Title: Research on sensor fault identification based on improved 1-v-r SVM classification method

Authors: Chun-ying Jiang; Li-cai Li; Chang-long Ye; Su-yang Yu

Addresses: Department of Mechatronics Engineering, Shenyang Aerospace University, Shenyang, 110136, China ' Department of Mechatronics Engineering, Shenyang Aerospace University, Shenyang, 110136, China ' Department of Mechatronics Engineering, Shenyang Aerospace University, Shenyang, 110136, China ' Department of Mechatronics Engineering, Shenyang Aerospace University, Shenyang, 110136, China

Abstract: The feature identification method based on artificial intelligence can significantly improve accuracy and effectiveness of sensor fault diagnosis. An improved support vector machine (SVM)-K-nearest neighbour (KNN) classification method that combines one-verse-rest (1-v-r) SVM and KNN was brought for sensor fault recognition. The method firstly constructs 1-v-r SVM training set by primary selection on training samples, and then classifies it using 1-v-r method. It re-classifies indivisible samples with KNN algorithm. Fault diagnosis experiment on photoelectric encoder sensor verifies that it can determine current fault belongs to which type of common sensor faults. The experiment also compared SVM-KNN with one-verse-one (1-v-1) SVM and bintree SVM. Results show that it has better classification accuracy and classification speed.

Keywords: sensor faults; fault identification; multi-class classification; SVM; support vector machines; feature identification; sensor fault diagnosis; fault recognition; K-nearest neighbour; KNN; photoelectric encoder sensors.

DOI: 10.1504/IJAMC.2016.080965

International Journal of Advanced Media and Communication, 2016 Vol.6 No.2/3/4, pp.235 - 245

Available online: 10 Dec 2016 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article