WO2019044496A1

WO2019044496A1 - Method and device for monitoring state of rolling bearing

Info

Publication number: WO2019044496A1
Application number: PCT/JP2018/030301
Authority: WO
Inventors: 正嗣北井
Original assignee: Ntn株式会社
Priority date: 2017-08-31
Filing date: 2018-08-14
Publication date: 2019-03-07
Also published as: JP2019045241A

Abstract

According to the present invention, the remaining service life of a rolling bearing is estimated. A method for monitoring the state of a rolling bearing according to the present invention comprises: a step for creating a classifier by means of machine learning; and a step for monitoring the state of the rolling bearing. The step for creating the classifier includes: steps (S11-S13) for calculating a set of features about first to third vibration data; and a step (S14) for creating the classifier by machine learning the first to third vibration data. The first vibration data is vibration data about the rolling bearing in a state in which damage of a first length is formed in the circumferential direction of a raceway surface. The second vibration data is vibration data about the rolling bearing in a state in which damage of a second length shorter than the first length is formed in the circumferential direction of the raceway surface. The third vibration data is vibration data about the rolling bearing for when the raceway surface is normal. The step for monitoring the state includes a step for using the classifier to calculate the remaining service life until the length of the damage reaches the second length.

Description

Method and apparatus for monitoring state of rolling bearing

The present invention relates to a method and a device for monitoring the state of a rolling bearing.

Conventionally, a method of monitoring the state of a rolling bearing is known. For example, Japanese Patent Application Laid-Open No. 2004-347401 (Patent Document 1) discloses a rolling bearing having a simple configuration efficiently and accurately by using a background noise component having no peak such as a rotational speed proportional component in a frequency spectrum. A diagnostic method of a rolling bearing capable of determining the presence or absence of an abnormality is disclosed.

Unexamined-Japanese-Patent No. 2004-347401

Japanese Patent Laying-Open No. 2004-347401 (Patent Document 1) discloses a configuration for detecting damage when periodical peaks occur in the frequency spectrum of vibration data of a rolling bearing.

In order to properly determine the rolling bearing replacement time, it is necessary to estimate the remaining life of the rolling bearing according to the degree of damage. However, in Japanese Patent Laid-Open No. 2004-347401 (Patent Document 1), the estimation of the remaining life of the rolling bearing according to the degree of damage is not taken into consideration.

The present invention has been made to solve the problems as described above, and its object is to estimate the remaining life of a rolling bearing according to the degree of damage.

The rolling bearing state monitoring method according to the present invention includes the steps of creating a classifier by machine learning and monitoring the rolling bearing state. The rolling bearing includes a rotating wheel, a stationary wheel, and a plurality of rolling elements. The plurality of rolling elements are disposed between the rotating wheel and the stationary wheel, and move along the raceway surface of the stationary wheel as the rotating wheel rotates. The step of creating a classifier includes the steps of calculating a feature amount set of the first to third vibration data, and machine learning of the feature amount set of the feature amount set of the first to third vibration data to create a classifier. And step. The first vibration data is vibration data of the rolling bearing in a state in which the damage of the first length is formed in the circumferential direction of the raceway surface. The second vibration data is vibration data of the rolling bearing in a state in which damage of a second length smaller than the first length is formed in the circumferential direction of the raceway surface. The third vibration data is vibration data of the rolling bearing when the raceway surface is normal. The number of times of loading until the first length of damage is formed in the circumferential direction of the raceway surface is the first number of times of loading. The number of times of loading until the second length of damage is formed in the circumferential direction of the raceway surface is the second number of times of loading less than the first number of times of loading. The step of monitoring the state includes the steps of calculating a feature amount set of the fourth vibration data, calculating the first to third precisions, and calculating the remaining life of the rolling bearing. The fourth vibration data is vibration data of the rolling bearing measured during monitoring. The first to third relevance factors are calculated by the classifier. The first to third adaptation rates are adaptation rates between the feature amount set of the fourth vibration data and the feature amount sets of the first to third oscillation data. The remaining life of the rolling bearing is calculated using the first and second number of loads and the first to third fitness factors. The remaining life is the number of remaining loads of the rolling bearing until the first length of damage is formed in the circumferential direction of the raceway surface.

According to the state monitoring method of a rolling bearing according to the present invention, it is possible to estimate the remaining life of the rolling bearing by using a classifier created by machine learning of a feature amount set of vibration data.

