JP2019132809A - Vibration analytical device of rolling bearing, and inspection method of rolling bearing - Google Patents

Vibration analytical device of rolling bearing, and inspection method of rolling bearing Download PDF

Info

Publication number
JP2019132809A
JP2019132809A JP2018017508A JP2018017508A JP2019132809A JP 2019132809 A JP2019132809 A JP 2019132809A JP 2018017508 A JP2018017508 A JP 2018017508A JP 2018017508 A JP2018017508 A JP 2018017508A JP 2019132809 A JP2019132809 A JP 2019132809A
Authority
JP
Japan
Prior art keywords
vibration
rolling bearing
value
feature amount
vibration waveform
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
JP2018017508A
Other languages
Japanese (ja)
Inventor
育彦 榊原
Ikuhiko Sakakibara
育彦 榊原
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NTN Corp
Original Assignee
NTN Corp
NTN Toyo Bearing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NTN Corp, NTN Toyo Bearing Co Ltd filed Critical NTN Corp
Priority to JP2018017508A priority Critical patent/JP2019132809A/en
Publication of JP2019132809A publication Critical patent/JP2019132809A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

To provide a vibration analytical device of a rolling bearing capable of improving productivity by reducing man-hours when determining a determination method.SOLUTION: A vibration analytical device of a rolling bearing includes a feature amount calculation part for calculating a feature amount Q of a vibration waveform obtained from a vibration sensor, when rotating a workpiece which is the rolling bearing, and a determination processing part for inputting the feature amount into a deep learning model 8A having a multi-layer perceptron structure, and for performing quality determination of the bearing by using a probability value (non-defective unit probability, defective unit probability) determined from an output value from the deep learning model 8A.SELECTED DRAWING: Figure 5

Description

この発明は、転がり軸受の振動分析装置および転がり軸受の検査方法に関し、特に、転がり軸受を回転させた際に発生する振動を分析することにより、良否判定を行なう転がり軸受の振動分析装置および転がり軸受の検査方法に関する。   The present invention relates to a rolling bearing vibration analysis apparatus and a rolling bearing inspection method, and more particularly to a rolling bearing vibration analysis apparatus and a rolling bearing that perform pass / fail judgment by analyzing vibrations generated when the rolling bearing is rotated. It relates to the inspection method.

転がり軸受を回転させた際に発生する振動を分析して転がり軸受の欠陥を発見する欠陥検査が従来から知られている。このような欠陥検査では、加速度センサや速度センサなどの振動センサから得られた振動成分に対して、実効値の算出やパルスカウントによるパルス数の計数などにより数値化を行なう。その後、数値化した各項目を対応するしきい値と比較し、しきい値を超えた場合に不良品と判定する。この際に数値化される項目の数は一つではなく複数の場合が多い。   A defect inspection for detecting a defect of a rolling bearing by analyzing vibration generated when the rolling bearing is rotated is conventionally known. In such a defect inspection, a vibration component obtained from a vibration sensor such as an acceleration sensor or a speed sensor is digitized by calculating an effective value or counting the number of pulses by pulse counting. Thereafter, each digitized item is compared with a corresponding threshold value, and when the threshold value is exceeded, it is determined as a defective product. In this case, there are many cases where the number of items to be quantified is not one but plural.

例えば、特開平１０−２２１１６１号公報に開示された検査装置では、加速度センサからの振動信号を、ＦＦＴによる前処理を施した上で２４帯域分のピーク値として数値化し、各々の帯域において数値と定められたしきい値とを比較することによって良否判定を行なっている。   For example, in the inspection apparatus disclosed in Japanese Patent Laid-Open No. 10-221161, the vibration signal from the acceleration sensor is digitized as a peak value for 24 bands after preprocessing by FFT, The quality is judged by comparing with a predetermined threshold value.

また、特開平６−３０７９２０号公報に開示された振動解析装置では、振動データを３つのコンパレータを用いてパルスとし、パルスの計数値として振動データの数値化を行ない、各々の計数値に対して独自のアルゴリズムを適合させることにより、良否判定を行なっている。   Further, in the vibration analyzing apparatus disclosed in Japanese Patent Laid-Open No. 6-307920, vibration data is converted into pulses using three comparators, and vibration data is digitized as a pulse count value. Pass / fail judgment is performed by adapting an original algorithm.

特開平１０−２２１１６１号公報JP-A-10-221161 特開平６−３０７９２０号公報JP-A-6-307920

従来の方法で良否判定を行う為には、複数の数値化項目に対してそれぞれにしきい値を設定する必要があるが、適切なしきい値を決定するために多くの工数を必要とする。   In order to perform pass / fail judgment by the conventional method, it is necessary to set a threshold value for each of a plurality of digitized items, but a large number of man-hours are required to determine an appropriate threshold value.

