Inspection automatique et sans contact de la rugosité des pièces usinées

Abstract

In mechanical engineering, the parts surface quality (roughness and texture) is a very important quality requirement. Especially in sectors such as aerospace and automotive, this requirement is critical because of its impact on component reliability and fatigue. Indeed, crack initiation is directly related to the surface quality. The roughness measurement is usually done 'off line' in a post metrology center with instruments and techniques that generally consume a lot of resources (time, equipment and operators). Non-contact measurement systems (eg. confocal crayon) proposed today in the industry are not only very expensive but also very bulky for use in a machining station. The main goal of this thesis is to develop contactless strategies to evaluate the surface roughness automatically, which can be integrated directly into a machining center. These strategies must be based on empirical models and algorithms to estimate, with acceptable precision, the surface finish of machined parts. The first part of this work focuses on the development of a roughness estimation model for milled parts derived from a 2D vision system. Under the preliminary work, we noted that predictive models available in literature are not always suitable for high-speed machining. They underestimate the arithmetic roughness (Ra) compared to experimental measurements. Considering this information, we have developed and validated an empirical model that allows an evaluation of the arithmetic roughness Ra from an image captured by a camera and a priori knowledge of machining parameter ‘feed per tooth '. This model also takes into consideration the type of material machined and the cutting tool used (coatings and geometry). It was demonstrated that other roughness parameters such as amplitude profile parameters (Rz,Rt,Rk,Rq) and the spacing parameter Rsm may also be estimated using similar models. The main disadvantage of the proposed models in our first work was the necessity to know a specific machining parameter used to shape the surface: the feed per tooth. To overcome this limitation, a new system vision and models completely independent of of the machining parameters were developed. Only prior knowledge of the machined material is maintained. The second part of our work presents the development of new descriptors from the image of the machined surface taken with a digital microscope. The goal is to implement a system of post-treatment having the potential to assess quickly and contactless, the surface roughness of the machined parts. The study has also shown that other roughness parameters (Ra, Rq, Rp, Rt and Rz) may be estimated using only features extracted from images and models without needing to know the machining parameters used for generate the surface.

Date	22 May 2013
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Victor Songmene (Supervisor) & Antoine Tahan (Co-supervisor)