of the bearing 32 in the radial direction, and the value of the difference is expressed as a voltage value. In addition, the composite signal (C5C7) of the detection coils C5 and C7 represents the difference between the detection signals of the detection coils C5 and C7. Through this difference, the displacement of the rear side of the rotor 36 in the radial direction is detected. That is, the difference represents the amount of wear of the bearing 33 in the radial direction, and the value of the difference is expressed as a voltage value.

[0181] Signal processing circuits 52e and 52f are composed of, for example, a filter circuit, a rectifier circuit, and an integrator circuit. The filter circuit is a bandpass filter with a passband (e.g., several hundred Hz to 2 kHz) that allows harmonic components to pass through. Signal processing circuit 52e is connected to detection coils C1 and C3, while signal processing circuit 52f is connected to detection coils C5 and C7. Signal processing circuits 52e and 52f perform prescribed signal processing (filtering, rectification, AC-DC conversion) on the corresponding composite signals (C1, C3, C5, C7), converting the composite signals (C1, C3, C5, C7) from AC to DC.

[0182] The A / D converter 54A is connected to the signal processing circuits 52 c to 52 f and the offset processing unit 59 , converts the analog signals inputted thereto into digital signals, and outputs the digital signals to the control unit 55A.

[0183] The control unit 55A controls the overall operation of the device 5A. The control unit 55A is comprised of, for example, a processor such as a CPU 55a, volatile memory such as a RAM 55b that functions as a workspace for the CPU 55a, and nonvolatile memory such as a ROM 55c that stores various information, including the estimation program and other control programs. The control unit 55A includes an acquisition unit 550, an estimation unit 551, a display control unit 552, and a signal generation unit 553.

[0184] This estimation program runs in the control unit 55A, and this estimation program cooperates with the hardware resources of the present device 5A to implement the present method described below. Furthermore, by having the processor (CPU 55a) constituting the control unit 55A execute this estimation program, it enables the processor to function as the acquisition unit 550, estimation unit 551, display control unit 552, and signal generation unit 553, thereby enabling the processor to execute this method. Furthermore, by having a computer execute this estimation program, it enables the computer to function as the present device 5A.

[0185] The signal generation unit 553 generates an amplitude signal (Sa2) based on the composite signal (C2C4) and the composite signal (C6C8). The specific operation of the signal generation unit 553 will be described later.

[0186] The amplitude signal (Sa2) contains almost no information about the displacement of the rotor 36 in the thrust direction. Similar to the amplitude signal (Sa1), it functions as information indicating the amplitude of the fundamental wave component. In the second embodiment, the amplitude signal (Sa2) is generated based on the composite signal (C2C4) and the composite signal (C6C8). The method for generating the amplitude signal (Sa2) will be described later. The amplitude signal (Sa2) is an example of the second amplitude signal in the present invention.

[0187] The storage unit 56A stores information necessary for the operation of the own device 5A (for example, the first learning model M21, the second learning model M22, offset information, etc.) The storage unit 56A is, for example, a nonvolatile memory such as an EEPROM or a flash memory.

[0188] Figure 18 It is a schematic diagram showing an example of information (first learning model M21 and second learning model M22) stored in the storage unit 56A.

[0189] The first learning model M21 is a model that has been machine-learned to output the radial wear state when inputting the composite signals (C1-C3), composite signals (C5-C7), and amplitude signal (Sa2). Similar to the first learning model M11, the first learning model M21 is pre-generated by the machine learning device 6 and stored in the storage unit 56A.

[0190] The second learning model M22 is a model that has been machine-learned to output the thrust wear state when the differential signal (Sd) and the amplitude signal (Sa2) are input. The second learning model M22, similar to the second learning model M12, is pre-generated by the machine learning device 6 and stored in the storage unit 56A.

