JP2009103525A - Method for diagnosing abnormality of tooth plane of gear and apparatus using same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a gear abnormality diagnostic apparatus using a gear diagnostic technique and its diagnostic method capable of accurately detecting and determining abnormality of a single gear item due to dents in a tooth plane of a gear and failures in the degree of roughness (surface roughness) in the tooth plane on the basis of time-series analysis of gear vibration signals when the gear is engaged and driven. <P>SOLUTION: Vibration and acceleration signals generated are measured when the single gear item is driven to perform time-series analysis on the vibration and acceleration signals. By determining a crest factor and the kurtosis of the residual between their estimated signals and observed signals and monitoring their changes, it is possible to determine (diagnose) gear abnormalities such as very small dents, etc. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a method for inspecting whether there is a defect in surface roughness of a gear tooth surface or a dent.

  Conventionally, 100% gears are currently inspected before assembling on the final production line. As an inspection method, gear abnormalities are discriminated by dents, foreign substances, etc. with a simple gear inspection machine using manual operation, vibration and sound of a skilled worker. However, since most of the judgment depends on the experience and five senses of the operator such as listening, it is difficult to perform a quantitative and accurate examination. In addition, when abnormal gears that could not be detected at this inspection stage were discovered by abnormal drive sound in the final inspection after being assembled as a transmission, it would require enormous labor such as disassembling all of them and searching for abnormalities. . Defective surface roughness and dents on the tooth surface can cause noise during driving, especially during the manufacturing process of gears (gears) and during transport between processes, colliding with other gears and processing equipment, etc. It is difficult to prevent the occurrence of a dent due to an accidental cause (a concave portion and a convex portion are formed around it).

In addition, the following patent document 1 can be mentioned as a well-known technique relevant to this invention.

JP 2005-91232 A

  As described above, the meshing dent detection method for gears (gears) according to the prior art and the apparatus thereof are inspections by abnormal driving sound in the final inspection after assembly as a transmission, and when a defect is discovered. Requires a tremendous amount of effort, such as disassembling all of them and finding abnormalities.

  For this reason, it takes a lot of time to fix the problem, and the cost for this is enormous every year, which is a great economic loss.

  The present invention has been made in view of the above points, and its purpose is to make a time series of gear vibration acceleration signals at the time of driving by engaging the abnormality of a single gear due to surface roughness defects or dents on the gear. It is an object of the present invention to provide a new gear tooth surface abnormality diagnosis method capable of detecting and diagnosing with high accuracy by analysis and a gear tooth surface abnormality diagnosis apparatus using the same.

  In the first aspect of the present invention, which achieves the above object, it occurs when an autoregressive AR (p) model for time series analysis is provided and the master gear and the diagnosis object gear are meshed and rotated by a drive motor. The vibration acceleration is measured (observed) with an acceleration sensor installed at one place, and an estimated value of vibration acceleration is calculated from the vibration acceleration signal measured by the acceleration sensor using the autoregressive AR (p) model. The difference (residual) of the two signals between the measured (observed) vibration acceleration signal and the estimated value calculated by the autoregressive AR (p) model is calculated, and the difference (residual) of the two signals in the time series analysis is calculated. By determining the crest factor or kurtosis of the statistics of the difference, and monitoring the changes in these statistics, it is possible to determine the surface roughness of the meshing tooth surface of the gear to be diagnosed and abnormalities of the gear tooth surface such as minute dents. (Diagnosis) An abnormality diagnosis method of the gear tooth surfaces, characterized in that possible.

  According to the first aspect of the present invention, a change in the residual (deviation) between the estimated value of the acceleration signal calculated from the autoregressive AR (p) model of the time series analysis and the actual measurement value (measurement value) is monitored, so that the gear teeth It is also possible to capture irregular random signal changes with no apparent difference between the normal gear and the vibration acceleration waveform with no apparent dent due to surface dents or slight surface roughness failure of the tooth surface.

  According to claim 2 of the present invention, there is an autoregressive AR (30) model having an order of 30 in time series analysis, and vibration acceleration generated when the master gear and the diagnosis object gear are meshed and rotated by a drive motor. Is measured by an acceleration sensor installed in one place, an estimated value of vibration acceleration is calculated from the vibration acceleration signal measured by the acceleration sensor by the autoregressive AR (30) model, and the vibration acceleration signal measured by the acceleration sensor is calculated. And the estimated value calculated by the autoregressive AR (30) model, the difference (residual) of the two signals is calculated, and the crest factor of the statistic of the difference (residual) of the two signals of the time series analysis Alternatively, by determining the kurtosis and monitoring changes in these statistics, it is possible to discriminate (diagnose) abnormalities in the gear tooth surface, such as the surface roughness of the meshing tooth surface of the gear to be diagnosed and minute dents. An abnormality diagnosis method of the gear tooth surfaces to.

