Piston Pump Fault Diagnosis Methods Based On EMD And Neural Networks

Concrete pump truck is a kind of construction machinery, which has complex load and bad working environment.Faults often happen on its internal components for fatigue or oil contamination. Once the concrete pump truck appears accidents, people and construction will face a great harm. According to statistical analysis, piston pump is the most important fault source of concrete pump truck. As the pumping power source, piston pump is not only a rotating machinery,but also a reciprocating machinery and machine hydraulic converter component.When it working,there is not only vibration between mechanical parts,but also shock caused by working medium. So it is very difficult to diagnose.This study analyzes the structural characteristics and movement rules of piston pumps, points out the locations and vibration frequencies of common faults, and obtains power spectrum and envelope spectrum by traditional spectrum analysis method, but fails to find out the fault features due to the serious medium impact. So this paper uses the core theory of Hilbert-Huang transform, the EMD decomposition, to decompose vibarion signals of five common faults and normal state from test bed, obtains signals of various bands in accordence with their own frequency somponents, establishes the time series AR model, and sets the model parametres as the characteristic parametres for fault diagnosis and input for the subsequent nueral network. Neural network has strong capabilities of nonlinear mapping and paralle processing. It uses the weight vector in network to simulate the memory process of brain neurons, so that it can store the characteristic information from training samples more stably, and identify the new input by available information. This study uses the newer Fuzzy ARTMAP to learn and classify the characteristic parametres of piston pump of 6 states, and the results show that the method is effective for diagnosis. This paper also discusses the influence of the number of parametres to the classification efficiency of neural network, and achieves the number reduction of parametres without affecting the accuracy of diagnosis.