An intelligent assessment method for wind turbine gearbox health

Abstract

The present invention belongs to the technical field related to intelligent operation and maintenance of high-end equipment, and discloses a method for intelligently evaluating the health of a wind turbine gearbox. The method includes: S1 collecting vibration signals generated by wind turbine gears, constructing an MsXGB model and evaluating the equipment status membership; S2 forming a wind turbine gearbox fault impact evaluation committee, in which experts in the committee evaluate the degree of impact of the fault on the equipment, convert the evaluation into a Pythagorean fuzzy evaluation set, quantify the authority level of the experts in the committee, construct a fault impact fuzzy evaluation system, and determine the impact weight of each fault; S3 constructing a relationship between equipment health indicators, calculating the equipment health indicators, and thereby dividing the corresponding health states and processing, thereby realizing intelligent evaluation of the health of wind turbine gearboxes. Through the present invention, the impact of different fault conditions on equipment behavior is measured at a finer granularity, which can then accurately reflect the current health status of the equipment, improve maintenance efficiency, and provide better flexibility.

Description

Technical Field

[0001] The present invention belongs to the technical field related to intelligent operation and maintenance of high-end equipment, and more specifically, relates to an intelligent health assessment method for a wind turbine gearbox. Background Art

[0002] Wind turbine gearboxes, as critical components of large wind turbines, primarily transmit the power generated by the wind's rotation of the impeller to the generator, generating electricity. They also convert the relatively low speed of the impeller input into the required speed required by the generator. Complex equipment structures and harsh operating environments make wind turbine gearboxes susceptible to damage. Common faults include tooth shedding, cracked and broken teeth, and multiple coupled faults. These often impact equipment operation, hinder industrial production, and, in severe cases, even threaten human life and property. Therefore, accurately and robustly assessing the health of wind turbine gearboxes is crucial to effectively avoid equipment downtime, reduce maintenance costs, and ensure safe operation.

[0003] In recent years, many researchers have developed high-precision fault diagnosis methods. Currently, fault diagnosis methods can be roughly divided into two categories, namely physical model-based methods and data-driven methods. Physical model-based methods build models based on the failure mechanism of the equipment and then determine the current health status of the equipment. Data-driven methods use condition monitoring equipment to collect massive amounts of data from the equipment, and use equipment and service data such as equipment usage data, operating condition data, performance data, and parts replacement data to determine the health status of each component of the equipment. However, the two current diagnostic methods have the following defects:

[0004] (1) Most of the physical models currently established focus on the degradation pattern of a certain performance parameter and lack consideration of the joint degradation of multiple performance parameters, resulting in low accuracy of the constructed health status discrimination model;

[0005] (2) Data-driven methods can only qualitatively determine the location and type of equipment failures, but cannot quantitatively assess the impact of each component failure on the overall behavior of the equipment. This results in all component failures being treated as equally important, and the monitoring results cannot provide a device-specific description of the equipment's health status, nor can they provide hierarchical guidance for subsequent maintenance decisions.

[0006] These issues significantly limit the practical application of fault diagnosis methods. Therefore, there is an urgent need to develop a wind turbine gearbox health assessment method that can accurately diagnose the various faults currently present in the equipment and assess the impact of different fault types on equipment behavior, providing layered guidance for subsequent wind turbine gearbox maintenance decisions. Summary of the Invention

[0007] In response to the above-mentioned defects or improvement needs of the prior art, the present invention provides an intelligent health assessment method for a wind turbine gearbox to solve the problem of low accuracy in judging the health status of the equipment.

[0008] To achieve the above objectives, according to the present invention, a method for intelligently evaluating the health of a wind turbine gearbox is provided, the method comprising the following steps:

[0009] S1 presets and simulates the fault conditions of wind turbine gears, collects vibration signals generated by the operation of wind turbine gears under various fault types, and uses the time domain and frequency domain characteristics of the collected vibration signals to build and train the MsXGB model; the model is used to evaluate the equipment status membership;

[0010] S2 establishes a wind turbine gearbox fault impact assessment committee. Experts on the committee assess the impact of faults on the equipment and convert the assessment into a Pythagorean fuzzy evaluation set. Based on the evaluation set, the authority of the committee experts is quantified. A fault impact fuzzy evaluation system is constructed and used to determine the impact weight of each fault.

