Fan gear box fault diagnosis and early warning system based on fusion of oil and vibration parameters

Abstract

The invention discloses a fan gearbox fault diagnosis and early warning system based on oil and vibration parameter fusion, and the system comprises an oil parameter and vibration signal collection module which is used for collecting oil parameters and vibration signals of a fan gearbox in real time through an oil sensor and a vibration sensor; the data fusion optimization module is used for optimizing the fusion weight of the oil and vibration data by adopting a firefly algorithm; the feature extraction and subset selection module is used for performing feature extraction and optimization on the fused oil parameters and vibration signals; the fault diagnosis module is used for inputting the optimized feature vector into a fault diagnosis model; the fault early warning generation module is used for generating a fault early warning signal in real time according to the output of the fault diagnosis model; and the closed-loop updating module is used for forming a closed-loop optimization-diagnosis-feedback iteration mechanism. According to the invention, fault diagnosis and early warning of the fan gearbox based on oil and vibration parameter fusion are realized for users.

Description

Technical Field

[0001] The present invention relates to the technical field of fan gearbox fault and early warning diagnosis, and particularly to a fan gearbox fault diagnosis and early warning system based on the fusion of oil fluid and vibration parameters. Background Art

[0002] With the continuous growth of the scale of wind power equipment and the continuous improvement of the installed capacity, as a typical high-load continuous operation equipment, the stability and reliability of the operation state of wind turbines have attracted increasing attention in the industry. As the core transmission component connecting the wind wheel and the generator, the operation state of the fan gearbox directly affects the energy efficiency output and safety level of the fan. Due to the long-term operation of the gearbox under complex loads and high-speed meshing conditions, typical fault types such as tooth surface wear, bearing damage, abnormal lubrication, and oil fluid pollution are likely to occur. Once a serious fault occurs in the gearbox, it will not only lead to the shutdown and maintenance of the unit, but also may cause significant economic losses and energy consumption. Therefore, carrying out research on high-precision and intelligent fault diagnosis and early warning technologies for gearboxes has become an important topic in the field of wind power operation and maintenance.

[0003] At present, conventional fan gearbox fault diagnosis methods mainly rely on vibration signal analysis, and some systems combine oil fluid analysis data for auxiliary judgment. In such methods, the data-driven method based on vibration signals has the advantages of high sensitivity and fast response. High-frequency fault information is widely extracted by acceleration sensors, and early fault identification is achieved by combining means such as spectrum analysis and envelope demodulation. However, when facing multi-source noise interference, complex load conditions, and early weak damage, this method is prone to problems such as characteristic signal masking and fuzzy fault classification, and there is a certain risk of false alarms and missed alarms. On the other hand, as an important means to reflect the internal wear state of the gearbox, oil fluid monitoring can provide information such as temperature, pressure, viscosity, and metal particle concentration, and has an advantage in directly diagnosing problems such as poor lubrication and foreign object intrusion. However, its data changes relatively gently and the timeliness is insufficient, making it difficult to independently support the real-time response ability to sudden faults.

[0004] Some existing studies have begun to attempt to combine vibration analysis with oil monitoring and propose a multi-source fusion diagnosis scheme. For example, some systems fuse and input the two types of data by setting weights according to rules, or perform joint modeling after manually extracting features. However, there are three core problems with such methods: First, the weight allocation in the data fusion process often depends on empirical settings and lacks an adaptive optimization mechanism, which is likely to cause one type of information to be overly dominant and ignore another key feature; Second, the feature selection lacks intelligent optimization, resulting in high input redundancy, large computational complexity, and limited diagnostic accuracy; Third, the output results of the fault diagnosis model are mostly static classification labels, lacking a dynamic feedback mechanism linked to the actual operating state of the equipment, and it is difficult to meet the adaptive early warning requirements of the fan during long-term operation. Especially in the actual scenario where the wind farm environment is complex and the climate is changeable, the existing systems often cannot accurately capture abnormal trends or dynamically adjust the diagnostic strategy, thus reducing the overall intelligent operation and maintenance level.