It is a figure which shows collectively the sectional view of the rolling bearing monitored by the state monitoring apparatus which concerns on embodiment, and a state monitoring apparatus. It is a functional block diagram of the state monitoring apparatus of FIG. It is a figure which shows the feature-value set calculated by the control part. It is a flowchart which shows the outline | summary of the estimation process of the remaining life of the rolling bearing performed by a control part. It is a figure which shows typically a mode that load area center vicinity is normal. It is a figure which shows typically a mode that damage of the damage level 1 has arisen in the load area center vicinity. It is a figure which shows typically a mode that the damage of the damage level 2 has arisen in the load area center vicinity. It is a figure which shows the result of having actually measured the relationship between the effective value of the vibration data of a rolling bearing, and the load frequency. It is a flowchart which shows concretely the flow of the process performed in a learning period. It is a flowchart which shows concretely the flow of the process performed in a monitoring period. It is a figure which shows the relationship between the load frequency of a rolling bearing, and the remaining life estimated by the state monitoring apparatus which concerns on embodiment. The relationship between the damage level 0 to 2 and the remaining life in each of the case where the rolling bearing is actually operated to form a damage and the case where the damage is artificially formed is shown together.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same or corresponding portions are denoted by the same reference characters and description thereof will not be repeated.

FIG. 1 is a view showing a cross-sectional view of a state monitoring device 1 and a rolling bearing 10 monitored by the state monitoring device 1 according to the embodiment. As shown in FIG. 1, the rolling bearing 10 includes an inner ring 12, an outer ring 14, and a plurality of rolling elements 18. Rolling bearing 10 includes, for example, self-aligning roller bearings, tapered roller bearings, cylindrical roller bearings, and ball bearings. The rolling bearing 10 may be a single row or multiple rows.

The inner ring 12 is a stationary ring externally fitted to the non-rotating shaft 11. The outer ring 14 is a rotating ring provided on the outer peripheral side of the inner ring 12 and integrally rotating with a rotating body (not shown). Each of the plurality of rolling elements 18 is interposed between the inner ring 12 and the outer ring 14 while being held at equal intervals to the adjacent rolling elements 18 by a cage (not shown).

The state monitoring device 1 monitors the state of the rolling bearing by measuring the physical quantity of the rolling bearing 10 by the vibration sensor. Examples of physical quantities of the rolling bearing 10 measured by the vibration sensor include acceleration, velocity, displacement, sound, AE (Acoustic Emission), and power.

A load area that receives a radial load from the plurality of rolling elements 18 is formed on the upper side in the vertical direction (X-axis direction) of the outer peripheral surface (track surface) of the inner ring 12 which is a stationary wheel. The radial load is maximum at the center of the load area. Therefore, damage to the raceway surface often occurs first at the center of the load area.

The vibration of the outer ring 14 may increase in accordance with the progress of the damage generated on the raceway surface. As a result, for example, an adverse effect such as an abnormal contact between a component supported by the rolling bearing 10 and a component adjacent thereto may occur. In order to use the rolling bearing 10 as long as possible, it is necessary to estimate the remaining life of the rolling bearing 10 according to the degree of damage to the raceway surface, and to appropriately determine the replacement time of the rolling bearing 10.

Therefore, in the embodiment, a set of feature quantities and a plurality of machine-learned sets are machine-learned by machine-learning feature sets of vibration data of a rolling bearing in which damage having different lengths in the circumferential direction is formed on the raceway surface. A classifier is calculated to calculate the matching rate with the feature quantity set of. The remaining life of the rolling bearing 10 can be estimated by using the classifier.

FIG. 2 is a functional block diagram of the state monitoring device 1 of FIG. As shown in FIG. 2, the state monitoring device 1 includes a vibration sensor 20, an LPF (Low Pass Filter) 31, a BPF (Band Pass Filter) 32, a HPF (High Pass Filter) 33, and a control unit 40. , Storage unit 50. The vibration sensor 20 measures vibration data of the rolling bearing 10, and outputs the measured data to the control unit 40, the LPF 31, the BPF 32, and the HPF 33.

The LPF 31 outputs vibration data of a frequency smaller than a predetermined frequency to the control unit 40. In the embodiment, the LPF 31 outputs vibration data of a frequency of less than 1000 Hz to the control unit 40.

