Monitoring the additive manufacturing technique of fused deposition modelling: a new approach using electret microphone and signal processing

• Ano de defesa: 2024
• Autor(a) principal: Lopes, Thiago Glissoi [UNESP]
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Fabricação por Filamento Fundido; Monitoramento de Processos; Segmentação de Sinais; Fused Filament Fabrication; Process Monitoring; Signal Segmentation
• Link de acesso: https://hdl.handle.net/11449/258339; http://lattes.cnpq.br/5963545027959566; http://lattes.cnpq.br/1455400309660081; http://lattes.cnpq.br/8166874558095198

The Fused Deposition Modeling process, also known as Fused Filament Fabrication (FFF), involves building three-dimensional objects through the successive layering of molten plastic filaments. Despite its extensive use, monitoring and diagnosing this process remains complex, requiring sophisticated methods to ensure high quality and efficiency. The first layer in the FFF process is very important, as it serves as the foundation for the entire 3D printed object. This layer sets the base for subsequent layers, influencing the overall print quality, dimensional accuracy, and structural integrity of the final product. This makes the first layer a focus for FFF process monitoring studies, given that if any defect could be detected during this fabrication, the fabrication process can be canceled avoiding further losses. The present thesis addresses signifi-cant gaps in the current body of research on FFF process monitoring by integrating advanced signal processing methods with computational intelligence tools. By utilizing an electret micro-phone mounted on the extruder, the research performs comprehensive signal processing and feature extraction, including spectral analysis, Root Mean Square (RMS) values, Ratio of Power (ROP), Counts statistic, Kurtosis, and Skewness. The use of machine learning algorithms, par-ticularly Support Vector Machine (SVM) and Neural Networks, enhances the system's classifi-cation and diagnostic capabilities. Spectral analysis was pivotal in identifying frequency values linked to defects and in extracting features that effectively distinguished between different ma-chine states. The implementation of machine learning algorithms, notably SVM, achieved an outstanding 100% accuracy in classifying normal and irregular machine conditions based on the extracted features. The thesis also details the development of the fff_segmenter script. This innovative and Free and Open Source Software (FOSS) tool automates the segmentation of monitoring acoustic signals using control signals from the X and Y stepper motor axes and the signal's sampling rate, significantly reducing errors and time consumption associated with man-ual segmentation methods. The script’s segmentation results can be easily into other MATLAB or Octave applications. In conclusion, this thesis presents a comprehensive framework for en-hancing the monitoring and diagnostic capabilities of the FFF process. By integrating low-cost sensors such as the electret microphone, advanced signal processing, machine learning, and open-source tools, it provides a clear framework for FFF monitoring, paving the way for future advancements in additive manufacturing.