Configurable ultrasonic flaw classification of oil pipelines

Abstract These two papers present an innovative method of configurable flaw classification and volume estimation in oil pipelines. In part I, the ultrasonic image acquisition system is introduced and surface and volume of the flaw are estimated with fuzzy image processing. A number of real figures illustrate the system performance. The flops calculation reveals that this fuzzy estimator could be integrated in a real time flaw detection system. In part II, at first, the dynamic detection of interesting points, i.e. as feature points at different levels of images, is proposed using wavelet transform. Furthermore, a guided searching strategy is used for the best matching from the coarse level to a fine level. Then, an error reduction neural network classifies the type of flaw using the wavelet transform as well as area and volume which were produced in Part I. Finally, a statistical error analyser verifies the system outputs by distinct inputs. The numerical results indicate that significant advantages are achieved over the other neural network based flaw characterisation schemes, whereas the accuracy of the predicted flaw profile can be controlled by the resolution of the network. Keywords: Nondestructive testingUltrasonicsFeature extractionWavelet transformNeural classificationStatistical modeling Acknowledgements This Research is funded by Pipeline and Communications Bureau of National Iranian Oil Company and is supported by the Chancellor of Sharif University of Technology (SUT), the Research Deputy Chancellor of SUT. The authors would like to thank Dr H. R. Amindavar and Mr E. Norouzi for their valuable comments.