This paper presents an improved genetic local algorithm by incorporating the simulatedannealing technique into the perturbation process of the genetic local search algorithm and proposes an improved-genetic-local-search-algorithm-based inverse algorithm for two-dimensional defect reconstruction from the magnetic-flux-leakage signals. In the algorithm, a radial-basis-function neural network is utilized as a forward model, and the improved genetic local search algorithm is used to solve the optimization problem in the inverse problem. Experiments are presented to compare the proposed inverse algorithm with both the canonical-genetic-algorithm-based inverse algorithm and the geneticlocal-search-algorithm-based inverse algorithm. The results demonstrate that the proposed inverse algorithm is more accurate and robust to the noise.
INTRODUCTIONThe magnetic-flux-leakage (MFL) method has established itself as the most widely used in-line inspection technique for the evaluation of gas-and-oil pipelines. One important challenge in MFL NDE is the determination of the defect parameters such as the length, width, or defect shape on the basis of the information contained in the measured signals. Iterative methods are commonly used approaches for solution of inverse problems (see [1] and references therein). These methods involve solving a well-behaved forward problem in a feedback loop. Traditionally, numerical models such as the finite-element model (FEM) have been used to represent the forward process. However, iterative methods using the numerical-based forward models are computationally expensive. Neural networks are utilized for solving inverse problems in NDE [1-4] and used to represent the forward process in iterative methods [1]. Huang et al.[3] described the use of a wavelet-basis-function neural network to predict three-dimensional defect profiles. Ramuhalli et al. [4] used two neural networks in feedback configurations. In this method, a forward network is used as the forward model, and an inverse network is used to predict the profile for given measured MFL signals. Ramuhalli et al. [1] proposed a neural-network-based iterative inversion algorithm using neural networks as the forward model. In the algorithm, the problem of defect profile reconstruction from MFL signals is formulated as an optimization problem, where the defect profile is updated using a combination of gradient descent and simulated annealing to minimize the error between the measured signal and the model-predicted signal. Li et al. presented a genetic local search algorithm (GLSA) for reconstructing the profiles of 3D defects from eddy-current NDE signals [5].This paper presents an improved genetic local search algorithm (IGLSA) by incorporation of the simulated-annealing technique into the perturbation process of the GLSA and proposes an IGLSA-based inverse algorithm for 2D defect reconstruction from MFL signals. In the algorithm, a radial-basis-function neural network (RBFNN) is utilized as the forward model, and the IGLSA is used to solve the o...