The BPF 32 outputs vibration data of a frequency within a predetermined range to the control unit 40. In the embodiment, the BPF 32 outputs vibration data of a frequency of 1000 Hz or more and 5000 Hz or less to the control unit 40.

The HPF 33 outputs vibration data of a frequency higher than a predetermined frequency to the control unit 40. In the embodiment, the HPF 33 outputs vibration data of a frequency higher than 5000 Hz to the control unit 40.

The control unit 40 machine-learnings the feature value set of the vibration data of the rolling bearing in which the damage whose circumferential length is different in the circumferential direction is formed in the learning period, and creates a classifier. The classifier calculates the matching rate between a certain feature amount set and each machine-learned feature amount set. Classifiers can include, for example, support vector machines, neural networks, naive Bayes, and decision trees. The control unit 40 calculates a feature amount set of vibration data measured at the sampling time in the monitoring period, and calculates the remaining life of the rolling bearing. The control unit 40 includes a computer such as a central processing unit (CPU).

The storage unit 50 stores, for example, vibration data, feature quantities calculated from the vibration data, and a classifier created by machine learning.

FIG. 3 is a diagram showing a feature amount set calculated by the control unit 40. In the embodiment, the feature amount set includes the feature amounts F1 to F60 shown in FIG.

Referring to FIGS. 2 and 3, the control unit 40 determines each of the time domain (F1 to F5), the frequency domain (F6 to F10), and the quefrance domain (F11 to F15) from the vibration data from the vibration sensor 20. In the region, the effective value, the maximum value, the crest factor, the kurtosis, and the skewness are calculated as feature amounts.

From the vibration data from the LPF 31, the control unit 40 determines the effective value, maximum value, crest factor, peak in each of the time domain (F16 to F20), the frequency domain (F21 to F25), and the quefrance domain (F26 to F30) The degree and skewness are calculated as feature quantities.

From the vibration data from the BPF 32, the control unit 40 determines the effective value, maximum value, crest factor, and peak in each of the time domain (F31 to F35), the frequency domain (F36 to F40), and the quefrance domain (F41 to F45) The degree and skewness are calculated as feature quantities.

From the vibration data from the HPF 33, the control unit 40 determines the effective value, maximum value, crest factor, peak in each of the time domain (F46 to F50), the frequency domain (F51 to F55), and the quefrance domain (F56 to F60) The degree and skewness are calculated as feature quantities.

FIG. 4 is a flowchart showing an outline of the estimation processing of the remaining life of the rolling bearing performed by the control unit 40. Hereinafter, the step is simply described as S. As shown in FIG. 4, the control unit 40 creates a classifier at S1. S1 is a process performed in the learning period. The control unit 40 calculates the remaining life of the rolling bearing using a classifier at S2. S2 is a process performed in the monitoring period. The classifier created by machine learning may be created at the start of the monitoring period, and the learning period and the monitoring period may not be continuous.

In the embodiment, the circumferential length (damage length) of the damage formed on the raceway surface is classified into three levels of damage levels 0 to 2. FIG. 5 is a view schematically showing how the center of the load area is normal. FIG. 6 is a view schematically showing how damage at damage level 1 occurs near the center of the load area. FIG. 7 is a view schematically showing how damage at level 2 damage occurs near the center of the load area. In FIGS. 5 to 7, the center of the load area of the rolling bearing 10 is enlarged and shown, and the outer peripheral surface of the inner ring 12 and the inner peripheral surface of the outer ring 14 are straight for ease of illustration. It is drawn on.

In the embodiment, as shown in FIG. 5, the damage level is 0 when the vicinity of the center of the load area is normal. The condition that the center of the load area is normal means that no damage occurs near the center of the load area or the length of the damage is smaller than a threshold. The ratio of the damage length L1 of damage level 1 shown in FIG. 6 to the damage length L2 of damage level 2 shown in FIG. 7 is 40% or more and 60% or less. As shown in FIG. 7, when a plurality of rolling elements 18 pass damage level 2 damage, two or more rolling elements 18 simultaneously receive no load from the inner ring 12 and the outer ring 14. .

FIG. 8 is a diagram showing the result of actually measuring the relationship between the effective value of the vibration data of the rolling bearing 10 and the number of loads. As shown in FIG. 8, in the embodiment, when the number of times of loading is 0, N1, and N2 (> N1), the lengths of damage occurring in the center of the load area are the damage levels 0 to 2 Do. The number of times of loading is the total number of times that the plurality of rolling elements 18 loaded a radial load to the center of the load area (passed through the center of the load area) after the rolling bearing 10 started operation.

