Scribd
Upload
8 views2 pages

Gear Crack Detection Methodology

The document discusses a polynomial model for analyzing time series data of gear signals, where coefficients indicate the similarity of the signals and can be used to assess local tooth defects. It describes experimental tests conducted on a test rig to gather data, using sensors to measure vibrations and imaging systems to track crack growth in gear teeth. The study focuses on characterizing 2D transverse cracks, which are common in spur gears, by measuring their lengths through image processing techniques.

Uploaded by

pandunugraha04
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views2 pages

Gear Crack Detection Methodology

The document discusses a polynomial model for analyzing time series data of gear signals, where coefficients indicate the similarity of the signals and can be used to assess local tooth defects. It describes experimental tests conducted on a test rig to gather data, using sensors to measure vibrations and imaging systems to track crack growth in gear teeth. The study focuses on characterizing 2D transverse cracks, which are common in spur gears, by measuring their lengths through image processing techniques.

Uploaded by

pandunugraha04
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

time series, x(t) and y(t), one may establish a polynomial model with degree p, which can be

expressed by

x t = c0 + c1 y t + c 2 y 2 t + + c p y p t + e(t ) (10)

If the two time series are similar, the polynomial coefficients and residual e(t) would be small.

Otherwise, they will be large. Therefore, one can gauge the difference between the two by looking

at the magnitude of coefficients. Li and Yoo substituted the measured gear signal into x(t) and the

ideal gear signal into y(t), (The ideal gear signal was obtained by using tooth averaging to extract

the part of signal that is repeated from one tooth mesh to the next.) and used the sum of

coefficients’ magnitude as an indicator of the local tooth defect.

The aforementioned gear condition indices include “condition index”, such as the Figures

of Merit, Peak-to-Valley, RMS, and NA4, which produces a single condition number from gear

vibration to reflect the overall condition of a gear, and “defect profile”, such as the energy operator,

tooth average residual, and narrow band demodulation, which produces a time series to give a

tooth-to-tooth account of the condition. Local defects show up as local aberrations in them. In this

study, the maximal value of a defect profiler while a tooth is in mesh is taken as an indicator for the

condition of that tooth. A third class of algorithms, including “model prediction error”, “correlation

coefficient discriminant”, and “polynomial discriminant function”, can be used as either a condition

index or a defect profiler.

3. Experimental work
Gear run-to-fail tests were conducted with a RPI test rig to provide calibration and test data

for the proposed method. The actual experimental setup diagram is shown in Fig. 1. It has a 40HP

AC driver connected to a 75HP regenerative DC Motor, acting as a dynamometer, via the test

gearbox. The gearbox contains a 23 teeth pinion and a 54 teeth gear. The gear face width is 0.5 inch

7
and the pressure angle is 20 degree. Two types of sensors, an accelerometer and two optical rotary

encoders, were used to measure the spur gearbox translational vibration and rotational

displacements. The accelerometer is a PCB (model number W352A78) piezoelectric accelerometer

with a bandwidth of 10kHz mounted on top of the gear housing, right above the bearing of the

gearbox. Its output was lowpass filtered at 10kHz. Synchronized averaging, which attenuates

noises that are not repeated from one gear rotation to the next, of the accelerometer output was

performed with the encoder attached to the pinion shaft and an average was recorded once every

1,536 to once every 4,609 pinion revolutions throughout a test.

To document and keep track of crack growth, an imaging system was setup to take enlarged

digital pictures of the side and face of pinion teeth periodically from which crack size is estimated

using image processing software. For example, test 1 data contains 473 synchronized averages of

gear vibration and 47 sets of images (hence crack sizes).

The tests were setup to produce a 2D transverse crack like the one shown in Figure 2, which

is a common failure mode of spur gears with a backup ratio greater than one. This failure mode

was chosen also because 2D cracks are geometrically less complicated than 3D cracks and

therefore are easier to characterize in terms of size. In this study, the size of the crack is the length

of the crack in the direction across the tooth thickness.

To determine the actual crack length from a zoomed-in image of a small area around the

crack, two reference marks are made on the side of the tooth near the crack as shown in Figure 3.