  According to the invention of claim 2, the gear wavefront is monitored by monitoring the change in the residual (deviation) between the estimated value of the acceleration signal calculated from the autoregressive AR (30) model of time series analysis and the actual measurement value (measurement value). It is also possible to capture changes in random signals with irregularities with no apparent difference between a normal gear with no apparent dent and a vibration acceleration waveform due to a slight surface roughness of the tooth surface.

  According to a third aspect of the present invention, there is an autoregressive AR (30) model having an order of 30 in time series analysis, the master gear and the gear to be diagnosed are meshed and incorporated in the gear tooth surface abnormality diagnosis device body, and the drive motor Built-in sensor amplifier that amplifies the vibration acceleration signal detected by this acceleration sensor, and the vibration acceleration signal that is detected by this acceleration sensor. The vibration acceleration calculated by the autoregressive AR (30) model after being taken into the signal processing PC via a module type signal input / output unit that converts the digital signal into, for example, 12 to 24 bits by an A / D converter Calculate the difference (residual) of the two signals between the estimated value of the signal and the measured value of the vibration acceleration signal measured by the acceleration sensor. The crest factor or kurtosis of the statistic is obtained from the difference (residual) between the two signals in the time series analysis, and the change in these statistic is stored in the storage device 33, displayed on the display 50, or displayed on the printer 60. An abnormality diagnosis of a gear tooth surface characterized by being able to determine (diagnosis) an abnormality of the gear tooth surface such as the surface roughness of the meshing tooth surface of the gear to be diagnosed or a minute dent by monitoring the output Device.

  In the invention of claim 3, the gear wavefront is monitored by monitoring the change in the residual (deviation) between the estimated value of the acceleration signal calculated from the autoregressive AR (p) model of time series analysis and the actual measurement value (measurement value). It is also possible to capture changes in random signals with irregularities with no apparent difference between a normal gear with no apparent dent and a vibration acceleration waveform due to a slight surface roughness of the tooth surface.

  The gear tooth surface diagnosis method and gear tooth surface diagnosis device according to the present invention, in particular, apply a time series model analysis method that can capture small fluctuations in random signals as changes in mathematical model parameters. This is effective for detecting gear tooth surface abnormalities such as tooth surface roughness (surface roughness) and minute dents (for example, dents of 3 μm or less) that were very difficult to detect by analysis and amplitude analysis.

The abnormality diagnosis method newly proposed in the present invention is based on a residual factor (deviation) calculated from a time-series model of measured vibration acceleration and an actual measurement signal, which is a statistic such as a crest factor and a kurtosis. The abnormality of the gear is detected by monitoring the quotient obtained by dividing the fourth-order product factor around the average by σ ⁴ (σ: standard deviation). In the present invention, at the on-site gear single product inspection stage, anyone can easily and accurately determine a gear tooth surface abnormality such as a dent or a tooth surface roughness (surface roughness) with a quantitative numerical value (PC) A personal computer) -based gear dent diagnosis system is provided.

  The diagnosis method (method) based on the time series analysis has a higher diagnostic sensitivity than the conventionally used statistical analysis of gear abnormality, and the tooth surface of the gear of 3 μm or less, which has not been detected at all so far, is very small. It was found that abnormalities such as dents can be sufficiently discriminated. In addition, it is a very effective diagnostic technique because it can accurately determine a gear abnormality due to a tooth surface roughness (tooth surface roughness) defect, which is difficult to determine by statistical analysis, with sufficient accuracy.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

