JP7442001B1 - Comprehensive failure diagnosis method for hydroelectric power generation units - Google Patents

Abstract

The present invention enables diagnosis of different types of hydroelectric power generation units and improves the accuracy rate of diagnosis. The present invention is a comprehensive failure diagnosis method for a hydroelectric power generation unit, which acquires power plant layout data, unit layout data, equipment layout data, FMEA failure mode layout data, and diagnostic workflow layout data. Step 1 of performing preparatory work before scan diagnosis, and analyzing power plant location data and unit location data using various diagnostic methods, including big data model algorithms, rule inference, fault case matching methods, and fault tree analysis methods. Step 2 of starting a fault diagnosis to diagnose equipment layout data; Step 3 of acquiring the results of each diagnosis method; Step 4 of performing a fault diagnosis fusion decision on the diagnosis results of multiple diagnosis methods; and the final step. step 5 of obtaining a diagnostic result. [Selection diagram] Figure 1

Description

The present invention relates to the technical field of equipment failure diagnosis for a hydraulic power generation unit, and more particularly to a comprehensive failure diagnosis method for a hydraulic power generation unit.

The working environment of hydroelectric power generation units is complex, with frequent changes in operating conditions, and the working conditions are affected by hydraulic, mechanical, and electrical elements.Moreover, various elements are interconnected and difficult to disassemble. The difficulty of diagnosing power generation unit failures has been significantly improved. Currently, failure analysis and diagnosis for hydropower units is often based on well-defined failure mechanisms and expert experience. There are relatively few types of failures with clear mechanisms, and most of them are based on mechanism diagnosis using a single element, which lacks the quantity and accuracy of diagnosis and cannot cover the various complex operating conditions of work situations. Home experience diagnosis is highly dependent on the level and experience of experts, and it is difficult to perform reliable and effective diagnostic analysis for different types of hydropower units or unexperienced faults. Therefore, performing accurate and effective fault diagnosis for hydropower units has long been a problem and difficulty in the industry.

Patent Document 1 discloses an equipment failure diagnosis method based on FMECA and the main body, and the adopted diagnosis method combines the analytical thinking of FMECA, extracts knowledge in the field of equipment failure diagnosis, and provides the basis for constructing the failure diagnosis main body. shall be. For equipment to be diagnosed, first analyze the equipment structure and functional flow. In the case of general industrial equipment, the structure is divided into the overall equipment level, system level, component level, and parts level. We summarize the equipment corresponding to each level structure of the equipment to be diagnosed, analyze the equipment workflow based on the equipment structure, and evaluate the possible impact on equipment at the same level or higher level if equipment at each level breaks down. I have summarized. The above diagnostic process only adopts a single diagnostic tool, the diagnostic accuracy is not high, and misdiagnosis is easy to occur.

Chinese patent document CN115685972A

In view of the technical problems existing in the background art, the comprehensive fault diagnosis method for hydropower units provided by the present invention enables diagnosing different types of hydropower units, and integrates various diagnostic methods. Finally, a diagnostic result with a high probability is obtained and the accuracy rate of diagnosis is improved.

In order to solve the above technical problem, the present invention is realized using the following technical means,
It is a comprehensive failure diagnosis method for hydroelectric power generation units,
Step 1 of performing preparatory work before failure scan diagnosis to obtain power plant location data, unit location data, equipment location data, FMEA failure mode location data, and diagnostic workflow location data;
Initiating fault diagnosis to diagnose power plant layout data, unit layout data, and equipment layout data using various diagnostic methods including big data model algorithms, rule inference, fault case matching methods, and fault tree analysis methods. 2 and
Step 3 of obtaining the results of each diagnostic method;
Step 4: performing a fault diagnosis fusion decision on the diagnosis results of the plurality of diagnosis methods;
Step 5 of obtaining a final diagnosis result.

Preferably, the substeps of step 1 are:
Step 1.1 of acquiring power plant location data including a power plant code, a power plant location data related unit information table, and a power plant location data ID;
Step 1.2 of acquiring unit placement data including a unit code, unit placement data related equipment list, and unit placement data ID;
Step 1.3 of acquiring equipment layout data including equipment name, equipment code, and installed system equipment information table associated with the equipment layout data;
Step 1.4 of obtaining all failure mode information, equipment logic information .

Preferably, in step 2, each diagnosis method constitutes one diagnosis node, and the diagnosis node includes a big data model algorithm node, a rule node, a fault case library node and a fault tree analysis method node, and when diagnosing, the big data Diagnosis is performed by entering the model algorithm node, rule node, failure case library node, and fault tree analysis method node, and sequentially diagnosing the power plant layout data, unit layout data, and equipment layout data.

Preferably, the result table of step 3 includes a big data model algorithm result table, a rule result table, a fault case library result table, and a fault tree analysis method result table.

