A gas turbine fault diagnosis method and system

Abstract

The present invention discloses a gas turbine fault diagnosis method and system. Based on complex fault factors, typical fault modes and multi-source symptom information data are obtained to complete the construction of a fault diagnosis ontology knowledge base model; the functional requirements of ontology query and ontology-based data access are realized; the historical data of the gas turbine is automatically cleaned to obtain benchmark data that meets the training parameters; the benchmark data that meets the training parameters is processed to build a multivariate benchmark model of the unit; the multivariate benchmark model is processed to obtain early detection information of abnormal states of the gas turbine; based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and fault knowledge, a diagnostic decision model based on a structural causal model is constructed. Based on the diagnostic decision model, all candidate faults are compared, the cause of the fault is obtained, and a maintenance strategy is provided. The present invention can obtain the cause of the fault in complex machinery, save manpower and material resources, and improve work efficiency.

Description

Technical Field

[0001] The invention belongs to the field of mechanical fault diagnosis and relates to a gas turbine fault diagnosis method and system. Background Art

[0002] As a typical complex system involving the mutual transformation of multiple media such as gas, water, steam, and oil, as well as multiple energy forms such as heat, cold, machinery, and electricity, the gas turbine combined cycle unit has close connections between its components through material flow, energy flow, and the accompanying information flow. The state parameters are high-dimensional and nonlinear. During operation, the unit may experience various faults such as gas path and mechanical vibration, and energy efficiency anomalies such as improper operation adjustment and reduced operating efficiency caused by changes in operating conditions. The faults are diverse, hierarchical, and transmissible at different time and space scales, forming a complex causal transmission chain. Summary of the invention

[0003] The purpose of the present invention is to solve the problems in the prior art and to provide a gas turbine fault diagnosis method and system. Based on the causes of complex faults, a fault diagnosis model is established and the fault diagnosis model is evaluated. All candidate faults can be obtained, the causes of the faults can be obtained, and maintenance strategies can be provided.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A gas turbine fault diagnosis method, comprising:

[0006] Based on the theoretical analysis of fault systems and the fault characteristics of gas turbines, typical fault modes and multi-source symptom information data at each structural and functional level are obtained;

[0007] Process the acquired information data and construct a fault diagnosis ontology knowledge base model;

[0008] Based on the constructed fault diagnosis ontology knowledge base model, the functional requirements of ontology query and ontology-based data access are realized;

[0009] According to the operating characteristics of the gas turbine, the historical data is automatically cleaned to obtain benchmark data that meets the training parameters;

[0010] Process the acquired benchmark data that meets the training parameters to build a multivariate benchmark model for the unit;

[0011] Process the constructed multivariate benchmark model to obtain early detection information of abnormal conditions of gas turbines;

[0012] Based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and the fault knowledge, a diagnostic decision model based on the structural causal model is constructed;

[0013] Counterfactual reasoning is applied to the constructed decision model to compare all acquired candidate faults, derive the fault causes and provide maintenance strategies.

[0014] A further improvement of the present invention is:

[0015] Obtain typical failure modes and multi-source symptom information at each structural and functional level. Specifically: Based on FMEA and FTA theory, analyze the multi-source heterogeneous failure data of the gas turbine, and obtain typical failure modes and multi-source symptom information data at each structural and functional level.

[0016] Process the acquired information data and construct a fault diagnosis ontology knowledge base model, including:

[0017] According to ontology theory, semantic web technology and data semantic modeling method, a fault diagnosis ontology knowledge base model is constructed based on the acquired fault modes and information.

[0018] Based on the constructed fault diagnosis ontology knowledge base model, the functional requirements of ontology query and ontology-based data access are realized; including: reasoning on the ontology based on OWL axioms and SWRL, and realizing the functional requirements of ontology query and ontology-based data access through SPARQL language.

[0019] Automated cleaning includes: steady-state identification, working condition division, and benchmark sample screening;

[0020] The working condition division is based on the K-means clustering algorithm.

[0021] The obtained benchmark data that meets the training parameters is processed to construct a multivariate benchmark model of the unit, including: determining a benchmark value and establishing a multivariate benchmark model of the unit based on a conditional variable autoencoder.

