Pipeline defect detection and sizing based on MFL data using immune RBF neural networks

Corrosion inspection tools, or pipeline pigs, based on Magnetic Flux Leakage (MFL) are commonly used in oil-gas pipeline inspections. One of the difficult areas in applying this type of technology is how to recognize and quantify the corrosion characteristics. A Radial Basis Function Neural Network (RBFNN) has been found to be a suitable technique for such purposes. The RBFNN has excellent local closing ability and the Immune algorithm has self-organization ability. Utilizing the advantages of both, an Immune RBFNN (IRBFNN) algorithm was proposed to process the MFL data. In this paper, steps of the algorithm are shown in detail, and the model built was applied to lab data to assess this technique’s ability to determine the location and size of the corrosion spots on the pipeline. A segment of metal pipeline was taken as an object for inspection. Several corrosion spots were artificially made on it. The sizes of the square corrosion spots were first computed by using standard 3D finite-element analysis and this was then compared to the MFL data collected. These sizes were regarded as training inputs to the neural network. Experimental results show that the method successfully identified the location and size of corrosion on the pipe. It is fast in convergence, responds to the geometrical characteristics of corrosion correctly and, thus provides a promising new method for accurately detecting and sizing oil-gas pipeline corrosion.