[0005] In addition, in terms of the application of optimization algorithms, existing technologies mostly use general intelligent optimization methods such as traditional particle swarm and genetic algorithms to search the feature space. These algorithms perform stably in a static environment, but when faced with the nonlinear, multi-fluctuation, and multi-temporal characteristics of the fan gearbox operation data, problems such as slow convergence, low accuracy, and getting stuck in local optima often occur. On the other hand, existing research on fault early warning mostly stays at the level of setting fixed thresholds and triggering alarms based on single-point indicators, lacking a deep mechanism that integrates multi-dimensional weights, dynamic score calculation, and confidence factor evaluation, and it is difficult to implement an interpretable and controllable hierarchical early warning system.

[0006] Therefore, how to provide a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil and vibration parameters is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil and vibration parameters. The present invention (wait a moment to expand the specification abstract).

[0008] The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil and vibration parameters according to an embodiment of the present invention includes:

[0009] An oil parameter and vibration signal acquisition module, configured to collect the oil parameters and vibration signals of the fan gearbox in real time through an oil sensor and a vibration sensor;

[0010] A data fusion optimization module, configured to optimize the fusion weights of the oil and vibration data by using a firefly algorithm;

[0011] A feature extraction and subset selection module, configured to perform feature extraction and optimization on the fused oil parameters and vibration signals;

[0012] A fault diagnosis module, which is used to input the optimized feature vector into a fault diagnosis model;

[0013] A fault warning generation module, which is used to generate a fault warning signal in real time according to the output of the fault diagnosis model;

[0014] A closed-loop update module, which is used to form a closed-loop optimization-diagnosis-feedback iteration mechanism;

[0015] Optionally, the modules are implemented by the following methods:

[0016] S1. The oil parameters of the fan gearbox are collected in real time through an oil sensor, and the vibration signals of the fan gearbox are collected in real time through a vibration sensor, and the collected oil parameters and vibration signals are preprocessed;

[0017] S2. The firefly algorithm is used to optimize the fusion weights of the oil parameters and vibration signals, and the influence of the data source on the final diagnosis result is adjusted;

[0018] S3. The bat algorithm is used to extract and optimize the features of the fused oil parameters and vibration signals, and the optimal feature subset is selected;

[0019] S4. The feature vectors of the oil and vibration data optimized by the firefly algorithm and the bat algorithm are input into the fault diagnosis model to generate an optimized feature vector, and the fault type is identified;

[0020] S5. According to the output of the fault diagnosis model, a fault warning signal is generated in real time, and fault information and maintenance suggestions are provided to the operator;

[0021] S6. Continuously monitor the oil and vibration data of the fan gearbox, generate real-time monitoring data, and dynamically adjust the warning threshold according to the fault warning signal;

[0022] S7. According to the change of the real-time monitoring data, adjust the optimization strategies of the firefly algorithm and the bat algorithm to optimize the fusion and feature extraction processes.

[0023] Optionally, the oil parameters specifically include the oil temperature, pressure, pollutant concentration and viscosity of the fan gearbox, and the vibration signals specifically include vibration frequency, vibration amplitude and vibration acceleration;

[0024] Optionally, S1 includes the following specific steps:

[0025] S11. Preprocess the collected oil parameters, use median filtering to denoise the oil parameters, filter out the outliers caused by sensor noise, use Kalman filtering to smooth the oil temperature and pressure parameters, and use linear interpolation to fill in the missing oil parameters;

[0026] S12. Preprocess the collected vibration signals. Filter the vibration signals using a band-pass filter to remove low-frequency and high-frequency noises, retain the effective vibration information within the frequency range, extract the instantaneous vibration amplitude using Hilbert transform, and perform detrending processing. Use the IQR-based outlier detection method to detect and remove abnormal vibration data points.

