Several Methods to Achieve Hydraulic Pump Information Fusion Fault Diagnosis

Abstract: According to the dispersion and fuzziness of hydraulic pump fault characteristics, an information fusion fault diagnosis method based on vibration and pressure sensors is proposed. Based on the full analysis of the mechanism of the looseness of the hydraulic head, the vibration signal and the pressure signal are denoised by wavelet to effectively extract the characteristic of the ball looseness. The different types of feature parameters are fused with the feature layer, and the BP neural network with principal component analysis and improved algorithm is used to diagnose the looseness of hydraulic head ball looseness. Experiments show that based on different types of sensor information fusion fault diagnosis method can effectively achieve the diagnosis of weak faults in the hydraulic pump. Introduction Hydraulic pump is the heart of the hydraulic system, its fault diagnosis is an important part of hydraulic system fault diagnosis. Due to the compressibility of the fluid, the fluid-structure interaction between the pump source and the servo system and the intrinsic mechanical vibration of the hydraulic pump itself make the mechanism of the hydraulic pump complicated, the extraction of the fault feature difficult and the fuzziness of the fault diagnosis. A large number of hydraulic pump fault diagnosis data show that the fault signal detected by the pump outlet is often submerged by the interference signal, and the single fault detection signal often shows strong ambiguity. It is difficult to improve the effective fault characteristics by the conventional signal processing method. From the perspective of fault diagnosis, any kind of diagnostic information is fuzzy and inaccurate. For any kind of diagnostic object, using single information to reflect its state behavior is incomplete. If the same Objects of multi-dimensional fault redundancy information to be comprehensive utilization, the system can be more reliable and more accurate monitoring and diagnosis. In this paper, the piston pump ball loose failure mode, the vibration sensor and pressure sensor at the outlet of the hydraulic pump for fault detection, signal processing by wavelet analysis noise reduction, the use of principal component analysis to extract effective fusion information, improved algorithm BP neural Network to achieve a weak signal pump or more effective fault diagnosis. 1, hydraulic pump ball loose failure mechanism analysis Due to manufacturing error or pressure pump in the course of the work of the pressure impact, often make the plunger ball and socket Shen concave deformation of the ball head and ball socket gap, resulting in the plunger Loose ball failure. 1.1 based on the vibration signal failure mechanism analysis Hydraulic pump cylinder in the process of rotation, the plunger reciprocating in the cylinder. When the cylinder turned a certain angle, after the top dead center piston into the suction area, the ball head and the plunger occurred a collision; when the cylinder rotation after the top dead center, the ball began to move toward the plunger, the ball head And the relative movement of the plunger occurs; when the transfer of the discharge area, the high pressure oil on the plunger, the plunger quickly to the ball head direction, resulting in another impact. Cylinder rotation for a week, the ball head and the plunger occurred two collisions, through the transmission shaft and bearing energy transfer to the shell, so the ball loose vibration frequency of the shaft frequency of 2 times. 1.2 Based on the failure mechanism of pressure signal Analysis of the ball loose at the hydraulic pump outlet pressure pulsation also has an impact. When the cylinder turned to the top dead center, the ball head to the direction of the plunger movement, when the oil discharge tank into the unloading area, the ball and the plunger has not yet collided, then the role of high pressure oil, the plunger But also to the ball head direction, the ball head and ball nest collisions, resulting in vibration impact at the same time, the collision through the plunger on the high-pressure oil to produce a pressure pulsation, so the ball caused by loosening pump outlet pressure pulsation frequency and pump The same axial frequency, we can see from the above analysis, if the ball and the ball nest gap is small, the relative velocity of the ball and the plunger is not large, resulting in a small collision energy. When the gap increases, the vibration energy will increase, and with the time-varying periodic changes in the shell vibration energy is usually distributed at 2 times the axial frequency; for the pressure pulse signal, the energy is mainly distributed in the shaft frequency Department. 1.3 ball loose fault diagnosis system For loose ball failure, the vertical direction of the hydraulic pump outlet installed two acceleration sensor ax, a. Vibration detection, a pressure sensor P detects the pump pressure pulsation. As the vibration signal and pressure signal detected at the outlet of the hydraulic pump are frequently inundated by the interference signal, in order to extract the fault feature, the detection signal of the sensor is subjected to wavelet denoising processing. 