[0191] While the machine learning device 6 generates the first learning model M21 and the second learning model M22, the testing device 7 generates the amplitude signal (Sa2) using the same method as the amplitude signal (Sa2) described later. Furthermore, the signal generating device 8 generates a simulated amplitude signal (Sa2) through calculation.

[0192] Amplitude signal generation method (generation process)

[0193] Next, a method for generating (generating processing) the amplitude signal (Sa2) in this embodiment will be described.

[0194] The signal generation unit 553 periodically acquires the composite signal (C2C4) and the composite signal (C6C8) during the operation of the pump 1. As described above, each of the acquired composite signals (C2C4, C6C8) is a DC digital signal.

[0195] Next, the signal generator 553 weights each composite signal (C2C4, C6C8) and adds the weighted composite signals (C2C4, C6C8) to generate an amplitude signal (Sa2). Weighting is performed by multiplying the voltage value of each composite signal (C2C4, C6C8) by a weighting coefficient. The generated amplitude signal (Sa2) is acquired by the acquisition unit 550.

[0196] Figure 19is a schematic diagram illustrating weighting performed by the signal generating unit 553. Figure 19 (a) shows the state before weighting, Figure 19 (b) shows the state after weighting.

[0197] In this figure, the vertical axis represents the signal level (voltage), and the horizontal axis represents the displacement of the rotor 36 in the thrust direction (the amount of wear on the bearings 32 and 33). For ease of explanation, the state before weighting is shown with a thin line in (b) of the figure. For ease of explanation, the changes in voltage values ​​are emphasized in this figure.

[0198] As described above, the voltage value of each composite signal (C2C4, C6C8) decreases in a curve shape relative to the displacement of the rotor 36 in one direction in the thrust direction, and remains almost constant relative to the displacement in the other direction. Moreover, this curve-shaped change indicates the displacement information of the rotor 36 in the thrust direction. Therefore, if the composite signal (C2C4) and the composite signal (C6C8) are simply added together, the added signal contains the displacement information of the rotor 36 in the thrust direction. As a result, Figure 19 As shown in (a), the change in the voltage value of the summed signal relative to the position of the rotor 36 (single-dot chain line) becomes a slightly convex curve and is not constant. In addition, the detection sensitivity (voltage value change) of the front detection coils C2 and C4 and the rear detection coils C6 and C8 to the displacement of the rotor 36 is affected by the initial position and magnetic circuit determined by the model of this pump 1. Therefore, the detection sensitivity of the detection coils C2 and C4 may be different from the detection sensitivity of the detection coils C6 and C8. Therefore, as Figure 19 As shown in (a) of the figure, a state may also occur in which the change of the composite signal (C2C4) is greater than the change of the composite signal (C6C8). In this case, as shown in (a) of the figure, the voltage value of the added signal is in the shape of a curve and changes in an inclined direction to the upper right in the front-to-back direction. The above-mentioned "weighting coefficient" is a coefficient that is multiplied by the voltage value of each composite signal (C2C4, C6C8) to offset the change in the voltage value based on this magnetic flux change (displacement information). That is, weighting is performed in a manner that offsets (reduces) the curved changes (and similar inclined changes) in the voltage value of each composite signal (C2C4, C6C8). The weighting coefficient is set to, for example, "1.0" or a decimal less than "1.0" (except 0: for example, Figure 19 In the example shown, "0.5" is set for the composite signal (C2C4), and "1.0" is set for the composite signal (C6C8).