In the embodiment, the term “remaining life” refers to the number of times of loading (the number of remaining loads) from a certain measurement time during status monitoring until damage of damage level 2 is formed. In FIG. 8, the remaining life NL0 of the rolling bearing 10 at the damage level 0 (the number of loads is 0) is N2. The remaining life NL1 of the damage level 1 rolling bearing is N2-N1. The remaining life NL2 of the damage level 2 rolling bearing is zero.

FIG. 9 is a flowchart specifically showing the flow of the process (S1 in FIG. 4) performed in the learning period. As shown in FIG. 9, in S11, the control unit 40 calculates a feature amount set of vibration data of a rolling bearing whose damage length is damage level 2, and advances the process to S12.

At S12, the control unit 40 calculates a feature amount set of vibration data of the rolling bearing whose damage length is damage level 1, and advances the process to S13.

In S13, the control unit 40 calculates a feature amount set of vibration data of a rolling bearing whose damage length is damage level 0, and advances the process to S14.

In S14, the control unit 40 performs machine learning on the feature amount set of damage levels 0 to 2 to create a classifier, and advances the process to S15. The classifier calculates the matching rate between a certain feature set and the feature set for each damage level.

In S15, the control unit 40 associates the feature amount sets of damage levels 0 to 2 with the remaining lives NL0 to NL2 and stores them in the storage unit 50, and ends the processing.

S11 to S13 may be performed before S14, and need not be performed in the order shown in FIG. Also, S11 to S13 may be performed not simultaneously but concurrently.

FIG. 10 is a flowchart specifically showing the flow of the process (S2 in FIG. 9) performed in the monitoring period. The process shown in FIG. 10 is executed at each sampling time by a main routine (not shown).

As shown in FIG. 10, the control unit 40 calculates a feature amount set of vibration data (measurement data) measured at sampling time in S21, and advances the process to S22. In S22, the control unit 40 uses the classifier created in the learning period to calculate the relevance ratios R0 to R2 between the feature data set of the measurement data and each of the three feature data sets that have been machine-learned. Go to

The matching rate R0 is a matching rate between the feature quantity set of measurement data and the feature quantity set of damage level 0. The matching rate R1 is a matching rate between the feature quantity set of measurement data and the feature quantity set of damage level 1. The matching rate R2 is a matching rate between the feature amount set of measurement data and the feature amount set of damage level 2. The sum of the relevance factors R0 to R2 is one.

In S23, the control unit 40 calculates the remaining life NL until the length of the damage at the sampling time becomes the length of the damage level 2 using the following equation (1), and returns the process to the main routine.

NL = NL0 × R0 + NL1 × R1 + 0 × R2 (1)
If remaining life NL is smaller than the threshold, control unit 40 notifies the user of, for example, a message prompting replacement of the rolling bearing.

Below, the result of the experiment which estimated the remaining life of the rolling bearing of a rolling bearing is shown using the state monitoring apparatus which concerns on embodiment. In the experiment, a cylindrical roller bearing NU224 manufactured by NTN Co. was operated at a rotational speed of 1500 min ^-1 , and vibration acceleration in the vertical direction was measured at a sampling speed of 50 kHz for 20 seconds every two hours. The radial load was 90 kN.

FIG. 11 is a view showing the relationship between the number of times of loading of the rolling bearing and the remaining life estimated by the state monitoring device according to the embodiment. As shown in FIG. 11, the remaining life corresponding to the state of damage caused to the raceway surface which develops with the increase of the number of times of loading is estimated.

In the embodiment, the case where the inner ring is a stationary ring and the outer ring is a rotating ring has been described. In the rolling bearing to be monitored by the state monitoring device according to the embodiment, the inner ring may be a rotating ring and the outer ring may be a stationary ring.

In the embodiment, the case has been described in which the length of the damage is divided into three levels, and the feature amount set of each level is machine-learned in the learning period. The length of injury may be divided into four or more levels. Also, the feature value set of vibration data may include feature values other than the effective value, maximum value, crest factor, kurtosis, and skewness, for example, Short Time Fourier Transform (STFT) data May be included.

In the embodiment, the case where machine learning is performed on the damage formed by actually operating the rolling bearing and the feature amount set in the state where the damage is formed has been described. The raceway surface of the rolling bearing may be artificially damaged, and the feature amount set in the state in which the damage is formed may be machine-learned. When the damage is artificially formed without actually operating the rolling bearing, the number of loads until the damage is formed may be set by past operation results or simulation of the same type or similar rolling bearing. it can.