[0011] S3 uses the device state membership and the fault impact weight to construct a device health index relationship, calculates the device health index, divides the device into corresponding health states according to the health index, and performs corresponding processing on the device according to the health state, thereby realizing intelligent evaluation of the health of the wind turbine gearbox.

[0012] Further preferably, the MsXGB model includes a Count XGB model and a Locate XGB model, the CountXGB model is used to predict the number of faults, and the Locate XGB model is used to predict the probabilities corresponding to various faults.

[0013] Further preferably, the evaluation of the membership degree includes the following steps:

[0014] S11 extracts the time domain and frequency domain features of the vibration signal, takes the number of faults as output, uses the time domain and frequency domain features of the vibration signal as input to train the Count XGB model, and calculates the number of faults using the Count XGB model;

[0015] S12 uses the time domain and frequency domain features of the vibration signal as input and the fault state as output to train the Locate XGB model, modifies the loss function in the Locate XGB model so that its output is the fault probability of the corresponding fault state, that is, the fault state membership, and calculates the fault state membership using the Locate XGB model;

[0016] S13 combines the fault number obtained by the Count XGB model with the fault status membership result obtained by the Locate XGB model to obtain the device status membership.

[0017] Further preferably, the device state membership is calculated according to the following relationship:

[0018]

[0019]

[0020] Where p is the sample number, the sample is sensor data, p = 1, 2, ..., P, P is the number of samples, f p is the time domain and frequency domain characteristics of the vibration signal, MEMB(f p ) is the device status membership, is the predicted number of failures after the tth iteration of the Count XGB model, Z c (f p ) is the fault probability distribution, c is the fault state number, C is the number of fault states, The output of the previous layer of Softprob for the p-th sample input to the Locate XGB model for fault state c.

[0021] Further preferably, in step S2, quantifying the authority level of the experts in the committee according to the evaluation set comprises the following steps:

[0022] S21 The chief expert determines the best expert and the worst expert based on the Pythagorean fuzzy evaluation set;

[0023] S22 uses the BWM method to give the fuzzy Best-to-Others matrix and Others-to-Worst matrix of expert professionalism level;

[0024] S23 uses the BWM maximum value model to calculate the authority level of the expert group.

[0025] Further preferably, in step S23, the maximum value model is performed according to the following relationship:

[0026] minε

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] λ k ≥0

[0033] Among them, α is the possibility level predefined by the decision maker, ε represents the optimal consistency level of fuzzy preferences, and λ k is the weight of the k-th expert, λ B is the most authoritative expert weight, λ W is the weight of the least authoritative expert, and The best to others matrix of the kth expert authority fuzzy preference given by the chief expert The median and upper boundary values ​​of opinions in , and The others to worst matrix of the kth expert authority fuzzy preference given by the chief expert The median and lower bound values ​​of opinions.

[0034] Further preferably, in step S2, the construction of the fault impact fuzzy evaluation system is performed according to the following steps:

[0035] S24 For any j-th fault, construct and calculate the graph fuzzy set of the fault importance;

[0036] S25 calculates the knowledge metric of the graph fuzzy set of the j-th fault;

[0037] S26 constructs the relationship of the fault impact weight using the knowledge metric, and calculates the impact weight of each fault based on it.