2, the wavelet signal denoising The working environment of the hydraulic pump is generally harsh, its working conditions are greatly affected by the environment, usually detected at the pump signal contains a lot of noise. The test shows that the pressure signal and the vibration signal detected at the outlet of the hydraulic pump show the following characteristics: ① The spectrum of the signal is very wide, the waveform is disorderly and the regularity is poor; ② The time-varying and non-stationary performance is obvious. Therefore, it is very difficult to extract fault features based on these two signals, so it is necessary to de-noising the detected signals. Wavelet analysis is a more effective method of signal processing. It can analyze signals in both time and frequency domain, and can effectively distinguish the abrupt part and noise in the signal to realize the signal denoising. Pump outlet vibration signal and its wavelet denoised signal, select the wavelet denoising global threshold of 1.049. Obviously, many interference signals are contained in the detection signal, so it is difficult to simply use the detected vibration signal for effective fault diagnosis. In order to eliminate the influence of interference, the wavelet processing can effectively eliminate the noise contained in the vibration signal at the pump outlet, which is good for the extraction of fault features. 3, information fusion fault diagnosis method Information fusion is the intelligent synthesis of multi-source information, resulting in more accurate than a single source of information, fault tolerance and robustness of the stronger estimates and judgments '2'. Because the information detected at the outlet of the hydraulic pump is weak and easily inundated by disturbance, it is difficult to effectively detect the weak fault feature by using the detection signal of a single sensor. The information fusion fault diagnosis process, that is, the vibration signal and the pressure signal are denoised by wavelet, the statistical analysis is used to extract the effective feature information, the principal component analysis (PCA) is used to effectively decouple the correlation between fault features and reduce the fault The dimension of feature, BP neural network with improved algorithm is used to diagnose the looseness of the hydraulic head. 3. Feature Level Information Fusion Feature level state attribute fusion is that multiple types of sensor data are preprocessed to perform feature extraction and data registration. That is, sensor data is transformed into sensor data to transform each sensor input data into a unified data representation. Through the eigenvector normalization processing can realize the information fusion data registration. In this paper, the mean value of vibration signal and pressure signal, the peak factor, the energy value of the characteristic frequency and the power spectrum amplitude, the fourth moment and so on are extracted as the eigenvectors of the ball head looseness fault. 3.2 Select the principal components in the new sample space, the successive calculation of the sensor index of the composite index of the main contribution to the composition. The main contribution index of composite index threshold of 85%, according to the contribution of the composite index select the first few principal components, as the next step of information fusion. For the normal hydraulic pump and 4 kinds of ball loose failure, each 100 samples were selected. Because of the high degree of significance, it shows that these four groups of eigenvectors have very obvious differences. Therefore, the different fault degree of such faults can be diagnosed. The structure of BP neural network is chosen. The four kinds of ball loose faults which are normal and set in the hydraulic pump are trained on the precision of training error. The optimized weight matrix of BP network is obtained by improving the algorithm learning and training. In practical use, the weight matrix of BP neural network and its improved algorithm are used to effectively diagnose multiple faults. Where output node 1 represents the output value of the neural network when the hydraulic pump is normal, node 2 represents the output value of the neural network when the gap is 6 μm, node 3 represents the output value of the 9 μm neural network for the gap, and node 4 represents the output value of the neural network for the gap 12 μm Output value, node 5 represents the output value of the neural network when 15μm. The use of BP neural network and its improved algorithm can effectively diagnose different clearance size ball loose failure. 4. Conclusions In this paper, the vibration signals and pressure signals at the outlet of the hydraulic pump are used to effectively extract the fault features by wavelet denoising. By using PCA analysis, the dimensionality of the information fusion eigenvector is reduced to a great extent. The diagnostic test proves that the PCA re- The combined eigenvector can achieve multi-fault diagnosis. Introducing additional momentum term in BP algorithm to obtain the optimal learning rate, and improving the BP algorithm to effectively diagnose the ball clearance looseness of different clearance size.