The aim of the paper is to develop an algorithm based on fuzzy logic (FL) systems for reconstructing cracks shapes, which will be faster and simpler to learn than neural networks, especially in case of large training set.
The inverse model in defectoscopy can be considered as reverse model of the whole measurement process (crack‐sensor‐output). The most important disadvantage of the inverse neural models with dynamic networks is that the performance of training is disappointing due to large training set and many inputs of such networks. The paper proposes the FL as the substitute of neural network. The typical ANFIS networks are sufficient only for simulating systems with small number of inputs. For this reason the paper developed the learning algorithm that produces relatively small number of rules and it can be used in case of systems with hundred of inputs and thousands of training pairs.
This paper provides details about algorithm for reconstructing cracks profiles that produces relatively small number of rules of the fuzzy system. The basis rule of inverse model with moving window for one‐ and multi‐frequency method is described. The results of profile identification in 2D and 3D space for real and simulated data are presented as well.
Generally, the algorithm proposed in this paper can be widely used for simulating multi‐input systems, which are described by a large training set.