[0038] More preferably,

[0039] The graph fuzzy set of fault importance is performed as follows:

[0040]

[0041]

[0042] Among them, PF j is the voting picture fuzzy matrix of the chief expert and assistant experts on the severity of the j-th fault state, λ *VF ,λ *KN ,λ *VA ,λ *Rov The voting weights of experts who choose “yes”, “neutral”, “no” and “abstain” are respectively, R(PF j ) is the vote of the chief expert and assistant experts on the severity of the j-th fault state Picture fuzzy matrix PF jThe voting hesitation degree is denoted by VF, KN, VA and Rov, which represent the options of approval, neutrality, opposition and abstention for the question “the jth fault state is a serious fault”. a and A are the expert number and number of experts who choose the approval option, b and B are the expert number and number of experts who choose the neutral option, d and D are the expert number and total number of experts who choose the opposition option, e and E are the expert number and total number of experts who choose the abstention option, and Ω is the value of the expert number and total number of experts who choose the abstention option. C.E The self-confidence coefficient given by the chief expert;

[0043] The knowledge measurement is performed as follows:

[0044] K(PF j )=1-0.5(E(PF j )+R(PF j ))

[0045]

[0046] Among them, K(PF j ) is PF j The knowledge measurement results of E(PF j ) is PF j The information entropy value, d n (PF j ,PF min ) is PF j and PF min The Hamming distance between n (PF j ,PF max ) is PF j and PF max Hamming distance between max is the maximum voting picture fuzzy matrix, PF min is the minimum voting picture fuzzy matrix, PF max =(1,0,0), PF min =(0,0,1).

[0047] Further preferably, in step S26, the relationship between the fault impact weight is as follows:

[0048]

[0049] in, is the optimal impact weight of the jth fault state, j is the fault state number, C is the number of fault states, K(PF j ) is PF j The knowledge measurement results.

[0050] Further preferably, in step S3, the device health index is calculated according to the following relationship:

[0051]

[0052] Among them, HI is the health of the equipment, z j is the state membership of the j-th fault, g is the Benefit criterion number, is the Non-Benefit criterion number, where all fault states are Non-Benefit criterion, j is the fault state number, is the number of predicted failure states, is the optimal impact weight of the j-th fault state.

[0053] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art:

[0054] 1. The present invention constructs the MsXGB model. The Count XGB and Locate XGB models in the MsXGB model can diagnose the complex fault category of the equipment. The SoftProb in the Locate XGB outputs the device state membership, which can quantitatively measure different current fault states.

[0055] 2. The GKM constructed in this invention can accurately measure the authority of the expert panel in terms of expert knowledge, experience, and capabilities. On this basis, it redistributes the expert voting weights to build a comprehensive and robust fault impact fuzzy evaluation system, determine the impact weights of different fault types on equipment behavior, and output objective and stable evaluation results.

[0056] 3. The wind turbine gearbox health assessment method of the present invention can output the overall health of the equipment and the single fault damage index, comprehensively reflecting the health of the wind turbine gearbox. At the same time, it can accurately determine the type of fault that is currently most harmful to the health of the equipment, quickly identify the most severely damaged parts, guide the maintenance sequence, and improve maintenance efficiency.

[0057] 4. After considering maintenance efficiency, maintenance costs, and technical constraints, the present invention provides a hierarchical maintenance and post-monitoring strategy for equipment in different health states from the aspects of maintenance level, maintenance method, and maintenance sequence. This strategy can provide technical guidance for operation and maintenance engineers and efficiently and economically handle the complex and changeable health conditions of wind turbine gearboxes. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of a method for evaluating the health of a wind turbine gearbox constructed according to a preferred embodiment of the present invention;

[0059] Figure 2 yes Figure 1 Schematic diagram of the MsXGB equipment status membership evaluation model involved in the wind turbine gearbox health evaluation method;

[0060] Figure 3 yes Figure 1 Schematic diagram of the GKM fault impact weight evaluation model involved in the wind turbine gearbox health assessment method.

[0061] Figure 4 The equipment in each state constructed according to the preferred embodiment of the present invention adopts corresponding maintenance and post-monitoring strategies. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0063] See also Figure 1 、 Figure 2 and Figure 3 The wind turbine gearbox health assessment method provided by the present invention mainly includes the following steps:

[0064] Step 1: Determine the fault condition to be simulated based on the typical fault mode of the wind turbine gearbox, and place vibration sensors according to the vibration characteristics under the fault type, and then collect the vibration signals generated by the equipment operation under different fault conditions.

