In the aviation industry, the performance of hydraulic system directly affects the safety and passengers' life, and the hydraulic pump is the power source of the hydraulic system, so the condition monitoring and fault diagnosis of hydraulic pump is especially important.Bearing failure is one of the common failure mode of hydraulic pump, the additional vibration caused by bearing fault relative to the hydraulic pump of the inherent vibration is weak, so it is difficult to separate the fault information from the signal.So far, the fault diagnosis of hydraulic pump bearing failure is lack of effective methods.In this paper, in the frequency domain and the frequency domain feature extraction, aims to solve the problem of bearing feature extraction difficult and use the integrated BP network to solve multiple fault diagnosis and recognition and robustness problems.

1 hydraulic pump bearing fault feature extraction For the mechanical system, such as defective is bound to cause additional vibration system.Vibration signals is dynamic, it contains rich information, very suitable for fault diagnosis.But if the additional vibration signals due to the inherent or interference of flooding, fault signal interference is very big and then how to extract useful signal from vibration signal is very critical.

Flow according to the tribology theory, when the bearing surface on the inner ring and outer ring raceway and roller damage, raceway surface smooth damage, when the roller rolled damage points, will produce a vibration.Assumptions for rigid bearing parts, not considering the influence of the contact deformation, roller along the groove for pure roll.

The Hilbert transform is used in signal analysis and time domain signal envelope, in order to achieve to smooth power spectrum to highlight the fault information.Define the signal: as the best envelope.Cepstrum envelope model is obtained from the sensor signal spectrum analysis, and then the cepstrum envelope signals are extracted, and duality to highlight the fault information, the signal-to-noise ratio for small provides basis to extract fault features.

2 the principle of integrated BP network for fault diagnosis The organizational structure of the neural network is determined by solving the field problem characteristics.Because of the complexity of the fault diagnosis system of neural network was applied to diagnosis system design, will be a large-scale neural network organization and the study question.In order to reduce the complexity of the work, reduce the number of network learning time, this article will set is decomposed into several logical fault diagnosis knowledge independent subsets, each set rules is decomposed into several subsets, then according to the rules set to organize network.Each rule subset is a logically independent sub network mapping, the relationship between rules subset, through the network of power system matrix said.Each network independently using BP learning algorithm for learning training respectively.Due to the subnet of the decomposed much smaller than the original network and localization problem, which has greatly reduced the training time.Using integrated BP network for fault diagnosis of hydraulic pump bearing from information processing capability, mechanism of nonlinear characteristics of neurons and BP algorithm.

3, the neural network research of robustness Robustness of the neural network is refers to the neural network for fault tolerance.It is well known that the human brain with fault tolerance, the features of individual neurons in the brain damage will not make it to the overall performance of the serious degradation, this is because the brain every concept is not only stored in one neuron, but scattered in many neurons and their connections.The brain can learn by again, make through part of the neuron damage and forgotten knowledge expression in the rest of the nerve yuan again.Because neural network is a simulation of biological neural networks, so the biggest characteristic of neural network is the associative memory function, the neural network can be combined by previous knowledge, in the part of the loss of information or part of the information uncertainty conditions, with the rest of the information to make correct diagnosis.Table 2 shows the bearing some input features in six characteristic information is not correct or uncertain cases, correct diagnosis and identification of success.

Neural network robust statistics in table 1 The input characteristics of the uncertain element in the diagnosis of success rate A characteristic parameter uncertainty was 100% The two characteristic parameters uncertainty of 94% Three characteristic parameters uncertainty of 76% Four characteristic parameters uncertainty of 70% Five characteristic parameters uncertainty of 20% Six characteristic parameters uncertainty of 8% Can be seen from table 1, the use of integrated neural network for fault diagnosis can be in the case of a lost a lot of information (nearly half of the characteristic parameters uncertainty) still can make the right judgment is of very high success rate (76% ~ 100%) and integrated neural network has strong ability

5 conclusion Because neural network has self-learning, self-organization, associative memory and other functions determine the neural network method is very suitable for research of fault diagnosis.In this paper the vibration signals of the frequency domain and the frequency domain as characteristic parameter, using the integrated BP network to realize the multiple fault diagnosis of hydraulic pump bearing and recognition.The test results show that the method has a high success rate and robustness.