محمدرفيع حميدنيا

Polymer matrix composites, due to their high strength-to-weight ratio, corrosion resistance, an‎d versatile design capabilities, have found widespread applications in aerospace, automotive, an‎d civil engineering industries. One of the main challenges in using these materials is the phenomenon of fatigue, which can lead to premature failure an‎d a reduction in structural lifespan. Traditional fatigue prediction methods, such as stress–life curves an‎d stiffness degradation models, face significant limitations because they rely on costly experimental conditions an‎d are unable to simultaneously consider all influencing parameters. In the present study, using a dataset from the U.S. Department of Energy at Michigan State University consisting of over 1,500 experimental samples, an artificial neural network (ANN) model was developed to predict the fatigue life of polymer matrix composites. Various network architectures were trained by combining different numbers of hidden layers an‎d neurons an‎d by using different data partitions, employing the Levenberg–Marquardt algorithm. The data were divided into training, validation, an‎d testing subsets, an‎d the model’s performance was eva‎luated using statistical metrics such as mean squared error, correlation coefficient, an‎d stress–life plots. Additionally, a Non-dominated Genetic Algorithm was used to optimize the number of neurons an‎d hidden layers, an‎d the results were compared. To investigate the effect of fiber orientation on the neural network performance, fiber angles were introduced both directly an‎d through their sine an‎d cosine components as input parameters. The results showed that the artificial neural network could accurately predict the fatigue life of glass/epoxy polymer composites by capturing the nonlinear relationships among input parameters such as fiber angle, stacking sequence, fiber volume fraction, an‎d mechanical properties. In many cases, the correlation coefficient was close to one, an‎d the prediction error was very low. Comparison with classical models also demonstrated that the proposed method achieved higher accuracy an‎d better generalization capability under various loading an‎d stacking configurations.

عمر خستگي كامپوزيت , شبكه عصبي مصنوعي , الگوريتم ژنتيك نامغلوب

fatigue life of composites , artificial nueral network , non dominated sorted genetic algorithm

Mohammad Rafie Hamidnia

Dr. Fatollah Taheri-Behrooz