(Diagnosis method)
The gear tooth surface abnormality diagnosis apparatus 1 shown in FIG. 1 includes a gear tooth surface that can be incorporated by meshing a master gear B that is a normal gear serving as a reference device and a gear C to be inspected, and rotated by a drive motor D. An acceleration sensor A for installing and measuring (observing) vibration acceleration generated when the gears are rotationally driven according to a start command from the abnormality diagnosis device main body 2 and the keyboard 40. And a sensor amplifier E that amplifies the vibration acceleration signal detected by the acceleration sensor A and an A / D converter that incorporates the amplified vibration acceleration signal. The output unit F and the digital signal that is the measured vibration acceleration signal are captured and stored in the auxiliary storage device (HDD, FDD) 33b. Time series analysis autoregressive AR (p) model that is temporarily stored in the main storage device 33 and provided to the central processing unit 30 under the management of the control unit 32 according to the program. The difference (residual) of the two signals between the estimated value and the measured value calculated by the above is calculated, and the crest factor or kurtosis of the parameter which is a statistic of the difference (residual) of the two signals in the time series analysis is calculated. A change in the parameter (the crest factor or kurtosis of the statistic of the difference between the two signals (estimated value and measured value) of the estimated value and the measured value) is stored in an auxiliary memory by an instruction from the keyboard 40 or the mouse 40b as an input device. A signal processing PC (personal computer) G that is stored in the device 33b, displayed on the display 50, which is an output device, or output to the printer 60, is monitored by monitoring changes in the parameters. Minute dents or tooth surface roughness of the abnormality determination of the gear tooth surfaces of defects or the like (the surface roughness of the tooth surface) (diagnosis) is abnormal diagnosis method of the gear tooth surfaces, characterized in that it is.

  FIG. 14 shows a block diagram of the gear tooth surface diagnostic apparatus 1. Although not shown in FIGS. 1 and 14, the power of the signal processing PC G, the gear tooth surface abnormality diagnosis device main body 2, and the like are supplied with necessary power by each power source connected to a commercial power source.

  The gear tooth surface abnormality diagnosis device main body 2 of the gear tooth surface abnormality diagnosis device 1 will be described in detail. A drive motor D having a motor pulley 6 on a motor shaft 5 on a base 3 having an anti-vibration leg 4 on the lower surface portion. Is attached to the base 3 with bolts via a vibration-proof rubber (not shown). An upper plate 9 supported and fixed to left and right support frames 7 and 8 fixed on the base 3 with bolts or the like through vibration-proof rubber (not shown) at left and right end portions on the base 3. Is provided. These anti-vibration rubbers (not shown) prevent the vibration of the drive motor D itself from propagating to the upper plate 9 via the base 3 and the left and right support frames 7 and 8. The acceleration acceleration of the drive motor D itself is not detected by the acceleration sensor A attached to one side of the side surface 20 (made of iron-based material).

  A first support member 10 penetrating the upper plate 9 is fixed to the upper plate 9 with a bolt or the like at a substantially central portion of the upper plate 9. The center portion of the first support member 10 is hollow, the drive shaft 11 penetrates the hollow portion in the vertical direction, and the vicinity of the upper and lower end portions of the drive shaft 11 is at the upper and lower end portions of the first support member 10. It is supported by a provided bearing and is rotatably supported. A driven pulley 15 corresponding to the motor pulley 6 is fixed to the lower end portion of the drive shaft 11, and a belt 16 (a belt capable of quiet power transmission with little slippage is desirable between the motor pulley 6 and the driven pulley 15. A belt). Furthermore, a male screw is provided at the upper end portion of the drive shaft 11, and a diagnosis target gear C is detachably provided by a nut 17.

  Further, a second support member (made of iron-based material) 20 penetrating the upper plate 9 is fixed to the upper plate 9 with a bolt or the like in parallel with the first support member 10 and the drive shaft 11. It is provided. Similarly to the first support member 10, the second support member 20 has a hollow central portion, and the hollow shaft is penetrated by the driven shaft 21 in the vertical direction, and is provided in the vicinity of the upper and lower end portions thereof. It is supported by a fixed bearing and is rotatable. A male screw is provided at the upper end of the driven shaft 21, and the master gear B is detachably provided by a nut 24. Further, a second support member 20 that rotatably supports the driven shaft 21 to which the master gear B is attached and is fixed by a bolt or the like through the upper plate 9 is attached to the drive shaft 11. Thus, the driven shaft 21 can be moved closer to or away from the center shaft so that the center distance can be moved, and the engagement between the master gear B and the diagnosis target gear C can be adjusted appropriately.

  In the gear diagnosis, a master gear B serving as a reference device and a gear C to be inspected are assembled on the gear tooth surface abnormality diagnosis device main body 2 as shown in FIG. Diagnosis was made by attaching the acceleration sensor A with a magnet base 27 at one location on the side of the iron-based material 20. The acceleration sensor A attached by the magnet base 27 may be a side surface of the first support member 10. The place where the acceleration sensor A is attached is preferably close to the source of vibration acceleration generated when the master gear B and the gear C to be inspected are engaged. Here, the master gear (normal gear) B is a gear (gear) having a normal gear tooth surface, which is a reference device having a good surface finish without a dent confirmed by an inspector in advance.

