Desenvolvimento de ferramenta computacional baseada em Multiple Instance Learning para análise de vibrações mecânicas através de uma câmera digital

Abstract

In the engineering field, there is a need for monitoring and evaluating equipment that is in constant operation. A type of mechanism can cause problems in some components, so it can also cause a failure in the mechanical element, varying early as changes, a change in the operations applied to the equipment can occur. However, there is a high cost of acquisition, installation and people well prepared to perform this type of maintenance. Thus, it is important to study and develop new techniques that enable this type of operation to prevent premature failures. With this in mind, a low-cost method is proposed, without the need for only a low cost to carry out the filming of the equipment in operation and a computer for the treatment of the image and the analysis of data, so that it can be thought without contact. In this work, a spectrum is developed in Python to identify the positioning of an object of interest in the footage, in order to later make a frequency analysis. It is expected to compare the results obtained through the imaging technique, with a professional accelerometer model, so that frequency values can be found satisfactorily close. With this in mind, tests were carried out on a cantilevered beam. An impulsive input was applied with an impact hammer so that it was possible to perform a vibrating motion at equispaced points of vibration. This was performed differently, both by the accelerometer and by the imaging technique, later performed as frequency.