[0065] Specifically, several NI-cDAQ-9174 / 9234 vibration sensors were fixed to the surfaces of various components of the wind turbine gearbox. The sensors were set to a sampling frequency of 10240 Hz and a sampling duration of 100 seconds to collect vibration signals generated by the equipment under different fault conditions.

[0066] In step 2, the vibration signals collected under various fault conditions are used as samples to train the constructed MsXGB model, and then the trained MsXGB model is used to evaluate the equipment state membership of the wind turbine gearbox.

[0067] Specifically, MsXGB includes two modules: Count XGB and Locate XGB.

[0068] The Count XGB module uses a regression model to obtain the number of equipment failures. First, the time domain and frequency domain features of the vibration signal are extracted, and the time domain and frequency domain features f with the corresponding fault number label m∈(1,...,M) are converted into p(p=1,...,P) is used to train the Count XGB model. The number of faults predicted by Count XGB is:

[0069]

[0070] in, and are the regression results of the number of failures of sample p after the t-1th and tth iterations respectively. t C (f p ) is the prediction result of the t-th decision tree in the Count XGB model.

[0071] When Count XGB applies the addition strategy, the objective function of the decision tree model added in each round is determined by using the second-order Taylor expansion formula:

[0072]

[0073] Among them, m p is the number of failures of sample p, g p is the first-order gradient, h p is the second-order gradient, is the loss function between the regression result of sample p and the number of faults after the tth iteration of the Count XGB model, Ω(F t C ) is the complexity of the t-th decision tree model, and constant is a constant.

[0074] Locate XGB uses a multi-classification model to calculate the probability distribution of each device fault as the fault state membership. p ∈(1,...,C)’s time domain and frequency domain features f p (p=1,...,P) trains the LocateXGB model. The fault category label calculated by the Locate XGB model is:

[0075]

[0076] in, and are respectively the fault label prediction results of sample p after the t-1th and tth iterations, F t L (f p ) is the calculation result of the t-th decision tree model of the Locate XGB model.

[0077] The Locate XGB classification model also uses the additive strategy. To enable the model to identify multiple faults simultaneously, the loss function of the model is modified:

[0078]

[0079] Among them, l L (·) is the cross entropy loss function used in classification research, is the cumulative score of sample p on the C-th fault after t-1 iterations, is the score of sample p on the C-th fault after the t-th iteration, Represents the higher-order terms of Taylor expansion. The fault probability values ​​of each sample are obtained through SoftProb as the fault state membership:

[0080]

[0081] Among them, θ c are the model parameters under each fault category, The value of the type C fault is normalized by SoftProb. The fault number obtained by the Count XGB module is combined with the fault status membership result obtained by the Locate XGB module to obtain the device status membership:

[0082]

[0083] Among them, MEMB(f p ) is the device status membership, is the number of failures, Z c (f p ) is the failure probability distribution.

[0084] Step 3: Establish a wind turbine gearbox fault impact assessment committee. Committee experts use natural semantics to evaluate the impact of each fault on equipment behavior, and then convert it into a Pythagorean fuzzy evaluation set.

[0085] Specifically, the natural semantic evaluation of the wind turbine gearbox fault impact evaluation committee experts is converted into a Pythagorean fuzzy evaluation set, see Table 1.

[0086] Table 1

[0087]

[0088] In Table 1, the fault severity is divided into 6 levels. The right column shows the Pythagorean fuzzy preference corresponding to the fault severity, which consists of the lower boundary value of the preference opinion, the median value of the preference opinion, and the upper boundary value of the preference opinion. The higher the value, the more serious the fault.

[0089] In step 4, based on the Pythagorean fuzzy evaluation set given by the committee experts, the proposed Group Knowledge Measurement (GKM) quantifies the authority of each expert, constructs a comprehensive and robust fault impact fuzzy evaluation system, and determines the impact weight of different types of faults on equipment behavior.