[0027] Optionally, S2 includes the following specific steps:

[0028] S21. Optimize the fusion weights of the oil parameters and vibration signals through the firefly algorithm, calculate the relative contribution degrees of the oil parameters and vibration signals in fault diagnosis, and ensure that the sum of all weights is 1 and each weight value is within the range of [0, 1] during the optimization process;

[0029] S22. Initialize the weights and positions of the firefly individuals, set the initial brightness and attractiveness. The brightness represents the fitness of the individual, and the attractiveness is related to the distance from the optimization target and the fitness evaluation. Define the fitness function of the brightness as evaluating the weighted eigenvalue of the oil parameters and vibration signals, and calculate the fitness of each individual:

[0030]

[0031] where, F i is the brightness of the i-th firefly, W i is the fusion weight of the i-th oil parameter or vibration signal, x i is the normalized eigenvalue of the i-th oil parameter or vibration signal, α i is the adjustment factor of the i-th oil parameter or vibration signal, β is the balance adjustment factor, n is the number of oil parameters and vibration signals, is the error used to measure the weight and normalization process, i is the index number of the oil parameters and vibration signals, (W i ·x i ) 2 is the square operation on W i and x i ;

[0032] S23. According to the update rule of the firefly algorithm, perform iterative optimization. In each iteration, update the weights and positions of the oil and vibration signals;

[0033] S24. Through multiple iterations, the firefly algorithm will update the fusion weights of the oil and vibration data until the maximum number of iterations N max is reached, generating optimized fusion weights that can achieve an optimal balance between the oil parameters and vibration signals;

[0034] S25. After each iteration, monitor the convergence of the algorithm by evaluating the change of the objective function, and adjust the learning rate and the weight update step size to ensure the stability and convergence speed of the optimization process.

[0035] Optionally, the S3 includes the following specific steps:

[0036] S31. Extract features from the fused oil parameter and vibration signal data, optimize the features using the bat algorithm, initialize the feature positions and velocities of the bat individuals, and set the initial brightness and frequency. The brightness represents the fitness of the bat, and the frequency is used to control the flight speed and search step size of the bat. The feature positions and velocities are used to update the search process of the features. Define a new fitness function as:

[0037]

[0038] where, is the brightness of the i-th bat, i is the index number of the oil parameter and vibration signal, λ j is the adjustment factor of the j-th feature, W j is the weight coefficient of the j-th feature, x j is the normalized value of the j-th feature, p is the exponent controlling the influence of each feature, j is the feature number, γ is the balance factor, m is the total number of features, is to measure the difference between the sum of all feature weights and 1, |W j ·x j | p is the absolute value of the weighted normalized eigenvalue W j ·x j and perform a power operation;

[0039] S33. Perform iterative optimization through the flight update rule of the bat algorithm. In each iteration, update the feature positions and velocities of the oil parameter and vibration signal;

[0040] S34. Update the feature set and feature weights through multiple iterations until the maximum number of iterations N max is reached, and generate the optimal feature subset and feature weights.

[0041] Optionally, the S4 includes the following specific steps:

[0042] S41. Input the feature vectors of the oil parameter and vibration data optimized by the firefly algorithm and the bat algorithm into the fault diagnosis model, use the decision tree algorithm to identify the fault type, and construct a classifier model according to the training set to minimize the objective function:

[0043]

[0044] where, is the parameter obtained by solving the optimization algorithm, θ is the parameter set to be optimized, n is the number of oil parameters and vibration signals, m1 is the number of fault types, j is the index of the fault type, i is the index of the oil parameter and vibration signal number, W j is the weight of the j-th type of fault, X i is the feature vector of the i-th sample, μ j is the mean vector of the j-th type of fault, is the brightness of the i-th bat, F i is the brightness of the i-th firefly, is the regularization factor, (X i -μ j ) 2 is the feature vector X of the i-th sample i and the mean vector μ of the fault type j j The square difference between them, is the brightness in the bat algorithm and the brightness F in the firefly algorithm i The square difference between them;

[0045] S42. According to the optimized feature vector, each feature is input into the fault diagnosis model, and the conditional probability of the fault type is calculated. It will process the input feature vector according to the parameters obtained during the training process, evaluate the probability of each fault type occurring, evaluate the contribution of each feature to the fault type, and calculate the probability value for each fault type. The output result is the posterior probability distribution of the fault type under the given features, and the most likely fault type is determined.

