Fault detection for internal sensors of mobile robots based on support vector data description

Abstract

Fault detection and diagnosis is an important issue for mobile robots, especially for the case that the dynamics of fault models are unknown, where the samples of fault models are difficult to obtain. Support vector data description (SVDD) is an useful tool for model construction based only on one class of samples. This paper presents a fault detection method for mobile robots internal sensors based on SVDD. It assumes that only the samples from the normal model are available. The presented method firstly builds an compact hypersphere for these normal samples based on SVDD, then a new test data is validated with the obtained hypersphere. Simulation results of mobile robot fault detection show the accuracy of the method.