CN-121659251-B - Emulsion pump fault intelligent diagnosis method and system based on multidimensional entropy feature fusion

CN121659251BCN 121659251 BCN121659251 BCN 121659251BCN-121659251-B

Abstract

The invention relates to the technical field of fluid machinery fault diagnosis, and discloses an intelligent emulsion pump fault diagnosis method and system based on multidimensional entropy feature fusion, wherein the method comprises the steps of constructing a multidimensional physical field state monitoring space and synchronously collecting vibration acceleration, outlet pressure and flow time sequence signals; the invention effectively solves the problems that the characteristic distortion of the entropy value and the weak fault characteristic of liquid-machine coupling are masked caused by non-Gaussian impact noise by fusing the fractional number mapping and the multiple physical fields, and improves the robustness and the accuracy of fault diagnosis of an emulsion pump under complex working conditions.

Inventors

ZHANG WENQUAN
CHEN JINGJING
CHEN JIALE
DAI HONGBING
ZHU LOU

Assignees

南京六煤机械有限公司

Dates

Publication Date: 20260508
Application Date: 20260206

Claims (11)

1. The intelligent emulsion pump fault diagnosis method based on multidimensional entropy feature fusion is characterized by comprising the following steps of: Constructing a multidimensional physical field state monitoring space, synchronously acquiring an original vibration acceleration time sequence signal, an original outlet pressure time sequence signal and an original flow time sequence signal from the multidimensional physical field state monitoring space, performing adaptive variation modal decomposition and effective sensitive component screening on the original vibration acceleration time sequence signal, and obtaining a reconstructed vibration signal after reconstruction; Performing phase space reconstruction and quantile mapping on the reconstructed vibration signal to generate a quantile symbol sequence, calculating normalized vibration quantile arrangement entropy characteristics according to the quantile symbol sequence, extracting pressure fluctuation entropy and flow pulsation variance from the filtered pressure signal and the filtered flow signal, and fusing the normalized vibration quantile arrangement entropy characteristics, the pressure fluctuation entropy and the flow pulsation variance to construct a multidimensional fault feature vector; and inputting the multidimensional fault feature vector into a pre-constructed multi-classification support vector machine model, outputting an emulsion pump running state label, and executing a hierarchical closed-loop control strategy according to the emulsion pump running state label.
2. The intelligent diagnosis method for the fault of the emulsion pump based on the multi-dimensional entropy feature fusion according to claim 1, wherein the execution method for the adaptive variation modal decomposition comprises the following steps: performing heuristic variable decomposition on the original vibration acceleration time sequence signal to obtain heuristic eigenmode function components, performing joint judgment and iterative optimization on the heuristic eigenmode function components, and determining the optimal decomposition layer number; And executing final variation modal decomposition on the original vibration acceleration time sequence signal according to the optimal decomposition layer number to obtain an eigenmode function component set.
3. The intelligent diagnosis method for emulsion pump faults based on multidimensional entropy feature fusion according to claim 2, wherein the screening method for effective sensitive components comprises the following steps: And calculating a cross correlation coefficient between each eigenmode function component in the eigenmode function component set and the original vibration acceleration time sequence signal, and marking the eigenmode function components with the cross correlation coefficient larger than or equal to the effective characteristic screening threshold value as effective sensitive components.
4. The intelligent diagnosis method for emulsion pump failure based on multi-dimensional entropy feature fusion according to claim 3, wherein the method for performing heuristic variant modal decomposition comprises: setting a decomposition layer number initial value K, and executing variation modal decomposition operation on the original vibration acceleration time sequence signal based on the decomposition layer number initial value K to obtain K heuristic eigenmode function components.
5. The intelligent diagnosis method for emulsion pump faults based on multi-dimensional entropy feature fusion according to claim 4, wherein the method for carrying out joint judgment and iterative optimization on heuristic eigenmode function components comprises the following steps: checking the time domain waveform of each heuristic eigenmode function component to obtain a waveform observation result, wherein the waveform observation result comprises waveform distortion and waveform distortion; Performing center frequency distribution evaluation on the heuristic eigenmode function components to obtain a center frequency distribution evaluation result, wherein the center frequency distribution evaluation result comprises mode aliasing occurrence and mode aliasing non-occurrence; and performing iterative optimization based on the waveform observation result and the center frequency distribution evaluation result.
