Fan blade damage identification and fault prediction method based on twinborn model residual device

Through the online data-driven twin model and fuzzy mathematical principles, a twin residual device is designed for wind turbine blade fault diagnosis, solving the problem that existing technology is difficult to cope with complex environments and massive data, and achieving accurate identification of blade fault categories and degrees and forecast, improving the operating safety of wind turbines.

CN118940148AInactive Publication Date: 2024-11-12SHANGHAI JIAQI NETWORK TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410983839.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing wind turbine blade fault diagnosis technology is difficult to cope with complex external environment changes and fault feature extraction under massive data, and cannot effectively predict blade faults and safety hazards, and cannot target actuator faults, model uncertainty and interference effects. Targeted design.

Method used

Design a twin model based on online data-driven, design a twin residual device using real-time external output acceleration signals, combine fuzzy mathematical principles to establish a logical mapping relationship between fault types and fault signs, and build an intelligent fault identification method that takes into account model uncertainty and system interference. , realize health monitoring and failure degree prediction of blade surface damage.

Benefits of technology

It realizes intelligent judgment of wind turbine blade fault categories and accurate prediction of fault degree, reduces the misjudgment rate of fault diagnosis, and improves the operating reliability and safety of wind turbines.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940148A_ABST
    Figure CN118940148A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent fault diagnosis method for judging the health state of a blade of a wind driven generator in real time and predicting the damage degree of the blade of the wind driven generator to cope with complex external environment changes. The method comprises the following steps of establishing a data-driven twinborn model, generating a twinborn model residual device, extracting fault symptom information based on residual errors, and training an intelligent fault diagnosis device. The method comprises the following steps: establishing a fault-free model sand hole fault model, a crack fault model and a fracture fault model which accord with a vibration state (simulating an external environment wind load) by utilizing input-output data under the condition of considering model uncertainty and matching and non-matching interference influences, and extracting fault symptom information based on residual statistical characteristics in various different fault states; and establishing a logic mapping relation between a fault symptom fuzzy subset and a fault type based on an expert, and automatically forming a fan blade surface damage intelligent health state and health degree monitoring and prediction based on a twin model residual device through a self-learning mechanism on the basis of not giving experience knowledge and an identification function of a related mode in advance. The operation reliability of the fan is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis of wind turbine systems, and specifically to an intelligent fault diagnosis and system health monitoring method for wind turbine blade surface damage identification based on an online data-driven twin model. Background Art

[0002] Wind turbines are electrical equipment that converts wind energy into electrical energy. They are an environmentally friendly and renewable energy form. As one of the core components of wind turbines, blades not only affect the wind energy capture efficiency of the generator in terms of initial design, shape optimization, and installation safety, but also the state of the blades during operation is an important factor affecting the economic feasibility and rate of return of wind power projects. Therefore, regular maintenance and inspections, and timely detection of blade wear, cracks, and other potential problems can help detect and resolve blade failures in advance, maximize the performance and life of wind turbines, and are a research hotspot for accelerating the intelligent transformation of power grid infrastructure and the construction of smart microgrids.

[0003] Real-time online fault diagnosis and prediction technology has a good ability to predict potential faults and safety hazards of wind turbines, and can realize the state prediction and health management (Prognostics and Health Management, PHM) of the operating wind turbine system, reduce safety hazards, and improve the reliability of wind turbine operation. However, with the rapid growth of the number of wind turbines installed, the current data monitoring system cannot meet the development needs of wind power technology. Establishing an efficient intelligent fault diagnosis and prediction system to realize online state monitoring has important theoretical significance and engineering application value for preventing major accidents and ensuring the safety of the power grid.

