Dynamic Stiffness Prediction in Cracked Cantilever Beams Using Enhanced ANN Techniques

This research paper presents an in-depth investigation into the application of Artificial Neural Networks (ANN) for predicting the dynamic stiffness characteristics of cantilever beams affected by cracks of varying depths and positions. Structural failure due to vibration remains a critical concern, and this study focuses on how crack-induced changes in modal stiffness compromise structural reliability. A significant research gap identified in previous studies is the limited consideration of crack location; this work addresses that gap by analyzing how both crack depth and position influence the natural frequency of the beam. Using modal analysis on cantilever beams with identical dimensions but different crack configurations, it was observed that deeper cracks and cracks near the fixed support led to a marked reduction in natural frequency, making the structure more vulnerable to resonance. Cracks located at the free end, however, showed negligible impact. To model and predict these dynamic behaviors, two ANN models were developed: a feedforward backpropagation network optimized using the Levenberg-Marquardt algorithm, and a hybrid ANN model integrated with Principal Component Analysis (PCA) for noise reduction and feature optimization. Simulation results served as the training dataset for these models, and both networks demonstrated high prediction accuracy for unseen crack scenarios. The study not only provides insight into the physical behavior of cracked cantilever beams but also offers a robust predictive tool for structural health monitoring, with practical implications across aerospace, civil, and mechanical engineering domains. Keywords: Cantilever Beam, Crack Detection, Modal Analysis, Natural Frequency, Artificial Neural Network, Levenberg–Marquardt Algorithm, Principal Component Analysis