Cascade-Correlation Neural Network for Sensor Fault Detection and Data Recovery with On-line Learning

摘要

Cascade-Correlation (CC) is a new architecture and supervised learning algorithm for artificial neural networks. The fundamental theory of the Cascade-Correlation neural network is firstly introduced, then a novel method based on the Cascade-Correlation neural network with on-line learning is proposed, which is used in sensor fault detection and data recovery, and the specific procedure of the method is described in detail. Finally, this method is applied to a six-component force/torque sensor, and compared with Back Propagation (BP) neural network predictor, the experimental results show that the proposed method has higher prediction and recovery accuracy and consumes less time than a BP neural network. Therefore, the proposed method is suitable and very effective for sensor fault detection and short-term data recovery.