また、複数の計数値に対して独自のアルゴリズムを作成して良否判定を行なう方法では、最適な判定結果を得る為の計数値の組み合わせの数が膨大となるため、分類アルゴリズムの作成が困難であるといった課題がある。   In addition, in the method of making a pass / fail judgment by creating an original algorithm for a plurality of count values, it is difficult to create a classification algorithm because the number of combinations of count values for obtaining an optimum judgment result becomes enormous. There is a problem that there is.

この発明は、上記の課題を解決するためのものであって、その目的は、判定方法を決定する際の工数が削減され、生産性が向上した転がり軸受の振動分析装置および転がり軸受の検査方法を提供することである。   The present invention is for solving the above-described problems, and its object is to reduce the number of steps for determining the determination method and improve the productivity of the rolling bearing vibration analysis apparatus and the rolling bearing inspection method. Is to provide.

この発明は、要約すると、転がり軸受の振動分析装置であって、転がり軸受を回転させた際に振動センサから得られた振動波形の特徴量を算出する特徴量算出部と、特徴量を多層パーセプトロン構造のディープラーニングモデルに入力し、ディープラーニングモデルの出力値から求められる確率値を用いて軸受の良否判定を行なう判定部とを備える。   In summary, the present invention relates to a vibration analysis device for a rolling bearing, wherein a feature amount calculation unit that calculates a feature amount of a vibration waveform obtained from a vibration sensor when the rolling bearing is rotated, and the feature amount are converted into a multilayer perceptron. A determination unit that inputs to the deep learning model of the structure and determines the quality of the bearing using a probability value obtained from the output value of the deep learning model.

好ましくは、特徴量算出部は、振動波形の実効値と、振動波形のパルスカウント値と、振動波形のピーク値と、振動波形の振幅変調波の振幅の大きさとを特徴量として算出する。実効値は、帯域が互いに異なる複数フィルタをそれぞれ通過した複数の振動波形の実効値を含む。パルスカウント値は、複数の異なるカウントレベルによってそれぞれ計数された複数のパルスカウント値を含む。   Preferably, the feature amount calculation unit calculates the effective value of the vibration waveform, the pulse count value of the vibration waveform, the peak value of the vibration waveform, and the amplitude of the amplitude-modulated wave of the vibration waveform as the feature amount. The effective value includes effective values of a plurality of vibration waveforms that have passed through a plurality of filters having different bands. The pulse count value includes a plurality of pulse count values each counted by a plurality of different count levels.

この発明は、他の局面では、転がり軸受の検査方法であって、転がり軸受を回転させた際に振動センサから得られた振動波形の特徴量を算出するステップと、特徴量を多層パーセプトロン構造のディープラーニングモデルに入力し、ディープラーニングモデルの出力値から求められる確率値を用いて軸受の良否判定を行なうステップとを備える。   In another aspect, the present invention provides a method for inspecting a rolling bearing, the step of calculating a feature amount of a vibration waveform obtained from a vibration sensor when the rolling bearing is rotated, and the feature amount of the multilayer perceptron structure. A step of inputting into the deep learning model and determining the quality of the bearing using a probability value obtained from the output value of the deep learning model.

好ましくは、特徴量を算出するステップは、振動波形の実効値と、振動波形のパルスカウント値と、振動波形のピーク値と、振動波形の振幅変調波の振幅の大きさとを特徴量として算出する。実効値は、帯域が互いに異なる複数フィルタをそれぞれ通過した複数の振動波形の実効値を含む。パルスカウント値は、複数の異なるカウントレベルによってそれぞれ計数された複数のパルスカウント値を含む。   Preferably, the step of calculating the feature amount calculates the effective value of the vibration waveform, the pulse count value of the vibration waveform, the peak value of the vibration waveform, and the amplitude of the amplitude-modulated wave of the vibration waveform as the feature amount. . The effective value includes effective values of a plurality of vibration waveforms that have passed through a plurality of filters having different bands. The pulse count value includes a plurality of pulse count values each counted by a plurality of different count levels.

本発明によれば、振動による転がり軸受の良否判定において、しきい値設定や検査アルゴリズム作成の工数を削減することによる、生産性の向上が期待できる。   According to the present invention, improvement in productivity can be expected by reducing the number of steps for setting a threshold value and creating an inspection algorithm in the quality determination of a rolling bearing due to vibration.

本実施の形態で用いられる分類器の概念を示す図である。It is a figure which shows the concept of the classifier used by this Embodiment. 検査対象である転がり軸受と測定系とを示した図である。It is the figure which showed the rolling bearing and measuring system which are inspection objects. 振動分析装置の構成を示すブロック図である。It is a block diagram which shows the structure of a vibration analyzer. 特徴量算出部の構成を示すブロック図である。It is a block diagram which shows the structure of a feature-value calculation part. 良否確率算出部の構成を示す図である。It is a figure which shows the structure of a pass / fail probability calculation part. 多層パーセプトロンの各ノードを示した図である。It is the figure which showed each node of a multilayer perceptron.

