RESUMO
This work focuses on acoustic analysis as a way of discriminating mineral oil, providing a robust technique, immune to electromagnetic noise, and in some cases, depending on the applied sensor, a low-cost technique. Thus, we propose a new method for the diagnosis of the quality of mineral oil used in electrical transformers, integrating a ferroelectric-based hydrophone and an acoustic transducer. Our classification solution is based on a supervised machine learning technique applied to the signals generated by an in-home built hydrophone. A total of three statistical datasets entries were collected during the acoustic experiments on four types of oils. The first, the second, and third datasets contain 180, 240, and 420 entries, respectively. Eighty-four features were considered from each dataset to apply to two classification approaches. The first classification approach is able to distinguish the oils from the four possible classes with a classification error less than 2%, while the second approach is able to successfully classify the oils without errors (e.g., with a score of 100%).