[0022] The constructed multivariate benchmark model is processed to obtain early detection information of abnormal conditions of the gas turbine, including: using the reconstruction probability as a characteristic indicator of abnormal detection to determine whether the operating condition of the gas turbine is in an abnormal state. If the tested operating condition is an abnormal state, the abnormal parameters of the operating condition are obtained by calculating the difference vector between the reconstructed output and the original input parameters.

[0023] Based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and the fault knowledge, a diagnostic decision model based on the structural causal model is constructed; including: guided by the obtained early detection information of the gas turbine, using the semantic search of the ontology to obtain the fault, cause and symptom information of the abnormal state, using the attribute relationship between the fault and symptoms of the abnormal state to construct a causal graph network, combining the introduced hidden variable nodes and probability assignment to construct a diagnostic decision model based on the structural causal model.

[0024] Counterfactual reasoning is applied to the constructed decision model, and all candidate faults obtained are compared to obtain the cause of the fault; including: counterfactual reasoning is performed according to the topological structure of the structural causal model, and all candidate faults are sorted based on the two indicators of sufficiency and necessity of the fault alternative set to obtain the cause of the fault; based on the diagnostic reasoning results, ontology knowledge is used again to trace the fault back to form a fault causal chain, diagnose the most likely fault and its evolution process, and provide maintenance strategies according to the cause of the fault.

[0025] A gas turbine fault diagnosis system, comprising:

[0026] An information data acquisition module, which acquires typical failure modes and multi-source symptom information data at various structural and functional levels based on a theoretical analysis of the failure system and the failure characteristics of the gas turbine;

[0027] A first construction module, the first construction module is used to process the acquired information data and construct a fault diagnosis ontology knowledge base model;

[0028] A knowledge base model processing module, which realizes the functional requirements of ontology query and ontology-based data access based on the constructed fault diagnosis ontology knowledge base model;

[0029] A data cleaning module, wherein the data cleaning module is used to automatically clean the historical data according to the operating characteristics of the gas turbine to obtain benchmark data that meets the training parameters;

[0030] A data processing module, the data processing module is used to process the acquired benchmark data that meets the training parameters and construct a multivariate benchmark model of the unit;

[0031] A multivariate benchmark model processing module, the multivariate benchmark model processing module is used to process the constructed multivariate benchmark model to obtain early detection information of abnormal state of the gas turbine;

[0032] A second building module, wherein the second building module builds a diagnostic decision model based on a structural causal model based on the early detection information of the abnormal state of the gas turbine and the causal relationship and uncertainty between the fault knowledge;

[0033] A comparison module is used to apply counterfactual reasoning to the constructed decision model, compare all acquired candidate faults, derive the cause of the fault and provide a maintenance strategy.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention discloses a gas turbine fault diagnosis method and system. Combining system theory and the fault characteristics of gas turbine units, the FMEA and FTA technologies are used as guidance to obtain typical fault modes and multi-source symptom information at various structural and functional levels from multi-source heterogeneous fault data. The semantic modeling method of domain knowledge and data based on ontology theory and semantic web technology is studied to meet the requirements of multi-granularity semantic modeling of fault knowledge. According to the advantages of two reasoning methods based on OWL axioms and SWRL in ontology reasoning, the functional requirements of ontology query and ontology-based data access are realized through SPARQL language. According to the operating characteristics of the gas turbine, the automatic cleaning process of historical data is studied, including the steps of steady-state discrimination, operating condition division, benchmark sample screening, etc. In view of the problem of variable operating conditions of gas turbines, a multivariate benchmark model of the unit is established based on an improved conditional variable autoencoder, and the reconstruction probability is used as a characteristic indicator to realize the early detection of abnormal state of the unit. Research on gas turbine fault diagnosis and maintenance decision-making is carried out. Considering the causal relationship and uncertainty between fault knowledge, the diagnosis decision model based on the structural causal model is studied by introducing hidden variable nodes and AND-OR gate logic. Taking the maximum causal explanation of evidence as the semantics of diagnosis, counterfactual reasoning is applied to diagnosis, and the strength of the causal explanation of the fault cause to the evidence is measured from the perspectives of sufficiency and necessity. The model is evaluated from three aspects: sensitivity, reliability, and interpretability, confirming the advantages of counterfactual reasoning in diagnosis compared with reasoning based on posterior probability. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 A flow chart of a method for diagnosing a fault of a gas turbine according to the present invention;

[0038] Figure 2 Another flow chart of the gas turbine fault diagnosis method of the present invention;

