Enhancing Damage Detection Using Reptile Search Algorithm-Optimized Neural Network and Frequency Response Function

Purpose: This study aims to enhance structural damage detection in steel beams by optimizing Artificial Neural Network (ANN) training using the Reptile Search Algorithm (RSA). The hybrid ANN-RSA model is proposed to address limitations such as local optima entrapment and constrained convergence accuracy. Frequency Response Function (FRF) data is utilized as input to improve the precision of damage location prediction. Methods: FRF data was obtained from experimental vibration analysis and validated through Finite Element (FE) modeling. Damage scenarios were simulated by introducing notches of identical dimensions at varying positions on steel beams. The RSA-optimized ANN model's performance was compared against alternative models, including Particle Swarm Optimization (ANN-PSO), Artificial Bee Colony (ANN-ABC), and Genetic Algorithm (ANN-GA). Key metrics like prediction accuracy, regression values, and computational efficiency were evaluated. Results: The hybrid ANN-RSA model demonstrated superior performance, achieving a regression value nearing 1 and reducing prediction errors by approximately 15% compared to ANN-PSO, ANN-ABC, and ANN-GA models. For all damage scenarios, the model exhibited consistent accuracy with error margins below 1%, outperforming competitors in both computational speed and precision. The ANN-RSA model required fewer iterations for convergence, highlighting its computational efficiency. Comparative analysis showed the RSA method effectively optimized ANN parameters, improving damage localization reliability and accuracy. Conclusion: The integration of RSA into ANN training significantly improves the accuracy and computational efficiency of damage prediction in structural health monitoring. The use of FRF data as input ensures minimal information leakage and enhances model precision.