Experimental crack identification of API X70 steel pipeline using improved artificial neural networks based on whale optimization algorithm

Abstract

Intelligent systems have recently received recognition for their ability to solve extremely complicated and multidimensional problems. Artificial Neural Networks (ANN) has quite a lot of success in overcoming such issues, but some limitation can be found. The present study discusses in detail the application of the WOA-ANN hybrid model for predicting the crack length based on different input values, i.e. strains, stresses, and displacements, to test the accuracy of the presented technique. The proposed technique is compared with GA-ANN, AOA-ANN, and WOABAT-ANN. Coupled metaheuristic optimization algorithms with ANN aim to increase its effeciency. The connectivity between neurons carries some weight. Neurons are also connected to some biases. Connection weights and biases are modified to give the smallest possible error function based on the input values, and corresponding target output values supplied. Back Propagation (BP) is the usual name for this approach. The investigated approach is related to real engineering applications and controls the structures’ state. Standard ASTM test specimens are chosen to study the evolution of fracture mechanics parameters. Next, an analytical model is developed by simulating the tests using the Finite Element Method (FEM) and validated with experimental results. FEM is used to analyse the tensile failure process of the one-sided notch samples with the mesoscopic GTN damage model and extract the data required for WOA-ANN. After collecting the database, our model is ready for predicting different scenarios. The obtained results using WOA-ANN are efficient compared to other techniques