Intelligent robotic die polishing system through fuzzy neural networks and multi-sensor fusion

Abstract

Traditional die polishing is a labor intensive field requiring skilled machinists. A robotic die polishing system was designed and demonstrated conceptually in the Computer Integrated Manufacturing Laboratory of the Pennsylvania State University. Multiple vision sensors are employed to capture the images of the die texture for the robotic die polishing system. For each vision sensor, a multiple net invariant network (MNIN) model is employed to accommodate orientation changes and achieve shift invariance. And the multiple decisions from different MNIN models are fused together by using a trained ANN. Due to slow convergence of the ANN, fuzzy modeling is used to accelerate the training speed. The proposed system not only discerns patterns and create strategies for polishing rough-machined dies that initially have a range of unpredictable surface finishes due to machining variations including differences from tool changes and spindle vibrations, but also is capable of fusing multiple sensory information.