USO DE SENSORES DIGITAIS E MACHINE LEARNING PARA MONITORAMENTO DA AERAÇÃO DE GRÃOS ARMAZENADOS

Abstract

Digital temperature and relative humidity sensors emerge as an alternative for monitoring the grain mass temperature inside silos, as they are simple to install and use, simpler and less costly to maintain, and are easy to associate with the automation of the aeration system. Thus, the objective of the research was to study the efficiency of digital temperature sensors guiding different aeration strategies of stored grains, to verify the efficiency and applicability of digital sensors of intergranular relative humidity in the storage of grains, as well as to estimate energy efficiency of the aeration of grains. using Machine learning. The experiment was carried out in a storage unit in two metallic silos. The aeration in these silos was manually coordinated in the silo containing thermocouples, and was automated in the silo with a digital thermometry system. In addition to grain mass temperature data, the quality of sunflower grains was monitored during 90 days of storage in three silo heights (upper, middle and lower thirds). To evaluate the efficiency of intergranular relative humidity sensors, two samples of soybean grains were analyzed in the upper, middle and lower third of the silo, for grain mass temperature, intergranular relative humidity, equilibrium moisture content estimated according to data collected and moisture content of the sampled grains. The moisture content values obtained by estimating the system using the hygroscopic equilibrium equation and by the oven method were compared. The estimation of the equilibrium moisture content of the different equations was compared with the method of direct determination and estimated by the system. To investigate the aeration management energy efficiency and propose an estimation model based on machine learning, a sunflower grain storage database was used. The database consisted of information about the average temperature of the sensors, average temperature of the silo, external ambient temperature, occurrence of aeration, if any: cooling, heating, direct heating during aeration, and the energy efficiency of the aeration process. The models prediction by the algorithms and the obtaining of the clusters, in unsupervised machine learning, respectively, were processed by the Weka 3.8.5. The aeration strategies influenced the grain mass temperature and, consequently, the quality of the stored product, mainly in the moisture content. No influence of aeration strategies was identified for protein, oil and iodine content of the stored sunflower oil. The automation of aeration contributes to the use of better climatic conditions to carry out the process, compared to the manual system used. The aeration strategies adopted: cooling and conservation, proved to be efficient to soften the grain mass temperature in the automated process. The silo with automated aeration better preserved the quality of stored sunflower seeds. The equilibrium moisture content estimated by the data provided by the sensors did not differ from the moisture content determined by the oven. Digital temperature and relative humidity sensors proved to be efficient, as they contribute to the estimation of equilibrium moisture content with satisfactory accuracy. The model for estimating the energy efficiency of the aeration process proved to be efficient, identifying that during the aeration of stored sunflower grains the energy efficiency was 97.78%. The proposed aeration efficiency assessment model has applicability for use in predictive analysis of the process. Among the classifier algorithms tested in the SVM-Poly presented the best metrics and indicators, which is recommended for implementation in neural networks.