[0039] Figure 3 A flow chart for obtaining failure mode and multi-source symptom information data;

[0040] Figure 4 The FTA analysis diagram of compressor surge fault of the present invention;

[0041] Figure 5 It is the fault knowledge ontology class and attribute relationship diagram of the present invention;

[0042] Figure 6 is a data processing flow chart of the present invention;

[0043] Figure 7 This is the network structure diagram of CVAE;

[0044] Figure 8 The gas turbine intelligent diagnosis flow chart of the present invention;

[0045] Fig. 9 This is a structural diagram of the fault diagnosis system of the gas turbine of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0049] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0050] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0051] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0052] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0053] See also Figure 1 and Figure 2 , Figure 1 and Figure 2 A gas turbine fault diagnosis method is disclosed, comprising:

[0054] S101, based on the theoretical analysis of the fault system and the fault characteristics of the gas turbine, obtain typical fault modes and multi-source symptom information data at each structural and functional level.

[0055] Gas turbine generator set is a complex system involving the mutual conversion of multiple media such as gas, water, steam, oil, and multiple energy forms such as heat, cold, machine, and electricity. Failures in different structural units show characteristics such as diversity, emergence, conditionality, transmission, and symptom complexity. According to the characteristics and laws of the occurrence and development of gas turbine failures, FMEA and FTA are used to scientifically guide fault analysis and acquisition.

[0056] See also Figure 3 and Figure 4 , taking the gas turbine compressor equipment as the object and the compressor surge failure as an example, the FTA and FMEA analysis process is explained. Table 1 is the compressor surge FMEA analysis table.

[0057] Table 1 Compressor surge FMEA analysis table

[0058]

[0059] S102, processing the acquired information data to construct a fault diagnosis ontology knowledge base model.

[0060] According to ontology theory, semantic web technology and data semantic modeling method, a fault diagnosis ontology knowledge base model is constructed based on the acquired fault modes and information.

[0061] The fault knowledge ontology is a semantic representation of the knowledge of the causal chain of fault evolution, which is divided into five categories: fault mode, fault symptom, fault cause, fault impact and maintenance measures. The fault mode is the top event of the fault tree analysis, the fault cause is the bottom event of the fault tree analysis, the fault impact and maintenance measures are inevitably related to the fault cause, and the fault symptom is the evidence for identifying the occurrence of the fault mode and cause. The logical relationship between the fault knowledge ontology classes and attributes is as follows: Figure 5 As shown in the figure, by introducing the object attributes of "hasCause", "hasEffect", "hasSymptom" and "hasMaintenance", the expression of causal relationship between faults is enriched, and the final ontology knowledge base is obtained.

[0062] S103, based on the constructed fault diagnosis ontology knowledge base model, realize the functional requirements of ontology query and ontology-based data access.

[0063] Ontologies are reasoned based on OWL axioms and SWRL, and the functional requirements of ontology query and ontology-based data access are realized through SPARQL language.

[0064] OWL (Ontology Web Language) is used to describe rich and complex knowledge such as concepts, concept sets, and relationships between concepts in the Semantic Web.

[0065] According to the different ways of semantic interpretation, OWL can be divided into OWL Full and OWL DL. There is an irreconcilable contradiction between the expressive power and reasoning power of ontology languages. Among them, OWL Full is the most complete OWL language, which interprets semantics with RDF graph structure and has strong expressive power, but lacks the decidability of reasoning. OWL DL is a direct semantic interpretation based on Description Logics (DL). Compared with OWL Full, OWL DL sacrifices some semantics and restricts the constructors used by OWL, but guarantees the completeness and decidability of reasoning.

[0066] OWL DL is selected as the ontology description language in the field of health maintenance because it maximizes the balance between the two and is closer to human natural language.

[0067] Taking the data of a compressor fouling period as an example, the effect of ontology reasoning based on SWRL (Semantic Web Rule Language) rules in fault identification is verified. The upper limit of the compressor outlet temperature benchmark is 420.5, the value type is float, and the lower limit is 415.3, the value type is float; the upper limit of the compressor efficiency benchmark is 92.2, the value type is float, and the lower limit is 89.6, the value type is float; the upper limit of the compressor outlet pressure benchmark is 1.45, the value type is float, and the lower limit is 1.35, the value type is float.