Preferably, the substeps of step 4 are:
step 4.1 of filtering the diagnostic results of each diagnostic method to obtain the results with the maximum probability of different failure modes for each diagnostic method;
Each diagnostic method includes step 4.2 of performing a fusion calculation process on the filtered results to obtain the final failure mode and its probability.

Preferably, the fusion calculation in step 4.2 is divided into two types: Type 1, where there is only one failure mode in the tool results, and Type 2, where there are multiple failure modes in the tool results. ,
For type 1, where there is only one failure mode in the tool results, it is performed as shown in Table 1.

Preferably , for the second type in which the tool results include multiple failure modes, it is necessary to calculate using a fused decision model based on DS evidence theory.

Preferably, the process of fault tree analysis method includes:
Divide the failure into three levels: a failure top event, an intermediate event, and a basic event, a step 1 in which the failure top event includes a plurality of intermediate events, and the intermediate event includes a plurality of basic events;
Step 2: connecting failure events at each level through logic gates to form a tree structure for failure analysis;
Each level of failure event includes step 3 corresponding to a corresponding failure symptom.

Preferably, the process of rule inference is
S1, which establishes a rule based on empirical data on the relationship between known failure modes and failure signs, and determines the probability of occurrence of the failure mode based on the input failure sign data;
S2: constructing an analytical model for the time-series data of the equipment and analyzing whether or not related failure signs have occurred;
S3 encapsulating the equipment symptom data parsed in S2 into a JSON structure input to the rule interface and requesting the rule REST interface established in S1;
Processing the processing REST interface request by the rule execution engine, inputting data to the rule, and determining and outputting the probability of failure mode occurrence.

The process of big data model algorithm is to first input tagged fault samples, then perform data preprocessing, and then train with a classification model. Including outputting results by the model.

The beneficial effects of the present invention are as follows:
The comprehensive failure diagnosis method for hydroelectric power generation units provided by the present invention makes it possible to diagnose different types of hydropower generation units, and by combining various diagnostic methods, it is possible to finally obtain a diagnosis result with a high probability. and improve the diagnostic accuracy rate.

The present invention will be further explained below using the accompanying drawings and examples.
FIG. 1 is a diagnostic flowchart of the present invention. FIG. 2 is a diagnostic flowchart according to the first embodiment of the present invention.

A preferred method is a comprehensive failure diagnosis method for hydroelectric power generation units, as shown in Figure 1.
Step 1: Perform preparatory work before failure scan diagnosis, obtain power plant layout data, unit layout data, equipment layout data, FMEA failure mode layout data, and diagnostic workflow layout data;
Failure mode and effects analysis (FMEA) is a standard method for all unit-level and equipment-level failure-related information features such as failure modes, failure symptoms, failure types, failure causes, characteristic parameters, severity, occurrence, and detectability. A diagnostic analysis that describes and establishes failure modes and effect analysis of a unit system according to the structure and functional coupling relationships between equipment systems or parts,
Step 1.1: Obtain power plant location data including a power plant code, power plant location data related unit information table, and power plant location data ID;
The power plant arrangement data related unit information table is all unit arrangement data included in a certain power plant.

Step 1.2: Obtain the unit placement data including the unit code, unit placement data related equipment list, and unit placement data ID;
The unit arrangement data related equipment list is all equipment arrangement data included in a certain unit.

Step 1.3: Obtain equipment layout data that includes equipment name, equipment code, and system equipment information table introduced in relation to the equipment layout data;
The system equipment information table introduced in association with equipment location data includes all sub-equipments included in a certain equipment.

Step 1.4. Obtain all failure mode information and equipment logic information .

Step 2: Perform fault diagnosis and diagnose power plant layout data, unit layout data, and equipment layout data using various diagnostic methods, and the diagnosis methods include big data model algorithm, rule inference, and failure case matching. The diagnosis method adopted by the present invention is a conventional diagnosis method, including the method and fault tree analysis method, and the function introduction is as follows.
Diagnostic method 1, the fault tree analysis method (FTA), is a quantitative analysis method that can help engineers determine the reliability and safety indicators of equipment and determine appropriate preventive measures. Fault tree analysis methods are typically used in combination with other analysis methods such as FMEA (Failure Modes and Effects Analysis) to increase the accuracy and reliability of the analysis. The process of fault tree analysis divides faults into three levels: fault top event, intermediate event, and basic event. The fault top event includes multiple intermediate events, and the intermediate event includes multiple basic events. Fault events are connected by logic gates to form a tree structure for fault analysis, and each level of fault event corresponds to a corresponding fault symptom. The specific analysis process is
S1 creating an FTA logic model tree based on information provided by accidents that can occur or that have occurred in the system, or based on the FMEA;
S2 generating a model tree in which the created logic model tree is instantiated as an FTA;
The FTA instantiated model tree uses its own fault diagnosis algorithm to combine the algorithm model, diagnosis rules, scripts and measurement point data to complete the analysis and deduction of the entire fault tree, thereby calculating the probability of failure occurrence. S3 to obtain.