FIG. 12 shows the relationship between the damage levels 0 to 2 and the remaining life in the case where the rolling bearing is actually operated to form a damage and the case where the damage is artificially formed. In FIG. 12, a broken line C20 indicates the correspondence when the rolling bearing is actually operated to form a damage, and a broken line C21 indicates the correspondence when the damage is artificially formed. As shown in FIG. 12, the change of the remaining life with respect to the change of each damage level 0 to 2 is the same tendency at the broken lines C20 and C21. Even when artificially formed damage is used, the remaining life of the rolling bearing can be estimated with the same degree of accuracy as when using the damage formed by actually operating the rolling bearing.

As described above, according to the state monitoring device according to the embodiment, the remaining life of the rolling bearing can be estimated by using the classifier created by machine learning of the feature amount set of the vibration data.

It should be understood that the embodiments disclosed herein are illustrative and non-restrictive in every respect. The scope of the present invention is shown not by the above description but by the scope of claims, and is intended to include all modifications within the scope and meaning equivalent to the scope of claims.

1 state monitoring device, 10 rolling bearing, 11 shaft, 12 inner ring, 14 outer ring, 18 rolling element, 20 vibration sensor, 40 control unit, 50 storage unit.

Claims

Creating a classifier by machine learning;
Monitoring the state of the rolling bearing,
The rolling bearing is
With a rotating wheel,
With the stationary wheel,
And a plurality of rolling elements disposed between the rotating wheel and the stationary wheel and moving along a raceway surface of the stationary wheel as the rotating wheel rotates.
The step of creating the classifier comprises
Calculating a feature amount set of the first vibration data of the rolling bearing in a state in which the damage of the first length is formed in the circumferential direction of the raceway surface;
Calculating a feature amount set of the second vibration data of the rolling bearing in a state where a damage of a second length smaller than the first length is formed in the circumferential direction;
Calculating a feature amount set of third vibration data of the rolling bearing when the raceway surface is normal;
Machine learning each feature amount set of the first to third vibration data to create the classifier;
The number of times of loading on the raceway surface until the first length of damage is formed in the circumferential direction is a first number of times of loading,
The number of times of loading on the raceway surface until the second length of damage is formed in the circumferential direction is a second number of times of loading less than the first number of times of loading,
The step of monitoring the condition of the rolling bearing is:
Calculating a feature quantity set of the fourth vibration data of the rolling bearing measured during monitoring;
Calculating, using the classifier, first to third fitness factors of the feature quantity set of the fourth vibration data and the feature quantity sets of the first to third vibration data, respectively;
The remaining life of the rolling bearing until the damage of the first length is formed in the circumferential direction of the raceway surface using the first and second number of loadings and the first to third fitness factors A method of monitoring the state of a rolling bearing, comprising the steps of: calculating.
The first length is a length at which the two rolling elements are simultaneously unloaded when two adjacent rolling elements out of the plurality of rolling elements pass the damage of the first length. The state monitoring method of the rolling bearing according to claim 1.
The sum of the first to third precisions is 1,
The step of calculating the remaining life is a sum of values obtained by multiplying the second and third fitness factors, the number of remaining loads obtained by subtracting the number of second loads from the number of first loads, and the number of first loads. The rolling bearing state monitoring method according to claim 1, wherein the remaining life is calculated.
The method of monitoring the condition of a rolling bearing according to any one of claims 1 to 3, wherein each of the first and second lengths of damage formed in the circumferential direction is an artificially generated damage. .
The damage of the first length formed in the circumferential direction is a damage formed by operating the rolling bearing until the number of times of loading on the raceway surface becomes the first number of loading,
The damage of the second length formed in the circumferential direction is a damage formed by operating the rolling bearing until the number of times of loading on the raceway surface becomes the number of times of second loading. A method of monitoring the state of a rolling bearing according to any one of items 1 to 3.
The state of the rolling bearing according to any one of claims 1 to 5, wherein each feature value set of the first to fourth vibration data includes an effective value, a maximum value, a crest factor, a kurtosis, and a skewness. How to monitor.
A state monitoring device that monitors the state of the rolling bearing using the method of monitoring the state of the rolling bearing according to any one of claims 1 to 6.