  This time, as the gear C to be inspected, three types (normally without dent) or abnormal (with dent, poor tooth surface roughness (surface roughness)) determined by the sensory inspection inspector ( A type: 30 teeth, standard pitch circle diameter 52.93, total tooth height 4.753, D type: 53 teeth, standard pitch circle diameter 99.86, total tooth height 3.982, HA type: external dimensions, etc. Was the same as the A type, and the gears with different tooth surface roughness were selected and diagnosed.

  In this diagnosis, the drive motor D is driven at a constant rotation speed of 500 rev / min by a start command from the keyboard 40, and at that time, the radial vibration acceleration of the gear that occurs when the master gear B and the gear C to be diagnosed are engaged with each other. Is detected by the acceleration sensor A. Thereafter, the detection signal (measurement signal) is amplified by the sensor amplifier E, converted into a digital signal of 12 to 24 bits, for example, by an A / D converter built in the module type signal input / output unit F, and the bus line 34. To the signal processing PC (personal computer) G. At that time, the sampling frequency of the A / D conversion of the module type signal input / output unit F is 5.12 kHz. Then, the measurement signal is temporarily stored in the main storage device 33 under the management of the control unit 32 according to the program stored in the auxiliary storage device 33b, provided to the central processing unit 30, and via the calculation unit 31 Signal processing, analysis, and diagnosis are performed by the programmed autoregressive AR (p) model of time series analysis. That is, the difference between the two signals (residual) between the estimated value calculated by the autoregressive AR (p) model of time series analysis and the measured value is calculated, and is a statistic of the difference between the two signals (residual). The parameter crest factor or kurtosis is obtained, and the change of the parameter (the crest factor or kurtosis of the statistic of the difference (residual)) between the two signals of the estimated value and the measured value is input as a keyboard 40, According to the instruction of the mouse 40b, it is stored in the auxiliary storage device 33b, displayed on the display 50, output to the printer 60, and monitored, so that the minute dents on the tooth surface and the roughness of the tooth surface (surface (Roughness) Anomalies (diagnosis) of gear tooth surfaces such as defects were determined. Of course, the signal processing PC (personal computer) G may be a desktop computer or a notebook computer as long as the storage capacity and processing speed are appropriate.

(Time-series model analysis method)
In this analysis, an attempt is made to express a vibration acceleration signal generated when the gears are engaged with each other by an AR (autoregressive model) model of time series analysis.

The autoregressive AR (p) model for time series analysis is that the output (x _n ) at a certain point in time is a linear combination of past outputs (x _n-1 , x _n-2 --- x _np ) and the current input ( This model represents a system obtained as the sum of ω _n ), and is generally used when performing time series analysis, and is expressed as follows. Here, p is the order, a _i is a model parameter (regression coefficient), and ω _n is a white noise sequence (white noise white noise).

  When only an output such as vibration acceleration from the vibration system is available at the time of measurement (measurement), an analysis based on a time series model in which the input is regarded as a white noise signal (white noise white noise) is effective.

On the other hand, the determination of the order p of the autoregressive AR (p) model for time series analysis occurs when the gear to be inspected meshes with the normal gear, and the time series data x _n and the order p of the measured (measured) vibration acceleration. The difference (residual, deviation) between the value estimated by the autoregressive AR (p) model of time series analysis and the measured value is calculated while increasing the autocorrelation function (ACF) and partial autocorrelation ( The partial autocorrelation function (PACF) may be selected so as to become white noise (white noise). Here, the white noise is a signal having an average of zero and a variance showing a normal steady distribution, and assuming that autocorrelation is zero except when the time lag is zero.

  The optimal order p of the autoregressive AR (p) model obtained by this time series analysis is 30. FIG. 2 shows the autocorrelation (ACF) and deviation of the estimated value calculated from the time series model of the autoregressive AR (30) and the residual (deviation) of the signal actually measured when the normal gear and the gear to be measured are engaged. Autocorrelation (PACF) is shown. At this time, there is no peak in the ACF and PACF values, and the closer to perfect noise, the better the accuracy of the model.