[0068] The compressor outlet temperature is 418.9 degrees Celsius, the outlet pressure is 1.31Mpa, the compressor efficiency is 89.1%, and the gas turbine power is 209.4MW. According to the data of the characteristic parameters and the assignment of the reference interval, two signs of a decrease in the compressor outlet pressure and a decrease in the compressor efficiency can be obtained, and it can be judged that the gas turbine is under load.

[0069] In order to represent the diagnostic report after ontology reasoning in this case, a diagnostic report ontology was created, which is divided into parameter report class, symptom report class, fault report class and working condition report class. The SWRL rules are as follows: Rule 1 means: if the characteristic parameter corresponding to the fault symptom is abnormal, the fault symptom is judged to be present; Rules 2 and 3 mean: if the value of the characteristic parameter exceeds the reference interval, the parameter is judged to be abnormal; Rule 4 means: if the symptom corresponding to the fault mode appears and the condition for the fault to occur meets the current working condition, the fault mode is judged to be present; Rule 5 means: if the gas turbine power exceeds 5MW, the gas turbine is judged to be in a load condition.

[0070] The fault identification based on SWRL reasoning is as follows: SWRL rule 1 is used to accurately detect the occurrence of the two fault signs of "compressor outlet pressure drop" and "compressor efficiency drop"; SWRL rule 2 is used to accurately detect the abnormalities of the two characteristic parameters of "compressor outlet pressure" and "compressor efficiency"; SWRL rule 4 is used to automatically infer the occurrence of the "compressor fouling" fault by combining the results of operating condition judgment and symptom judgment.

[0071] In addition, the symptom of “compressor outlet pressure fluctuation” was also detected during the reasoning process. This is because pressure fluctuation cannot be eliminated by only judging the limit values ​​of the characteristic parameters and the reference interval. However, SWRL Rule 4 uses the operating conditions as supplementary information to eliminate the occurrence of compressor surge failure.

[0072] The results show that by configuring SWRL rules by domain experts and carrying out ontology semantic reasoning based on SWRL, fault identification can be automatically performed, so that the knowledge system based on ontology representation has intelligent reasoning ability. At the same time, the participation of domain experts in knowledge classification is reduced, and the efficiency of knowledge management is improved.

[0073] In addition, the ontology knowledge base supports query rules based on SPARQL. By using the SPARQL language to define the association pattern of information, you can retrieve the target information that meets the user's needs in the ontology. The typical query statement form of SPARQL is shown in Formula 1:

[0074] SELECT? x WHERE{? x rdfs:subClassOf:F01.} (1)

[0075] For example:

[0076] SELECT? fault? symptom

[0077] WHERE {? faultrdfs:label'compressor fouling'. (2)

[0078] ? fault fd:hasSymptom? symptom.}

[0079] The results of the SPARQL query are displayed in Protégé. Through multiple pattern matching of two triples, the signs of compressor fouling failure are found to be: "increased compressor outlet temperature", "decreased compressor efficiency" and "decreased compressor outlet pressure".

[0080] S104, automatically cleaning the historical data according to the operating characteristics of the gas turbine to obtain benchmark data that meets the training parameters.

[0081] See also Figure 6 , Figure 6 The data processing process was announced; among them, automated cleaning includes: steady-state judgment, working condition division, and benchmark sample screening; the working condition division is based on the K-means clustering algorithm.

[0082] S105, processing the acquired benchmark data that meets the training parameters to construct a multivariate benchmark model of the unit.

[0083] Determine the benchmark value and establish a multivariate benchmark model of the unit based on the conditional variable autoencoder.

[0084] The operating state model of the gas turbine is the conditional probability distribution logp of the characteristic indicators X and Y under the condition of the operating variable C. θ(x, y|c) for monitoring. Therefore, based on the standard VAE (Variational Auto-Encoder) structure, conditional variables are introduced to improve the original network structure, and a conditional variational auto-encoder (CVAE) model is established to predict variable working conditions. The network structure of CVAE is as follows Figure 7 shown.