[0090] Specifically, the chief expert determines the best and worst experts based on the Pythagorean fuzzy evaluation set given by the assistant experts. At the same time, the fuzzy Best-to-Others matrix of the expert professional level is given using the Best Worst Method (BWM) method. Others-to-Worst Matrix The authority level of the expert group is calculated through the following BWM maximum value model

[0091] minε

[0092]

[0093]

[0094]

[0095]

[0096]

[0097] λ k ≥0

[0098] Among them, α is the possibility level predefined by the decision maker, ε represents the optimal consistency level of fuzzy preferences, and λ k is the weight of the k-th expert, λ B is the most authoritative expert weight, λ W is the weight of the least authoritative expert, and The best to others matrix of the kth expert authority fuzzy preference given by the chief expert The median and upper boundary values ​​of opinions in , and The others to worst matrix of the kth expert authority fuzzy preference given by the chief expert The median and lower bound values ​​of opinions.

[0099] Based on the expert authority assessment, the experts' opinions on the question of "whether the fault has a significant impact" are summarized in a graph fuzzy environment. For example, a chief expert and k assistant experts in the expert group are asked to evaluate whether the jth fault has a significant impact on the device behavior. If expert A and a senior expert in the expert group choose the option "yes", expert B chooses the option "neutral", expert D chooses the option "no", and expert E chooses the option "abstention", the graph fuzzy set of each expert's opinions on the importance of the jth fault is:

[0100]

[0101]

[0102] Among them, PF j is the voting picture fuzzy matrix of the chief expert and assistant experts on the severity of the j-th fault state, λ *VF ,λ *KN ,λ *VA ,λ *Rov The voting weights of experts who choose “yes”, “neutral”, “no” and “abstain” are respectively, R(PF j ) is the vote of the chief expert and assistant experts on the severity of the j-th fault state Picture fuzzy matrix PF j The voting hesitation degree is denoted by VF, KN, VA and Rov, which represent the options of approval, neutrality, opposition and abstention for the question “the jth fault state is a serious fault”. a and A are the expert number and number of experts who choose the approval option, b and B are the expert number and number of experts who choose the neutral option, d and D are the expert number and total number of experts who choose the opposition option, e and E are the expert number and total number of experts who choose the abstention option, and Ω is the value of the expert number and total number of experts who choose the abstention option. C.E The self-confidence coefficient given by the chief expert;

[0103] Graphical fuzzy set of the importance of the j-th fault The knowledge measure of is defined as follows:

[0104] K(PF j )=1-0.5(E(PF j )+R(PF j ))

[0105]

[0106] Among them, E(PF j ) is the entropy value of the graph fuzzy set, d n (PF j ,PF min ) is the PF of the graph fuzzy set j With PF minHamming distance between max =(1,0,0), PF min =(0,0,1). R(PF j ) is the abstention rate information of the graph fuzzy set, also known as the hesitation degree. The impact weight of each fault is defined as follows:

[0107]

[0108] in, is the optimal impact weight of the jth fault state, j is the fault state number, C is the number of fault states, K(PF j ) is PF j The knowledge measurement results.

[0109] Step 5: Combine the calculated equipment status membership and fault impact weight information using the Ratio system method to obtain the equipment health index, thereby conducting an overall quantitative assessment of the health of the wind turbine gearbox equipment. If the assessment results determine that the equipment is not in a healthy state, a maintenance and post-monitoring plan is given according to the stage it is in; otherwise, the process ends.

[0110] Specifically, after obtaining the device state membership and fault impact weights, the key to ensuring robust device health assessment results is to scientifically aggregate the fault state information and comprehensively and objectively measure the impact of the benefit and cost criteria on the overall device behavior. The Ratio system method is used to analyze and obtain device health indicators.

[0111]

[0112] Among them, HI is the health of the equipment, z j is the state membership of the j-th fault, g is the Benefit criterion number, is the Non-Benefit criterion number, where all fault states are Non-Benefit criterion, j is the fault state number, is the number of predicted failure states, is the optimal impact weight of the j-th fault state.

[0113] After obtaining the equipment health index, the committee experts divided the health status of wind turbine gearboxes into five states: healthy state, sub-healthy state, unhealthy state, emergency state and dangerous state. The health indicators of the equipment without any faults, the equipment with a single non-serious fault, the equipment with a single serious fault, the equipment with one non-serious fault and one serious fault at the same time, and the equipment with two serious faults at the same time are used as the state division boundaries.