[0004] In order to meet the requirements of wind turbine systems in complex and changing environments, under the premise of closed-loop stability, the blade system status monitoring is realized based on data-driven system models and real-time monitoring data. Traditional signal analysis and processing methods for fault identification are difficult to cope with the fault feature extraction of massive and complex data. In addition, in the face of the multiple impacts of actuator failure, model uncertainty, matching and non-matching interference on the blade system, the existing fault diagnosis technology has not been designed in a targeted manner and cannot meet the requirements of fault status diagnosis and prediction of complex wind turbine blade systems. Summary of the invention

[0005] In response to the above-mentioned existing technical problems, the present invention aims to disclose a real-time intelligent fault category diagnosis and blade damage degree fault identification method for wind turbine blade health management caused by complex external environment changes. This method is based on a simplified model that can approximately describe the system characteristics, and takes into account the influence of model uncertainty, matching and non-matching interference. It uses real-time external output acceleration signals to design an online residual device, and designs a twin residual device fault diagnosis method to achieve intelligent identification of blade fault categories and prediction of blade fault degrees.

[0006] The technical solution to realize the present invention is to ignore the influence of nonlinear factors, use input and output data to establish a fault-free model, a sand hole fault model, a crack fault model and a fracture fault model that conform to the characteristics of the blade system under the vibration state (simulating the wind force of the external environment), and study the statistical characteristics of the model residuals of different fault types through a large number of experiments, and design an intelligent fault identification method based on statistical characteristics. According to fuzzy mathematics and related principles, a logical mapping relationship between expert-based fault types and fuzzy subsets of fault symptoms is established, and a fuzzy relationship with self-learning ability is used to construct an online model establishment and residual generation mechanism under the influence of model uncertainty and system interference, so as to realize an intelligent health monitoring and identification method for surface damage of wind turbine blades.

[0007] The present invention provides an intelligent fault identifier and a fault degree prediction method based on a twin model to generate a residual device, the steps of the method are:

[0008] (1) Establishment of the blade twin model in a healthy state

[0009] An excitation signal is designed to excite the blade system, simulate the output response signal under the influence of wind external loads, and establish a random system twin model driven by input-output data to describe the system characteristics of the wind turbine blades under a fault-free state. The established wind turbine blade model in a healthy state can realize the real-time generation, data statistics and analysis of the twin residual generator under a fault state while meeting the requirements of model accuracy and generalization.

[0010] (2) Generation and data analysis of online digital twin residuals under various fault conditions

[0011] The same excitation signal is used to excite the blade system under different fault states. It is assumed that the model order and model structure remain unchanged compared to the system running in a healthy state. The real-time signal acquisition device is used to realize input and output data acquisition and analysis. The recursive least squares method with forgetting factor is used to online identify the system model under different fault states. The twin model residual device under different fault states based on the verification results is designed. After a large number of residual device generation experiments under various fault states, the residuals under different fault states are analyzed, and online statistics and analysis are performed on each residual result.

[0012] (3) Establish a nonlinear mapping relationship (fuzzy relationship) between fault categories and twin residual models

[0013] Extract the fault signs that depend on the residual model, and establish a nonlinear mapping relationship between the fault sign space (residual statistical characteristics analysis) and the fault space. Use the system online residual generator in different states to generate real-time residuals as the fuzzy description of the input variables of the fuzzy reasoning system. At the same time, the corresponding fault type is used as the fuzzy description of the output variables of the fuzzy reasoning system. The shape of the membership function is obtained by analyzing the relevant data of the residual variables obtained in real time, so as to achieve the purpose of influencing the fuzzy segmentation of the input and output spaces, and realize the automatic formation of the required decision area through the self-learning mechanism without the need to give the empirical knowledge and recognition function of the relevant patterns in advance, which has a direct impact on the fuzzy reasoning performance, reasoning accuracy and fault degree estimation.

[0014] Each step in the present invention is determined according to the following experimental principles and methods:

[0015] In step (1), the excitation signal is designed to be transmitted to a small vibration table and an impedance head through a signal acquisition device to excite the blade stump, thereby simulating the external load of wind force on the wind turbine blade. The response output signal is transmitted through an acceleration sensor. Finally, the output acceleration response signal of the blade system in a vibrating state is obtained in real time through the signal acquisition device.

