Desenvolvimento de plataformas analíticas portáteis baseados em línguas optoeletrônicas e narizes eletrônicos para análise discriminativa

Abstract

Artificial olfactory systems have been widely employed for the detection and discrimination of chemical compounds in various analytical applications, including quality control, environmental monitoring, diagnosis, and food analysis. However, their performance remains limited by high instrumental cost and constraints in the generation and processing of chemical information. In colorimetric systems, limitations in signal acquisition may reduce response resolution and hinder the discrimination of complex samples. In this context, the development of low-cost and more efficient approaches for chemical information extraction remains highly relevant in Analytical Chemistry. This thesis aimed to develop and evaluate two artificial olfactory systems based on a portable electronic nose and a paper-based colorimetric sensor array with multichannel signal acquisition, integrated with chemometric tools for chemical pattern recognition. The low-cost electronic nose was constructed using semiconductor sensors coupled to an open-source microcontroller platform, while the colorimetric system employed a paper-based sensor array combined with an AS7341 spectral sensor. In both cases, optimization of experimental conditions led to more stable and reproducible responses. The electronic nose enabled the separation of volatile organic compounds, whereas the colorimetric sensor array allowed the discrimination of acid–base systems and eleven types of food samples. Compared with RGB-based approaches, the multichannel system exhibited higher resolution, sensitivity, and robustness against experimental interferences. PCA and HCA confirmed the presence of multivariate chemical signatures capable of discriminating complex systems. Overall, the results demonstrate that both approaches are promising low-cost analytical tools for screening and quality control. Future work includes extension to different matrices and integration with machine learning methods to improve classification and predictive performance