Specifically, step S1 includes the following substeps,
Based on the accidents that can occur or have occurred in the system, select the system failure with the highest impact as the top event. and S11, which binds the logic equipment and the fault diagnosis tool (algorithm model, diagnosis rule, script, and measurement point) for the top event of the fault tree;
Next, S12 step-by-step decomposes the cause of system failure into intermediate events and binds logic equipment and failure diagnosis tools (algorithm model, diagnosis rules, scripts and measurement points) for the top event of the fault tree;
Events that are impossible or unnecessary to decompose are called basic events, that is, subordinate events. and S13 of setting a failure diagnosis tool and default probability of occurrence for the lower event;
S14, in which the nodes of each fault tree are connected using a logical relationship calculation method (for example, logical product, logical sum), thereby completing the drawing of a complete logic fault tree;
After the drawing of the fault tree is completed, if the integrity check is passed, the creation of the logic fault is completed in step S15.

Specifically, step S2 includes the following substeps,
S21 converting the logic model into an instantiated model by selecting instantiated stations and equipment;
S22 converting logic equipment, measurement points, etc. bound to the node into physical equipment, physical measurement points, etc. bound to the node during the instantiation process;
S22 converting the logic algorithm model into a physical algorithm model;
S23, if a subtree is referenced by a node of the fault tree, the subtree also needs to be converted into an instantiated tree;
Then, the model name of the logic fault tree is converted to the instantiated model name, and here the creation of the fault tree instance is completed in S24.

Specifically, step S3 includes the following substeps,
S31, where the instantiated fault tree can be used formally and can be called through the result interface or automatically executed through a periodic task;
S32: obtaining the diagnostic algorithm bound to the fault tree node, calling the corresponding diagnostic interface based on different types of diagnostic algorithms, obtaining the diagnostic result, and setting the node alarm result and probability;
If the fault tree is bound to the faulty node, S33 repeats S32;
The failure probability obtained in step S32 is substituted into the fault tree, and if no fault diagnosis algorithm is bound to the node, the default setting failure probability is obtained. Based on the FTA fault diagnosis algorithm, such as minimum cut set, minimum path set, etc., analyze and deduce whether the top event of the fault tree has failed and the probability that the failure will occur S34.

Diagnostic method 2, the process of rule inference is:
S1: Establish rules in the rule placement tool based on empirical data on the relationship between known failure modes and failure symptoms, and determine the probability of failure mode occurrence based on the input failure symptom data. After the rules are released, provide REST interface services.

A rule placement proposal consists of two parts: data design and logic design.

Data design requires defining input data, output data, and intermediate calculation result data for fault diagnosis rules, including names and types of data variables. If desired, multiple related data can be encapsulated into complex variable types. Only defined data variables can be used in rule logic. Define business parameters in rule logic as constants to improve readability and maintainability.

In logic design, you can create a series of rule blocks containing data processing processes and decision logic by dragging various rule elements, including conditional judgments, data assignments, data processing functions, basic four arithmetic calculations, etc., according to the processing flow for input data. Build.

When the rule is released, specify the input data of the rule and the data that needs to be output after the rule execution is completed from the data variables in the data design above, and the released REST interface receives the input data in JSON format. , return the result to JSON format.

S2: Build an analytical model for the time-series data of the equipment, and analyze whether related failure signs have occurred.

S3, according to the input requirements specified at the time of rule release, encapsulates the equipment symptom data analyzed in S2 into the JSON structure input to the rule interface, and requests the rule REST interface established in S1.

S4: The REST interface request is processed by the rule execution engine serving as the rule placement back end, data is input to the rule, and the probability of failure mode occurrence is determined and output based on the designed logic.

A rule execution engine compiles each rule block into a rule tree, executes these rule trees sequentially according to the order of each rule block within the rule, and updates associated data variables. Finally, the specified output data is output, including failure modes and their occurrence probabilities.

For a particular failure or problem, a visualized knowledge structure diagram is formed by organizing, summarizing, and classifying related information. This structure diagram typically includes the causes, performance, elimination methods, solutions, etc. of the failure to assist in understanding and resolving a particular failure or problem.

Diagnosis method 3 is the fault case matching method, which stores and organizes the information of fault cases through the fault case library, and when a new fault occurs, the system automatically compares it with the data in the fault case library and selects the most similar Find failure cases, extract useful information from them, and diagnose them. Search for time-series data of the corresponding measurement point based on information such as the time period, equipment, and unit that require diagnosis. The diagnostic process is as follows.
S1. Sign calculation is performed based on time series data of measurement points, and each sign corresponds to a different calculation logic. Determine what kind of symptoms occurred during the period and the reliability of those symptoms.

S2: After grouping according to the corresponding failure mode based on the calculated symptoms, the reliability of the calculated symptoms is normalized according to a fixed weight to obtain the reliability of each failure mode.

