PSI - Issue 68

Morteza Khomami Abadi et al. / Procedia Structural Integrity 68 (2025) 1312–1318 Morteza Khomami Abadi et al. / Structural Integrity Procedia 00 (2025) 000–000

1313

2

1. Introduction Stress Intensity Factors (SIFs) are used to calculate the stress near the crack tip, which is critical for predicting failure and assuring structural component safety. Previous numerical methods like the Finite Element Method (FEM), Meshless Methods, Boundary Element Method (BEM), etc. are widely used for extracting SIFs. Also, new methods such as the extended finite element method, Isogeometric Analysis (IGA), and Peridynamic, were developed to determine stress intensity coefficients. In addition to numerical methods, some experimental methods such as Digital Image Correlation (DIC), Fracture Toughness Testing, Acoustic Emission Monitoring, etc. and analytical are employed to determine these coefficients. However, using the mentioned methods for problems with complexity in geometry, loading and boundary conditions is considered very time-consuming and expensive. Advances in recent years in computational intelligence led to the replacement of previous computational methods with new machine learning techniques to determine stress intensity factors. In this research, a statistical method based on the Group Method of Data Handling (GMDH) neural network is presented to solve the limitations of numerical and experimental methods in estimating stress intensity factors in plates and beams. The previous studies on stress intensity factor extraction methods are presented as follows. Tada et al. formulated stress intensity factors in terms of the energy release rate for cracked beams and plates. In this study, direct formulation was presented for various problems such as the center and single edge cracked, two and three dimensional specimen with crack, etc. Kienzler and Herrmann derived a simple formula for stress intensity factors in rectangular and circular cross sections based on the energy release rate. An artificial neural network (ANN) based on the acoustic emission (AE) data on compact tension specimens was trained to identify the stress intensity factor in the microcrack to fracture interval by Kim et al. The results show the ANN is a promising approach for estimating the material's stress intensity factor. A new two-level finite element method based on the symplectic series as global functions is presented by He and Su [4] to determine stress intensity factor, in which the conventional finite element shape functions as local functions are developed. The evaluation of stress intensity factors in cracked T-section beams is presented Ricci and Viola. Kumar calculated stress intensity factors for specific boundary conditions by taking into account the limitations in the structural dimension. These limitations include geometric conditions such as sample dimensions, crack dimensions, loading conditions, boundary conditions, etc. An isogeometric analysis method is developed for the stress intensity factors in curved crack problems by Choi and Cho. In the isogeometric method, Non-Uniform Rational B-Spline basis functions in CAD system is directly utilized in the response analysis. The stress intensity factors in pavement cracking by neural networks based on semi-analytical FEA are extracted by Wu et al. In this study a database was generated by analyzing varieties of pavement structures using elastic semi-analytical FEA program; secondly, from the results in the database, neural network (NN) based SIF equations were developed for new SIFs. The evaluation procedures of the stress intensity factor in the opening mode are established for the Euler– Bernoulli cracked beam based on the GMDH neural network, and the ABAQUS software by Alijani et al. A framework for the automated development of mechanics-guided handbook SIF solutions is presented based on interpretable machine learning via genetic programming for symbolic regression (GPSR) by Merrell et al. A prediction model based on an artificial neural network is presented to determine stress intensity factors for pressurized pipes with an internal crack by Seenuan et al. An artificial neural network (ANN) model was developed to enhance KI prediction accuracy in pressurized pipes. Predictions from the ANN model were compared with precise finite element analysis (FEA). In this research, a new statistical method based on neural network is presented to determine stress intensity factors. In this method, a new algorithm is introduced. The inputs of this algorithm are the results of SIFs extracted from Design of Experiments (DOD) and simulation in Abaqus software. After the training and prediction process by the GMDH neural network, the output of the algorithm is presented in the form of explicit equations to determine the stress intensity factors without limitation. This study aims to improve the accuracy and efficiency of SIF calculations, making them more applicable to engineering problems. This method solves the limitations of other numerical, statistical and laboratory methods in the complexity of geometry, loading, boundary conditions, etc. In addition, it significantly reduces the time and cost of calculations compared to other methods. By applying this method, researchers and engineers can better predict fracture in materials and structures, leading to safer and more reliable designs across various industries such as civil infrastructure, aerospace, automotive, etc.