Abstract

To improve the surface properties of fiber-reinforced polymer composites, one method is to employ thermal spray to apply a coating on the composite. For this purpose, it uses a metal mesh serving as an anchor between the composite and the coating to increase adhesion. However, the metal mesh is covered by resin, and getting an acceptable coating is only possible through an optimum exposure of the metal mesh by sand blasting prior to coating. Therefore, this study aims to develop a closed-loop control system to inspect and blast the parts properly. Specifically, this approach takes the top-view images from a microscope as the inputs. A convolutional neural network (CNN) is trained to correlate these images with the corresponding exposure levels of the metal mesh, measured by a destructive method. Then, this trained CNN model can estimate the exposure level by only analyzing the top-view images. Finally, it serves as a feedback mechanism to guide the subsequent sandblasting operations. The state-of-the-art has only examined the sand-blasted composites manually, requiring expertise and experience. This method automates the inspection process efficiently with an inexpensive portable digital microscope. The experimental results show that the method can distinguish the status of surface preparation successfully, and it is practical in closed-loop control. This study also has applications to various fields of manufacturing for defect detection and closed-loop control.