以下、本発明の実施の形態について図面を参照しつつ説明する。なお、以下の図面において同一または相当する部分には同一の参照番号を付し、その説明は繰返さない。   Embodiments of the present invention will be described below with reference to the drawings. In the following drawings, the same or corresponding parts are denoted by the same reference numerals, and description thereof will not be repeated.

転がり軸受の良否判定を行う為には、振動波形の複数の数値化項目に対してそれぞれにしきい値を設定する必要があるが、適切なしきい値を決定するために多くの工数を必要とする。   In order to judge the quality of rolling bearings, it is necessary to set a threshold value for each of the numerical values of the vibration waveform, but a lot of man-hours are required to determine an appropriate threshold value. .

また、複数の計数値に対して独自のアルゴリズムを作成して良否判定を行なう方法では、最適な判定結果を得る為の計数値の組み合わせの数が膨大となるため、分類アルゴリズムの作成が困難であるといった課題がある。   In addition, in the method of making a pass / fail judgment by creating an original algorithm for a plurality of count values, it is difficult to create a classification algorithm because the number of combinations of count values for obtaining an optimum judgment result becomes enormous. There is a problem that there is.

そこで、本実施の形態では、機械学習を用いて分類器を作成する。図１は、本実施の形態で用いられる分類器の概念を示す図である。この分類器は、振動センサから得られた振動信号を、複数の帯域の実効値、カウントレベルを変えた場合のパルスカウント値、振動波形のピーク値、および振幅変調の大きさなどと言った複数の特徴量として数値化し、それらの特徴量を入力値、良品と不良品である確率値を出力とする多層パーセプトロン構造のディープラーニングモデルを用いて軸受の良否分類を行なうことを特徴とする。   Therefore, in this embodiment, a classifier is created using machine learning. FIG. 1 is a diagram showing a concept of a classifier used in the present embodiment. This classifier uses multiple vibration signals obtained from vibration sensors such as effective values in multiple bands, pulse count values when the count level is changed, peak values of vibration waveforms, and amplitude modulation magnitudes. It is characterized in that the quality of the bearings is classified using a deep learning model of a multilayer perceptron structure in which the feature values are converted into numerical values, and the feature values are input values and the probability values of good and defective products are output.

図２は、検査対象である転がり軸受と測定系とを示した図である。図２に示す測定系において検査対象となる転がり軸受の振動測定とその良否判定を行なう。ワークホルダ１にワーク２が配置される。ワーク２は転がり軸受であって、外輪２ａと鋼球２ｂと内輪２ｃと保持器２ｄとを含む。回転軸３が内輪２ｃに挿入されアキシャル加圧された状態で、図示しないモータなどによって回転軸３を一定速度（例えば１８００ｒｐｍ）で回転させる。振動分析装置５は、外輪２ａに接触させた振動センサ４からの振動信号を分析し、ワーク２の良否判定を行なう。振動センサ４としては、たとえば加速度センサを使用することができる。   FIG. 2 is a diagram showing a rolling bearing and a measurement system to be inspected. In the measurement system shown in FIG. 2, the vibration measurement of the rolling bearing to be inspected and the quality determination thereof are performed. A work 2 is placed on the work holder 1. The work 2 is a rolling bearing and includes an outer ring 2a, a steel ball 2b, an inner ring 2c, and a cage 2d. In a state where the rotation shaft 3 is inserted into the inner ring 2c and is axially pressurized, the rotation shaft 3 is rotated at a constant speed (for example, 1800 rpm) by a motor or the like (not shown). The vibration analyzer 5 analyzes the vibration signal from the vibration sensor 4 brought into contact with the outer ring 2a, and determines whether the workpiece 2 is good or bad. As the vibration sensor 4, for example, an acceleration sensor can be used.

図３は、振動分析装置の構成を示すブロック図である。振動分析装置５は、振動センサから得られる電圧信号をデジタルデータの振動波形に変換するＡＤ変換部６と、振動波形から波形の特徴量を算出する特徴量算出部７と、算出された特徴量から良否確率を算出する良否確率算出部８と、得られた良否確率から最終的な合否判定を行う判定処理部９と、得られた結果を外部に出力する結果出力部１０とを含む（図３）。   FIG. 3 is a block diagram showing the configuration of the vibration analyzer. The vibration analyzer 5 includes an AD conversion unit 6 that converts a voltage signal obtained from the vibration sensor into a vibration waveform of digital data, a feature amount calculation unit 7 that calculates a feature amount of the waveform from the vibration waveform, and a calculated feature amount. Includes a pass / fail probability calculation unit 8 that calculates a pass / fail probability from a result, a determination processing unit 9 that performs a final pass / fail determination from the obtained pass / fail probability, and a result output unit 10 that outputs the obtained result to the outside (FIG. 3).

