Support vector data description for calibration monitoring of remotely located microrobotic system

Abstract

Product miniaturization and high precision are current trends in industry. The microrobotic system developed in this study is particularly suited for the handling of small parts, a task that requires accurate positioning within extremely small tolerances. To ensure consistent positional accuracy, the performance of the robot needs to be monitored regularly. This paper proposes the use of a nonparametric support vector data description (SVDD) method to monitor a microrobotic system and detect out-of-calibration conditions. By fitting a hypersphere around calibration data, SVDD envelops the calibration data within a high-dimensional space with the smallest possible volume. The SVDD monitoring scheme manifests clear benefits over other one-class classification methods, i.e., principal component analysis (PCA) data description and nearest-neighbor data description. The non-parametric nature of SVDD would be quite useful when sample distribution is abnormal or when no prior knowledge about the distribution is available. In addition, the use of SVDD in calibration monitoring permits a flexible decision boundary that uses various nonlinear kernel functions. To overcome the technical limitations of remote calibration of a robot, in this work the microrobotic system has been integrated with Web-accessible sensor networks for remote monitoring and accuracy quantification. This integration provides many technical benefits. Robot engineers do not have to be on-site, and the positioning accuracy of the robot can be quantified and verified over the Web. Geographical barriers can also be overcome, and the robot can be merged with automated information networks for more efficient production planning.