PSI - Issue 57

Mohammad F. Tamimi et al. / Procedia Structural Integrity 57 (2024) 121–132 125 Mohammad F. Tamimi & Mohamed Soliman/ Structural Integrity Procedia 00 (2023) 000 – 000 5 processing, the network dispatches an output from the nodes on the output layer. Each neuron in the input layer represents an input parameter and the output of a neuron o in layer m is computed as (Hambli 2010): = ( ) (3) =∑ −1 −1 + =1 (4) where denotes the activation function generally considered a sigmoid function (e.g., hyperbolic tangent and logistic functions (Dresia et al., 2019)), is the number of connections to the previous layer, −1 are the weights of each connection, and is the bias factor. The Levenberg-Marquardt algorithm (LMA) has been chosen for training the ANN in this paper (Hagan & Menhaj, 1994). Feedforward artificial neural networks employing LMA are utilized in this paper as a surrogate for the FE analysis, offering the crack driving parameter for a specific set of input parameters. 5. Case Study The discussed sensitivity evaluation method is applied to stiffened panels with T- and L-shaped stiffeners subjected to axial tensile fatigue loading. These panels investigated experimentally by Mahmoud and Dexter (2005), allow for validation of the adopted crack growth prediction approach. The panels were fabricated with a length of 3,454 mm, a width of 1,626 mm, and a main panel thickness of either 13 mm or 9 mm. The specimens were fabricated with four L (L101×76×8) or bulb T stiffeners (HP160×9) spaced at certain intervals. A center, through-thickness, initial crack was machined in the panels and tracked during the tests. A constant amplitude cyclic load was applied to all the specimens resulting in a stress range of 55 MPa with a load ratio of 0.2 (Mahmoud and Dexter, 2005). A general layout of the stiffened panels is shown in Figure 2. The main panels were fabricated utilizing A572 Gr. 50 steel (ASTM A572/A572M, 2015), whereas the stiffeners were made using grade AH36 steel (ASTM A131-14, 2014).

Fig. 2 . Layout of the investigated stiffened panels (adapted from Mahmoud and Dexter (2005)). In this paper, the crack driving parameter (i.e., J -integral) is computed using 3D finite element models of the stiffened panels in the ABAQUS environment, modeled using four-node shell elements (S4R element) (Simulia, 2018). The contour integral method is used to quantify the crack driving parameter and analysing the crack tip behaviour based on Brocks and Scheider (2001). The loading and support conditions are applied at opposing ends of the panel as shown in Figure 3. In this paper, the initial analyses of crack propagation were carried out without