振動センサ４で得られた電圧信号は、ＡＤ変換部６で１６ｂｉｔのデジタル波形に変換され特徴量算出部７へ送られる。   The voltage signal obtained by the vibration sensor 4 is converted into a 16-bit digital waveform by the AD converter 6 and sent to the feature quantity calculator 7.

図４は、特徴量算出部の構成を示すブロック図である。特徴量算出部７は、複数の帯域フィルタ７Ａ，７Ｂ，７Ｇと、実効値算出部７Ｃ，７Ｄと、パルスカウント計数部７Ｈ，７Ｉと、ピーク値検出部７Ｌと振幅変調計測部７Ｎとを含む。   FIG. 4 is a block diagram illustrating a configuration of the feature amount calculation unit. The feature amount calculation unit 7 includes a plurality of band filters 7A, 7B, 7G, effective value calculation units 7C, 7D, pulse count counting units 7H, 7I, a peak value detection unit 7L, and an amplitude modulation measurement unit 7N. .

特徴量算出部７では、通過帯域の異なる複数の帯域フィルタを通過させた波形に対して、実効値算出部（７Ｃ，７Ｄ）によってそれぞれ実効値（７Ｅ，７Ｆ）を算出し、特徴量Ｑとして、実効値（７Ｅ）と実効値（７Ｆ）とを得る。複数の帯域フィルタは、例えば、８０〜４００Ｈｚを通過させる帯域フィルタ（７Ａ）と４００〜６ｋＨｚを通過させる帯域フィルタ（７Ｂ）を含む。   In the feature amount calculation unit 7, the effective value (7E, 7F) is calculated by the effective value calculation unit (7C, 7D) with respect to the waveform that has passed through a plurality of band filters having different pass bands, and the feature amount Q is obtained. The effective value (7E) and the effective value (7F) are obtained. The plurality of band filters include, for example, a band filter (7A) that passes 80 to 400 Hz and a band filter (7B) that passes 400 to 6 kHz.

さらに、特徴量算出部７では、帯域フィルタ（７Ａ）および帯域フィルタ（７Ｂ）とは通過帯域が異なるフィルタ（７Ｇ）を通過させた波形に対して、カウントレベルが異なる複数のパルスカウント計数部、例えばパルスカウント計数部（７Ｈ）とパルスカウント計数部（７Ｉ）でカウントしきい値を超える振動パルスの数を計測し、特徴量として、パルスカウント値（７Ｊ）とパルスカウント値（７Ｋ）を得る。帯域フィルタ（７Ｇ）は、例えば、１ｋＨｚから５ｋＨｚを通過させる帯域フィルタとすることができる。   Further, in the feature amount calculation unit 7, a plurality of pulse count counting units having different count levels with respect to a waveform that has passed through a filter (7G) having a different pass band from the band filter (7A) and the band filter (7B), For example, the pulse count counting unit (7H) and the pulse count counting unit (7I) measure the number of vibration pulses exceeding the count threshold value, and obtain the pulse count value (7J) and the pulse count value (7K) as feature quantities. . The band filter (7G) can be, for example, a band filter that passes 1 kHz to 5 kHz.

さらに、特徴量算出部７では、帯域フィルタ（７Ｇ）を通過させた波形に対して、ピーク値検出部（７Ｌ）で、特徴量として、ピーク値（７Ｍ）を得る。   Further, the feature value calculation unit 7 obtains a peak value (7M) as a feature value by the peak value detection unit (7L) for the waveform that has passed through the bandpass filter (7G).

さらに、特徴量算出部７では、帯域フィルタ（７Ｇ）を通過させた波形に対して、振幅変調計測部（７Ｎ）で、特徴量として、振幅変調波の振幅の大きさ（７Ｏ）を得る。   Further, in the feature quantity calculation unit 7, the amplitude (70) of the amplitude modulation wave is obtained as a feature quantity in the amplitude modulation measurement unit (7N) with respect to the waveform passed through the bandpass filter (7G).

