10th WorldS4 2026

Authors - Erika Haydee Rubio-Camara, Oscar May Tzuc, Elsy Maria Rosales-Uc, Fran-cisco Gilberto Herrera-Chale, Roman A. Canul-Turriza, M. Jimenez Torres
Abstract - Mechanical vibration energy harvesting has emerged as a promising strategy for supporting sustainable energy generation in industrial environments, where machinery and transportation systems continuously produce recoverable vibrational energy. This study presents the development and evaluation of predictive models based on deep multilayer perceptrons (DMLP) and Convolutional Neural Networks (CNNs) for estimating the energy potential associated with mechanical vibrations under industrial operating conditions. A simulation frame-work was implemented using experimentally reported operational ranges, including vibration frequencies between 10 and 50 Hz, amplitudes from 0.01 to 0.03 m, and temperatures between 25 and 45 °C. The analysis considered piezoelectric, electromagnetic, and triboelectric harvesting mechanisms to evaluate model adaptability under different scenarios. The predictive framework was implemented using TensorFlow and validated through a 10-fold cross-validation strategy combined with hyperparameter optimization. Results indicate that both architectures achieve high predictive capability for estimating harvested energy; however, CNN models consistently outperformed Deep MLP models, obtaining lower prediction errors and higher stability across validation folds. The superior performance of CNNs is associated with their ability to capture localized patterns and structured relationships within vibration-related data. The proposed method-ology demonstrates the feasibility of integrating artificial intelligence techniques into vibration-based energy harvesting systems for industrial applications. Furthermore, the study provides a computational framework for evaluating operational conditions, optimizing harvesting performance, and supporting the design of sustainable self-powered monitoring systems.