| Summary: | 碩士 === 國立臺北科技大學 === 自動化科技研究所 === 104 === In electronics industry, the automatic insertion technology of electronic component plays an important role. Traditionally, it is necessary to identify the geometrical shape of insertion objects to find a proper insertion pose. Even though some of insertion tasks can be done by an 4-DOF robot arm (SCARA), this SCARA robot has more workspace constraints than an six-DOF robot. In this research, we aim to solve the problem of automatic insertion for transformers.Transformers are difficult to insert because there are 6 pins in a transformer to be inserted at the same time. Since a SCARA robot is not suitable in this problem, we use a 6-DOF robot to validate the experiments. In this research, we propose a three-layer method including vision, motion, and decision layers. The vision layer is to extract important features of transformers for the decision layer. The motion layer is constructed by using Fuzzy C-means (FCM) to find representative insertion patterns. The decision layer based on Support Vector Machine (SVM) is used to predict the insertion pose for transformers. By training a great number of transformers, the hierarchical SVMs learn the relationship of vision features of transformer pins and the corresponding insertion poses. The result showed that the accuracy rate of the proposed method on five hundred testing transformers is up to 87%.
|
|---|