S3: Calculate the degree of similarity between the symptoms that occurred above and the symptoms corresponding to each failure case in the failure case library using a method of measuring a plurality of similarities.

S4: Finally, the reliability and similarity of the failure modes are combined to obtain the reliability of each failure mode, that is, the probability of failure occurrence.

The process of diagnosis method 4, big data model algorithm is as follows.
S1: Search for time-series data of corresponding measurement points based on the KKS instantiated equipment tree according to the time period and unit equipment that require diagnosis.

S2: Perform table conversion processing on time series data (convert pivot table, pivot, id, time, and v format data into id1, id2, id3, . . . formats).

S3: After performing the table conversion process, the times may not match (some measurement points have data at the same time, and some measurement points do not), so the active power measurement points and unit water head Perform previous value correction for the measurement point.

S4, perform stability screening on the data based on the active power measurement points, and the specific logic is as follows:
S4.1. Set window t and wave threshold M, and check whether the unit within the window at time t is operating stably, that is, whether the wave amplitude of active power (maximum value - minimum value within the window) is greater than M. Determine whether it is small,
S4.2. Slide the window back and repeat decision logic 1.

S5: The reason why the previous value correction is performed on all empty data and the previous value correction is performed in two steps is that it is necessary to guarantee the authenticity of the data.

S6: Select measurement point data that exceeds the sensor range.

S7: A threshold value filter is applied to the target measurement point, and when the target measurement point exceeds the threshold value set by the user, marking is performed.

S8: The marked data is stored in a data table, and the table is used for data exchange between the model development department and the fault diagnosis module.

Further, diagnosis method 1-4 is a quantitative diagnosis, and if the diagnosis cannot be made by quantitative diagnosis, diagnosis is performed using the knowledge graph diagnosis method, and the diagnosis process is as follows.
S1. Based on the knowledge about hydropower units and faults, design the knowledge and relationship models used for diagnosis, and the knowledge nodes are system, subsystem, unit, failure case library, FMEA (failure modes and effects analysis) library, FTA. (Fault tree analysis method) Contains libraries, early alarm information, measurement points, etc., and organizes the related relationships between knowledge to construct a conceptual model of hydropower fault diagnosis knowledge graph, specifically,
First, knowledge graphs are used to describe various entities or concepts existing in the real world and their relationships, ultimately arranging a huge semantic network diagram. Here, nodes represent entities or concepts, and edges are arranged with attributes or relationships.
The second is the node (or body). Represents concepts, entities, attributes, or events. The main body can be, for example, one main transformer, one worker, etc., or it can be an abstract concept, such as a big data platform, knowledge graph, power generation unit, etc.
The third is an edge (or arc). A section corresponds to a "relationship" and represents some kind of connection that exists between nodes. Edge labels represent the type of relationship or relationship between entities (class relationship, configuration relationship, etc.).

In the fault diagnosis knowledge graph for hydropower units, nodes include systems, parts, equipment, failure case libraries, FMEA (Failure Modes and Effects Analysis) libraries, FTA (Fault Tree Analysis) libraries, early alarm information, measurement points, etc. , and relationships such as class relationships, order relationships, causal relationships, and configuration relationships between nodes together constitute a conceptual model of a fault diagnosis knowledge graph (FaultDiagnosisKG Conception Model).

S2, classify knowledge. This part is very important, as it prepares for later knowledge services, knowledge calculation, knowledge inference and knowledge visualization, etc., and nodes are divided into systems, equipment, failures, methods, data, users, etc. Relationships are classified into class relationships, composition relationships, aggregation relationships, instance relationships, order relationships, association relationships, dependence relationships, causal relationships, related parties, etc.

S3, subdivide the data structure of knowledge nodes, that is, organize related attributes useful for failure analysis targeting knowledge nodes, such as units (equipment KKS code, equipment specifications, operating hours, equipment type, manufacturer, etc.), failure cases (fault case code, fault case name, short description of the fault case, failure case source type, failure case related images, etc.), failure (fault encoding, failure mode, failure phenomenon description, fault level, fault effect, etc.), failure processing ( processing measures, processing steps, processing effects), failure characteristics (feature data, graphical features, failure data samples, data samples after failure recovery), etc.

S4, construct a tool using the knowledge graph, construct a conceptual knowledge model in the graph database Neo4j, and create a concept-level knowledge template using triples (subject, predicate, object), such as (unit, association, fault), ( Hydro turbine, class relationship, unit), (failure cause, causal relationship, failure), (failure characteristic, related relationship, failure sign), (failure, related person, defect recorder), (alert, related relationship, unit), etc. Convert to

S5: Preparing for knowledge extraction. Configure mapping settings for knowledge nodes of structured type, arrange tasks such as knowledge source placement and knowledge annotations for semi-structured and unstructured data, and create knowledge extraction tasks and task execution parameters (task execution frequency, execution coverage range, knowledge coverage method, etc.).