特徴量算出部７で算出された特徴量は、良否確率算出部８に送られる。
なお、特徴量パラメータは上記に例示された６個に限定されるものではない。例えば、実効値を算出する為の帯域フィルタの帯域を５０〜３００Ｈｚと３００〜１．８ｋＨｚと１．８ｋＨｚ〜１０ｋＨｚの３帯域としても良いし、５０Ｈｚから１０ｋＨｚまでを１／３オクターブ間隔で分割した２４帯域としても良い。また、特徴量として音声認識の分野でよく使用されるメル周波数ケプストラム計数（ＭＦＣＣ）を用いても良い。 The feature amount calculated by the feature amount calculation unit 7 is sent to the pass / fail probability calculation unit 8.
Note that the feature amount parameters are not limited to the six exemplified above. For example, the band of the band filter for calculating the effective value may be three bands of 50 to 300 Hz, 300 to 1.8 kHz, and 1.8 kHz to 10 kHz, and 50 Hz to 10 kHz is divided at 1/3 octave intervals. There may be 24 bands. Further, a mel frequency cepstrum count (MFCC) often used in the field of speech recognition may be used as the feature amount.

図５は、良否確率算出部の構成を示す図である。良否確率算出部８は、特徴量算出部７で得られた特徴量Ｑを入力とし、良品と不良品に対応した出力値を算出する多層パーセプトロン構造のディープラーニングモデル８Ａと、多層パーセプトロンからの出力値を良品と不良品それぞれの確率値に変換する確率値算出部８Ｂとを含む（図５）。   FIG. 5 is a diagram illustrating a configuration of the pass / fail probability calculation unit. The pass / fail probability calculation unit 8 receives the feature quantity Q obtained by the feature quantity calculation unit 7 and inputs a deep learning model 8A having a multi-layer perceptron structure that calculates output values corresponding to a non-defective product and a defective product, and an output from the multi-layer perceptron And a probability value calculation unit 8B that converts the values into the probability values of the non-defective product and the defective product (FIG. 5).

ディープラーニングモデル８Ａの多層パーセプトロン構造は、例えば入力層６点、中間層９点×２層、出力層２点とすれば良い。   The multilayer perceptron structure of the deep learning model 8A may be, for example, 6 input layers, 9 intermediate layers × 2 layers, and 2 output layers.

図６は、多層パーセプトロンの各ノードを示した図である。多層パーセプトロンの各ノードは、下式（１）に示されるように、前層の各ノードからの入力ｘ（ｘ１〜ｘｎ）と重み係数ｗ（ｗ１〜ｗｎ）の積を合計した値にバイアスｂを足し合わせ、さらに活性化関数ｆ（ｘ）を適用した値を出力値とする。 FIG. 6 is a diagram showing each node of the multilayer perceptron. Each node of the multi-layer perceptron, as shown in the following formula (1), the sum of the product of the input x (x 1 ~x n) and the weight coefficient w (w 1 to w n) from each node of the previous layer A value obtained by adding the bias b to the value and further applying the activation function f (x) is defined as an output value.

Figure 2019132809
Figure 2019132809

重み係数ｗとバイアスｂの値は誤差逆伝搬法による学習で算出される。活性化関数ｆ（ｘ）は下式（２）に示すシグモイド関数σを使用する。   The values of the weighting coefficient w and the bias b are calculated by learning by the error back propagation method. As the activation function f (x), a sigmoid function σ shown in the following equation (2) is used.

Figure 2019132809
Figure 2019132809

多層パーセプトロン構造のディープラーニングモデル８Ａからの２つの出力値は、確率値算出部８Ｂで下式（３）に示すソフトマックス関数にて確率値に変換される。   Two output values from the deep learning model 8A having a multilayer perceptron structure are converted into probability values by the probability value calculation unit 8B using a softmax function expressed by the following expression (3).

Figure 2019132809
Figure 2019132809

尚、中間層の点数や層数は設計的項目であり、本実施例の条件に限定されるものではない。   Note that the number of intermediate layers and the number of layers are design items and are not limited to the conditions of this embodiment.

多層パーセプトロン構造のディープラーニングモデル８Ａ内での演算に使用する係数は予め、特徴量と正解を関連づけたデータを、誤差逆伝搬法を用いたディープラーニングの学習プログラムに与えて算出しておく。   The coefficient used for the calculation in the deep learning model 8A having the multi-layer perceptron structure is calculated in advance by giving data that associates the feature quantity and the correct answer to the deep learning learning program using the error back propagation method.

図３の判定処理部９は、良否確率算出部８から入力される確率値と予め決められたしきい値との大小を比較して判定を行なう。例えば、良品確率値に対してしきい値を設け、しきい値を上回るものを良品とし、しきい値を下回るものを不良品とする。   The determination processing unit 9 in FIG. 3 makes a determination by comparing the probability value input from the pass / fail probability calculation unit 8 with a predetermined threshold value. For example, a threshold value is provided for the non-defective product probability value, a product that exceeds the threshold value is regarded as a non-defective product, and a product that is below the threshold value is regarded as a defective product.

結果出力部１０は、判定処理部９から受け取った判定結果の画面表示や他の機器へ出力を行なう。   The result output unit 10 displays the determination result received from the determination processing unit 9 on the screen and outputs it to other devices.