[0199] In this case, the smaller the weighting coefficient is (the closer it is to 0), the smaller the change in the weighted voltage value is. Therefore, for example, when the change in the voltage value of the composite signal (C2C4) is smaller than the change in the composite signal (C6C8), the value of the weighting coefficient of the composite signal (C2C4) is set to a value greater than the value of the weighting coefficient of the composite signal (C6C8) (a value close to "1.0"). In addition, for example, when the two changes are almost the same, the values ​​of the two weighting coefficients are set to the same value. Furthermore, for example, when the voltage value of the added signal changes in an inclined manner, the value of the weighting coefficient on the side with a higher contribution to the inclination is set to a value smaller than the value of the weighting coefficient on the side with a lower contribution to the inclination (a value close to "0"). The result of weighting is that, as Figure 19 As shown in (b), in the composite signal (C2C4) that has been weighted less than "1.0", the change in voltage value (i.e., displacement information) becomes smaller, and the change in voltage value is corrected to be almost constant (close to linear) regardless of the position of the rotor 36. Therefore, the curved and tilt-like changes in the voltage value of the added signal also become smaller, and the change is almost constant. By setting the weighting coefficients in this way, the change in voltage value of each composite signal (C2C4, C6C8) becomes smaller, and the curved changes (tilt-like changes) also become smaller. The weighting coefficients are pre-measured under specified reference driving conditions, for example, before the pump 1 is shipped, and are stored in the storage unit 56A.

[0200] It should be noted that, in the present invention, weighting may also be performed by using a function that cancels out displacement information. In this case, the storage unit 56A stores the function instead of the weighting coefficient.

[0201] In addition, in the present invention, the weighting coefficients of the front-side composite signal (C2C4) and the rear-side composite signal (C6C8) may be different or the same.

[0202] Operation of shielded pump (motor bearing wear state estimation device (2))

[0203] The operation of the pump 1 in the second embodiment is similar to that of the pump 1 in the first embodiment, except that the method for generating the amplitude signal (Sa2) differs from the method for generating the amplitude signal (Sa1) in the first embodiment, and the amplitude signal (Sa2) used in the radial wear state estimation process (ST1) and the thrust wear state estimation process (ST2) differ from the amplitude signal (Sa1) in the first embodiment. Specifically, the pump 1 in the second embodiment repeats the radial wear state estimation process (ST1) and the thrust wear state estimation process (ST2) periodically (for example, every few seconds). In this case, the estimation unit 551 uses the amplitude signal (Sa2) instead of the amplitude signal (Sa1).

[0204] Summary (2)

[0205] According to the second embodiment described above, each of the plurality of detection coils C1 to C8 outputs a detection signal indicating a change in magnetic flux corresponding to a change in the mechanical position of the rotor 36 relative to the stator 37 .

[0206] The multiple detection coils C1 to C8 include: multiple detection coils C1, C3, C5, and C7, which detect changes in magnetic flux in the radial direction; and multiple detection coils C2, C4, C6, and C8, which detect changes in magnetic flux in the thrust direction. The detection signal includes a fundamental component based on the drive frequency of the motor unit 3. The present device 5A includes an acquisition unit 550, an estimation unit 551, a signal generation unit 553, and a storage unit 56A. The storage unit 56A stores a first learning model M21 and a second learning model M22. The first learning model M21 is machine-learned to output the radial wear status when input with the composite signal (C1, C3), the composite signal (C5, C7), and the amplitude signal (Sa2). The second learning model M22 is machine-learned to output the thrust wear status when input with the differential signal (Sd) and the amplitude signal (Sa2). The amplitude signal (Sa2) is generated based on the composite signal (C2, C4, C6, C8). With this configuration, the present device 5A uses the amplitude signal (Sa2), which primarily includes the amplitude (signal level) of the fundamental wave component that increases or decreases proportionally with changes in the driving conditions, as well as the composite signals (C1, C3, C5, C7) representing the wear amount (wear state) of the bearings 32 and 33, and the differential signal (Sd) as input data for the learned model. This allows the device 5A to estimate the wear state corresponding to the driving conditions. Therefore, even if the driving conditions change, the present device 5A can estimate the wear state corresponding to the driving conditions without requiring manual mechanical intervention.

