Application of neural networks to segment images of peen formed work pieces

• Grant number: 10/05483-1
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: May 01, 2010
• End date: November 30, 2011
• Field of knowledge: Engineering - Mechanical Engineering - Manufacturing Processes
• Principal Investigator: Flavius Portella Ribas Martins
• Grantee: João Pedro Prospero Ruivo
• Host Institution: Escola Politécnica (EP). Universidade de São Paulo (USP). São Paulo , SP, Brazil

Abstract

Peen forming is a cold forming technique commonly used by the aeronautical industry to produce plates and stiffned panels exhibiting aerodinamical shapes. However, the development of controlled such processes requires proper knowledge of the relationships among their intervening variables. Coverage, the shotted area to total area ratio, one of the most important such variables, is usually estimated by human visual inspectors supplied with magnifying graded lenses. In order to improve such a tedious and failure prone technique, a new computer vision method, based on a trained MLP neural network, is proposed. Applied to optically magnified images of peen formed pieces, it is expected that the referred method could properly identify the shotted areas and, as a consequence, be the kernel of an automatic method of peen forming coverage estimation. (AU)