[0046] Optionally, the S5 includes the following specific steps:

[0047] S51. According to the parameters output by the fault diagnosis model, a fault warning score is generated in real time. If Ψ j >T threshold , it is determined that the warning of the j-th type of fault is triggered:

[0048]

[0049] Among them, Ψ j is the comprehensive warning score of the j-th type of fault, j is the index of the fault type, n is the total number of features, i is the index of the i-th input feature, ω ij is the weighting coefficient of the feature i for the fault type j, θ i is the optimal parameter vector The parameter weight corresponding to the i-th feature in, is the brightness of the i-th bat, F i is the brightness of the i-th firefly;

[0050] S52. Comprehensive warning score Ψ for the j-th type of fault j Perform confidence evaluation. Based on the parameters participating in the scoring, comprehensively analyze the credibility of the j-th type of fault score. If the values corresponding to most high-weight features are significantly higher than the remaining features and the brightness difference is stable, it indicates that the recognition result of the fault has high consistency and reliability, and a confidence factor is generated;

[0051] S53. According to the warning score and the confidence factor, the system automatically sets the severity level of the fault, and determines the response time window and recommended maintenance priority according to the level;

[0052] S54. Generate a fault diagnosis report according to the scoring result. The report includes the fault type number, comprehensive score, confidence factor, severity level, and recommended disposal plan, and push the report to the wind turbine remote monitoring platform.

[0053] Optionally, the S6 includes the following specific steps:

[0054] S61. The system continuously collects the oil fluid parameters and vibration signals during the operation of the wind turbine gearbox, constructs a time series data containing multi-dimensional features, and constructs a monitoring matrix that changes over time. Each column is a feature vector at a moment;

[0055] S62. Input the real-time monitoring matrix into the fault diagnosis model, and conduct a comprehensive analysis of the current operating state in combination with the optimal parameter vector, optimized feature vector, and comprehensive warning score of the fault;

[0056] S63. Calculate the average fluctuation degree of all features in time according to the change amplitude between the optimized feature vector and the feature vector at the previous moment, and use it as the current operating state fluctuation coefficient;

[0057] S64. Dynamically adjust the warning threshold corresponding to the current fault type according to the current operating state fluctuation coefficient and the comprehensive warning score of the fault in the previous cycle. When the operating state fluctuates greatly, automatically increase the warning sensitivity and correspondingly lower the warning threshold. When the operating state fluctuates little, correspondingly increase the threshold;

[0058] S65. If the system detects that the operating state fluctuations continuously increase in multiple consecutive time periods and the fault score always exceeds the current warning threshold, automatically trigger a high-priority warning mechanism and fix the current warning threshold of the fault at the minimum acceptable value;

[0059] S66. Store the current optimized feature vector, comprehensive warning score of the fault, optimized feature vector and comprehensive warning score of the fault, and warning threshold in the database in each sampling period.

[0060] Optionally, the S7 includes the following specific steps:

[0061] S71. During the operation of the system, based on the optimal parameter vector, optimized feature vector, comprehensive early warning score of faults, firefly brightness, and bat brightness collected at the current moment, a continuous monitoring data window is constructed to track and evaluate the performance in different time periods;

[0062] S72. When the change amplitude of the fusion weight and brightness difference detected in multiple consecutive cycles is lower than the set convergence determination value, it is determined that the current optimization process has tended to be stable, and then the system switches to the low-frequency update mode;

[0063] S73. If it is detected that the eigenvalue fluctuation increases significantly, the early warning score continuously exceeds the early warning threshold, or the output parameters of the fault diagnosis model show a divergent trend, the system automatically triggers the optimization strategy reconstruction mechanism to dynamically adjust the core control parameters of the firefly algorithm and the bat algorithm;

[0064] S74. For the firefly algorithm, the system adjusts the attraction parameter and the individual movement step size according to the average fluctuation amplitude of the optimized oil fluid parameters and vibration data feature vectors of the firefly algorithm and the bat algorithm in the current sampling cycle. When the system detects a significant increase in the eigenvalue change, it automatically increases the attraction parameter, making the individuals with higher brightness in the algorithm have a stronger attraction to the remaining individuals;