[0207] In addition, according to the second embodiment described above, the signal generating unit 553 generates an amplitude signal (Sa2) by weighting the composite signal (C2C4) and the composite signal (C6C8) respectively to offset the magnetic flux change based on the wear of the bearings 32 and 33 in the thrust direction, and adding the weighted composite signal (C2C4) and the composite signal (C6C8). According to this structure, the displacement information in the thrust direction is almost offset in the amplitude signal (Sa2). Therefore, the present device 5A can use the amplitude signal (Sa2) that mainly includes the amplitude (signal level) of the fundamental wave component that increases and decreases in proportion to the increase or decrease of the driving condition as input data for the learned learning model. Therefore, the present device 5A can estimate the wear state corresponding to the driving condition without being affected by the displacement information.

[0208] Note that, in the second embodiment described above, the signal generating unit 553 may generate an average value of the two composite signals (C2C4, C6C8) as the amplitude signal instead of generating the sum of the two composite signals (C2C4, C6C8).

[0209] Furthermore, in the second embodiment described above, the voltage value of the sum of the composite signal (C2C4) and the composite signal (C6C8) relative to the position of the rotor 36 varies by approximately 7% to 10%. Therefore, even if the signal generator 553 generates the amplitude signal (Sa2) without weighting the composite signals (C2C4, C6C8), the impact on the displacement information estimation is likely to be limited. Therefore, the signal generator 553 may generate the amplitude signal (Sa2) without weighting the composite signals (C2C4, C6C8). In this configuration, the influence of the displacement information remains greater in the amplitude signal (Sa2) than in the second embodiment, resulting in a slight decrease in the accuracy of the wear state estimation as wear in the thrust direction increases. However, as described above, the change in the sum signal exhibits a mountainous shape, so the estimation accuracy decreases in the direction of slightly more sensitive wear state detection. Therefore, the present device 5A can fully function as a device for estimating the wear state of the bearings 32 and 33. Therefore, even with this configuration, even if the driving conditions change, the present device 5A can estimate the wear state corresponding to the driving conditions at a level required by the present device 5A without requiring human intervention in mechanical operations.

[0210] Other implementations

[0211] It should be noted that in each of the embodiments described above, regarding the initial position of the rotor 36, the center of the rotor 36 in the thrust direction may not coincide with the center of the stator 37 due to manufacturing errors, positional tolerances, etc. In other words, for example, in the initial position, the rotor 36 may be arranged offset relative to the stator 37 in the thrust direction.

[0212] In addition, in each embodiment described above, the number of detection coils C1 to C8 is not limited to “8” as long as the present invention can be implemented.

[0213] Furthermore, in each of the embodiments described above, the control unit 55, 55A may also be composed of a processor such as a DSP (Digital Signal Processor) or a (GP) GPU ((General Purpose) Graphics Processing Unit) instead of the CPU 55a.

[0214] In the above-described embodiments, the present method is executed by the control unit 55 or 55A. Alternatively, the present method may be executed by an external computing device (eg, a computer) connected to the present device 5 or 5A.

[0215] Furthermore, in each of the embodiments described above, part or all of the information stored in ROM 55c may be stored in storage units 56 and 56A. Alternatively, part of the information stored in storage units 56 and 56A (e.g., first learning models M11 and M21, and second learning models M12 and M22) may be stored in ROM 55c. In the latter case, ROM 55c can function as the storage unit in the present invention.

[0216] Modes for Carrying Out the Invention

[0217] Next, the embodiments of the present invention understood from the above-described embodiments will be described below by citing the terms and reference numerals described in the embodiments.

