Residual Strength Prediction of Aluminum Panels with Multiple Site Damage Using Artificial Neural Networks

Abstract

Multiple site damage (MSD) cracks are small fatigue cracks that may accumulate at the sides of highly loaded holes in aging aircraft structures. The presence of MSD cracks can drastically reduce the residual strength of fuselage panels. In this paper, artificial neural networks (ANN) modeling is used for predicting the residual strength of aluminum panels with MSD cracks. Experimental data that include 147 unique configurations of aluminum panels with MSD cracks are used. The experimental dataset includes three different aluminum alloys (2024-T3, 2524-T3, and 7075-T6), four different test panel configurations (unstiffened, stiffened, stiffened with a broken middle stiffener, and bolted lap-joints), many different panel widths and thicknesses, and the sizes of the lead and MSD cracks. The results presented in this paper demonstrate that a single ANN model can predict the residual strength for all materials and configurations with high accuracy. Specifically, the overall mean absolute error for the ANN model predictions is 3.82%. Furthermore, the ANN model residual strength predictions are compared to those obtained using the most accurate semi-analytical and computational approaches from the literature. The ANN model predictions are found to be at the same accuracy level of these approaches, and they even outperform the other approaches for many configurations.

A Novel Assisted Artificial Neural Network Modeling Approach for Improved Accuracy Using Small Datasets: Application in Residual Strength Evaluation of Panels with Multiple Site Damage Cracks

“…A variety of analytical and computational approaches can theoretically be used for estimating the residual strength of panels with MSD cracks. These approaches include analytical, semi-analytical (empirically corrected), computational and data-driven, and the accuracy of these approaches varies significantly [27]. Figure 1 illustrates the configuration of a typical panel having a lead crack along with adjacent MSD cracks, which is being considered here.…”

“…Hijazi et al [27] developed a single ANN model to predict the residual strength of panels with MSD made of any of the three different aluminum alloys (2024-T3, 2524-T3 and 7075-T6) and compared the ANN predictions accuracy to that of the three empirically-corrected models of Smith et al [30,44,45]. Their results have shown that both the ANN model and the empirically-corrected linkup models are able to give residual strength predictions with reasonably high accuracy where the average error for both approaches is about 4%.…”

“…The ability of the ANNs to learn by example, makes them particularly useful for modeling highly complicated and non-linear processes, since the development of analytical models for such processes is extremely difficult. ANN modeling has been successfully used in a wide variety of engineering applications including automatic control [2], solar energy systems [3,4], traffic and transportation [5], image processing [6], biomechanics [7], materials science and engineering [8][9][10], manufacturing [11,12], fracture mechanics and fault detection [13][14][15][16][17][18][19][20][21][22][23][24][25][26][27].…”

“…The reason behind this relatively high ANN prediction error can partly be attributed to the small size of the dataset used for training the network (only about 40 data points). The authors of this study, Hijazi et al [27], have recently used ANN modeling for residual strength prediction of panels with MSD. A dataset comprising 147 different configurations was assembled form many literature sources where 97 data points were used for training, 25 data points for validation and 25 data points for testing.…”

“…Situations where only small experimental datasets are available may be encountered in many fields because generating large datasets is time-and-cost prohibitive [14,27,35,36]. Therefore, different methods have been developed and implemented in ANN modeling to cope with the small dataset conditions.…”