[0016] Using the obtained real-time input and output data, the SISO fan blade system is approximated as a second-order linear system under the premise of ignoring the influence of nonlinear factors, and the autoregressive moving average model (ARMA) is used.

[0017] A(z -1 )y k =z -d B(z -1 ) k +ξ k

[0018] Describe the system. In the formula, u k and kThey represent the input (vibration signal) and output (acceleration) of the blade system, d (d ≥ 1) is the pure delay, ξ k is a random perturbation, z -1 is a backshift operator, and

[0019]

[0020] In the experiment, a modulated sinusoidal signal with both amplitude and frequency attenuated is used.

[0021] u k =A m e -k / 4000 sin(2πf max e -k / 3900 kT0),A m =1,f max =30Hz

[0022] The blade system is excited, where A m =1 and f max =30Hz are the initial values ​​of the input excitation signal, T0 is the sampling period, and the designed input signal is combined with the above experimental device to excite the blade system. The unknown parameters in the model are optimized and estimated by using the input and output data in a noisy environment combined with the least squares identification algorithm with a forgetting factor.

[0023] The generalization and accuracy of the model are important criteria for testing the performance of the established model. The model is verified using signals with different initial and attenuation frequencies from the excitation input.

[0024]

[0025] Analyze the quality of data fitting. In the formula, y is the real-time transmission angular velocity output, Represents the model estimation output. The above formula can be used to verify the accuracy of the model establishment achieved by the data-driven model identification method adopted.

[0026] Furthermore, in step (2), the optimization function in the random system is

[0027]

[0028] The identification strategy for the target can be used to establish a system model with uncertainty, where E(·) is the mathematical expectation operator, y k is the actual output of the system under wind load and noise environment, The output is predicted by a mathematical model fitted with the blade system input-output data in a fault-free state.

[0029] The present invention intends to use the real-time output data (acceleration signal) of the blade system under different states through the same excitation function and the output of the model established in step (1) to obtain the system twin residual generator

[0030]

[0031] Based on the generated real-time residual, fault signs based on the residual signal are extracted (the present invention intends to extract the residual mean under different fault states as the sign information). Combined with the basic idea of ​​fault diagnosis, a fuzzy logic strategy that is good at judging and processing fuzzy phenomena is adopted to establish a mapping relationship from the fault symptom space to the fault space, so as to achieve accurate identification of the fault category and estimation of the fault degree, so as to make the optimal decision in line with economic benefits for the wind turbines in operation.

[0032] Furthermore, the implementation of the fuzzy inference engine in step (3) is divided into two parts: the formation of fuzzy rules and the determination of fuzzy implication relations. In the formation of fuzzy rules, the determination of fuzzy relations and their synthesis adopt Euclidean vectors.

[0033] Y={y1,y2,y3,…,y n}

[0034] Define the different fault types in the blade system, where n is the total number of fault types and y is the output language variable. Similarly, for the fault symptoms of the blade system, the Euclidean vector is used to describe:

[0035] X={x1,x2,x3,…,x m}

[0036] Wherein, the subscript m is the total number of fault symptoms extracted, and x represents the input language variable.

[0037] According to the principle of fuzzy mathematics, the logical relationship between the fuzzy subsets of fault types and the fuzzy subsets of fault symptoms can be described as:

[0038] Y = X * R,

[0039] Among them, * is the fuzzy operator, and R is the fuzzy diagnosis matrix of the diagnostic expert based on experience and knowledge, which can be expressed as:

[0040]

[0041] The above fuzzy relationship is represented by the membership function μ R (x i ,y j ) completely characterizes the element x in the non-empty domain X. i and the jth element y in the non-empty domain Y iFor the membership degree of fuzzy relation R, μ R (x i ,y j )∈[0,1],i=1,2,…,m,j=1,2,…,n.