[0218] A first embodiment of the present invention is a motor bearing wear state estimation device (e.g., present device 5, 5A), which estimates the wear state of bearings (e.g., bearings 32, 33) supporting a rotating shaft (e.g., rotating shaft 31) of a motor (e.g., motor unit 3) of a shielded pump (e.g., present pump 1) based on respective detection signals of a plurality of detection coils (e.g., detection coils C1 to C8) that detect magnetic flux changes corresponding to mechanical position changes of a rotor (e.g., rotor 36) relative to a stator (e.g., stator 37), wherein the plurality of detection coils are respectively mounted on the stator, the detection signals include a fundamental wave component based on a drive frequency of the motor, and the plurality of detection coils include: a plurality of radial detection coils (e.g., detection coils C1, C3, C5, C7) that detect magnetic flux changes in the radial direction of the rotating shaft; and a plurality of thrust detection coils (e.g., detection coils C2, C4, C6, C8) that detect magnetic flux changes in the radial direction of the rotating shaft. The magnetic flux changes of the rotating shaft in the thrust direction, the plurality of radial detection coils include: a first radial detection coil (for example, detection coils C1 and C3) constituting one group; and a second radial detection coil (for example, detection coils C5 and C7) constituting another group, the plurality of thrust detection coils include: a first thrust detection coil (for example, detection coils C2 and C4) constituting one group; and a second thrust detection coil (for example, detection coils C6 and C8) constituting another group, the motor bearing wear state estimation device includes: a storage unit (for example, storage unit 56, 56A) which stores a learned first learning model (for example, first learning model M11, M21) and a learned second learning model (for example, second learning model M12, M22), the learned first learning model is used as an input for a first differential signal (for example, a composite signal (C1)) representing the difference between the detection signals of each of the first radial detection coils. C3)), a second differential signal (e.g., a composite signal (C5C7)) representing the difference between the detection signals of the respective second radial detection coils, and an amplitude signal (e.g., an amplitude signal (Sa1, Sa2)) representing the amplitude of the fundamental wave component, are input to the second learning model, and machine learning is performed in a manner such that the wear state in the thrust direction is output when a thrust differential signal (e.g., a differential signal (Sd)) representing the difference between the detection signals of the respective first thrust detection coils and a second composite signal (e.g., a composite signal (C6C8)) synthesized from the detection signals of the respective second thrust detection coils is input, and the amplitude signal is input; an acquisition unit (e.g., an acquisition unit 550) that acquires the first differential signal, the second differential signal, the thrust differential signal, and the amplitude signal;and an estimating unit (e.g., estimating unit 551) that estimates the wear state by inputting the first difference signal, the second difference signal, and the amplitude signal acquired by the acquiring unit into the first learning model, and inputting the thrust difference signal and the amplitude signal acquired by the acquiring unit into the second learning model, wherein the amplitude signal is a first amplitude signal (Sa1) generated based on the detection signals output from the first radial detection coil and / or the second radial detection coil, or a second amplitude signal (Sa2) generated based on the first composite signal and the second composite signal.

[0219] According to this configuration, even if the driving conditions are changed, the device can estimate the wear state corresponding to the driving conditions without requiring human intervention in mechanical operations.

[0220] The second embodiment of the present invention is a motor bearing wear state estimation device according to the first embodiment, wherein the motor bearing wear state estimation device has a signal generating unit (for example, signal generating unit 53, 553), and the signal generating unit generates the first amplitude signal (for example, amplitude signal (Sa1)) based on the detection signal of the first radial detection coil and / or the second radial detection coil, or generates the second amplitude signal (for example, amplitude signal (Sa2)) based on the first composite signal and the second composite signal.

[0221] According to this configuration, the acquisition unit can acquire the amplitude signal as input data of the learning model at any timing.

[0222] The third embodiment of the present invention is a motor bearing wear state estimation device (for example, the present device 5) according to the second embodiment, wherein the signal generating unit (for example, the signal generating unit 53) generates the first amplitude signal (for example, the amplitude signal (Sa1)) by adding the respective detection signals of the first radial detection coil and / or the second radial detection coil.

[0223] According to this structure, the present device can estimate the thrust direction wear state and the radial direction wear state based on the driving conditions without being affected by the displacement information.

[0224] A fourth aspect of the present invention is the motor bearing wear state estimation device according to the third aspect, wherein the signal generation unit adds the detection signals of all the radial detection coils.

[0225] According to this structure, the present device can accurately estimate the thrust direction wear state and the radial direction wear state according to the driving conditions without being affected by the displacement information.

