This paper presents a non-destructive testing approach using a magnetic flux leakage technique for the inspection of oil pipelines. This technique is based on the magnetic field saturation of the pipe walls, by pumping a so-called "electronic pig" through the pipeline, from one compression station to the next, which may be located several kilometers away. The inspection pig is fitted with a circunferential array of magnetic flux leakage detectors, which respond to anomalies and defects in the pipe wall. The objective of this paper is to detect and identify several types of defects by processing the responses detected by the sensors. Velocity variation and vibration of the electronic pig are some of the parameters taken into account in this research.A data pre-processing stage is described, including a low-pass anti-aliasing filter to remove high frequency (noisy) components of the detected signals. An interpolation algorithm is used to modify the spatial data sampling frequency, reducing the effects of variations of the pig velocity, so that the distance between two adjacent data samples becomes independent of the velocity. After being detected, the defects are classified by employing a neural network specially designed for this application using a set of measured training data. Experiments have been carried out using data obtained with a test pipeline having 14 inches in diameter and 2,410 meters in extension, whose defects were previously known. Experimental results obtained are shown to verify the efficiency of the proposed approach. This work has been developed in cooperation with Petrobras (Petróleo Brasileiro S. A.).
Key words: artificial intelligence, computer processing and simulation
