CN-122021836-A - Diagnosis method and system for health state of water lubrication bearing

CN122021836ACN 122021836 ACN122021836 ACN 122021836ACN-122021836-A

Abstract

The invention belongs to the technical field of intelligent operation and maintenance of water-lubricated bearings, in particular to a method and a system for diagnosing the health state of a water-lubricated bearing, which are used for acquiring film thickness, vibration, temperature and noise parameters in the running process of the bearing, extracting current values and variation trends and generating a structural state parameter tuple; the method comprises the steps of obtaining parameters, generating natural language state description by using a large language model, searching candidate fault modes and associated paths thereof in a knowledge graph by taking abnormal symptoms in a tuple as query conditions to form search results, determining current symptom combination characteristics according to types, quantity and trends of the abnormal symptoms, selecting matched reasoning mode templates, filling the tuple and the search results into the templates to generate reasoning instructions, guiding the large language model to obtain a core diagnosis conclusion, searching the knowledge graph by taking the natural language description as the query conditions to extract a knowledge subgraph of the best matching fault modes, and inputting the core conclusion, the state description and the knowledge subgraph into the large language model to generate a comprehensive diagnosis report.

Inventors

LIANG PENG
WANG CONG
PAN WEI
GUO FENG
GAO HAOJIE
LI SHUYI
ZHANG XIAOHAN
Hou Jitao

Assignees

青岛理工大学

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. A method for diagnosing the health condition of a water lubricated bearing, comprising the steps of: obtaining film thickness, vibration, temperature and noise parameters in the running process of the water lubrication bearing, generating a structural state parameter tuple, and generating natural language description aiming at the current health state of the water lubrication bearing by using a large language model; searching in a pre-constructed knowledge graph by utilizing a structural state parameter tuple to obtain an associated candidate fault mode and an associated path thereof to form a structural search result, determining the current symptom combination characteristic according to the structural state parameter tuple, and selecting an reasoning mode template matched with the characteristic from a template library; Searching in a knowledge graph to obtain a best matching fault mode by taking the obtained natural language description as a query condition, and extracting a knowledge subgraph related to the fault mode; And inputting the core diagnosis conclusion, the natural language description and the knowledge subgraph into a large language model to generate a comprehensive diagnosis report.
2. The method for diagnosing the health state of a water lubricated bearing according to claim 1, wherein the construction of the knowledge graph comprises the steps of: acquiring design and manufacturing data, operation monitoring data, maintenance data and field literature data of the water lubricated bearing as multi-source knowledge data; Aiming at unstructured text in maintenance data, a large language model is adopted for information extraction, the large language model is guided through a prompt word template to identify and output key knowledge elements of fault modes, symptom manifestations, root causes and maintenance actions, a 'head entity-relation-tail entity' form knowledge graph triplet is formed, and an initial confidence coefficient attribute is given to each triplet; And importing the design and manufacturing data, the operation monitoring data, the field literature data and the extracted triples into a map database, and determining quantitative attributes of the entity and the relationship through entity alignment, relationship disambiguation and conflict resolution, wherein the quantitative attributes at least comprise a typical threshold value, a development speed, a confidence coefficient and a delay time.
3. The method for diagnosing a health state of a water lubricated bearing according to claim 2, wherein the knowledge graph comprises a physical unit, a relational unit, and an attribute unit; the entity unit comprises a component entity, a fault mode entity, a symptom parameter entity and a maintenance action entity; The relationship unit comprises causal relationship, accompanying relationship, characterization relationship and disposal relationship between the connection entities; the attribute unit assigns each entity and each relation a quantified attribute including a typical threshold, development speed, confidence.
4. The method for diagnosing the health of a water lubricated bearing according to claim 1, wherein the construction of the template library comprises the steps of: Obtaining a plurality of diagnosis reasoning modes by analyzing a knowledge correlation structure in the knowledge graph; aiming at each diagnosis reasoning mode, a structured prompt template comprising a system role instruction, a step-by-step reasoning chain and an output format requirement is designed; And testing and optimizing the prompt templates by using the historical case data stored in the knowledge graph to form a special prompt template library of the water lubrication bearing.
5. The method for diagnosing the health state of the water lubricated bearing according to claim 1, wherein the best matching failure mode is retrieved from a knowledge graph by taking the obtained natural language description as a query condition, specifically: extracting symptom entities from the natural language description; in the knowledge graph, starting from each extracted symptom entity, searching out all fault mode entities possibly connected to the extracted symptom entity; Searching a common fault mode entity capable of simultaneously associating all symptom entities, determining the matching degree according to the path length between the symptom entities and the fault mode entity, and taking the common fault mode entity with the shortest path as the best matching fault mode.
6. The method for diagnosing the health state of the water lubricated bearing according to claim 1, wherein the film thickness parameter is obtained by acquiring the position change of the outer surface of the paddle shaft relative to the inner surface of the bearing through at least three eddy current displacement sensors distributed in the water lubricated bearing along the same circumferential section, calculating the eccentricity and the offset angle of the axis of the paddle shaft in the water lubricated bearing, and further determining the minimum value of the lubrication film thickness on the circumferential section as the film thickness parameter.
7. The method for diagnosing the health state of the water lubricated bearing according to claim 1, wherein the noise parameters are obtained by respectively acquiring comprehensive noise close to the bearing position and background noise far from the bearing position through the main microphone and the reference microphone, calculating a coherence function between two paths of noise signals, and obtaining a pure bearing noise signal by identifying frequency components of the comprehensive noise highly coherent with the background noise and filtering the comprehensive noise.
8. A method of diagnosing a health condition of a water lubricated bearing according to claim 1, wherein the abnormal symptoms in the tuple of structured state parameters are identified by comparing the current value of each parameter with a preset health threshold in the knowledge graph, and mapping the parameter to a corresponding symptom parameter entity in the knowledge graph as an abnormal symptom when the current value of the parameter exceeds the health threshold.
9. A method of diagnosing a health condition of a water lubricated bearing according to claim 1, wherein the integrated diagnostic report contains at least fault type, fault level, fault cause, maintenance recommendation and remaining service life.
10. A system for diagnosing the health of a water lubricated bearing, comprising: The data acquisition and processing module is configured to acquire film thickness, vibration, temperature and noise parameters in the running process of the water lubrication bearing, extract the current numerical value and the change trend of each parameter, generate a structural state parameter tuple and generate natural language description aiming at the current health state of the water lubrication bearing by using a large language model; The knowledge retrieval and instruction reasoning module is configured to utilize the structural state parameter tuple to retrieve in a pre-constructed knowledge graph to acquire an associated candidate fault mode and an associated path thereof to form a structural retrieval result, determine the current symptom combination characteristic according to the structural state parameter tuple and select an reasoning mode template matched with the characteristic from a template library, fill the structural state parameter tuple and the structural retrieval result into the selected reasoning mode template to generate a reasoning instruction containing domain logic, and obtain a structural core diagnosis conclusion under the guidance of the reasoning instruction by the large language model; The knowledge visualization module is configured to search a knowledge graph to obtain a best matching fault mode by taking the obtained natural language description as a query condition, and extract a knowledge subgraph related to the fault mode; And the comprehensive report generation module is configured to input the core diagnosis conclusion, the natural language description and the knowledge subgraph into the large language model to generate a comprehensive diagnosis report.