S6: Extract and update knowledge. Execute knowledge extraction tasks on a timed or periodic basis and store instantiated data in a graph database based on a pre-created hydropower fault diagnosis knowledge graph conceptual model, and when encountering entity definition mismatches, etc. Then, artificial means are assisted to perform knowledge integration, and finally form a knowledge graph for hydropower fault diagnosis.

S7, perform knowledge calculation and inference. Based on the fault diagnosis inference rules and related models, knowledge calculations or knowledge inferences are performed on the formed hydropower fault diagnosis knowledge graph to obtain fault diagnosis results and failure analysis process records.

S8: Visualize knowledge. Display the fault diagnosis results and analysis process to the user in an appropriate format.

Each diagnosis method constitutes one diagnosis node, and the diagnosis node includes a big data model algorithm node, a rule node, a fault case library node, and a fault tree analysis method node. Diagnosis is performed by entering the failure case library node and the fault tree analysis method node, and sequentially diagnosing the power plant location data, unit location data, and equipment location data.

The specific diagnostic process is shown in Table 2.

Step 3. Obtain the results of each diagnosis method, the result table includes a big data model algorithm result table, a rule result table, a fault case library result table, and a fault tree analysis method result table.

Step 4: Make a fault diagnosis fusion decision based on the diagnosis results of multiple diagnosis methods,
Step 4.1: Perform filter processing on the diagnostic results of each diagnostic method, obtain the results with the maximum probability of different failure modes for each diagnostic method,
Step 4.2: Perform fusion calculation processing on the filtered results of each diagnostic method to obtain final failure modes and their probabilities.

The fusion calculation in step 4.2 is divided into two types : the first type, in which the tool result has only one failure mode , and the second type, in which the tool result has multiple failure modes.
For type 1, where there is only one failure mode in the tool results, it is performed as shown in Table 3.

Step 4.2.1 , For the second type where the tool results have multiple failure modes, it is necessary to calculate using a fusion decision model based on DS evidence theory. A document related to DS evidence theory is a method for selecting a moving target defense optimal strategy based on DS evidence theory disclosed in Chinese patent document CN110166437B.

Step 5: Obtain the final diagnosis result.

The final diagnosis results include rule diagnosis results, failure case matching results, fault tree analysis method matching results, model matching results, and knowledge inference results, and finally the above final diagnosis results are aggregated and integrated. .

Specifically, detailed result information of various diagnosis results is displayed on a failure diagnosis detail screen. It includes rule diagnosis results, failure case matching results, fault tree analysis method matching results, model matching results, and knowledge inference results, and finally aggregates and integrates various diagnosis results to evaluate the diagnosis probability and historical reliability of various models. Synthesize and give a fault diagnosis decision-making result through fusion decision.

First, displaying rule results:
Detailed pricing explanation, based on the failure mode, the maximum probability calculated for each failure mode is taken as the result, and if there are multiple failure modes, multiple modes are taken [1],
Explaining the diagnosis basis, which failure mode was hit, and listing the symptom name and occurrence situation [2]
When based on signs, assume that mass imbalance hits a certain combination of signs in Article 110, list the reliability of each sign and the reliability of the combination,
When based on feature parameters, it is assumed that the failure mode is a series of feature parameter calculations connected, and the calculation results and calculation method of the feature parameters of the group will be presented.

Second, display of failure case results:
Detailed pricing explanation, based on the failure mode, the maximum probability calculated for each failure mode is taken as the result, and if there are multiple failure modes, multiple modes are taken [1],
It can be divided into two types: diagnostic basis explanation, mechanism signs, and characteristic parameters [2].
When based on symptoms, the symptom range involved in this calculation is a list of what has occurred and what has not occurred, and the occurrence of the failure mode is estimated based on the combination of symptoms.
If it is based on feature parameters, list the feature parameters involved in this calculation.

Third, displaying the fault tree analysis method results:
Detailed value-taking explanation, based on the fault tree analysis method, it is possible to take the failure mode hit by the tree of each fault tree analysis method and view all paths hit in that failure mode according to the fault effect [1]. ,
Diagnosis basis explanation, failure mode nodes of bound algorithms/rules, and occurrence/absence are listed in the tree of the fault tree analysis method [2].

Fourth, displaying knowledge graph results:
Forming a complete association graph of this inference based on detailed value explanations, symptoms or failure modes [1];
It can be divided into two types: diagnostic basis explanation, mechanism signs, and characteristic parameters [2], and specifically,
When based on symptoms, if a symptom is involved in this calculation, obtain the relevant symptom from the symptom result table, infer the corresponding failure mode, failure cause, and treatment measures based on the symptom,
When based on feature parameters, if no symptoms are found in the current calculation, it is necessary to find the corresponding failure mode by activating the rule ID of feature parameter calculation, and infer the related failure cause treatment measures etc. through the failure mode. There is.

