| dc.description.abstract | It is estimated that by 2050 more than 500 million tonnes of polymeric materials will be produced and, if no action is taken, the oceans will have more weight in polymers than fish. It is based on these statements that the motivations for carrying out this work emerge. A solution to reduce the accumulation of these polymers is the use of biodegradable materials, which degrade with the environment. To determine the percentage of biodegradability of these materials, it is necessary to use standardized conditions, through the application of standards and laboratory tests. However, these tests are long since the time to perform them varies from 90 to 180 days. During this time it is necessary to maintain a precise control over the variables involved in the process, requiring a lot of involvement from the operators. Therefore, the objective of this work is the definition of an automated equipment model to determine the percentage of biodegradability of materials, using Internet of things and machine learning techniques. The work carried out in this area does not include the development of a fully automated machine, putting in doubt the humidity control methodology of the test environments. In addition, no record was found in the literature available for prototypes or equipment that use Machine Learning techniques, making this the great differential of the research carried out in this dissertation. The proposal aims to work on three fronts: evaluation of low cost sensors for real applications, when different measurement sensors of CO2 were evaluated, verification of the effectiveness of a precise control for soil moisture, when it was carried out a study of different control configurations and, finally, implementation of Machine Learning algorithms, which aim to predict the results of the biodegradability percentage of polymeric materials with less testing time. In this context, this research proposes the IoTBIO model. IoTBio was evaluated by assembling a prototype and simulations. In the end, it was possible to verify that the results of the percentage of biodegradability of an equipment automated by sensors improved significantly, when compared with the results of tests carried out in a non-automated way. ARIMA and Recurrent Neural Network algorithms were tested, more specifically the LSTM architecture. Only the Recurrent Neural Network was able to predict values with acceptable errors, because maximum errors of 13 % were reached between real values and predicted values for an execution of only 50 real test days, reducing by almost 1/3 the maximum time needed to perform these tests. The development of this work brought as main contributions the development of an automatic and intelligent prototype that uses Machine Learning algorithms to predict and reduce the costs of biodegradability tests, by reducing the total testing time and applying low added value sensors. The successful results of this dissertation resulted in the construction of three pieces of equipment to determine the percentage of biodegradability, which are available to serve industry and institutions at the Senai Institute for Innovation in Polymer Engineering in São Leopoldo. | en |