[0114] Among them, when the fault weight ranking is in the top 50%, the fault type is considered a serious fault, otherwise it is considered a non-serious fault. According to five different statuses, equipment maintenance and post-event monitoring suggestions are given in terms of maintenance level, maintenance method and maintenance sequence:

[0115] Healthy status: Adopt a post-maintenance strategy at the grassroots level and do not take any maintenance measures. Continuous status monitoring of the equipment is sufficient.

[0116] Sub-health: A grassroots-level, condition-based maintenance strategy is implemented, with on-site maintenance actions based on the results of the fault health assessment. For example, if the rolling elements of a bearing are slightly worn, lubrication can be added without stopping the machine. However, if the bearing is severely worn, the machine must be stopped to replace the smallest replacement unit. Continuous condition monitoring of the equipment is then performed.

[0117] Unhealthy status: A relay-level condition-based maintenance strategy is implemented. The repair shop determines which faulty parts require maintenance or scrapping based on the health assessment results. The equipment is shut down for maintenance or replacement. The order of repair is determined by the impact weight of the fault. Continuous condition monitoring is then performed on the equipment.

[0118] Emergency: A relay-level, condition-based maintenance strategy is implemented. The repair shop scraps the faulty component based on the health assessment results, shuts down the machine, and replaces it with a new one. The order of component repairs is determined by the impact weight of the fault. A scheduled maintenance strategy is then implemented for the equipment.

[0119] Dangerous status: Base-level maintenance strategy based on the situation, the entire equipment returned to the factory is shut down for maintenance to determine whether to replace parts or scrap the entire machine.

[0120] In order to further explain the present invention in detail, wind turbine gearbox fault simulation experimental data is used to verify the present method. The wind turbine gearbox fault simulation test bench consists of two ABB mv1008-225 motors (1.2kW), a gearbox, a flywheel, a data acquisition instrument and a computer. One of them is used as a prime mover to drive a multi-stage gearbox, and the other is used as an asynchronous generator to simulate various resistance torques and is connected to the driver through a coupling. In the experiment, the input speed of the device is set to 1400 rpm, and the speeds of the two meshing gear sets in the gearbox are 1184 rpm and 840 rpm respectively. By changing the output of the frequency converter provided to the servo motor drive, vibration signals under different loads are obtained.

[0121] The health of wind turbine gearboxes needs to be evaluated under fifteen fault conditions: one healthy state (H1), six single fault conditions (IFS1–IFS6), and eight compound fault conditions (CFS1–CFS8). The specific fault types and corresponding labels are shown in Table 2. The six common single fault conditions involved in this experiment include broken teeth, gear shedding, gear cracks, coupling looseness, bearing rolling element wear, and bearing outer ring wear.

[0122] Table 2

[0123]

[0124]

[0125] The gearbox vibration signal is acquired by an NI-cDAQ-9174 / 9234 vibration sensor with a sampling frequency of 10240 Hz. The sampling duration is set to 100 s, which means that 102,400 data points are obtained for each fault state (i.e., one sensor × 100 s × 10240 Hz). The data points for each fault state are divided into 510 samples, for a total of 7650 samples, which are each split into a 60% training set, a 10% validation set, and a 30% test set.

[0126] In order to further verify the effectiveness of the MsXGB equipment state membership evaluation model, this paper adopts three commonly used algorithm adaptation methods, including Dual-Extreme Learning Machines (Dual-ELM), Multi-label radial basis function (ML-RBF) and Multi-label k-nearest neighbor (ML-kNN), which are labeled as Method 1 to Method 3. The same training samples were used to eliminate the influence of sample selection on the state evaluation performance, and 25 repeatability experiments were performed. The composite fault diagnosis results of the three comparison methods and the MsXGB method are shown in Table 3. It can be seen from the table that the fault diagnosis accuracy of the present invention is significantly higher than that of the other three methods. It also further proves that the present invention has strong creativity and applicability and can be applied to actual industry.