本実施の形態の例では、６つの特徴量に対して、良品確率値に対する一つのしきい値で良否判定を行なうことができるため、しきい値を決定する為の工数が削減できる。また複数の特徴量を元に良否判定を行なう複雑なアルゴリズムを作成する必要が無く、アルゴリズム開発の工数を削減できる。   In the example of the present embodiment, the quality determination can be performed with respect to the six feature quantities with one threshold value for the non-defective product probability value, so that the man-hour for determining the threshold value can be reduced. In addition, it is not necessary to create a complex algorithm for determining pass / fail based on a plurality of feature quantities, and the man-hour for algorithm development can be reduced.

最後に、本実施の形態について総括する。図２、図３、図５を参照して、本実施の形態に係る転がり軸受の振動分析装置５は、転がり軸受であるワーク２を回転させた際に振動センサ４から得られた振動波形の特徴量Ｑを算出する特徴量算出部７と、特徴量を多層パーセプトロン構造のディープラーニングモデル８Ａに入力し、ディープラーニングモデル８Ａの出力値から求められる確率値（良品確率、不良品確率）を用いて軸受の良否判定を行なう判定処理部９とを備える。   Finally, this embodiment will be summarized. 2, 3, and 5, the vibration analysis device 5 of the rolling bearing according to the present embodiment has a vibration waveform obtained from the vibration sensor 4 when the work 2, which is a rolling bearing, is rotated. A feature amount calculation unit 7 that calculates a feature amount Q, and the feature amount are input to a deep learning model 8A having a multilayer perceptron structure, and a probability value (good product probability, defective product probability) obtained from the output value of the deep learning model 8A is used. And a determination processing unit 9 for determining the quality of the bearing.

好ましくは、特徴量算出部７は、図４に示すように、振動波形の実効値７Ｅ，７Ｆと、振動波形のパルスカウント値７Ｊ，７Ｋと、振動波形のピーク値７Ｍと、振動波形の振幅変調波の振幅の大きさ７Ｏとを特徴量Ｑとして算出する。実効値は、帯域が互いに異なる複数フィルタ７Ａ，７Ｂをそれぞれ通過した複数の振動波形の実効値７Ｅ，７Ｆを含む。パルスカウント値は、複数の異なるカウントレベルを有するパルスカウント計数部７Ｈ，７Ｉによってそれぞれ計数された複数のパルスカウント値７Ｊ，７Ｋを含む。   Preferably, as shown in FIG. 4, the feature quantity calculation unit 7 has effective values 7E and 7F of the vibration waveform, pulse count values 7J and 7K of the vibration waveform, a peak value 7M of the vibration waveform, and an amplitude of the vibration waveform. The amplitude 7O of the modulated wave is calculated as the feature quantity Q. The effective value includes effective values 7E and 7F of a plurality of vibration waveforms respectively passing through the plurality of filters 7A and 7B having different bands. The pulse count value includes a plurality of pulse count values 7J and 7K respectively counted by pulse count counting units 7H and 7I having a plurality of different count levels.

本発明によれば、振動による転がり軸受の良否判定において、しきい値設定や検査アルゴリズム作成の工数を削減することによる、生産性の向上が期待できる。   According to the present invention, improvement in productivity can be expected by reducing the number of steps for setting a threshold value and creating an inspection algorithm in the quality determination of a rolling bearing due to vibration.

今回開示された実施の形態は、すべての点で例示であって制限的なものではないと考えられるべきである。例えば、転がり軸受は玉軸受を例示したが、円筒ころ軸受、円錐ころ軸受、ニードル軸受、自動調心ころ軸受、その他の転がり軸受であってもよい。本発明の範囲は、上記した実施の形態の説明ではなくて特許請求の範囲によって示され、特許請求の範囲と均等の意味および範囲内でのすべての変更が含まれることが意図される。   The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. For example, although the ball bearing is exemplified as the rolling bearing, a cylindrical roller bearing, a tapered roller bearing, a needle bearing, a self-aligning roller bearing, and other rolling bearings may be used. The scope of the present invention is shown not by the above description of the embodiments but by the scope of claims for patent, and is intended to include meanings equivalent to the scope of claims for patent and all modifications within the scope.