2A and 2B, the horizontal axis represents the time lag value, and the vertical axis represents the ACF value (autocorrelation coefficient) and the PACF value (partial autocorrelation coefficient). As can be seen from FIG. 2, the time series model of the autoregressive AR (30) obtained has a sufficient accuracy because it is within the reliability limit. At this time, the parameters a ₁ , a ₂ , ----- and a ₃₀ of the autoregressive AR (30) model of the time series analysis are the maximum likelihood method (Maximum likelihood method) according to the program stored in the auxiliary storage device 33b. ).

In this time series analysis, the order P of the autoregressive AR (p) model, the residual (deviation) between the estimated value and the measured value, and the model parameters (a ₁ , a ₂ , ------ , A _n ) were calculated according to a program stored in the auxiliary storage device 33b, and the model order P was 30 and model parameters (a ₁ , a ₂ , -----, a ₃₀ ) were obtained.

  The number of data used for time series analysis was 32,768 points. As for the length of the data collection time, it is desirable to use data for a long time to some extent. Therefore, the rotation time of the motor D (500 rev / min) and the inspection object (diagnosis object) gear C is considered to be 6 to 8 seconds. . This corresponds to the time required for the inspection target gear C to rotate about three times. Furthermore, in this experiment of abnormality diagnosis of the gear tooth surface, since the transmission direction of force depends on the shape of the gear (helical gear helical gear), the rotational direction is counterclockwise (cw). Abnormality diagnosis was performed by measuring vibration acceleration signals in two directions (clockwise).

(Results of abnormality diagnosis by time series model analysis)
As described above, it has been found that the vibration acceleration signal when the inspection target gear C is engaged with the normal gear can be expressed as an autoregressive AR (30) model of time series analysis. Here, the difference between the estimated signal value calculated from the autoregressive AR (30) model of the time series analysis and the actual acceleration signal value (actual signal value or measured signal value) according to the program stored in the auxiliary storage device 33b. By monitoring (residual), it is determined whether or not there is an abnormality in the gear dents or the surface roughness of the gear teeth.

In other words, when the size of the dent is very small, the residual (deviation) between the estimated signal value and the measured signal value is small, but when there is a relatively large dent or tooth surface (surface roughness) defect, the actual measurement is performed. Nonlinear elements included in the signal increase and the residual (deviation) increases. The abnormality of the gear is diagnosed by detecting such a variation of the residual (deviation). FIG. 3 schematically shows a method for obtaining the residual. (However, y _i corresponds to x _i in equation (1).)

  In the present invention, the residual (deviation) by the time series analysis is represented by using the statistical parameters crest factor and kurtosis, and the abnormality of the gear is determined. The crest factor is an index of the wave height of the waveform, and the kurtosis is a value indicating how shocking the vibration waveform is or how sharp the waveform is. If there is a local defect in the rolling bearing or gear device, the signal becomes shocking, so it is actively used for diagnosis of the rolling bearing or gear device. However, in the diagnosis using only the statistics without using the time series analysis, the roughness of the gear tooth surface (surface roughness) in which there is no minute impact mark (for example, an impact mark of 3 μm or less) or a characteristic impact peak in the waveform. ) And other abnormalities were difficult to detect. (Refer to the description of FIG. 13 below)

(Abnormal diagnosis by statistical analysis)
Here, in order to compare the diagnostic method using the time series analysis according to the present invention with the conventionally used statistical analysis diagnostic method, the kurtosis, the skewness, and the crest factor ( parameter based on statistical analysis such as (crest factor). These parameters can be defined as follows:
In general, when the probability density function of the time series signal x (t) is p (x), the expected value or the average value μ of the time series signal x (t) can be expressed by the following equation.

The kth moment around the mean value of p (x) is

When the k-th moment of Equation (3) is expressed in a discrete system, the following equation is obtained. Here, N is the number of data, and k is the order of the moment.

Further, the standard deviation σ is expressed by the following equation.

Using the above, the skewness, the kurtosis (kurtosis, the fourth-order product factor around the mean divided by σ ⁴ (σ: standard deviation)), and the crest factor (crest factor) are as follows: Can be defined as However, Max peak in the equation (8) indicates the maximum value in the time series waveform.

(Typical vibration signal waveform for each gear state)
4, 5, and 6, with respect to the master gear B in FIG. 1, there are gears (A type, D type) without dents and gears with poor tooth surface roughness (tooth surface roughness) (HA). The measurement waveforms of vibration acceleration of typical time-series signals obtained when the molds are engaged and rotated at a constant speed are shown respectively.

