出版社：Academy & Industry Research Collaboration Center (AIRCC)
摘要：This paper describes a classification method for raw sensor data using a Fuzzy InferenceSystem to detect the defects in large LNG tanks. The data is obtained from a Magnetic FluxLeakage (MFL) sensing system which is usually used in the industry to located defects inmetallic surfaces, such as tank floors. A robotic inspection system has been developed inconjunction with the presented work which performs the same inspection tasks at much lowertemperatures than human operators would thus reducing the shutdown time significantly whichis typically of the order of 15-20 million Dollars per day. The main challenge was to come upwith an algorithm that can map the human heuristics used by the MFL inspectors in field tolocate the defects into an automated system and yet keep the algorithm simple enough to bedeployed in near real-time applications. Unlike the human operation of the MFL equipment, theproposed technique is not very sensitive to the sensor distance from the test surface and thecalibration requirements are also very minimal which are usually a big impediment in speedyinspections of the floor by human operator. The use of wavelet decomposition with Coifletwaves has been utilized here for deconvolving the essential features of the signal beforecalculating the classification features. This wavelet was selected to its canny resemblance withthe actual MFL signals that makes these wavelets very natural basis function fordecomposition..
