Predicting Fault-Tolerant Workspace of Planar 3R Robots Experiencing Locked Joint Failures Using Mixture Density Networks

There are currently two existing methods to compute the fault-tolerant workspace of a redundant robot arm for a given set of artificial joint limits. However, both of these methods are very computationally expensive. This article proposes using a mixture density network to learn the probability that a rotation angle belongs to the fault-tolerant rotation ranges. A difference filter is used to remove outlying rotation angles predicted by the network, and the remaining rotation angles are grouped together to generate the fault-tolerant workspace. Because this method is highly computationally efficient, it can be used alongside a genetic algorithm to compute the optimal artificial joint limits to maximize the area of the fault-tolerant workspace for a given robot arm. The predicted fault-tolerant workspace is compared to the actual fault-tolerant workspace, which proves the effectiveness of this algorithm. The computational speed of this proposed algorithm is roughly 390 times faster than the traditional method. Finally, a trajectory is placed within the fault-tolerant workspace predicted by the proposed method, and the experimental results show that this trajectory is tolerant to arbitrary joint failures.