  4 and 5, in the case of the gear determined to have no dent, a large disturbance in the vibration waveform was not seen, but in the case of the gear determined to have a dent, the vibration waveform was regular (periodic). Shocking peaks were observed. This is a typical phenomenon observed when a local defect such as a dent is present on the gear tooth surface (a regular (periodic) impact peak is observed). As shown in this figure, when there is a relatively large scratch on the gear, it is possible to determine the gear abnormality from the aspect of the measured waveform of the time signal (vibration acceleration of the time series signal).

  On the other hand, FIG. 6 shows a measurement waveform of vibration acceleration of a time series signal due to a poor tooth surface roughness (tooth surface roughness), which is generally compared with the waveform of a gear with a dent (FIGS. 4 and 5). It can be seen that the amplitude is about 3 times larger. However, there is no characteristic impact peak in the waveform, and apparently no difference from the waveform aspect of the gear A type and D type without dents is recognized.

  The gear abnormality diagnosis result using time series analysis and the diagnosis result by statistical analysis are shown below for comparison.

(Gear diagnosis by time series analysis of vibration acceleration signal)
Here, the result of performing a gear state diagnosis using the residual between the estimated signal calculated from the autoregressive AR (30) model described above and the actual acceleration signal (measured signal) is shown. In the evaluation, the residual is expressed using the crest factor and the kurtosis which are one of the statistical parameters.

  Using the A-type gear, the crest factor and kurtosis of the residual (deviation) calculated using 32,768 data points for 6.4 seconds measured in FIGS. 7 (a) and 7 (b) Shows the value normalized by dividing by the crest factor and the kurtosis of the residual in the gear (gear number 2) determined as having no dent. (Based on normal gear with gear number 2)

  From this figure, it can be seen that the gear abnormality determination sensitivity due to the crest factor is higher than the kurtosis. That is, when there is a diagnostic criterion for both (in this case, the diagnostic criterion value is a wave height factor of 1 when the value calculated for normal gear is 1, when it is 1.5 times larger than this, or in the case of kurtosis Is assumed to be abnormal when each value is larger than about twice the normal value) When determining whether there is a gear dent or not from the value, 17 out of a total of 23 samples at the standard crest factor FIG. 7A shows that there is a dent, although the size of the gear of FIG. 7 is small (in FIG. 7A), the standard wave kurtosis indicates that there are abnormalities in 10 samples (FIG. 7 ( b)).

  As a result of the determination, when a relatively large dent is present on the gear tooth surface as shown in FIG. 4 and a regular impact peak is observed in the signal waveform, an abnormality can be determined based on the kurtosis. If it is small, it is distinguished from the normal gear from the corresponding kurtosis and suggests that it is difficult to determine abnormality. (This is because only 10 samples out of the total of 23 samples are recognized as abnormal.) On the other hand, even if the crest factor is small in scratches such as dents, it is possible to determine the gear abnormality with high sensitivity. It can be seen that it is. (This is because it is recognized as abnormal in 17 samples out of a total of 23 samples.)

  On the other hand, FIGS. 8A and 8B show the standard crest factor and the standard kurtosis of the residual investigated for the D-type gear. From this figure, if the normal gear with gear number 1 is used as a reference, the crest factor and kurtosis of the residual should be used, as in the results seen with the A-type gear (FIGS. 7A and 7B). It can be seen that a gear abnormality can be determined. Moreover, it turns out that the sensitivity of the abnormality determination by a crest factor tends to be a little high compared with the sharpness similarly to the result seen in A type gear.

  FIGS. 9A and 9B show the results of diagnosis performed on the gear HA having a poor tooth surface roughness (tooth surface roughness). From this figure, when the normal gear with the gear number 6 is used as a reference, the crest factor and the sharpness are large for all gears, and the determination result is abnormal. For these gears, abnormalities were also judged in advance by a skilled inspector.

  On the other hand, in the case of the gear HA having a poor tooth surface roughness (tooth surface roughness), the determination result is the same, but unlike the A type and the D type, the abnormality determination based on the sharpness is more sensitive. Is recognized. This is because the gear HA type with poor tooth surface roughness (tooth surface roughness) does not have a significant impact peak seen in the waveform when there is a dent like the A type and D type gears. This is considered to be related to the fact that random irregular waveforms are shown.

  Based on the above results, by monitoring the residual between the estimated signal (estimated value) calculated from the time series model and the actual vibration acceleration signal (measured value), the impression mark and tooth surface roughness (tooth surface roughness) It turns out that the gear abnormality diagnosis by a defect etc. is possible.

