Error prevention in robotic assembly tasks by a machine vision and statistical pattern recognition method

Assembly errors can occur in a robotic assembly system. In this paper, a method that predicts an assembly error is proposed. It considers that assembly errors occur under the condition that the geometric trajectory of a mated part and the relational position and orientation of a base part are outside the allowable tolerance. A certain point, which is determined by using a physical light reflectance model of a mated part, is followed with two high-speed cameras. A statistical pattern recognition method in which explanatory variables are tracked points in a three-dimensional space is then employed to predict an assembly error. The proposed method is applied to a peg-in-a-hole assembly by a selective compliance assembly robot arm (SCARA)-type robot and its potential value is discussed.