Machine learning applied to hydro generator stator bars quality control

Abstract

In order to improve visual inspections carried out by quality control in the manufacture of hydro generator stator bars, as well as agreeing high levels of requirements from hydroelectric plant customers regarding the proper functioning of their machine components, is created the hypothesis that the use of machine learning would enable a semi-automatic tool to support visual inspection, which involves subjective decisions that depend entirely on the level of experience and commitment of its executor. The focus of the article is on the use of the Edge Impulse platform to develop the tool, since the software proves to be easy to use, provides several resources and provides access to people with low knowledge in the area of information technology due to its low level of requirement for programming languages and data processing. It was proposed to use the DOE (design of experiments) tool to analyze the parameters and observe the factors that influence the model final result. Graphics presented during the text help to understand how the quality, quantity and model of pictures influence machine learning. With that, in addition to obtaining the results, there was also a more comprehensive perspective on the Edge Impulse software.