Description

Diagnosis method and system for health state of water lubrication bearing Technical Field The invention belongs to the technical field of intelligent operation and maintenance of water lubricated bearings, and particularly relates to a method and a system for diagnosing the health state of a water lubricated bearing. Background The statements in this section merely mention background of the present disclosure and do not necessarily constitute prior art. The water lubrication bearing takes water as a lubricating medium, is generally applied to wading equipment such as ship propulsion, hydroelectric power generation, sea water pumps and the like, and the failure mode relates to multiple mechanisms such as abrasive particle abrasion, rubber aging, lubricating film cracking and the like. The diagnosis method based on the physical model is used for analyzing the characteristic frequency of the fault or the change rule of the physical parameter by establishing a dynamic model or a lubrication theoretical model of the bearing, so as to judge the type and the severity of the fault. The diagnosis capability of the method is highly dependent on the accuracy of the model, and factors such as nonlinearity, time variability, multi-field coupling and the like in the actual working condition make accurate modeling extremely difficult, so that the reliability of a diagnosis result is limited. The diagnosis method based on data driving realizes automatic identification of bearing fault types by carrying out feature extraction and mode training on historical monitoring data and constructing a classifier or regression model. Such data driven methods are essentially "black box" models, whose internal decision process lacks transparency. When the system outputs a 'fault code', an operation and maintenance person cannot know whether the conclusion is that the fault is caused by the abnormality of the parameters, and cannot know the association relation between the abnormalities, and the development stage of the current fault and the adaptive maintenance measures are difficult to grasp, so that on-site personnel need to conduct secondary research and judgment according to personal experience, and therefore, the guiding value of simply outputting the fault code on the operation and maintenance decision is extremely limited. Disclosure of Invention The invention provides a diagnosis method and a system for the health state of a water lubrication bearing, and provides a method capable of outputting comprehensive diagnosis reports containing phenomena, reasons, mechanisms and measures to help operation and maintenance personnel to change from 'passive maintenance' to 'active monitoring maintenance' aiming at the problem of the interpretability of bearing fault diagnosis 'only fault, not to say reason'. The first aspect of the invention discloses a method for diagnosing the health state of a water lubricated bearing, which comprises the following steps: obtaining film thickness, vibration, temperature and noise parameters in the running process of the water lubrication bearing, extracting the current numerical value and variation trend of each parameter, generating a structural state parameter tuple, and generating natural language description aiming at the current health state of the water lubrication bearing by using a large language model; searching in a pre-constructed knowledge graph by utilizing a structural state parameter tuple to obtain an associated candidate fault mode and an associated path thereof to form a structural search result, determining the current symptom combination characteristic according to the structural state parameter tuple, and selecting an reasoning mode template matched with the characteristic from a template library; Searching in a knowledge graph to obtain a best matching fault mode by taking the obtained natural language description as a query condition, and extracting a knowledge subgraph related to the fault mode; And inputting the core diagnosis conclusion, the natural language description and the knowledge subgraph into a large language model to generate a comprehensive diagnosis report. Further, the construction of the knowledge graph comprises the following steps: acquiring design and manufacturing data, operation monitoring data, maintenance data and field literature data of the water lubricated bearing as multi-source knowledge data; Aiming at unstructured text in maintenance data, a large language model is adopted for information extraction, the large language model is guided through a prompt word template to identify and output key knowledge elements of fault modes, symptom manifestations, root causes and maintenance actions, a 'head entity-relation-tail entity' form knowledge graph triplet is formed, and an initial confidence coefficient attribute is given to each triplet; And importing the design and manufacturing data, the operation monitoring data, the field literature data and the extr