１ ワークホルダ、２ ワーク、２ａ 外輪、２ｂ 鋼球、２ｃ 内輪、２ｄ 保持器、３ 回転軸、４ 振動センサ、５ 振動分析装置、６ ＡＤ変換部、７ 特徴量算出部、７Ａ，７Ｂ，７Ｇ 帯域フィルタ、７Ｃ，７Ｄ 実効値算出部、７Ｈ，７Ｉ パルスカウント計数部、７Ｌ ピーク値検出部、７Ｎ 振幅変調計測部、８ 良否確率算出部、８Ａ ディープラーニングモデル、８Ｂ 確率算出処理部、９ 判定処理部、１０ 結果出力部。   1 Work holder, 2 Work, 2a Outer ring, 2b Steel ball, 2c Inner ring, 2d Cage, 3 Rotating shaft, 4 Vibration sensor, 5 Vibration analyzer, 6 AD converter, 7 Feature quantity calculator, 7A, 7B, 7G Bandpass filter, 7C, 7D RMS value calculation unit, 7H, 7I Pulse count counting unit, 7L peak value detection unit, 7N amplitude modulation measurement unit, 8 pass / fail probability calculation unit, 8A deep learning model, 8B probability calculation processing unit, 9 determination Processing unit, 10 result output unit.

Claims (4)

転がり軸受を回転させた際に振動センサから得られた振動波形の特徴量を算出する特徴量算出部と、
前記特徴量を多層パーセプトロン構造のディープラーニングモデルに入力し、前記ディープラーニングモデルの出力値から求められる確率値を用いて軸受の良否判定を行なう判定部とを備える、転がり軸受の振動分析装置。 A feature amount calculation unit that calculates a feature amount of a vibration waveform obtained from the vibration sensor when the rolling bearing is rotated;
A vibration analysis device for a rolling bearing, comprising: a determination unit that inputs the feature amount into a deep learning model having a multilayer perceptron structure and determines a quality of the bearing using a probability value obtained from an output value of the deep learning model. 前記特徴量算出部は、振動波形の実効値と、振動波形のパルスカウント値と、振動波形のピーク値と、振動波形の振幅変調波の振幅の大きさとを前記特徴量として算出し、
前記実効値は、帯域が互いに異なる複数フィルタをそれぞれ通過した複数の振動波形の実効値を含み、
前記パルスカウント値は、複数の異なるカウントレベルによってそれぞれ計数された複数のパルスカウント値を含む、請求項１に記載の転がり軸受の振動分析装置。 The feature amount calculation unit calculates the effective value of the vibration waveform, the pulse count value of the vibration waveform, the peak value of the vibration waveform, and the amplitude of the amplitude modulation wave of the vibration waveform as the feature amount,
The effective value includes an effective value of a plurality of vibration waveforms respectively passing through a plurality of filters having different bands,
The vibration analysis device for a rolling bearing according to claim 1, wherein the pulse count value includes a plurality of pulse count values respectively counted by a plurality of different count levels. 転がり軸受を回転させた際に振動センサから得られた振動波形の特徴量を算出するステップと、
前記特徴量を多層パーセプトロン構造のディープラーニングモデルに入力し、前記ディープラーニングモデルの出力値から求められる確率値を用いて軸受の良否判定を行なうステップとを備える、転がり軸受の検査方法。 Calculating a feature value of a vibration waveform obtained from the vibration sensor when the rolling bearing is rotated;
A rolling bearing inspection method comprising: inputting the feature amount into a deep learning model having a multilayer perceptron structure, and determining a quality of the bearing using a probability value obtained from an output value of the deep learning model. 前記特徴量を算出するステップは、振動波形の実効値と、振動波形のパルスカウント値と、振動波形のピーク値と、振動波形の振幅変調波の振幅の大きさとを前記特徴量として算出し、
前記実効値は、帯域が互いに異なる複数フィルタをそれぞれ通過した複数の振動波形の実効値を含み、
前記パルスカウント値は、複数の異なるカウントレベルによってそれぞれ計数された複数のパルスカウント値を含む、請求項３に記載の転がり軸受の検査方法。 The step of calculating the feature amount calculates the effective value of the vibration waveform, the pulse count value of the vibration waveform, the peak value of the vibration waveform, and the amplitude of the amplitude modulation wave of the vibration waveform as the feature amount,
The effective value includes an effective value of a plurality of vibration waveforms respectively passing through a plurality of filters having different bands,
4. The rolling bearing inspection method according to claim 3, wherein the pulse count value includes a plurality of pulse count values respectively counted by a plurality of different count levels.
JP2018017508A 2018-02-02 2018-02-02 Vibration analytical device of rolling bearing, and inspection method of rolling bearing Pending JP2019132809A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2018017508A JP2019132809A (en) 2018-02-02 2018-02-02 Vibration analytical device of rolling bearing, and inspection method of rolling bearing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2018017508A JP2019132809A (en) 2018-02-02 2018-02-02 Vibration analytical device of rolling bearing, and inspection method of rolling bearing

Publications (1)

Publication Number Publication Date
JP2019132809A true JP2019132809A (en) 2019-08-08

Family

ID=67547409

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2018017508A Pending JP2019132809A (en) 2018-02-02 2018-02-02 Vibration analytical device of rolling bearing, and inspection method of rolling bearing

