Research on Fault Diagnosis of Hydraulic Pump Bearing Based on Neural Network

Abstract: The fault diagnosis method of hydraulic pump bearing based on integrated BP network is studied. Using the frequency domain and the inverse frequency domain for feature extraction, an integrated BP network is adopted to diagnose and identify the fault, which solves the problems of the fault feature of the hydraulic pump bearing and the difficulty of multi-fault identification. The experimental results show that the integrated BP network can effectively diagnose and identify multi-failure modes of hydraulic pump bearings and has strong robustness.

Key words: hydraulic pump; bearing fault; fault diagnosis; integrated BP network

In the aviation industry, the working performance of the hydraulic system directly affects the safety of the aircraft and the lives of passengers. The hydraulic pump is the power source of the hydraulic system. Therefore, the condition monitoring and fault diagnosis of the hydraulic pump are especially important. Bearing failure is one of the common failure modes of hydraulic pumps. Since the additional vibration caused by the bearing failure is weaker than the inherent vibration of the hydraulic pump, it is very difficult to separate the fault information from the signal. So far, there is still a lack of a very effective way to diagnose the fault of hydraulic pump bearing. In this paper, feature extraction in the frequency domain and the inverse frequency domain is proposed to solve the problem of bearing feature extraction difficult and to solve the problem of multi-fault diagnosis and identification and robustness by using integrated BP network.

1 hydraulic pump bearing fault feature extraction

For mechanical systems, any malfunctioning must cause additional vibration to the system. The vibration signal is a dynamic signal, it contains a wealth of information, it is suitable for fault diagnosis. However, if the additional vibration signals are submerged due to the interference of the inherent signals or outside interference to the fault signals, how to extract the useful signals from the vibration signals is very crucial.

According to tribology theory, when the bearing inner surface of the flow surface, the outer ring raceway and roller there is a damage, the smooth surface of the raceway is damaged, whenever the roller rolls over the injury point, there will be a vibration. Assuming bearing parts as rigid body, regardless of the impact of contact deformation, the roller along the raceway is pure roll, then the following damage vibration frequency:
When the inner raceway has a damage, its vibration pulse characteristic frequency is:

fI = frZ (1 + dcosα / D) / 2 (1)

When there is a damage on the outer raceway, the vibration pulse frequency is:

fo = frZ (1- dcosα / D) / 2 (2)

When there is a damage on the roller, its vibration pulse characteristic frequency is:

fR = frD (1-d2cosα / D2) / d (3)

Where: fr-inner ring speed frequency; D-bearing pitch circle diameter; d-roller diameter; α-contact angle; Z-roller number.
In order to overcome the difficulty that bearing fault signal is weak and easily submerged by natural vibration of hydraulic pump, the following characteristics with strong anti-jamming capability are selected as fault diagnosis characteristic parameters.
(1) Average energy characteristics of vibration The vibration acceleration signal measured on the pump body of the hydraulic pump is:
a (t) = {a1 (t), a2 (t), ..., an (t)}
It is the fault signal to pump the signal after transmission. According to statistical theory, the root mean square of vibration reflects the time domain information of vibration:

The characteristic parameter is that it represents the effective value of the vibration signal and reflects the average energy of the vibration.
(2) peak characteristics of the vibration signal

Pp = max {a (t)} (5)

It is a characteristic signal that reflects the periodic pulsation in the vibration signal.
(3) Cepstral envelope feature Let f (t) be the fault excitation signal and h (t) be the impulse response of the transmission channel. Their corresponding Fourier transforms have the following relations:

(6) for the following transformation:

In the formula, τ is called the frequency inversion; (τ) is cepstrum. It can be seen from the above equation that the characteristics of the fault excitation signal and the transmission channel are separated from each other. In general, the fault excitation signal and the transmission channel signal occupy different back-off sections, which can highlight the characteristics of the faulty vibration signal.
The Hilbert transform is used to find the envelope of the time-domain signal in signal analysis, in order to smooth the power spectrum and highlight the fault information. Definition signal: the best envelope. Cepstral envelope model is the essence of the signal obtained from the sensor cepstrum analysis, and then its cepstrum signal envelope extraction, which dual highlighted the fault information for the signal to noise ratio of small fault feature extraction provided in accordance with.

