Quantification and localization of cracks in steel specimens using improved artificial neural networks with experimental validation

Abstract

Cracks in any structure are undesirable since they frequently lead to the structure's fracture or failure. API X 70 steel is a critical component in pipeline manufacturing; additionally, it is prone to cracking due to the harsh working conditions it is subjected during its service operation (s). The current fracture detection methods either require disassembly of the tube's substructure for visual inspection or external excitation of the relevant area of the tube for subsequent dynamic analyses...etc; as a result, these approaches are highly complex and time consuming. For crack identification in pipeline steel, a simplified crack identification approach is provided in this work. The crack identification method is straightforward and based on simple stress, strain, and displacement measurements of load and absorbed energy at recognized places. Finite Element Analysis was used in the study, which was done using ABAQUS, a well-known commercial finite element tool.

Intelligent systems have recently been praised for solving complicated difficult, multidimensional issues. Artificial neural networks (ANN) have had a lot of success in solving these challenges, although they do have some limitations. The current work examines the application of the WOA-ANN hybrid model for crack length prediction using various inputs such as strains, stresses, and displacements to assess the technique's accuracy. The proposed method is, nevertheless, compared to GA-ANN, AOA-ANN, and WOABAT-ANN. The use of ANN in combination with metaheuristic optimization techniques aims to increase its significance. The weight of neuronal connection is significant. Some biases are also linked to neurons. Based on the input and goal output values supplied, connection weights and biases are changed to give the least possible error function. This method is commonly referred to as back propagation (BP). The explored approach is relevant to real-world engineering applications and regulates the status of structures. The evolution of fracture mechanics parameters is studied using standard ASTM test specimens. After modeling the tests with the Finite Element Method (FEM), the numerical model is then evaluated with experimental test analysis. With the mesoscopic GTN damage model, FEM is utilized to analyze the tensile failure process of one-sided notch samples and extract the data required for WOA-ANN. Our model is now ready to forecast various scenarios after collecting the data. When compared to other crack detection approaches, the findings produced utilizing WOA-ANN is effective