CN-121980343-A - Residual life prediction method and device for slurry pump, electronic equipment and medium

CN121980343ACN 121980343 ACN121980343 ACN 121980343ACN-121980343-A

Abstract

The invention relates to a method, a device, electronic equipment and a medium for predicting the residual life of a slurry pump, which belong to the technical field of fault diagnosis, wherein the method comprises the steps of preprocessing original data of a target slurry pump to obtain preprocessed original data, wherein the original data comprises mechanical dynamic data, process performance data, equipment state data and historical maintenance data; the method comprises the steps of carrying out fusion on preprocessed original data based on a Kalman filtering algorithm to obtain a global multi-mode fusion feature sequence, inputting the global multi-mode fusion feature sequence into a xLSTM model which is completely trained to obtain a fault classification result of a target slurry pump, constructing a knowledge graph based on the fault classification result, and predicting the residual life of the target slurry pump based on the knowledge graph. The method improves the accuracy of predicting the residual life of the slurry pump.

Inventors

LIU ZHIYANG
ZHOU LUOYU
LUO MINGZHANG
KANG PEI
LI HAORAN
LU ZHONGLI
CHEN XIN

Assignees

长江大学

Dates

Publication Date: 20260505
Application Date: 20260112

Claims (10)

1. A method for predicting remaining life of a slurry pump, comprising: Preprocessing original data of a target slurry pump to obtain preprocessed original data, wherein the original data comprises mechanical dynamic data, process performance data, equipment state data and historical maintenance data; Fusing the preprocessed original data based on a Kalman filtering algorithm to obtain a global multi-mode fusion characteristic sequence; Inputting the overall multi-mode fusion characteristic sequence into a xLSTM model which is completely trained, and obtaining a fault classification result of the target slurry pump; And constructing a knowledge graph based on the fault classification result, and predicting the residual life of the target slurry pump based on the knowledge graph.
2. The method of predicting remaining life of a slurry pump according to claim 1, wherein the mechanical dynamic data includes vibration acceleration, acoustic emissions, and pressure pulsation signals; The process performance data comprise slurry density, slurry viscosity, slurry pump flushing and slurry pump discharge pressure; The equipment state data comprises bearing temperature, oil temperature and motor current voltage; the historical maintenance data comprises maintenance records, design parameters and fault cases.
3. The method for predicting the residual life of a slurry pump according to claim 1, wherein the preprocessing the raw data of the target slurry pump to obtain preprocessed raw data includes: And carrying out standardization processing on the original data of the target slurry pump to obtain the preprocessed original data.
4. The method for predicting the residual life of a slurry pump according to claim 1, wherein the kalman filtering algorithm is used for fusing the preprocessed raw data to obtain a global multi-mode fusion feature sequence, and the method comprises the following steps: inputting the preprocessed original data into a state space model based on Kalman filtering to perform denoising and smoothing processing to obtain temporary data; And carrying out weighted fusion of covariance matrixes on the temporary data based on a federal Kalman filtering algorithm to obtain a global multi-mode fusion characteristic sequence.
5. The method for predicting the residual life of a slurry pump according to claim 1, wherein the step of inputting the global multi-modal fusion feature sequence into a fully trained xLSTM model to obtain a fault classification result of the target slurry pump comprises the steps of: training a first branch in a complete xLSTM model to process transient impact characteristics in a global multi-mode fusion characteristic sequence to obtain first characteristics; Training a second branch in the complete xLSTM model to process the working condition characteristics in the overall multi-mode fusion characteristic sequence to obtain second characteristics; training a third branch in a complete xLSTM model to process the gradual change trend feature in the overall multi-mode fusion feature sequence to obtain a third feature; and interacting and fusing the first feature, the second feature and the third feature through a attention mechanism in a training complete xLSTM model to obtain a fault classification result of the target slurry pump.
6. The method for predicting the remaining life of a slurry pump according to claim 1, wherein predicting the remaining life of a target slurry pump based on a knowledge graph comprises: Predicting the health index and the probability distribution of the residual life of the target slurry pump based on the knowledge graph; And determining the residual life of the target slurry pump according to the health index and the probability distribution of the residual life based on the health index.
7. The method for predicting the remaining life of a slurry pump according to claim 1, wherein the probability distribution formula of the remaining life of the target slurry pump is: Where RUL denotes the remaining life, p (RUL=t|Y 1:k ) denotes the probability density of the remaining life equal to t given all observations Y1: k from time 1 to time k, Y 1:k denotes the set of all observations from time 1 to time k, i.e. the historical health index data, t is the specific value of RUL, p (x k ∣Y 1:k ) denotes the posterior probability distribution of the system hidden state x k given all historic observations Y 1:k at time k, x k is the hidden state of the system at time k, p (RUL=t|x k ) is the conditional probability of the remaining life equal to t given the current hidden state x k Dx k is the integration of all possible implicit states x k .
8. A residual life prediction device for a slurry pump, comprising: The original data acquisition module is used for preprocessing the original data of the target slurry pump to obtain preprocessed original data, wherein the original data comprises mechanical dynamic data, process performance data, equipment state data and historical maintenance data; The fusion data acquisition module is used for fusing the preprocessed original data based on a Kalman filtering algorithm to obtain a global multi-mode fusion characteristic sequence; the fault classification module is used for inputting the overall multi-mode fusion characteristic sequence into a xLSTM model which is completely trained to obtain a fault classification result of the target slurry pump; And the residual life prediction module is used for constructing a knowledge graph based on the fault classification result and predicting the residual life of the target slurry pump based on the knowledge graph.
9. An electronic device comprising a memory and a processor, wherein, The memory is used for storing programs; The processor, coupled to the memory, is configured to execute the program stored in the memory to implement the steps in a method for predicting the remaining life of a mud pump as set forth in any one of claims 1 to 7.
10. A computer readable storage medium storing a computer readable program or instructions which when executed by a processor is capable of carrying out the steps of a method of predicting the remaining life of a mud pump as claimed in any one of claims 1 to 7.

