Search

In this paper, a computer vision inspection methodology is presented for the detection and segmentation of areas of wear by electric arc on the surface of electrical contacts using segmentation by instances. Cadmium is a toxic metal frequently used in the manufacture of alloys for electrical contact, so efforts have been made to replace or eliminate this element from the industry. Our work proposes an approach oriented to the automation of the inspection process, which reduces the subjectivity of the repetitive task of inspecting electrical contacts, allowing to compare two different alloys, such as one with Cadmium (Cd) and the other with Copper (Cu)., subjected to the same conditions and determine which of them presents less wear by electric arc. In this way you can know if the new part is better, equal or worse in terms of performance. Initially, an adaptation was made of the source code hosted in a github repository of mask R-CNN of the Matterport corporation, where variables, classes were modified, and directories were created according to our needs. The results obtained show that the masks generated by the algorithm adjust very well to contact wear, despite the irregular shape of the damage. However, the "charred" edge was probably not considered, because in training images it was a feature that few images presented, so segmentation may not be considering this factor. [ABSTRACT FROM AUTHOR]