RealSense: uma abordagem explorando aprendizado de máquina no monitoramento de temperatura e umidade em sistemas de compostagem

Abstract

Composting, as in "recycling of organic waste", is a way to recover nutrients and return them to their natural cycle, enriching the soil for agriculture and reducing the need for industrialized fertilizers. It is also a method to minimize waste, by properly disposing a material that would accumulate in dumps and landfills, creating foul odor and releasing methane gas and leachate, a liquid with the potential to contaminate soil and water. About 60% of the waste in the country today is composed of organic waste, which is deposited in dumps, generating a severe negative impact on the environment and public health. Throughout the composting process, being it an aerobic decomposition, there are variations in temperature and humidity. Without proper monitoring and control over these variables, biochemical reactions will not occur properly. There are studies targeting different types of soil using non-destructive methods to quickly and accurately measure humidity content, such as those based on capacitive sensors. Particularly in the case of composting, there is a need for further studies, considering the significant changes in its constitutive aspects over a single production cycle, adding to this the individual substrate characteristics of each windrow. Therefore, this work aimed to conceptualize a hardware and software proposal called RealSense, which explores machine learning techniques for the self-tuning of a capacitive sensor, considering its use in composting of different natures. The aim is to qualify the monitoring of humidity, providing greater control of proportions of different waste and aerator materials, resulting in a better activity for microscopic organisms that aid in decomposition and, consequently, generating a more suitable fertilizer. The evaluation of results considered the gravimetric analysis as validation basis for the proposed model. Through the analysis of the results, it was possible to conclude that the proposed technique is a fast and reliable way of measuring moisture in compost. It reached a correlation coefficient of 0.9939 between moisture content and the prediction.