Depth estimation of inner wall defects by means of infrared thermography
There two common methods dealing with interpreting data from infrared thermography: qualitatively and quantitatively. On a certain condition, the first method would be sufficient, but for an accurate interpretation, one should undergo the second one. This report proposes a method to estimate the def...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Monograph |
| Language: | English |
| Published: |
Faculty of Electrical Engineering
2009
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/9704/1/78120.pdf http://eprints.utm.my/id/eprint/9704/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Universiti Teknologi Malaysia |
| Language: | English |
| Summary: | There two common methods dealing with interpreting data from infrared thermography: qualitatively and quantitatively. On a certain condition, the first method would be sufficient, but for an accurate interpretation, one should undergo the second one. This report proposes a method to estimate the defect depth quantitatively at an inner wall of petrochemical furnace wall. Finite element method (FEM) is used to model multilayer walls and to simulate temperature distribution due to the existence of the defect. Five informative parameters are proposed for depth estimation purpose. These parameters are the maximum temperature over the defect area (Tmax-def), the average temperature at the right edge of the defect (Tavg-right), the average temperature at the left edge of the defect (Tavg-left), the average temperature at the top edge of the defect (Tavg-top), and the average temperature over the sound area (Tavg-so). Artificial Neural Network (ANN) was trained with these parameters for estimating the defect depth. Two ANN architectures, Multi Layer Perceptron (MLP) and Radial Basis Function (RBF) network were trained for various defect depths. ANNs were used to estimate the controlled and testing data. The result shows that 100% accuracy of depth estimation was achieved for the controlled data. For the testing data, the accuracy was above 90% for the MLP network and above 80% for the RBF network. The results showed that the proposed informative parameters are useful for the estimation of defect depth and it is also clear that ANN can be used for quantitative interpretation of thermography data. |
|---|