[0042] According to the above discussion on fuzzy relations and combined with the analysis of blade vibration simulation experimental results, a fault category identification based on experimental data analysis is designed. First, the twin model residual and system state are described as fuzzy linguistic variables based on the available wind turbine blade fault category database, namely:

[0043] X(x) = X(residual) = {very small, small, medium, large}

[0044] Y(y)=Y(system status)={no fault, sand hole, crack, break}

[0045] Accordingly, by analyzing the experimental results of the above-mentioned different fault types, it can be concluded that as the degree of system fault becomes more serious, the maximum model residual under faults such as sand holes, cracks and fractures increases accordingly. Therefore, the twin model residual generator can be selected as a fault sign to determine the health state and health level of the blade system.

[0046] In the model training process of the fuzzy intelligent fault diagnosis strategy for real-time identification of blade status, the empirical rule is adopted: "If the residual is small, the fault type is judged to be sand hole." Let x and y represent the fuzzy language variables "residual" and "fault" respectively. According to the relevant vibration experimental data and results of the fan blade system, the domains of the input variable x and output variable y of the fuzzy fault diagnosis machine are set as

[0047] X={x1,x2,…,x n}

[0048] Y={y1,y2,…,y m}

[0049] Where m and n are the total number of fuzzy segmentation domains. If Zadeh representation is used to describe the fuzzy set A with small residual

[0050]

[0051] B represents the fuzzy set of fault degree

[0052]

[0053] Using the fuzzy implication minimum operation (Mamdani)

[0054]

[0055] Based on the above-mentioned fuzzy implication relationship, the fuzzy reasoning process analysis is realized, and the patent adopts the generalized affirmative reasoning method, namely:

[0056] Input: If x is A′

[0057] Premise: If x is A, then y is B

[0058] Conclusion: y is B′

[0059] Among them, A, A', B, B' are all fuzzy languages. For the fuzzy premise "if x is A, then y is B", it expresses the fuzzy implication relationship between the fuzzy languages ​​A and B, which is recorded as

[0060] A→B.

[0061] In view of the fact that the domains of the input and output variables x and y of the fuzzy inference engine used in the patent are both discrete values ​​and there is no complex fuzzy relationship synthesis, the fuzzy implication A→B can be directly represented by the fuzzy matrix R.

[0062] According to the statistical results of the previous twin model and residual data analysis, combined with the established fuzzy relationship, the input residual ∈ [0, 0.8] is fuzzified into four levels: very small, relatively small, medium and relatively large, and the fault degree of the output variable ∈ [0, 100%] is fuzzified into four levels: no fault, sand hole, crack, and fracture. The fuzzy recognition rules are:

[0063] IF the residual is very small THEN the blade surface is fault-free

[0064] IF the residual is relatively small THEN the blade fault is sand hole

[0065] IF the residual is medium THEN the blade failure is crack

[0066] IF the residual is is large THEN the blade surface is broken

[0067] Using the distribution characteristics of residuals, a fuzzy inference system based on the Mamdani model is designed to realize the online twin residual generator generated according to real-time data in complex environments, monitor and accurately judge the health status of the wind turbine blade system. Determine the membership functions within different domains, and select rules to follow the distribution of the residual mean under different fault conditions to assist in fine-tuning. Whether the selection of membership functions conforms to the system characteristics determines whether the fuzzy output is a strictly increasing function of the input, and also determines whether the established nonlinear mapping relationship conforms to the principle that the degree of fault gradually increases with the increase of residuals and the statistical law of data. It should be noted that the selection of membership functions needs to ensure that the intersection of adjacent fuzzy quantities belongs to the specified range to ensure that the output strictly increases with the input.

[0068] The principle has been explained.

[0069] Beneficial effects:

[0070] The present invention discloses an intelligent fault diagnosis method and fault prediction strategy research based on a twin model residual device driven by online vibration data. The designed input vibration signal and the real-time output acceleration signal of the blade system are used to establish a fault-free mathematical model description that conforms to the characteristics of the blade system; while using a large number of residual data of different fault types, i.e., sand holes, cracks, and fractures to establish a fault file, the fuzzy relationship is determined, synthesized, and the shape of the membership function is determined, thereby affecting the fuzzy segmentation of the input and output variables, and a nonlinear mapping relationship between the residual variables and the fault category variables that conforms to the fault file is constructed, realizing a fuzzy reasoning mechanism with self-learning and self-adaptive capabilities. Finally, the fuzzy implication relationship of the generalized affirmative recursive method is used to realize intelligent fuzzy reasoning and fault degree prediction based on the residual generator. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a structural diagram of the present invention;