[0085] The goal of CVAE is to solve the conditional log-likelihood logp θ The value of the parameter when (x,y|c) is maximized. Under variable operating conditions, the logp θ (x) is expanded to the conditional log-likelihood logp θ (x,y|c), written as:

[0086] logp θ (x,y|c)=D KL (q φ (z|x,y,c)||p θ (z|x,y,c))+L ELBO (θ,φ; x, y, c) (3)

[0087] According to the basic principle of VAE, the variational lower bound L of the conditional log-likelihood ELBO (θ, φ; x, y, c) is expanded as shown in formula (4):

[0088] L ELBO (θ,φ;x,y,c)=E qφ(z|x,y,c) [log p θ (x,y|z,c)]-D KL (q φ (z|x,y,c)||p θ (z)) (4)

[0089] Assuming that the prior distribution of latent variables obeys isotropic multivariate Gaussian, the prior probability does not involve parameters, as shown in formula (5).

[0090] p(z)=N(0,I) (5)

[0091] The conditional likelihood of the observed variable p θ (x,y|z,c) obeys a multivariate Gaussian distribution, as shown in formula (6). The parameters are obtained from the output of the multi-layer neural network in the encoder, σ θ (z,c) is assumed to be a constant.

[0092] p θ (x,y|z,c)=N(μ θ (z,c),σ θ(z,c) 2 I) (6)

[0093] μ θ (z,c)=W θ [z,c]+b θ (7)

[0094] σ θ (z,c) = const (8)

[0095] Since the true posterior probability p θ The form of (z|x,y,c) is a multivariate Gaussian model with diagonal variance. In order to simplify the estimation of the posterior probability, the diagonal variance structure multivariate Gaussian distribution is also used to represent the encoder distribution q φ (z|x,y,c), as shown in formula (9).

[0096] q φ (z|x,y,c)=N(μ φ (x,y,c),σ φ (x,y,c) 2 I) (9)

[0097] The variational approximate distribution after reparameterization is shown in formula (10).

[0098] z=μ φ (x,y,c)+σ φ (x,y,c)⊙ε (10)

[0099] Among them, ε~N(0,1).

[0100] The parameters are also calculated by the neural network, as shown in formula (11).

[0101]

[0102] The CVAE model is trained using the benchmark samples screened by the data cleaning model to obtain a benchmark model under variable operating conditions of the gas turbine.

[0103] S106, processing the constructed multivariate benchmark model to obtain early detection information of abnormal state of the gas turbine.

[0104] The reconstruction probability is used as the characteristic indicator of anomaly detection to determine whether the operating condition of the gas turbine is in an abnormal state. If the tested operating condition is abnormal, the abnormal parameters of the operating condition are obtained by calculating the difference vector between the reconstructed output and the original input parameters.

[0105] Anomaly Detection Based on Reconstruction Probability

[0106] CVAE reconstructs the probability distribution of variables rather than the variables themselves, and contains more information about the variables, including the posterior distribution and generative likelihood of latent variables. The reconstruction probability is the reconstruction error term in the CVAE loss function. For the CVAE benchmark model, the smaller the reconstruction probability of the test data, the greater the difference between the data and the benchmark conditions. Therefore, the reconstruction probability E is chosen. qφ(z|x,y,z) [logp θ (x,y|z,c)] as a characteristic indicator for anomaly detection. The reconstruction probability not only considers the difference between the reconstruction and the original input, but also considers the impact of the latent variable variance on the reconstruction result, and has the potential to detect data fluctuation anomalies. Most importantly, the reconstruction probability measures the deviation of the current overall state from the benchmark, and is more sensitive to anomaly detection in multi-parameter coupled systems. The calculation formula for the reconstruction probability is shown in formula (12).

[0107]

[0108] The calculation logic of reconstruction probability is: first, input the test sample X, and obtain the parameter μ in the Gaussian distribution of the latent variable through the CVAE encoder φ and σ φ , and sample L points of the latent variable Z; secondly, the decoder calculates the parameter μ of the likelihood distribution of the reconstructed variable corresponding to each latent variable Z θ , σ θ ; Finally, the average value of the logarithmic likelihood of the test sample under the condition of latent variables is calculated. KDE statistics are used to reconstruct the probability distribution of the training samples, and the lower limit of the confidence interval corresponding to the confidence level equal to 95% is used as the threshold for anomaly detection, as shown in formula (13).

[0109]

[0110] In a combined cycle unit, the mutual coupling between thermodynamic parameters such as temperature and pressure, the disturbance of local parameters will be transmitted to other parameters in the system, and under the control and regulation of the system, the slight deviation of some parameters in the early stage of abnormality will be pulled to the normal level, affecting the location of parameter abnormality. Anomaly detection based on reconstruction probability can detect abnormal changes in the local system unit of the unit in advance. On this basis, the difference vector between the reconstructed output and the original input parameters is calculated, and the relative deviation between the two is used to further locate the abnormality of the parameters. The calculation of parameter deviation is shown in formula (14).