[0127] Table 3

[0128]

[0129]

[0130] Table 4 shows the state membership and fault count for each fault scenario obtained when the test set is input into the MsXGB model. Table 4 shows that the majority of samples corresponding to the 15 fault scenarios can be correctly identified. Combining the fault count predicted by Count XGB with the fault state membership calculated by the Locate XGB model, device health indicators can be accurately assessed.

[0131] Table 4

[0132]

[0133] The GKM method is used to combine the selection preferences and authority of each expert into an expert group knowledge measurement matrix, and finally the weight of the impact of each single fault on equipment behavior is determined. Table 5 shows the group knowledge measurement results when the chief expert's self-confidence coefficient Ω = 1.0.

[0134] Table 5

[0135]

[0136]

[0137] By combining fault state membership with impact weights using the Ratio system method, the impact of a single fault on equipment behavior and the overall health of the equipment under different fault conditions are determined, and the equipment health is classified into different states. Specifically, based on the Pythagorean fuzzy evaluation set provided by the committee's experts, the proposed GKM method quantifies the authority of each expert, constructing a comprehensive and robust fuzzy evaluation system for fault impact, and determining the impact weights and weight rankings of different fault types on equipment behavior. Fault types ranked in the top 50% are considered critical, while those ranked in the top 50% are considered non-critical.

[0138] like Figure 4 Then, according to the five different states, the system provides suggestions for equipment maintenance and post-event monitoring in terms of maintenance level, maintenance method and maintenance sequence, and adopts corresponding maintenance and post-event monitoring strategies for equipment in each state.

[0139] The wind turbine gearbox health assessment method proposed in this paper uses the MsXGB model in conjunction with the CountXGB and LocateXGB modules to obtain the device status membership. The GKM model then evaluates the degree of fault impact on the device. A health index is then derived based on the Ratio system method, combining the device status membership and the fault impact weight. This provides a comprehensive quantitative assessment of the wind turbine gearbox's health. This method measures the impact of different fault conditions on device behavior at a finer granularity, accurately reflecting the device's current health status and providing corresponding maintenance and post-event monitoring strategies, improving maintenance efficiency and flexibility.

[0140] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A wind turbine gearbox health intelligent assessment method, characterized in that: The method comprises the following steps: S1 presets and simulates the fault conditions of wind turbine gears, collects vibration signals generated by the operation of wind turbine gears under various fault types, and uses the collected vibration signals to build and train the MsXGB model; The model is used to evaluate the device status membership; S2 establishes a wind turbine gearbox fault impact assessment committee. Experts on the committee assess the impact of faults on the equipment and convert the assessment into a Pythagorean fuzzy evaluation set. Based on the evaluation set, the authority of the committee experts is quantified. A fault impact fuzzy evaluation system is constructed and used to determine the impact weight of each fault. S3 constructs a device health index relationship using the device state membership and the fault impact weight, calculates the device health index, divides the device into corresponding health states based on the health index, and performs corresponding processing on the device based on the health state, thereby realizing intelligent evaluation of the health of the wind turbine gearbox; In step S2, the construction of the fault impact fuzzy evaluation system is carried out according to the following steps: S24 For any j-th fault, construct and calculate the graph fuzzy set of the fault importance; S25 calculates the knowledge metric of the graph fuzzy set of the j-th fault; S26 constructs a relationship formula for the fault impact weight using the knowledge metric, and calculates the impact weight of each fault based on the relationship formula; In step S26, the relationship between the fault impact weight is as follows: in, is the optimal impact weight of the jth fault state, j is the fault state number, C is the number of fault states, K(PF j ) is PF j Knowledge measurement results; In step S3, the device health index is calculated according to the following relationship: Among them, HI is the health of the equipment, z j is the state membership of the j-th fault, g is the Benefit criterion number, is the Non-Benefit criterion number, where all fault states are Non-Benefit criterion, j is the fault state number, is the number of predicted failure states, is the optimal impact weight of the j-th fault state.

