Self-recovery method based on auto-associative neural network for intelligent sensors

• Published in: 2010 8th World Congress on Intelligent Control and Automation pp. 6918 - 6922
• Main Authors: Huang Guo-jian, Liu Gui-xiong, Chen Geng-xin, Chen Tie-qun
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2010
• Subjects: 9-ary tree; Artificial neural networks; Auto-Associative Neural Network; Intelligent sensors; Noise; Self-recovery; Testing; Training; Transducers
• DOI: 10.1109/WCICA.2010.5554231

In order to improve the self-recovery capabilities of the IEEE 1451 based intelligent sensors and enhance the level of sensors' intelligence, this paper presents a sensors fault detection and repair method based on Auto-Associative Neural Network (AANN). The error sum of squares (SSE) is introduced as a sensor fault evaluation factor on the basis of the inherent non-linearity & non-orthogonal of the AANN, besides a parallel 9-ary tree algorithm is proposed to locate multi-faulty transducers. The 9-ary tree algorithm can be further extended to estimate the correct value of the faulty transducers while the SSE is less than threshold. A 10-13-5-13-10 structured AANN is constructed to test the self-recovery capability of an insulator contamination status online monitoring networked intelligent sensor model. Results show that, the AANN can be trained within 2 seconds. By altering the corrected step of the 9-ary tree algorithm successively; this method can locate at least two faulty transducers synchronously, besides it can take appropriate strategy for recovering the drift failures and estimating their real value within 5 seconds.