Ultrasonic-Based Impact Energy Level Prediction in Composite Materials Using Sequential Backward Selection and Adaptive Neuro-Fuzzy Inference System

Abstract Accurately predicting impact energy levels in glass fiber-reinforced polymer (GFRP) composites is crucial for assessing material performance under varying impact conditions. This study presents a novel methodology that integrates sequential backward selection (SBS) and adaptive neuro-fuzzy inference system (ANFIS) to enhance the precision of impact energy estimation using non-destructive evaluation techniques. The proposed approach begins with the application of controlled impact energies to composite specimens, followed by ultrasonic inspection using the Mistras system to acquire B-scan and C-scan images. These images are subsequently converted into representative signals, from which key indicators are extracted. To optimize computational efficiency and improve predictive accuracy, SBS is employed to systematically select the most relevant features, minimizing redundancy and noise. The refined feature set is then used as input for an ANFIS model, which effectively captures nonlinear relationships between ultrasonic data and impact energy levels. The results demonstrate the potential of integrating advanced machine learning techniques with ultrasonic non-destructive evaluation for precise and reliable impact energy prediction in composite materials. This methodology provides a robust framework for structural health monitoring and predictive maintenance in industries where composite integrity is a critical concern.