[0226] The fifth embodiment of the present invention is a motor bearing wear state estimation device (for example, the present device 5A) according to the second embodiment, wherein the signal generating unit (for example, the signal generating unit 553) generates the second amplitude signal (for example, the amplitude signal (Sa2)) by weighting the first synthetic signal and the second synthetic signal respectively to offset the information showing the magnetic flux change based on the wear of the bearing in the thrust direction, and adding the weighted first synthetic signal to the second synthetic signal.

[0227] According to this configuration, the present device can estimate the wear state according to the driving conditions with little influence from the displacement information.

[0228] A sixth embodiment of the present invention is a bearing wear state estimation method (e.g., radial wear state estimation processing (ST1), thrust wear state estimation processing (ST2)), which is executed by a motor bearing wear state estimation device (e.g., present device 5, 5A), wherein the motor bearing wear state estimation device estimates the wear state of bearings (e.g., bearings 32, 33) of a rotating shaft (e.g., rotating shaft 31) supporting the rotor based on respective detection signals of a plurality of detection coils (e.g., detection coils C1 to C8) that detect magnetic flux changes corresponding to mechanical position changes of a rotor (e.g., rotor 36) of a motor (e.g., motor unit 3) of a shielded pump (e.g., present pump 1) relative to a stator (e.g., stator 37), wherein a plurality of The detection coils are respectively installed on the stators, and the detection signals include a fundamental wave component based on the driving frequency of the motor. The plurality of detection coils include: a plurality of radial detection coils (for example, detection coils C1, C3, C5, and C7) that detect the magnetic flux changes in the radial direction of the rotating shaft; and a plurality of thrust detection coils (for example, detection coils C2, C4, C6, and C8) that detect the magnetic flux changes in the thrust direction of the rotating shaft. The plurality of radial detection coils include: a first radial detection coil constituting one group; and a second radial detection coil constituting another group. The plurality of thrust detection coils include: a first thrust detection coil constituting one group; and a second thrust detection coil constituting another group. The motor shaft The bearing wear state estimation device includes a storage unit (for example, storage unit 56, 56A), which stores a learned first learning model and a learned second learning model, wherein the learned first learning model performs machine learning in a manner that outputs the wear state in the radial direction when a first differential signal representing the difference between the detection signals of each of the first radial detection coils, a second differential signal representing the difference between the detection signals of each of the second radial detection coils, and an amplitude signal representing the amplitude of the fundamental wave component are input, and the learned second learning model performs machine learning in a manner that outputs the wear state in the radial direction when a first synthetic signal representing the synthesis of the detection signals of each of the first thrust detection coils and an amplitude signal representing the amplitude of the fundamental wave component are input. When a thrust differential signal which is a difference between the detection signals of the respective first and second thrust detection coils and a second synthetic signal synthesized by the detection signals of the respective first and second thrust detection coils, and the amplitude signal are obtained, a method of outputting the wear state in the thrust direction is performed, and machine learning is performed, the amplitude signal is a first amplitude signal generated based on the detection signals of the respective first radial detection coils and / or the second radial detection coils, or a second amplitude signal generated based on the first synthetic signal and the second synthetic signal, the bearing wear state estimation method comprises: an acquisition step (for example, acquisition steps ST11 and ST21), the motor bearing wear state estimation device acquiring the first differential signal, the second differential signal, the thrust differential signal and the amplitude signal;and an estimation step (e.g., estimation steps ST12 and ST22), wherein the motor bearing wear state estimation device inputs the acquired first differential signal, the acquired second differential signal, and the acquired amplitude signal into the first learning model, and inputs the acquired thrust differential signal and the acquired amplitude signal into the second learning model to estimate the wear state.

[0229] According to this configuration, even if the driving conditions are changed, the device can estimate the wear state corresponding to the driving conditions without requiring human intervention in mechanical operations.

