Predicting composite laminates roughness: data-driven modeling approaches using force sensor data from robotic manipulators.

The development of autonomous finishing operations in manufacturing process has the potential to decrease the costs and increase the quality of the operations. In this context, robotic manipulators have been introduced in sanding and polishing applications. Inspired by the recent development in machine learning and robotics, this paper is focused on designing a system capable of estimating the surface roughness using only a force torque sensor integrated with a robotic manipulator that performs the sanding of fiberglass panels. We present an investigation into the usage of convolution neural networks on the force-torque data to produce a quantitative estimation of surface roughness. To validate the results obtained a profilometer is used to gather pre- and post-operation data. The establishment of a relationship between measured force data and post-operation surface roughness will be used to develop a prediction of the surface quality for sanding operation using robotic manipulators. This project intents to act as proof-of-concept that traditional robotic sensors, can be used beyond their original scope, minimize the complexity of robotic systems integrated into manufacturing processes. [ABSTRACT FROM AUTHOR]