(Measurement result of dent size on gear tooth surface)
FIG. 10 shows the result of examining a dent on the tooth surface of a typical A-type gear with a surface roughness measuring instrument. (If the roughness curve in the figure is above the reference line, there is a dent). From this figure, it was confirmed that there was a small dent of about 2 μm for gear number 7, 2.5 μm for number 8, 1 μm for number 19, and 1.5 μm for number 21. In this diagnosis, the gear number 7 is determined to be normal by the inspector, but the other gear numbers 8, 19, and 21 are determined to be normal because the inspector cannot confirm the presence of the dent. It was what was done. (In FIG. 7 (a), it can be seen that the crest factors for gear number 7, gear number 8, gear number 19, and gear number 21 are abnormal.)

  As described above, it is found that it is very difficult to determine whether or not there is a dent for a gear having a minute dent of 3 μm or less, based on the five senses and experience of a skilled inspector. On the other hand, in FIG. 7A, it can be read that the crest factors for gear number 7, gear number 8, gear number 19, and gear number 21 are abnormal.

(Comparison of gear judgment (diagnosis) results by time series analysis and statistical analysis)
FIGS. 11 to 13 show abnormality determination results by time series analysis and statistical analysis of vibration acceleration signals obtained for A-type, D-type, and HA-type gears.

  From these figures, it can be seen that the determination based on the time series analysis is generally more sensitive than the statistical analysis regardless of the gear form. In particular, as the degree of dents and surface roughness increases, the sensitivity of gear abnormality determination by time series analysis tends to increase, and a clear abnormality diagnosis can be performed.

  With respect to the A-type gear of FIG. 10, in the case of the number 7 and 21 gears whose dent size is smaller than 3 μm in the measurement value of the surface roughness measuring device, it is difficult to accurately determine whether or not the size is normal from statistical analysis. In the time series analysis, it can be seen that the abnormality determination can be made distinct from the normal case without the dent. 7 (a), it can be seen that each of the crest factors in the case of gear number 7 and gear number 21 is abnormal.

  Also, from FIG. 11, the kurtosis degree and the distortion degree that are conventionally used for abnormality determination of bearings, gears and the like (in the present invention, the judgment result by this shows almost the same result as the kurtosis degree. It can be seen that the sensitivity of abnormality determination using the crest factor is higher than that using the result of degree.

  Therefore, it can be said that it is more effective to determine the abnormality using the crest factor than the kurtosis when determining the gear dent abnormality.

  On the other hand, even in the case of abnormality determination in the D-type gear in FIG. 12, it can be seen that the sensitivity of abnormality determination is highest due to the crest factor of the residual between the estimated signal calculated from the time series model and the actual vibration acceleration signal.

  FIG. 13 shows the result of gear abnormality diagnosis by statistical analysis or time series analysis for an abnormal gear due to a poor tooth surface roughness (tooth surface roughness).

  From FIG. 13, in the case of the HA type gear in which almost no impact peak caused by the presence of a local defect such as an abnormal dent of the A type and D type gears is seen, the conventional statistical analysis method is normal. It can be seen that it is difficult to accurately determine a gear abnormality due to a poor tooth surface roughness (tooth surface roughness).

  On the other hand, in the method based on the time series analysis, it is possible to reliably capture an abnormality due to a poor tooth surface roughness (tooth surface roughness) with sufficient accuracy. This is a random signal with irregularity that does not have a clear difference between the normal gear and the waveform with no apparent dent by monitoring the difference between the estimated signal calculated from the time series model and the measured signal, that is, the change in the model. It is because the change of can be caught.

  In the present invention, a G-based gear of a signal processing PC that allows anyone to easily and accurately determine gear abnormalities such as dents and tooth surface roughness (tooth surface roughness) with quantitative numerical values at the on-site gear single item inspection stage. A dent diagnosis system 1 is provided. The abnormality diagnosis method newly proposed in the present invention performs gear abnormality detection by monitoring the crest factor and kurtosis of the residual between the estimated signal calculated from the time series model of the measured vibration acceleration and the actual measurement signal. .

  This diagnostic method has higher diagnostic sensitivity than abnormality diagnosis by statistical analysis that has been frequently used in the past, and it has been found that minute dent abnormality of 3 μm or less that could not be detected at all can be sufficiently discriminated. Further, it is convenient that a gear abnormality due to a defective tooth surface roughness (tooth surface roughness), which is difficult to determine by statistical analysis, can be accurately determined with sufficient accuracy.