Fifth, display of algorithm calculation results:
Detailed pricing explanation, based on the failure mode, the maximum probability calculated for each failure mode is taken as the result, and if there are multiple failure modes, multiple modes are taken [1],
Diagnosis basis explanation, based on each calculation sub-step, click to display appropriate basis [2].

Sixth, display of fusion calculation results:
Based on the calculation results of failure cases, fault tree analysis methods, big data model algorithms, rules, and knowledge graphs, we merge them based on failure modes, and take the data of the three failure modes with the highest probability for each tool and synthesize them. (Display strategy can be adjusted) and associates information fusion output such as failure signs, failure types, failure causes, characteristic parameters, severity, occurrence degree, detection degree, etc., to help operators make scientific judgments about equipment status. Provide evidence.

(Embodiment 1)
As shown in FIG. 2, the entire process of fault diagnosis will be explained in detail using transformer fault detection as an example.

Step 1: Obtain placement data 1. The implementation method of FMEA placement is as follows.
1.1. Build an FMEA standard template, add FMEA data to the standard, and select equipment that needs to be placed;
1.2. Arrange information such as failure modes, priorities, etc. corresponding to (multi-level) equipment according to different levels of systems, subsystems, equipment, parts, etc., and realize the association relationship between equipment and failure modes. And,
1.3. Automatically obtaining feature parameters and failure symptoms for all placed failure modes, or manually placing feature parameters corresponding to failure modes;
1.4. Arrange risk assessment information corresponding to all failure modes, including information such as severity (S), occurrence (O), detection degree (D), and finally form FMEA summary information to do.

Step 2. Perform diagnosis using multiple diagnostic methods (or use multiple diagnostic tools). 2. The method for realizing transformer failure case matching is as follows.
2.1. Manually create a transformer failure case and select the logic equipment, instantiated equipment, failure mode and failure time period.

2.2. According to the failure mode, search all the characteristic parameters related to the transformer failure mode through FMEA, and display the numerical value of this characteristic parameter through the symptom attribute column, data characteristic, and instantiated equipment and failure time period. You can take them all out.

2.3. Depending on the failure mode, the contents of the failure mode, such as the cause of failure, treatment measures, detection method, etc., can be searched and automatically brought out through FMEA.

2.4. Failure data samples can be generated and correlated across instantiated equipment and failure windows. Information contents such as alarms can also be associated.

2.5. Save and release cases.

2.6. In the case of case matching, matching is performed by calculating similarity based on the numerical values of feature parameters, and a matching algorithm called gray association analysis is used as the matching algorithm.

2.7. The case matching process first extracts all the failure modes in the logic equipment, calculates the feature parameters related to the failure modes based on the time period, and then matches them with the numerical values of the feature parameters in the case using a matching algorithm. Then, find the similarity order of all cases.

2.8. The case matching result displays the data feature parameters and case feature parameters, and the similarity order of all cases.

3. The implementation method for determining the transformer rule (rule inference) is as follows.
3.1. Rule construction [1]. Rule data design a). Create an input data object and select six measurement points related to the three-ratio method.

b) Create a data object for intermediate data set output, including variables to be output at the end and storage variables for intermediate data during data processing.

c) Set some business parameters to constants as necessary. The purpose of setting constants is to make data readable and to enhance data maintainability.

[2], Rule logic design a) Complete the logic design of the three-ratio method, and organize the execution process of the three-ratio method in a rule-based manner according to the order.
b), The general rule design process is as follows.
Processing the data, completing the processing of the data, such as extracting attribute values from the object structure, and processing the variables using functions to obtain the target variables.

Execute business logic according to conditional logic judgment, if...then...else structure.

The process of the three-proportion method rule is roughly as follows: taking the measurement point value, determining whether the operating conditions of the three-proportion method are met, calculating the three-proportion method, and calculating the ratio item. To provide data such as the cause of failure based on a combination of three ratio encoding.

[3], Rule quick test, test by providing input data and can also be run on physical measurement points to check whether the input results are as expected.

3.2. Scheduling a rule [1], Instantiating a rule, creating an instance and specifying the equipment to which the rule is applied.

[2], Rule Tasks,For rules that need to be executed periodically, a timing task can be created, and the task can contain multiple rule instances.

4. The method for realizing transformer knowledge graph judgment is as follows.
Build a knowledge system framework, classify knowledge based on needs, clarify the type and property information for each classification, predefine the relationships between knowledge, clarify each type of relationship, and manage and maintain it. conduct.

4.1. Knowledge Display Knowledge display determines the output target for which the graph is constructed, that is, the semantic description framework, Schema, entity naming and ID system of the knowledge graph,
Knowledge representation standard languages include RDFS and OWL, where RDFS provides simple descriptions for classes and attributes and provides a language for lexical modeling on RDF data. Richer definitions require a descriptive language from Sugi to OWL itself.