[0111]

[0112] In the formula, v k represents the standardization coefficient, which is obtained by the absolute deviation statistics between the reconstructed output of the training sample and the input signal. k and They represent the actual value and reconstructed value of parameter k respectively. Once the reconstruction probability anomaly is detected, the parameter anomaly is further located by comparing the deviation of each parameter in the system.

[0113] S107, based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and the fault knowledge, a diagnostic decision model based on the structural causal model is constructed.

[0114] Gas turbine intelligent diagnosis process Figure 8 As shown, guided by the obtained early detection information of the gas turbine, the semantic search of the ontology is used to obtain the fault, cause and symptom information of the abnormal state, and the attribute relationship between the fault and symptoms of the abnormal state is used to construct a causal graph network. Combined with the introduced latent variable nodes and probability assignment, a diagnostic decision model based on the structural causal model is constructed.

[0115] S108, applying counterfactual reasoning to the constructed decision model, comparing all acquired candidate faults, deriving the fault cause and providing a maintenance strategy.

[0116] According to the topological structure of the structural causal model, counterfactual reasoning is performed. Based on the two indicators of sufficiency and necessity of the fault alternative set, all candidate faults are sorted to obtain the cause of the fault. Based on the diagnostic reasoning results, ontology knowledge is used again to trace the fault back to form a fault causal chain, diagnose the most likely faults and their evolution process, and provide corresponding maintenance strategies based on the obtained fault causes.

[0117] See also Fig. 9 , Fig. 9 A gas turbine fault diagnosis system is disclosed, comprising:

[0118] An information data acquisition module, which acquires typical failure modes and multi-source symptom information data at various structural and functional levels based on a theoretical analysis of the failure system and the failure characteristics of the gas turbine;

[0119] A first construction module, the first construction module is used to process the acquired information data and construct a fault diagnosis ontology knowledge base model;

[0120] A knowledge base model processing module, which realizes the functional requirements of ontology query and ontology-based data access based on the constructed fault diagnosis ontology knowledge base model;

[0121] A data cleaning module, wherein the data cleaning module is used to automatically clean the historical data according to the operating characteristics of the gas turbine to obtain benchmark data that meets the training parameters;

[0122] A data processing module, the data processing module is used to process the acquired benchmark data that meets the training parameters and construct a multivariate benchmark model of the unit;

[0123] A multivariate benchmark model processing module, the multivariate benchmark model processing module is used to process the constructed multivariate benchmark model to obtain early detection information of abnormal state of the gas turbine;

[0124] A second building module, wherein the second building module builds a diagnostic decision model based on a structural causal model based on the early detection information of the abnormal state of the gas turbine and the causal relationship and uncertainty between the fault knowledge;

[0125] A comparison module is used to apply counterfactual reasoning to the constructed decision model, compare all acquired candidate faults, derive the cause of the fault and provide a maintenance strategy.

[0126] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A gas turbine fault diagnosis method, characterized in that: include: Based on the theoretical analysis of fault systems and the fault characteristics of gas turbines, typical fault modes and multi-source symptom information data at each structural and functional level are obtained; Process the acquired information data and construct a fault diagnosis ontology knowledge base model; Based on the constructed fault diagnosis ontology knowledge base model, the functional requirements of ontology query and ontology-based data access are realized; According to the operating characteristics of the gas turbine, the historical data is automatically cleaned to obtain benchmark data that meets the training parameters; Process the acquired benchmark data that meets the training parameters to build a multivariate benchmark model for the unit; The constructed multivariate benchmark model is processed to obtain early detection information of abnormal state of the gas turbine, including: using the reconstruction probability as an abnormal detection characteristic indicator to determine whether the working condition of the gas turbine is in an abnormal state, and if the tested working condition is in an abnormal state, the abnormal parameters of the working condition are obtained by calculating the difference vector between the reconstructed output and the original input parameters; Based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and the fault knowledge, a diagnostic decision model based on a structural causal model is constructed, including: guided by the obtained early detection information of the gas turbine, using the semantic search of the ontology to obtain the fault, cause and symptom information of the abnormal state, using the attribute relationship between the fault and the symptom of the abnormal state to construct a causal graph network, combining the introduced hidden variable nodes and probability assignment, to construct a diagnostic decision model based on the structural causal model; Counterfactual reasoning is applied to the constructed decision model to compare all the candidate faults obtained, derive the cause of the fault and provide a maintenance strategy, including: counterfactual reasoning is performed according to the topological structure of the structural causal model, all candidate faults are sorted based on the two indicators of sufficiency and necessity of the fault alternative set, and the cause of the fault is derived; based on the diagnostic reasoning results, ontology knowledge is used again to trace the fault back to form a fault causal chain, diagnose the most likely fault and its evolution process, and provide a maintenance strategy based on the cause of the fault.

