Monitoring fluidized bed dryers containing inert ABS particles using passive acoustic emissions and artificial neural networks

• Ano de defesa: 2024
• Autor(a) principal: Metzner, Willian Velloso
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Acoustic sensor; Coeficientes cepstrais de frequência mel; Fluid bed drying; Fourier transform; Mel frequency cepstral coefficients; Multi-layer perceptron; Perceptron multicamadas; Secadores leito fluidizado; Sensor acústico; Transformada de Fourier
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/74/74133/tde-26092024-150951/

This study aimed to monitor fluidized bed dryers with inert ABS plastic particles by applying the analysis of passive acoustic emissions recorded by a piezoelectric microphone, installed externally to the fluidized bed vessel. Audio features such as waveform, DFT and MFCC confirmed the existence of acoustic changes corresponding to variations in the agitation intensity of the inert particles. The MFCC coefficients were used as input neurons in a three-layer artificial neural network (ANN), developed to predict the dynamics of fluidization based on three case studies: air velocity of the fluidizing gas, liquid flow rate added, and drying time. The training and validation stages of the ANN converged after 15 epochs, through the minimization of the Loss function. The MFCC coefficients served as a robust basis for modeling by the neural network, displaying high predictive capacity with R² values above 0.8 in all case studies. This study highlighted that the application of passive acoustic signals and neural networks allows for real-time data acquisition of fluidized bed dryers with inert particles, being useful in monitoring the efficiency of the process and the quality of the product.