RDF (triad) is a basic data model, and its basic logic structure has three parts: subject (individual, class instance), predicate (attribute), and object (individual or data type instance). included.

Knowledge Acquisition,Currently, the failure mode information is mainly sourced from,the FMEA library, and for structured data knowledge extraction,,the logic table is usually mapped to RDF data through triplet mapping,,and triplet mapping allows each row of the logic table to be mapped to RDF data. is a rule that can be mapped to some RDF triplet. The triple rule includes two sections: subject mapping and multiple predicate-object mappings. Second, predicate-object mapping includes predicate mapping and object mapping.

4.2, Graph construction [1], Schema definition Schema clearly defines the entities, properties, and relationships in the knowledge graph, and clarifies the executable scope of the knowledge graph. Schema defines the knowledge graph. Equivalent to the body that constructs the graph.

Based on the application needs of fault diagnosis, taking user needs as the starting point and data statistics as evidence, different themes are constructed to meet users' needs.

[2] Construction of the main body graph Under the fault diagnosis application scene, the main spectrum is constructed for failure modes, symptoms, and equipment based on the Schema definition, and the spectrum is composed of entities, properties, and entities. Contains property relationships. Construct a graph of the defined body.

4.3. Knowledge fusion The goal of knowledge fusion is to establish communication between bodies so that the knowledge graphs between bodies can communicate with each other and realize their mutual operations.

Using graph algorithms and rules, it calculates and determines the relationship between bodies and automatically generates a new relationship graph.

5. Construction of transformer fault detection algorithm model After first inputting tagged fault samples, data preprocessing such as format conversion, time alignment, and empty data removal is performed. It is then trained by a classification model (eg, Naive Bayes, Support Vector Machine, etc.).

When making predictions, we perform the same preprocessing work on the data and output the results through a classification model.
The training data comes from samples accumulated during the unit's operation period, and includes a large amount of fault and normal data, which are compiled to build a sample library.

5.2. Perform label processing on the fault data, that is, add a fault type indicator, such as "high temperature discharge", "low temperature discharge", "partial discharge", etc., after each data. When running the model, replace it with a fault code (such as "30" or "46") that the program can recognize. Once the calculation output is complete, return to the failure explanation.

6. The implementation method of transformer FTA is as follows:
6.1. Create a transformer logic FTA, place the logic equipment and failure mode in the FTA node, place three ratio method rules in the high temperature overheating (higher than 700oC) node, etc., and perform failure diagnosis on the nodes that can be placed. Associate methods.

6.2. Once the logic FTA placement is complete, perform an integrity check, and once the check is complete, save and release.

6.3. Perform logic FTA instantiation to specific substations and units, such as instantiation to China Gezhou Dam Substation 2B transformer.

6.4. Perform logic FTA instantiation to generate FTA instances, and FTA instances can be re-edited, saved and released based on different factories and units.

6.5, FTA diagnosis, advanced application - Fault diagnosis sends request parameters KKS code, fault time period. The FTA matches the corresponding FTA instance based on the KKS, and launches the diagnostic method associated with all nodes on the FTA instance, returns to the diagnostic result after the diagnostic method is executed, and the diagnostic result is transferred to the FTA node. and displayed in red and green. Then, the probability of occurrence of the top event is estimated based on the failures fed back from each node.

Step 3 places failure modes, associated diagnostic processes and diagnostic tools in the workflow.

1. Creation Process In the creation process, the model name, model key, and model description are input. The model key here is not a pure number, but needs to be used when calling in a later interface, and is used as a key. If you enter OK and the save is successful, a blank process will be created.

2. Process design [1] Start process design It is necessary to drag the process node events of cases, rules, FTAs, knowledge maps, and algorithm models to the design panel and connect different elements with arrows (sequence flows).

For HTTP tasks, first for the execution sound monitor, the success or failure of the rest service is usually determined by what is written in the backend code, rather than relying solely on the HTTP Status Code. So the sound monitor running here is to put some variables in place so that the workflow knows if this HTTP service call was indeed successful. If it does not matter whether the call was successful or not, no placement is needed here.