Description

Residual life prediction method and device for slurry pump, electronic equipment and medium Technical Field The present invention relates to the field of fault diagnosis technologies, and in particular, to a method, an apparatus, an electronic device, and a medium for predicting a residual life of a slurry pump. Background In the field of petroleum drilling, a slurry pump is used as core equipment, and the running state of the slurry pump is directly related to the safety and efficiency of drilling operation. At present, the fault diagnosis technology in the field is mainly developed around a method based on single sensor signal analysis, such as adopting vibration signals to extract characteristics by combining EEMD decomposition, short-time Fourier transform or medium-high frequency band amplitude analysis and other means, and then performing fault recognition through CNN, BP neural network and other models. Although the method realizes fault detection to a certain extent, the method is limited by a single data source, and is difficult to comprehensively capture the equipment state under the complex working condition. To further enhance the diagnostic effect, some studies have attempted to introduce multisensor information fusion. For example, by extracting various statistical indexes of vibration signals as characteristics or adopting a multivariate fusion algorithm combining empirical mode decomposition and pseudo-phase diagram technology, more comprehensive fault information is expected to be obtained. In addition, methods exist for processing vibration timing data using timing processing techniques such as resonance demodulation in combination with fuzzy recognition, or conventional LSTM networks. These techniques enrich the feature expression to some extent, but still do not break through the key bottleneck. The main defects of the prior art are that firstly, the fusion level of multi-source heterogeneous data is shallow, the difference of space-time characteristics and physical association of different sensor data is not fully considered, so that the information utilization is insufficient, secondly, the dynamic time sequence modeling capability is limited, the traditional LSTM is difficult to effectively capture multi-scale fault characteristics under non-steady working conditions, early weak fault early warning capability is insufficient, and furthermore, the method is mostly limited to data driving, and the effective utilization of priori information such as equipment structure knowledge, fault mechanism and the like is lacking, so that the interpretation of diagnosis results is poor, and the reasoning capability is weak. In addition, the real-time performance and the self-adaptive capability of the existing method are also to be enhanced. Most model parameters are fixed with diagnostic threshold values, so that the dynamic change of the slurry pump in different working conditions and abrasion stages is difficult to adapt, and the performance of the slurry pump is reduced in field application. Meanwhile, the system is generally focused on post-hoc diagnosis, lacks the capability of real-time evaluation of health state, fault trend prediction and residual life estimation, and cannot effectively support predictive maintenance. The diagnosis link is disjointed from the operation and maintenance management, so that closed loop optimization of 'monitoring-diagnosis-decision-maintenance' is not formed, and the engineering application value of the technology is limited. In summary, the prior art lacks a method to improve the accuracy of fault diagnosis of the slurry pump, so as to further improve the accuracy of residual life prediction of the slurry pump. Disclosure of Invention In view of the foregoing, it is desirable to provide a method, apparatus, electronic device, and medium for predicting the remaining life of a slurry pump, so as to improve the accuracy of the remaining life prediction of the slurry pump. In order to achieve the above object, in a first aspect, the present invention provides a method for predicting remaining life of a slurry pump, comprising: Preprocessing original data of a target slurry pump to obtain preprocessed original data, wherein the original data comprises mechanical dynamic data, process performance data, equipment state data and historical maintenance data; Fusing the preprocessed original data based on a Kalman filtering algorithm to obtain a global multi-mode fusion characteristic sequence; Inputting the overall multi-mode fusion characteristic sequence into a xLSTM model which is completely trained, and obtaining a fault classification result of the target slurry pump; And constructing a knowledge graph based on the fault classification result, and predicting the residual life of the target slurry pump based on the knowledge graph. In one possible implementation, the mechanical dynamic data includes vibration acceleration, acoustic emissions, and pressure pulsati