  Although the embodiment of the present invention has been described above, the scope of the present invention is not limited to this, and it goes without saying that various modifications can be made without departing from the scope of the present invention.

  The present invention can be used in the industrial field where various gears are manufactured and sold.

The figure which shows the abnormality diagnosis apparatus of a gear tooth surface ACF value and PACF value of the residual between the estimated value and the actual measurement value by the AR (30) model Schematic diagram of how to calculate the residual (deviation) Typical time signal waveform in A-type gear Typical time signal waveform in D-type gear Typical time signal waveform in HA type gear Fig.7 (a) Standard crest factor of residual in A type gear Fig.7 (b) Standard kurtosis of residual in A type gear Fig.8 (a) Standard wave height of residual in D type gear Fig.8 (b) Standard kurtosis of residual in D type gear Fig.9 (a) Standard wave height factor of residual in HA type gear Fig.9 (b) Standard kurtosis of residual in HA type gear Measurement results of dents on the tooth surface of a typical A-type gear Comparison of statistical analysis results and time series analysis results for A-type gears Comparison of statistical analysis results and time series analysis results for D-type gears Comparison of statistical analysis results and time series analysis results for HA gears Block diagram of gear tooth surface diagnosis device

Explanation of symbols

A ... acceleration sensor, B ... master gear, C ... diagnostic gear, D ... drive motor, E ... sensor amplifier, F ... module type signal input / output unit, G ... signal processing PC (personal computer).

Claims (3)

  It has an autoregressive AR (p) model for time series analysis. When the master gear and the gear to be diagnosed are meshed and driven by a drive motor, the vibration acceleration generated is measured by an acceleration sensor installed in one place. The estimated value of vibration acceleration is calculated from the vibration acceleration signal measured by the acceleration sensor using the autoregressive AR (p) model, and is calculated using the vibration acceleration signal measured by the acceleration sensor and the autoregressive AR (p) model. The difference (residual) of the two signals from the estimated value obtained is calculated, the crest factor or the kurtosis of the statistic of the difference (residual) of the two signals in the time series analysis is obtained, and the change of these statistics An abnormality diagnosis method for a gear tooth surface, characterized in that a gear tooth surface abnormality such as a surface roughness of a meshing tooth surface of the gear to be diagnosed or a minute dent can be diagnosed by monitoring the gear.   A self-regression AR (30) model with an order of 30 in time series analysis. When the master gear and the diagnosis target gear are meshed and driven to rotate by the drive motor, the vibration acceleration generated is the acceleration installed at one place. An estimated value of vibration acceleration is calculated by the autoregressive AR (30) model from the vibration acceleration signal measured by the sensor, and the vibration acceleration signal measured by the acceleration sensor and the autoregressive AR (30 ) Calculate the difference (residual) of the two signals from the estimated value calculated by the model, obtain the crest factor or kurtosis of the statistic of the difference (residual) of the two signals of the time series analysis, An abnormality diagnosis method for a gear tooth surface, characterized by monitoring a change in statistics and diagnosing a gear tooth surface abnormality such as a surface roughness of the meshing tooth surface of the gear to be diagnosed or a minute dent. Occurs when it has an autoregressive AR (30) model with degree 30 of time series analysis, meshes with the master gear and the gear to be diagnosed, is incorporated in the gear tooth surface abnormality diagnosis device body, and is rotated by a drive motor. The vibration acceleration is measured (observed) by an acceleration sensor installed in one place, and a sensor amplifier that amplifies the vibration acceleration signal detected by the acceleration sensor and an A / D converter that incorporates the amplified vibration acceleration signal, for example, An estimated value of the vibration acceleration signal calculated by the autoregressive AR (30) model, which is taken into the signal processing PC via a module type signal input / output unit for converting into a 12-24 bit digital signal, and the acceleration The difference (residual) of the two signals from the measured value of the vibration acceleration signal measured by the sensor is calculated, and the two signals of the time series analysis are calculated. Seeking crest factor or kurtosis statistics by the difference (residual),
Changes in these statistics are stored in the storage device 33, displayed on the display 50, or output to the printer 60 for monitoring to monitor the surface roughness of the meshing tooth surface of the gear to be diagnosed and minute strikes. An apparatus for diagnosing an abnormality in a gear tooth surface, such as an abnormality in a gear tooth surface such as a mark.