2. The gas turbine fault diagnosis method according to claim 1, characterized in that: The obtaining of typical failure modes and multi-source symptom information at each structural and functional level is specifically as follows: based on FMEA and FTA theory, multi-source heterogeneous failure data of the gas turbine is analyzed to obtain typical failure modes and multi-source symptom information data at each structural and functional level.

3. The gas turbine fault diagnosis method according to claim 1, characterized in that: The acquired information data is processed to construct a fault diagnosis ontology knowledge base model, including: According to ontology theory, semantic web technology and data semantic modeling method, a fault diagnosis ontology knowledge base model is constructed based on the acquired fault modes and information.

4. The gas turbine fault diagnosis method according to claim 1, characterized in that: The functional requirements of ontology query and ontology-based data access are realized based on the constructed fault diagnosis ontology knowledge base model; including: reasoning on the ontology based on OWL axioms and SWRL, and realizing the functional requirements of ontology query and ontology-based data access through SPARQL language.

5. The gas turbine fault diagnosis method according to claim 1, characterized in that: The automated cleaning includes: steady-state identification, working condition division, and benchmark sample screening; The working condition division is based on the K-means clustering algorithm.

6. The gas turbine fault diagnosis method according to claim 1, characterized in that: The method processes the acquired benchmark data that meets the training parameters to construct a multivariate benchmark model for the unit, including: determining a benchmark value and establishing a multivariate benchmark model for the unit based on a conditional variable autoencoder.

7. A gas turbine fault diagnosis system, characterized in that: include: An information data acquisition module, which acquires typical failure modes and multi-source symptom information data at various structural and functional levels based on a theoretical analysis of the failure system and the failure characteristics of the gas turbine; A first construction module, the first construction module is used to process the acquired information data and construct a fault diagnosis ontology knowledge base model; A knowledge base model processing module, which realizes the functional requirements of ontology query and ontology-based data access based on the constructed fault diagnosis ontology knowledge base model; A data cleaning module, wherein the data cleaning module is used to automatically clean the historical data according to the operating characteristics of the gas turbine to obtain benchmark data that meets the training parameters; A data processing module, the data processing module is used to process the acquired benchmark data that meets the training parameters and construct a multivariate benchmark model of the unit; A multivariate benchmark model processing module, which is used to process the constructed multivariate benchmark model to obtain early detection information of abnormal state of the gas turbine; including: using the reconstruction probability as an abnormal detection characteristic indicator to determine whether the working condition of the gas turbine is in an abnormal state, if the tested working condition is in an abnormal state, by calculating the difference vector between the reconstructed output and the original input parameter, to obtain the abnormal parameters of the working condition; The second construction module, the second construction module constructs a diagnostic decision model based on a structural causal model based on the causal relationship and uncertainty between the early detection information of the abnormal state of the gas turbine and the fault knowledge; including: guided by the obtained early detection information of the gas turbine, using the semantic search of the ontology to obtain the fault, cause and symptom information of the abnormal state, using the attribute relationship between the fault and the symptom of the abnormal state to construct a causal graph network, combining the introduced hidden variable nodes and probability assignment, to construct a diagnostic decision model based on the structural causal model; A comparison module is used to apply counterfactual reasoning to the constructed decision model, compare all acquired candidate faults, derive the cause of the fault and provide a maintenance strategy, including: performing counterfactual reasoning according to the topological structure of the structural causal model, sorting all candidate faults based on the two indicators of sufficiency and necessity of the fault alternative set, and deriving the cause of the fault; based on the diagnostic reasoning results, using ontology knowledge again to trace the fault back to form a fault causal chain, diagnosing the most likely fault and its evolution process, and providing a maintenance strategy according to the cause of the fault.

Images