Country Status (1)

Country Link
JP (1) JP2019132809A (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS54104299A (en) * 1978-02-02 1979-08-16 Sumitomo Metal Ind Device for warning trouble of equipment having rotary unit
JPH05288599A (en) * 1992-04-10 1993-11-02 Asmo Co Ltd Machine operating sound inspecting device
JPH06115635A (en) * 1992-10-09 1994-04-26 Ishikawajima Harima Heavy Ind Co Ltd Diagnosis device for tower parking and the like
JPH06186136A (en) * 1992-12-18 1994-07-08 Ono Sokki Co Ltd Fault diagnostic apparatus for ball-and-roller bearing
JPH09243503A (en) * 1996-03-08 1997-09-19 Suzuki Motor Corp Structure non-defect/defect inspecting apparatus
US20050119840A1 (en) * 2003-01-10 2005-06-02 Rolls-Royce Plc Bearing anomaly detection and location
JP2018004473A (en) * 2016-07-04 2018-01-11 ファナック株式会社 Mechanical learning device for learning estimated life of bearing, life estimation device, and mechanical learning method

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS54104299A (en) * 1978-02-02 1979-08-16 Sumitomo Metal Ind Device for warning trouble of equipment having rotary unit
JPH05288599A (en) * 1992-04-10 1993-11-02 Asmo Co Ltd Machine operating sound inspecting device
JPH06115635A (en) * 1992-10-09 1994-04-26 Ishikawajima Harima Heavy Ind Co Ltd Diagnosis device for tower parking and the like
JPH06186136A (en) * 1992-12-18 1994-07-08 Ono Sokki Co Ltd Fault diagnostic apparatus for ball-and-roller bearing
JPH09243503A (en) * 1996-03-08 1997-09-19 Suzuki Motor Corp Structure non-defect/defect inspecting apparatus
US20050119840A1 (en) * 2003-01-10 2005-06-02 Rolls-Royce Plc Bearing anomaly detection and location
JP2018004473A (en) * 2016-07-04 2018-01-11 ファナック株式会社 Mechanical learning device for learning estimated life of bearing, life estimation device, and mechanical learning method

Similar Documents

Publication Publication Date Title
JP6882397B2 (en) Abnormal noise detection system, equipment, method and program
Kankar et al. Rolling element bearing fault diagnosis using autocorrelation and continuous wavelet transform
JP3484665B2 (en) Abnormality determination method and device
Kane et al. Critical evaluation and comparison of psychoacoustics, acoustics and vibration features for gear fault correlation and classification
US20190360942A1 (en) Information processing method, information processing apparatus, and program
CN110795843A (en) Method and device for identifying faults of rolling bearing
Skrickij et al. Diagnostic features for the condition monitoring of hypoid gear utilizing the wavelet transform
Kim Time frequency image and artificial neural network based classification of impact noise for machine fault diagnosis
Dong et al. Incipient bearing fault feature extraction based on minimum entropy deconvolution and K-singular value decomposition
CN109934136B (en) Rolling bearing fault diagnosis method based on Duffing vibrator and eigen mode component
Pichler et al. Data-driven vibration-based bearing fault diagnosis using non-steady-state training data
Ren et al. Shaking noise elimination for detecting local flaw in steel wire ropes based on magnetic flux leakage detection
JP3885297B2 (en) Abnormal sound determination device and abnormal sound determination method
JPH09257651A (en) Method for detecting damage of bearing
JP2019045484A (en) State monitoring method, and state monitoring device
Chi et al. Spectral DCS-based feature extraction method for rolling element bearing pseudo-fault in rotor-bearing system
Cong et al. Research on the order selection of the autoregressive modelling for rolling bearing diagnosis
JP2019132809A (en) Vibration analytical device of rolling bearing, and inspection method of rolling bearing
Abdelkader et al. Rolling bearing faults diagnosis based on empirical mode decomposition: Optimized threshold de-noising method
Jiang et al. Rolling bearing quality evaluation based on a morphological filter and a Kolmogorov complexity measure
JP4029405B2 (en) Abnormality determination method and apparatus
Lin et al. Condition-based shaft fault diagnosis with the empirical mode decomposition method
RU2331893C1 (en) Method of discrete component separation in signal spectre and device for its implementation
CN102830177B (en) Abnormality checking method and abnormality checking apparatus
Kamiel et al. Ball bearing fault diagnosis using wavelet transform and principal component analysis

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20210126

A977 Report on retrieval

Free format text: JAPANESE INTERMEDIATE CODE: A971007

Effective date: 20211122

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20211124

A02 Decision of refusal

Free format text: JAPANESE INTERMEDIATE CODE: A02

Effective date: 20220524