[0230] A seventh embodiment of the present invention is a bearing wear state estimation program that causes a computer to function as the motor bearing wear state estimation device according to the first embodiment.

[0231] According to this configuration, even if the driving conditions are changed, the device can estimate the wear state corresponding to the driving conditions without requiring human intervention in mechanical operations.

[0232] The eighth embodiment of the present invention is a shielded pump (for example, the present pump 1), wherein the shielded pump comprises: a motor (for example, the motor part 3), the motor having a rotor (for example, the rotor 36), a stator (for example, the stator 37) that rotates the rotor, and a rotating shaft (for example, the rotating shaft 31) that rotates together with the rotor; bearings (for example, bearings 32, 33) that support the rotating shaft; a plurality of detection coils (for example, detection coils C1 to C8) that detect changes in magnetic flux corresponding to changes in the mechanical position of the rotor relative to the stator; and a motor bearing wear state estimation device according to any one of the first to fifth embodiments (for example, the present device 5, 5A), which estimates the wear state of the bearing based on the detection signals of each of the plurality of detection coils.

[0233] According to this configuration, even if the driving conditions are changed, the pump can estimate the wear state corresponding to the driving conditions without requiring human intervention in mechanical operations.

[0234] Description of Reference Numerals

[0235] 1: Shielded pump

[0236] 3: Motor

[0237] 31: Rotation axis

[0238] 32: Bearing

[0239] 33: Bearing

[0240] 36: Rotor

[0241] 37: stator

[0242] 5: Motor bearing wear state estimation device

[0243] 53: Signal generation department

[0244] 550: Acquisition Department

[0245] 551: Estimation Department

[0246] 56: Storage

[0247] 5A: Motor bearing wear state estimation device

[0248] 553: Signal generation department

[0249] 56A: Storage

[0250] C1~C8: detection coil

[0251] M11: First Learning Model

[0252] M12: Second Learning Model

[0253] M21: First Learning Model

[0254] M22: Second Learning Model

Claims

1. A motor bearing wear state estimation device, which estimates the wear state of a bearing supporting a rotating shaft of a canned motor pump based on detection signals from a plurality of detection coils that detect changes in magnetic flux corresponding to changes in the mechanical position of the rotor relative to the stator of the motor, wherein: The plurality of detection coils are respectively installed on the stator. The detection signal includes a fundamental wave component based on the driving frequency of the motor, The plurality of detection coils include: a plurality of radial detection coils for detecting changes in the magnetic flux of the rotating shaft in a radial direction; and a plurality of thrust detection coils for detecting changes in the magnetic flux of the rotating shaft in a thrust direction; The plurality of radial detection coils include: a first radial detection coil constituting a set; and The second radial detection coil constituting another set, The plurality of thrust detection coils include: a first thrust detection coil constituting a set; and The second thrust detection coil constituting another group, The motor bearing wear state estimation device comprises: a storage unit storing a learned first learning model and a learned second learning model, wherein the learned first learning model is machine-learned in such a manner that, upon inputting a first differential signal representing a difference between the detection signals of the first radial detection coils, a second differential signal representing a difference between the detection signals of the second radial detection coils, and an amplitude signal representing the amplitude of the fundamental wave component, the model outputs the wear status in the radial direction; and the learned second learning model is machine-learned in such a manner that, upon inputting a thrust differential signal representing a difference between a first composite signal synthesized from the detection signals of the first thrust detection coils and a second composite signal synthesized from the detection signals of the second thrust detection coils, the model outputs the wear status in the thrust direction; an acquisition unit that acquires the first differential signal, the second differential signal, the thrust differential signal, and the amplitude signal; and an estimating unit that estimates the wear state by inputting the first differential signal, the second differential signal, and the amplitude signal acquired by the acquiring unit into the first learning model, and inputting the thrust differential signal and the amplitude signal acquired by the acquiring unit into the second learning model, The amplitude signal is a first amplitude signal generated based on the detection signal of each of the first radial detection coil and / or the second radial detection coil, or a second amplitude signal generated based on the first composite signal and the second composite signal.