[0072] Figure 2 is an experimental vibration analysis system and a piezoelectric accelerometer;

[0073] Figure 3 It is the installation and arrangement of experimental equipment;

[0074] Figure 4 It is the parameter convergence process of the data-driven model under the fault-free state;

[0075] Figure 5 is the model verification result under the fault-free state;

[0076] Figure 6 It is the model verification result under the sand hole fault state;

[0077] Figure 7 is the model verification result under crack failure state;

[0078] Figure 8 is the model verification result under fracture fault state;

[0079] Fig. 9 is the output response curve under different fault conditions;

[0080] Fig.10 It is a comparison diagram of the model residuals of the blade system under different fault conditions;

[0081] Fig.11 It is determined by the membership function corresponding to the fault symptom (model residual) and the fault type. DETAILED DESCRIPTION

[0082] Example:

[0083] The present invention provides a strategy for fault identification and fault degree prediction of wind turbine blade surface damage based on real-time generation of residuals by digital twin models. Figure 1 It is a structural diagram of the present invention.

[0084] In the embodiment of the present invention, the excitation signal is designed to be transmitted to a small vibration table (model: TIRA-GMBH-TV110) and an impedance head (model: KISTLER 8770A50, sensitivity 92.7mv / g) through a Lab Genius real-time signal acquisition device (model: LabGenius IM1208H), as shown in FIG2(a), to excite the blade stump, thereby simulating the external load of wind on the wind turbine blade. The output response signal is transmitted to an acceleration sensor (model: CA-YD-186, sensitivity ), Figure 2(b) shows the piezoelectric acceleration sensor used in the experiment. Finally, the output acceleration response signal of the blade system in the vibration state can be obtained in real time using the Lab Genius signal acquisition device. The experimental equipment and its installation method are shown in Figure 3 shown.

[0085] The unknown parameters in the model are optimized and estimated by using the obtained input and output signals combined with the least squares identification algorithm with forgetting factor. The parameter convergence process is as follows: Figure 4 In order to verify the generalization and accuracy of the model, another modulation signal with different initial frequency and attenuation frequency from the excitation signal is used to verify the established twin model. The model verification results are shown in Figure 5 As shown in the figure, the analysis of the experimental results shows that even if the established model is a second-order linear model, the output acceleration response is 0.7m / s 2 The model error is only 0.035m / s 2 , the model error is only 5%, and the mean square error is further used

[0086]

[0087] The data fitting was analyzed and the experimental results showed that the proposed model building strategy could achieve an accurate model verification result of MSE=0.1076.

[0088] In the process of establishing the fuzzy inference engine based on model residuals, firstly, a large number of experiments are conducted to obtain the real-time residual related experiments under various fault conditions, such as Figure 5-Figure 10 The fault information extraction is shown in Table 1. The experimental results of different fault types are shown in Figure 5-Figure 8 As shown in the figure, (a) and (b) in each figure correspond to the blade system output response and residual under the conditions of no fault, sand hole, crack, and fracture, respectively. Through analysis, it can be seen that under the condition of no fault ( Figure 5The maximum model residual (shown) is only 0.035m / s 2 As the system fault becomes more serious, the maximum model residuals under faults such as sand holes, cracks and fractures are 0.17m / s and 2 , 0.6m / s 2 and 0.985m / s 2 , the model residual increases accordingly, so the twin model residual generator can be selected as a fault sign to determine the health state and health level of the blade system.