2. The method for intelligently evaluating the health of a wind turbine gearbox according to claim 1, wherein: The MsXGB model includes a Count XGB model and a Locate XGB model. The Count XGB model is used to predict the number of faults, and the Locate XGB model is used to predict the probabilities corresponding to various faults.

3. The method for intelligently evaluating the health of a wind turbine gearbox according to claim 2, wherein: The evaluation of the membership degree includes the following steps: S11 extracts the time domain and frequency domain features of the vibration signal, takes the number of faults as output, uses the time domain and frequency domain features of the vibration signal as input to train the Count XGB model, and calculates the number of faults using the Count XGB model; S12 uses the time domain and frequency domain features of the vibration signal as input and the fault state as output to train the Locate XGB model, and modifies the Softmax function in the Locate XGB model into a Softprob function so that its output is the fault probability of the corresponding fault state, that is, the fault state membership. The fault state membership is calculated using the Locate XGB model; S13 combines the fault number obtained by the Count XGB model with the fault status membership result obtained by the Locate XGB model to obtain the device status membership.

4. The method for intelligently evaluating the health of a wind turbine gearbox according to claim 3, wherein: The device status membership is calculated according to the following relationship: Where p is the sample number, the sample is sensor data, p = 1, 2, ..., P, P is the number of samples, f p is the time domain and frequency domain characteristics of the vibration signal, MEMB(f p ) is the device status membership, is the predicted number of failures after the tth iteration of the Count XGB model, Z c (f p ) is the fault probability distribution, c is the fault state number, C is the number of fault states, is the output of the Locate XGB model before Softprob for fault state c.

5. A wind turbine gearbox health intelligent assessment method according to claim 1 or 2, characterized in that: In step S2, quantifying the authority level of the experts in the committee based on the evaluation set includes the following steps: S21 The chief expert determines the best expert and the worst expert based on the Pythagorean fuzzy evaluation set; S22 uses the BWM method to give the fuzzy Best-to-Others matrix and Others-to-Worst matrix of expert professionalism level; S23 uses the BWM maximum value model to calculate the authority level of experts in the expert group.

6. The method for intelligently evaluating the health of a wind turbine gearbox according to claim 5, wherein: In step S23, the maximum value model is performed according to the following relationship: minε l k ≥0 Among them, α is the possibility level predefined by the decision maker, ε represents the optimal consistency level of fuzzy preferences, and λ k is the weight of the k-th expert, λ B is the most authoritative expert weight, λ W is the weight of the least authoritative expert, and The best to others matrix of the kth expert authority fuzzy preference given by the chief expert The median and upper boundary values ​​of opinions in , and The others to worst matrix of the kth expert authority fuzzy preference given by the chief expert The median and lower bound values ​​of opinions.

7. The method for intelligently evaluating the health of a wind turbine gearbox according to claim 1, wherein: The graph fuzzy set of fault importance is performed as follows: Among them, PF j is the voting picture fuzzy matrix of the chief expert and assistant experts on the severity of the j-th fault state, λ *VF ,λ *KN ,λ *VA ,λ *Rov The voting weights of experts who choose "yes", "neutral", "no" and "abstain" are respectively, R(PF j ) is the vote of the chief expert and assistant experts on the severity of the j-th fault state Picture fuzzy matrix PF j The voting hesitation degree is , VF, KN, VA and Rov are the options of approval, neutrality, opposition and abstention for the question "the jth fault state is a serious fault", a and A are the expert number and number of experts who choose the approval option, b and B are the expert number and number of experts who choose the neutral option, d and D are the expert number and total number of experts who choose the opposition option, e and E are the expert number and total number of experts who choose the abstention option, Ω C.E The self-confidence coefficient given by the chief expert; The knowledge measurement is performed as follows: K(PF j )=1-0.5(E(PF j )+R(PF j )) Among them, K(PF j ) is PF j The knowledge measurement results of E(PF j ) is PF j The information entropy value, d n (PF j ,PF min ) is PF j and PF min The Hamming distance between n (PF j ,PF max ) is PF j and PF max Hamming distance between max =(1,0,0), PF min =(0,0,1).

Images