[2] Enter the + sign for the event to add a new event. For the event, select end. For the class, set the fixed value com. dhcc. flowable. common. listener. HttpTaskExecuteListener,
Enter + number for Name and add one field.
The name is the fixed value success
The expression is such that responseRes is fixed and the rest are arranged based on what the fields of the actual interface are equal to. For example, the return value of the actual interface is {statusCode: 0, data: [XXXXX]}, and if statusCode is represented by 0, here ${responseRes. statusCode==0}.
Enter + number for Name, add one field,
The name is fixed value errMsg,
In the expression, responseRes is fixed, and the rest are arranged based on the error information field of the actual interface. For example, if the return value of the actual interface is {statusCode: -7777, errorMessage: 'Error'}, here it is ${responseRes. errorMessage},
Once you have placed the execution sound monitor, you need to place other properties for the HTTP service task.
[3] Select GET or POST as the request method URL: Request ${requestHost} as a fixed value, fill in the following address based on the actual situation,
In the request header, authorization: ${authorization} is a fixed value, and if the interface needs to receive the body, place Content-Type: application/json on the second line, and if there is no need to receive the body, place Place Content-Type: application/x-www-form-urlencoded,
[4] Please note that the request body has a fixed value ${requestBody}, and the interface parameter must receive the variable requestBody and the corresponding value. The value is received as a parameter when calling the HTTP service.
The request timeout period is a fixed value of 100000 (or changed according to the actual situation),
The response variable name is a fixed value responseRes,
Save the response as JSON, fixed value true.

Step 4: Make a fusion decision and display the results. In the advanced application of fault diagnosis, select the start and end time of the information scan and optionally select all units of all stations or select a specified single unit of equipment for fault scan diagnosis. I do. The process management tool simultaneously launches various diagnostic tools in the background and displays their diagnostic results back to the front-end interface of the advanced fault diagnosis application. Detailed result information of various diagnostic results can be displayed on the fault diagnosis interface.

The above embodiments are only preferred embodiments of the present invention and should not be considered as limitations on the present invention, and the protection scope of the present invention covers the embodiments described in the claims and equivalents including technical features of the embodiments described in the claims. The scope of protection should be such substitution aspects. That is, equivalent substitution improvements within this range are within the protection scope of the present invention.

Claims (1)

Step 1 of performing preparatory work before failure scan diagnosis to obtain power plant location data, unit location data, equipment location data, FMEA failure mode location data, and diagnostic workflow location data;
Initiating fault diagnosis to diagnose power plant layout data, unit layout data, and equipment layout data using various diagnostic methods including big data model algorithms, rule inference, fault case matching methods, and fault tree analysis methods. 2 and
Step 3 of obtaining the results of each diagnostic method;
Step 4: performing a fault diagnosis fusion decision on the diagnosis results of the plurality of diagnosis methods;
step 5 of obtaining a final diagnosis result;
The substeps of step 1 are:
Step 1.1 of acquiring power plant location data including a power plant code, a power plant location data related unit information table, and a power plant location data ID;
Step 1.2 of acquiring unit placement data including a unit code, unit placement data related equipment list, and unit placement data ID;
Step 1.3 of acquiring equipment layout data including equipment name, equipment code, and installed system equipment information table associated with the equipment layout data;
step 1.4 of obtaining all failure mode information, equipment logic information;
In step 2, each diagnosis method constitutes one diagnosis node, and the diagnosis node includes a big data model algorithm node, a rule node, a fault case library node and a fault tree analysis method node, and at the time of diagnosis, the big data model algorithm node , enter the rule node, failure case library node, and fault tree analysis method node to perform diagnosis, sequentially diagnose power plant layout data, unit layout data, and equipment layout data,
The result table of step 3 includes a big data model algorithm result table, a rule result table, a fault case library result table, and a fault tree analysis method result table;
The substeps of step 4 are:
step 4.1 of filtering the diagnostic results of each diagnostic method to obtain the results with the maximum probability of different failure modes for each diagnostic method;
step 4.2 of performing a fusion calculation process on the filtered results of each diagnostic method to obtain a final failure mode and its probability;
The fusion calculation in step 4.2 is divided into two types: the first type, in which the tool result has only one failure mode, and the second type, in which the tool result has multiple failure modes.
For Type 1, which has only one failure mode in the tool results, it is performed as shown in Table 1,

For type 2, where the tool results have multiple failure modes, it is necessary to calculate using a fused decision model based on the D-S evidence theory.
The process of fault tree analysis method is
Divide the failure into three levels: a failure top event, an intermediate event, and a basic event, a step 1 in which the failure top event includes a plurality of intermediate events, and the intermediate event includes a plurality of basic events;
Step 2: connecting failure events at each level through logic gates to form a tree structure for failure analysis;
Each level of failure event includes step 3 corresponding to a corresponding failure symptom;
The process of rule inference is
S1, which establishes rules based on empirical data on the relationship between known failure modes and failure signs, and determines the probability of failure mode occurrence based on the input failure sign data;
S2: constructing an analytical model for the time-series data of the equipment and analyzing whether or not related failure signs have occurred;
S3 encapsulating the equipment symptom data parsed in S2 into a JSON structure input to the rule interface and requesting the rule REST interface established in S1;
processing the processing REST interface request by a rule execution engine, inputting data to the rule, determining and outputting a failure mode occurrence probability;
The process of big data model algorithm is to first input tagged fault samples, then perform data preprocessing, and then train with a classification model. A method for comprehensively diagnosing a failure of a hydroelectric power generation unit, the method comprising outputting a result using a model .