6. The intelligent diagnosis method for emulsion pump faults based on multi-dimensional entropy feature fusion according to claim 5, wherein the execution method for iterative optimization comprises the following steps: If modal aliasing or waveform distortion occurs, judging that the current decomposition layer number K is too large to cause excessive decomposition, terminating iteration and outputting K-1 as the optimal decomposition layer number; if modal aliasing does not occur and waveform distortion does not occur, calculating energy of a residual signal, wherein the residual signal is a difference value of a sum of an original vibration acceleration time sequence signal and all heuristic eigenmode function components; If the energy of the residual signal is larger than the preset noise reference energy, judging that the decomposition is incomplete, enabling K=K+1 and returning to execute heuristic variable decomposition mode decomposition; If the energy of the residual signal is less than or equal to the noise reference energy, the decomposition is judged to be sufficient, the iteration is terminated, and the current K value is output as the optimal decomposition layer number.
7. The intelligent diagnosis method for emulsion pump faults based on multi-dimensional entropy feature fusion according to claim 6, wherein the method for performing phase space reconstruction and quantile mapping on the reconstructed vibration signals comprises the following steps: Determining time delay of the reconstructed vibration signal by adopting a mutual information method, determining embedding dimension by adopting a Cao method, generating a reconstructed state vector sequence according to the time delay and the embedding dimension, and mapping the numerical value of each element in the reconstructed state vector sequence to a preset quantile interval to generate a quantile symbol sequence.
8. The intelligent diagnosis method for the fault of the emulsion pump based on the multi-dimensional entropy feature fusion of claim 7, wherein the calculation method for the normalized vibration quantile permutation entropy feature comprises the following steps of; Determining an arrangement mode of each quantile symbol vector in the quantile symbol sequence, counting the occurrence probability of each arrangement mode in the quantile symbol sequence, calculating quantile arrangement entropy according to the occurrence probability, and calculating normalized vibration quantile arrangement entropy characteristics based on the quantile arrangement entropy.
9. The intelligent diagnosis method for emulsion pump failure based on multi-dimensional entropy feature fusion according to claim 8, wherein the method for extracting pressure fluctuation entropy and flow pulsation variance from the filtered pressure signal and the filtered flow signal comprises: respectively carrying out low-pass filtering treatment on the original outlet pressure time sequence signal and the original flow time sequence signal to obtain a filtered pressure signal and a filtered flow signal; Calculating the statistical variance of the filtered flow signal in a preset sliding time window to obtain the flow pulsation variance.
10. The intelligent diagnosis method of emulsion pump faults based on multi-dimensional entropy feature fusion according to claim 9, wherein the training method of the multi-classification support vector machine model comprises the following steps: collecting historical multidimensional fault feature vector samples of the emulsion pump under different running states, marking fault types of the historical multidimensional fault feature vector samples, constructing a training sample set, and training a multi-classification support vector machine model by using the training sample set by adopting a radial basis function and a one-to-many classification strategy.
11. Emulsion pump fault intelligent diagnosis system based on multidimensional entropy feature fusion, for implementing the emulsion pump fault intelligent diagnosis method based on multidimensional entropy feature fusion according to any one of claims 1-10, characterized in that the system comprises: the signal reconstruction module is used for constructing a multidimensional physical field state monitoring space, synchronously collecting an original vibration acceleration time sequence signal, an original outlet pressure time sequence signal and an original flow time sequence signal from the multidimensional physical field state monitoring space, performing adaptive variation modal decomposition and effective sensitive component screening on the original vibration acceleration time sequence signal, and obtaining a reconstructed vibration signal after reconstruction; The feature fusion module is used for carrying out phase space reconstruction and quantile mapping on the reconstructed vibration signals to generate quantile symbol sequences, calculating normalized vibration quantile arrangement entropy features according to the quantile symbol sequences, extracting pressure fluctuation entropy and flow pulsation variance from the filtered pressure signals and the filtered flow signals, and fusing the normalized vibration quantile arrangement entropy features, the pressure fluctuation entropy and the flow pulsation variance to construct multidimensional fault feature vectors; The diagnosis control module is used for inputting the multidimensional fault feature vector into a pre-constructed multi-classification support vector machine model, outputting an emulsion pump running state label and executing a hierarchical closed-loop control strategy according to the emulsion pump running state label.

