Methods and systems for classifying the type and severity of defects in welds

• US 8,146,429 B2
• Filed: 08/03/2009
• Issued: 04/03/2012
• Est. Priority Date: 08/03/2009
• Status: Active Grant

• Current Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Original Assignee: Georgia Tech Research Corporation (University System of Georgia)
• Inventors: Ume, Ifeanyi Charles, Li, Renfu, Rogge, Matthew, Wu, Tsun-Yen
• Primary Examiner(s): Surin, J M Saint
• Application Number:
  US12/534,581
• Publication Number:
  US 20110023610A1
• Time in Patent Office: 974 Days
• Field of Search: 736/22, 735/97, 735/98, 735/99, 736/00, 736/02, 736/18, 736/27
• US Class Current: 73/622
• CPC Class Codes: G01N 2291/0258 Structural degradation, e.g. fatigue of composites, ageing of oils
• CPC Class Codes: G01N 2291/0427 Flexural waves, plate waves, e.g. Lamb waves, tuning fork, cantilever
• CPC Class Codes: G01N 2291/2675 Seam, butt welding
• CPC Class Codes: G01N 29/11 by measuring attenuation of acoustic waves
• CPC Class Codes: G01N 29/4445 Classification of defects
• CPC Class Codes: G01N 29/4481 Neural networks

Abstract

A method for determining the type of a defect in a weld may include determining a defect location and a corresponding defect signal by analyzing ultrasonic response signals collected from a plurality of measurement locations along the weld. The defect signal and the plurality of defect proximity signals corresponding to ultrasonic response signals from measurement locations on each side of the defect location may then be input into a trained artificial neural network. The trained artificial neural network may be operable to identify the type of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals and output the type of the defect located at the defect location. The trained artificial neural network may also be operable to determine a defect severity classification based on the defect signal and the plurality of defect proximity signals and output the severity classification.

Claims

1. A method for determining a type of a defect in a weld comprising:

determining a defect location and a defect signal corresponding to the defect location by analyzing ultrasonic response signals collected from a plurality of measurement locations along the weld;

inputting the defect signal and a plurality of defect proximity signals into a trained artificial neural network, wherein the plurality of defect proximity signals correspond to ultrasonic response signals from measurement locations on each side of the defect location and the trained artificial neural network is operable to;

identify the type of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals; and

output the type of the defect located at the defect location.

2. The method of claim 1 further comprising processing the defect signal and the plurality of defect proximity signals prior to inputting the defect signal and the plurality of defect proximity signals into the trained artificial neural network.

3. The method of claim 2 wherein the defect signal and the plurality of defect proximity signals are processed by dividing each signal into a plurality of segments and averaging each segment.

4. The method of claim 1 wherein the trained artificial neural network is operable to identify the type of the defect as a blow hole defect, a lack of penetration defect, a short leg defect, a lack of penetration/short leg defect or combinations thereof.

5. The method of claim 1 wherein the trained artificial neural network is further operable to:

determine a severity classification of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals; and

output the severity classification of the defect located at the defect location.

6. The method of claim 5 wherein the trained artificial neural network outputs a coded vector indicative of the type and severity classification of the defect.

7. The method of claim 1 wherein the trained artificial neural network is trained with a back propagation algorithm using a sample data set derived from destructively tested weld samples.

8. The method of claim 1 wherein the defect location is determined by:

filtering an ultrasonic response signal from each of the measurement locations to produce a filtered response signal for each of the measurement locations;

calculating an ultrasonic energy for each of the measurement locations with the filtered response signal corresponding to each of the measurement locations;

comparing the ultrasonic energy for each measurement location to the ultrasonic energy of adjacent measurement locations to identify potential defect locations, wherein, when the ultrasonic energy of a measurement location is less than the ultrasonic energy of the adjacent measurement locations, the measurement location is a potential defect location; and

analyzing fluctuations in the ultrasonic energy at measurement locations neighboring the potential defect locations to determine if a defect is present in the weld.

9. A method for determining a severity of a defect in a weld comprising:

determining a defect location and a defect signal corresponding to the defect location by analyzing ultrasonic response signals from a plurality of measurement locations along the weld;

inputting the defect signal and a plurality of defect proximity signals into a trained artificial neural network, wherein the plurality of defect proximity signals correspond to ultrasonic response signals from measurement locations on each side of the defect location and the trained artificial neural network is operable to;

determine a severity classification of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals; and

output the severity classification of the defect located at the defect location.

10. The method of claim 9 further comprising processing the defect signal and the plurality of defect proximity signals prior to inputting the defect signal and the plurality of defect proximity signals into the trained artificial neural network.

11. The method of claim 10 wherein the defect signal and the plurality of defect proximity signals are processed by dividing each of the signals into a plurality of segments and averaging each segment.

12. The method of claim 9 wherein the trained artificial neural network is trained with a back propagation algorithm using a sample data set derived from destructively tested weld samples.

13. The method of claim 9 wherein the defect location is determined by:

filtering an ultrasonic response signal from each of the measurement locations to produce a filtered response signal for each of the measurement locations;

calculating an ultrasonic energy for each of the measurement locations with the filtered response signal corresponding to each of the measurement locations;

comparing the ultrasonic energy for each of the measurement locations to the ultrasonic energy of adjacent measurement locations to identify potential defect locations, wherein, when the ultrasonic energy of a measurement location is less than the ultrasonic energy of the adjacent measurement locations, the measurement location is a potential defect location; and

analyzing fluctuations in the ultrasonic energy at measurement locations neighboring the potential defect locations to determine if a defect is present in the weld.

14. A defect classification system for identifying a type of a defect in a weld, the defect classification system comprising a controller, an acoustic signal generator, an acoustic signal detector, and a positioning device, wherein the acoustic signal generator, the acoustic signal detector and the positioning device are electrically coupled to the controller and the controller is programmed to:

induce ultrasonic signals at multiple measurement locations along the weld with the acoustic signal generator;

collect an ultrasonic response signal from each of the measurement locations with the acoustic signal detector and store each ultrasonic response signal in a memory operatively associated with the controller;

determine a defect location and a defect signal by analyzing the ultrasonic response signal from each of the measurement locations;

determine a plurality of defect proximity signals, wherein the plurality of defect proximity signals correspond to ultrasonic response signals from measurement locations on each side of the defect location;

input the defect signal and the plurality of defect proximity signals into a trained artificial neural network operatively associated with the controller, wherein the trained artificial neural network is operable to identify the type of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals; and

output the type of the defect located at the defect location.

15. The defect classification system of claim 14 wherein the trained artificial neural network outputs a coded vector indicative of the type of the defect.

16. The defect classification system of claim 14 wherein the trained artificial neural network is further operable to:

determine a severity classification of the defect located at the defect location based on the defect signal and the plurality of defect proximity signals; and

output the severity classification of the defect located at the defect location.

17. The defect classification system of claim 16 wherein the trained artificial neural network outputs a coded vector indicative of the type and severity classification of the defect.

18. The defect classification system of claim 14 wherein the controller is further programmed to process the defect signal and the plurality of defect proximity signals prior to inputting the defect signal and the plurality of defect proximity signals into the trained artificial neural network.

19. The defect classification system of claim 18 wherein the controller is programmed to process the defect signal and the plurality of defect proximity signals by dividing each signal into a plurality of segments and averaging each segment.

20. The defect classification system of claim 14 wherein the trained artificial neural network is trained with a back propagation algorithm using a data set derived from destructively tested weld samples.