[0089] In order to clearly display the experimental results, the data of the response period 0.017s-0.07s are selected to analyze the rationality of the twin model residual device as a fault symptom information extraction, such as Fig. 9 and Fig.10 As shown. After the output response curves and twin model residual curves under each fault state are enlarged, it can be more clearly and intuitively shown that the residual variable relationship of the wind turbine blade system in different health states is very different. The above experimental results can be summarized as follows: as the severity of the fault increases, the model residual will also increase. In addition, the broken parts will also cause serious random noise interference to the entire system.

[0090] Table 1 Residual statistical characteristics of different fault types

[0091]

[0092] In order to verify the above conclusions, the characteristic information of the blades under different states, including the model mean square error (MSE), residual mean (MRE), maximum residual error (MARE), residual variance (VARE) and other statistical characteristic functions, as shown in Table 1, this analysis process not only realizes the support Figure 5-Figure 10 In addition to the graphical curve results, what is more important is that it can determine the membership function in the intelligent diagnosis strategy, determine the expert rules that meet the system characteristics, reduce the misjudgment rate of fault diagnosis, and realize the intelligent fault diagnosis algorithm of the fan blade system in a complex environment with online self-learning and adaptive functions.

[0093] Table 2 Fuzzy logic fault classification and fault severity prediction of fan blades based on twin model residuals

[0094]

[0095] The fuzzy relationship in the fuzzy inference engine is determined. The input residual ∈ [0, 0.8] is fuzzified into four levels: very small, relatively small, medium and relatively large. The fault degree of the output variable ∈ [0, 100%] is fuzzified into four levels: no fault, sand hole, crack and fracture. The fuzzy inference system based on the Mamdani model is designed by Matlab using the established fuzzy rules and residual distribution characteristics to realize the real-time health monitoring and status management of the wind turbine blade system. Fig.11 Select membership functions within different domains.

[0096] According to the real-time output data of the blade and the model residual obtained by the second-order ARMA model established by data drive under the fault-free state, combined with the fuzzy logic fault diagnosis program established in Matlab, the state of the blade is tested when the model residuals are 0.032, 0.15, 0.40, 0.67, 0.75, 0.05, 0.34, and 0.53 respectively. The experimental results are shown in Table 2. The experimental results show that the intelligent fault identification algorithm used in the patent has an identification rate of almost 100% based on the accurate residuals that can be obtained.

[0097] The embodiments are only for illustrating the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution in accordance with the technical idea proposed by the present invention shall fall within the protection scope of the present invention.

Claims

1. The fan blade damage identification and fault prediction method based on the twin model residual device has the following improvements: 1) Establishment of real-time twin model. To cope with the complex external environment, a mathematical model description that conforms to the real-time characteristics of the system is established. Combined with the system model without faults, a twin model residualizer is established to statistically analyze the digital features of the model residuals under different fault conditions. 2) Intelligent nonlinear fuzzy logic mapping relationship. Establish the input-output fuzzy description of fuzzy reasoning, and use statistical characteristics to obtain the corresponding fuzzy segmentation, form the decision area through the self-learning mechanism, and obtain the fuzzy rules about the fault symptom space and the fault space.

2. According to the method described in claim 1, under the premise of ignoring model nonlinearity, matching and non-matching interference, the external wind load is simulated to excite the wind turbine blades to obtain the system acceleration output response, and the least squares method with a forgetting factor is used to establish a second-order linear model that conforms to the system characteristics. The ignored system characteristics are used as a real-time signal to generate a residual device, and the digital features of the residual under various fault states are used to further extract fault symptom information.

3. The method of claim 2, using fuzzy operators to approximate the fuzzy logic relationship between the fuzzy subsets of fuzzy fault types and the fuzzy subsets of fault symptoms. Using fuzzy language variables of fuzzy logic diagnosis mechanism based on the fault category database of wind turbine blade system: X(x) = X(residual) = {very small, small, medium, large} Y(y)=Y(system status)={no fault, sand hole, crack, break} The fuzzy intelligent fault diagnosis strategy of wind turbine blades is trained as a model to form an intelligent fault diagnosis mechanism with self-learning ability.

Citation Information

Cited By

  • Intelligent monitoring system and method for water pump vibration signal analysis and fault recognition

    CN121803480A