Description

Emulsion pump fault intelligent diagnosis method and system based on multidimensional entropy feature fusion Technical Field The invention relates to the technical field of fluid machinery fault diagnosis, in particular to an intelligent emulsion pump fault diagnosis method and system based on multidimensional entropy feature fusion. Background The emulsion pump is used as a power source of a hydraulic support system of a fully mechanized mining face, is called a heart of the hydraulic system, and the operation reliability of the emulsion pump is directly related to the production safety and the exploitation efficiency of the underground coal mine. In the extremely severe working condition environment in the pit, the emulsion pump runs in the high-pressure, high-flow and heavy-load state for a long time, and faults such as slider abrasion, cavitation, leakage of a distributing valve and the like are extremely easy to occur. With the advancement of intelligent construction of coal mines, the fault diagnosis technology based on data driving gradually replaces the traditional manual auscultation and regular dismantling and inspection, and becomes a key means for guaranteeing continuous and stable operation of a fully mechanized mining line. The Chinese patent with the grant bulletin number of CN116028849B discloses an emulsion pump fault diagnosis method based on a depth self-coding network, which is characterized in that the real-time parameters of an emulsion pump are monitored and compared with a set threshold value, and the parameters at the alarming moment are input into a depth self-coding network model for training, so that the real-time state is analyzed and diagnosed by utilizing historical data. The Chinese patent with the grant bulletin number of CN113378887B discloses a fault grading diagnosis method of an emulsion pump, which utilizes a stack automatic encoder and a Softmax classifier to construct a deep self-coding network, takes multi-source monitoring parameters as input to carry out self-adaptive feature learning, and combines an expert system to carry out fault positioning, so as to improve the accuracy of diagnosis through a grading strategy. The method utilizes the strong nonlinear fitting capability of the neural network to a certain extent, and improves the recognition efficiency of the fault mode. However, the prior art has remarkable limitations when facing complex non-stable working conditions of the fully mechanized mining face, and the contradiction phenomenon that false high-frequency alarm and hidden fault report coexist is caused frequently in field application. The physical root of this discrepancy is the non-gaussian statistical nature of the environmental noise downhole in the coal mine and the strongly coupled masking effect of the hydrodynamic signals. In actual production, when the coal mining machine performs cutting operation or the hydraulic support performs frame moving action, the environment field is filled with randomly generated high-amplitude non-Gaussian mechanical impact noise (such as transient impact of high-pressure liquid flow). These "large amplitude outliers" of non-faulty classes have extremely high energy densities in the time domain, which can instantaneously destroy the original arrangement topology of the time series. Existing diagnostic algorithms, if lacking robust suppression mechanisms for outliers, often misinterpret these external environmental impacts as dynamic mutations inside the device due to component loosening or wear, thereby inducing frequent false alarms. On the other hand, emulsion pumps are typical "liquid-machine coupled" systems, and the signal energy generated by early fluid side faults (e.g., cavitation bubble collapse, weak intra-valve clearance leakage) is extremely weak and has complex high frequency modulation characteristics. These weak high frequency fluid impact features are highly prone to being overwhelmed by motor rotor fundamental, pump body structure low frequency resonances, and the above-mentioned environmental noise (i.e., masking effects) with a single vibration transmission path or lack of fine signal resolution. The signal characteristics of coexistence of the strong noise and the weak fault lead the decision boundary of the fault characteristics in the characteristic space to become vague, so that the system is difficult to perceive early weak fluid faults, the system is forced to stop until the faults evolve into malignant accidents such as main shaft locking, gear fracture, pump head fracture and the like, and the optimal maintenance window is missed. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides an intelligent diagnosis method and an intelligent diagnosis system for emulsion pump faults based on multi-dimensional entropy feature fusion, which are characterized in that a multi-dimensional physical field monitoring space for vibration, pressure and flow collabo