2 integrated BP network fault diagnosis principle

The organizational structure of neural networks is determined by the domain characteristics of the problem solving. Due to the complexity of fault diagnosis system, the application of neural network to the design of fault diagnosis system will be a problem of organization and learning of large-scale neural network. In order to reduce the complexity of the work and reduce the learning time of the network, this dissertation decomposes the fault diagnosis knowledge set into several logically independent sub-sets, each sub-set is subdivided into a number of rule subsets, and then the network is organized according to the rule subsets. Each rule subset is a logically independent mapping of subnetworks, the relations between the subsets of rules, represented by the matrix of rights of the subnetworks. Each sub-network independently uses BP learning algorithm to study and train respectively. Since the decomposed sub-network is much smaller than the original network and the problem is localized, the training time is greatly reduced. The information processing capability of fault diagnosis of hydraulic pump using integrated BP network is derived from the nonlinear mechanism characteristics and BP algorithm of neurons, as shown in Fig.1.

Figure 1 BP network troubleshooting diagram

Each sub-network in Figure 2 is a BP network, each sub-network learning by the BP algorithm, the result of learning by the control network integration. BP network learning algorithm is as follows:

Figure 2 integrated BP network diagram

The value of each selected feature parameter (including energy feature, amplitude feature and cepstral envelope feature) x is mapped to a single node in the input / output layer of the neural network and is processed normally:

xi = 0.8 (x-xmin) / (xmax-xmin) +0.1 (8)

The purpose of formula (8) to regularize the eigenparameter to (0.1,0.9) is to avoid the problem that learning can not converge if the output value of Sigmoid function is extreme.
For the regular value obtained in equation (8), the following operation is performed to obtain the weighted value and the threshold of each neuron:

Where, j represents the current level, i represents the previous level, wij represents the connection weight; cj represents the current node threshold; fj represents the output.

3, neural network robustness research

The robustness of neural networks refers to the fault tolerance of neural networks to faults. It is well known that the human brain is fault tolerant and that the damage of individual neurons in the brain does not seriously degrade its overall performance because each of the concepts in the brain is not stored in just one neuron but spreads over many nerves Yuan and its connection. By learning again, the brain can re-express knowledge forgotten due to the damage of a part of neurons in the remaining neurons. As the neural network is a simulation of the biological neural network, the most important feature of the neural network is the function of "associative memory". That is to say, the neural network can be composed of the past knowledge, and under the condition of partial information loss or partial information uncertainty, The remaining feature information to make the correct diagnosis. Table 2 shows the success rate of correctly diagnosing and identifying some of the six characteristic information of the bearing with incorrect or indeterminate input characteristics.

Table 2 neural network robustness statistics

Input characteristic Uncertain element Diagnostic success rate One characteristic parameter Uncertainty 100%
Two characteristic parameters are not sure 94%
Three characteristic parameters are not sure 76%
Four characteristic parameters are not sure 70%
Five characteristic parameters are not sure 20%
The six characteristic parameters are not sure 8%

From Table 2, it can be seen that fault diagnosis using integrated neural network can still make correct judgments with a high success rate (76% -100%) when a large amount of information is lost (nearly half of the characteristic parameters are uncertain) Neural networks are very powerful
5 Conclusion

Because of the self-learning neural network, self-organization, associative memory and other functions determine the neural network method is very suitable for fault diagnosis research. In this paper, the vibration signals in the frequency domain and the scrambling frequency domain are taken as the characteristic parameters, and the multi-fault diagnosis and identification of the hydraulic pump bearing is realized by the integrated BP network. Experimental results show that this method has high success rate and robustness.