To calibrate the distance between the two marks, an image of the tooth is taken and the pixel counts

between the two marks, and across the tooth thickness are extracted with the image processing

software. The distance between the reference marks is then calculated from the pixel counts and

the known tooth thickness. To determine the crack length from a zoomed-in image later, one

Common questions

Powered by AI

The 2D transverse crack was chosen as the failure mode because it is a common issue in spur gears with a backup ratio greater than one. Additionally, 2D cracks are geometrically less complex than 3D cracks, making them easier to characterize in terms of size and behavior, which simplifies the analysis and study of crack growth .

To detect and quantify tooth mesh conditions over multiple gear cycles, the methodology involved the use of accelerometer and optical rotary encoder data to perform synchronized averaging. This digital signal processing technique helps in identifying local aberrations corresponding to defects. The test recorded synchronized averages of gear vibration every few thousand pinion revolutions, which allowed tracing progressive changes in gear condition attributable to factors like crack growth .

A condition index generates a single numerical value reflecting the overall condition of a gear by synthesizing gear vibration information. Examples include Figures of Merit, Peak-to-Valley, RMS, and NA4. A defect profile, on the other hand, provides a time series detailing the condition on a tooth-to-tooth basis. It is capable of highlighting local defects which appear as local deviations. Methods like energy operators or narrow band demodulation produce these defect profiles, offering a more granular view of gear health .

The crack growth was documented by setting up an imaging system to take enlarged digital pictures periodically of the side and face of pinion teeth. This allowed estimation of the crack size using image processing software. The crack size was represented by the length of the crack through the tooth thickness. Reference marks were made on the side of the tooth near the crack to calibrate measurements. Pixel counts between these marks and across the tooth thickness were extracted, and the known tooth thickness was used to calculate the distance between marks and thus the crack length .

Tooth averaging is used to extract ideal gear signals because it isolates the repetitive components of gear vibration signals that are consistent from one tooth mesh to the next, reducing noise and the influence of transient disturbances. By averaging these repetitive signals, a clearer picture of the baseline or ideal gear performance is obtained, permitting more accurate detection of discrepancies associated with defects when compared to measured signals .

Crack length measurements from digital images of gear teeth are calibrated using reference marks placed near the crack on the tooth surface. An image processing software extracts pixel counts between these marks and across the tooth thickness. With the known tooth thickness, these pixel counts can be converted into actual distances, allowing calculation of the crack length. This method ensures accuracy in tracking the propagation of faults within gears .

The experimental setup involved gear run-to-fail tests conducted with an RPI test rig designed to provide data for the proposed method. It utilized a 40HP AC driver connected to a 75HP regenerative DC motor via the test gearbox, which contained a 23 teeth pinion and a 54 teeth gear. Two types of sensors were used: a PCB piezoelectric accelerometer with a 10kHz bandwidth mounted on the gear housing above the gearbox bearing, and two optical rotary encoders for measuring translational vibration and rotational displacements. The accelerometer output was lowpass filtered at 10kHz, and synchronized averaging was performed with the encoder attached to the pinion shaft .

The polynomial model technique can assess the similarity between two time series by modeling the measured gear signal as x(t) and the ideal gear signal as y(t). A polynomial model with degree p is established, and the polynomial coefficients reflect the similarity. If the two time series are similar, the coefficients and residual e(t) will be small. Large coefficients and residuals imply differences between the time series. This technique indicates local tooth defects by using the sum of the coefficients' magnitude as a measure of difference. Local defects are identified by observing local aberrations in defect profiles like energy operators or tooth average residuals .

Synchronized averaging plays a crucial role in reducing noise in the experimental setup for gear condition monitoring. It attenuates noise that is not consistent across successive gear rotations, isolating the repetitive components of the signal that are indicative of gear condition. During the experiment, synchronized averaging was performed with the encoder attached to the pinion shaft, and averages were recorded at defined intervals of pinion revolutions, ensuring reliability of the vibration signal data .

The polynomial discriminant function can be employed as a model prediction error by using it to quantify the deviation between expected and observed gear vibration signals. It can serve to identify mismatches resulting from tooth defects. By applying the polynomial discriminant function to gear monitoring data, one can either confirm the predicted condition index or refine it based on emerging defect patterns. The degree and nature of these prediction errors guided by polynomial coefficients can then influence maintenance decisions .

You might also like