Deep Convolution Neural Networks for the Classification of Robot Execution Failures

Abstract

Deep convolution neural networks (DCNNs) are popular deep neural networks and are widely used in object recognition, handwriting recognition, image processing, and so on. In this paper, manipulator fault classifier based on DCNNs is proposed, and the sensor data from force and torque sensors are preprocessed and reconstructed into a new form that is suitable for the input of DCNNs. The experimental results show that the designed classifier can effectively distinguish time-series sensor data from the manipulator's normal state and various fault states. The proposed method aids in measurement, allowing the manipulator to recover from the fault state to normal working state, and is useful for enhancing the executive capability of manipulators.