Hydraulic pump fault diagnosis based on neural network and evidence fusion

ZHU Guan-lin1, WANG Zhao-qiang1*, WANG Yi-fan2, LI Zhi-feng3, SUN Chong-zhi4
(1.School of Mechanical and Automotive Engineering, Shanghai University of Engineering Science,
Shanghai 201620, China; 2.National Network of Zhejiang Electric Power Corporation Power Science
Research Institute, Hangzhou 310007, China; 3.Gansu Special Equipment Safety Technical
Inspection Center, Lanzhou 730020, China; 4.Gansu Province Special Equipment Inspection
and Testing Institute, Lanzhou 730050, China)

Abstract: Aiming at the problem that single sensor detection was susceptible to environmental interference, and it was difficult to accurately distinguish the type of hydraulic pump failure, the neural network classification recognition and evidence theory fusion technology were applied to hydraulic pump fault diagnosis. The particle swarm neural network (PSOBP) fault classification optimized by the adaptive adjustment method, the fusion paradox and the improvement of the failure problem in the DS evidence theory were studied. The cognitive factors and social factors were used to dynamically guide the particle optimization. A new conflict coefficient between two fault evidences was constructed by using the idea of gravitation. Furthermore, a multi-source sensor data fusion model which conformed to hydraulic pump fault diagnosis was proposed. Six operating states of hydraulic pumps were constructed through experiments and fault diagnosis tests were conducted respectively. The results indicate that the particle swarm neural network optimized by the adaptive adjustment method can improve the fault diagnosis accuracy of hydraulic pumps, reaching 93.50% and 93.67% respectively. The support degree of fusion diagnosis results is close to 1, which reduces the fuzziness of diagnosis.

Key words: plunger pump; hydraulic fault diagnosis; multisource sensors; neural network; evidence fusion