2. The motor bearing wear state estimation device according to claim 1, wherein: The motor bearing wear state estimation device has a signal generating unit that generates the first amplitude signal based on the detection signal of each of the first radial detection coil and / or the second radial detection coil, or generates the second amplitude signal based on the first composite signal and the second composite signal.

3. The motor bearing wear state estimation device according to claim 2, wherein: The signal generating unit generates the first amplitude signal by adding the detection signals of the first radial detection coil and / or the second radial detection coil.

4. The motor bearing wear state estimation device according to claim 3, wherein: The signal generating unit adds the detection signals of all the radial detection coils.

5. The motor bearing wear state estimation device according to claim 2, wherein: The signal generating unit generates the second amplitude signal by weighting the first composite signal and the second composite signal so as to cancel the change in magnetic flux caused by wear of the bearing in the thrust direction, and adding the weighted first composite signal and the second composite signal.

6. A method for estimating a bearing wear state, the method being performed by a motor bearing wear state estimating device, the motor bearing wear state estimating device estimating the wear state of a bearing supporting a rotating shaft of a canned motor pump based on detection signals from a plurality of detection coils that detect changes in magnetic flux corresponding to changes in the mechanical position of the rotor relative to the stator, wherein: The plurality of detection coils are respectively installed on the stator. The detection signal includes a fundamental wave component based on the driving frequency of the motor, The plurality of detection coils include: a plurality of radial detection coils for detecting changes in the magnetic flux of the rotating shaft in a radial direction; and a plurality of thrust detection coils for detecting changes in the magnetic flux of the rotating shaft in a thrust direction; The plurality of radial detection coils include: a first radial detection coil constituting a set; and The second radial detection coil constituting another set, The plurality of thrust detection coils include: a first thrust detection coil constituting a set; and The second thrust detection coil constituting another group, The motor bearing wear state estimation device includes a storage unit, which stores a learned first learning model and a learned second learning model. The learned first learning model performs machine learning in a manner that outputs the wear state in the radial direction when inputting a first differential signal representing the difference between the detection signals of each of the first radial detection coils, a second differential signal representing the difference between the detection signals of each of the second radial detection coils, and an amplitude signal representing the amplitude of the fundamental wave component. The learned second learning model performs machine learning in a manner that outputs the wear state in the thrust direction when inputting a thrust differential signal representing the difference between a first synthetic signal synthesized by the detection signals of each of the first thrust detection coils and a second synthetic signal synthesized by the detection signals of each of the second thrust detection coils, and the amplitude signal. The amplitude signal is a first amplitude signal generated based on the detection signal of each of the first radial detection coil and / or the second radial detection coil, or a second amplitude signal generated based on the first composite signal and the second composite signal. The bearing wear state estimation method comprises: an acquisition step, wherein the motor bearing wear state estimation device acquires the first differential signal, the second differential signal, the thrust differential signal, and the amplitude signal; and In the estimation step, the motor bearing wear state estimation device inputs the acquired first differential signal, the acquired second differential signal and the acquired amplitude signal into the first learning model, and inputs the acquired thrust differential signal and the acquired amplitude signal into the second learning model to estimate the wear state.

7. A bearing wear state estimation program, wherein: The bearing wear state estimation program causes a computer to function as the motor bearing wear state estimation device according to claim 1 .

8. A canned motor pump, wherein: The canned motor pump has: a motor including a rotor, a stator for rotating the rotor, and a rotating shaft for rotating together with the rotor; a bearing, the bearing supporting the rotating shaft; a plurality of detection coils configured to detect a change in magnetic flux corresponding to a change in the mechanical position of the rotor relative to the stator; as well as According to any one of claims 1 to 5, the motor bearing wear state estimation device estimates the wear state of the bearing based on the detection signals of each of the plurality of detection coils.

Images