[0065] S75. For the bat algorithm, the system dynamically adjusts the frequency adjustment parameter and the local perturbation amplitude according to the confidence factor. When the confidence factor is high, it indicates that the recognition tends to converge, and the system appropriately reduces the frequency adjustment range of the bat individuals and narrows the search perturbation amplitude. If the confidence factor decreases, the reverse operation is performed to expand the search frequency range;

[0066] S76. After completing the dynamic update of the firefly algorithm and the bat algorithm, restart the fusion weight optimization and feature extraction process to generate a new round of fusion feature vectors and input them into the fault diagnosis model to form a closed-loop optimization-diagnosis-feedback iteration mechanism.

[0067] The beneficial effects of the present invention are:

[0068] By constructing a fault diagnosis and early warning system for a wind turbine gearbox based on the fusion of oil fluid and vibration parameters, the present invention realizes the deep fusion of multi-source monitoring data, the accurate improvement of fault identification, and the dynamic optimization of the early warning mechanism, overcoming the deficiencies of the prior art in aspects such as insufficient extraction of fault features, inflexible response, and unintelligent early warning mechanism. Compared with the single monitoring method that only relies on vibration signals or oil fluid analysis, the present invention structurally integrates two types of data sources and adaptively optimizes their fusion weights through the firefly algorithm, effectively solving the problem that the weight setting in information fusion depends on experience and does not have the ability of dynamic adjustment, enabling the optimal information allocation of diagnostic input features under different operating states.

[0069] In addition, the present invention further introduces the bat algorithm to finely screen and subset optimize the fused features, which not only improves the representativeness of feature expression but also significantly reduces the interference of redundant features on the training process, thereby reducing the complexity of the fault diagnosis model and improving the generalization ability. In the diagnostic modeling link, the system inputs the optimized feature vector into the decision tree model for fault classification and outputs parameters to construct a comprehensive scoring function associated with the brightness difference of the firefly and bat algorithms, thus realizing the direct coupling between the early warning mechanism and the diagnostic result. Compared with the traditional method triggered by a fixed threshold, this scoring mechanism can not only reflect the changes at the feature level but also has the ability to dynamically adjust the early warning level according to the scoring change trend, further enhancing the system's perception and response to early risks.

[0070] During the operation of the system, the present invention constructs a real-time monitoring mechanism to continuously track the time-series changes of oil fluid and vibration data, and dynamically adjusts the early warning threshold according to information such as the fluctuation trend of the fault score and the feature response volatility. Such a dynamic adjustment mechanism breaks the limitation of the weak adaptability of the static model to complex scenarios, enabling the system to maintain the balance between diagnostic sensitivity and robustness during long-term operation. In addition, the system also has the ability to adaptively adjust the optimization strategy. By continuously tracking the optimization effect and feedback control algorithm parameters, it ensures that the firefly algorithm and the bat algorithm always maintain the optimal search ability at different stages, avoiding problems such as falling into local optima or excessive perturbation.

[0071] In summary, the system proposed by the present invention has achieved remarkable improvements in the accuracy of fault diagnosis, the real-time performance of early warning response, and the adaptive ability of the optimization algorithm, realizing the technological leap from "passive diagnosis" to "active perception and dynamic optimization", which not only improves the operating safety level of the wind turbine gearbox but also provides more stable and efficient technical support for the intelligent operation and maintenance of the wind farm. Brief Description of the Drawings

[0072] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the accompanying drawings:

[0073] Figure 1 is a flowchart of a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters proposed by the present invention;

[0074] Figure 2 is a schematic diagram of a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters proposed by the present invention;

[0075] Figure 3 is a data flow diagram of a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters proposed by the present invention. Detailed implementation manners

[0076] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0077] Refer to Figures 1 - 3 , a fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters, includes:

[0078] An oil fluid parameter and vibration signal acquisition module, used to collect the oil fluid parameters and vibration signals of the fan gearbox in real time through an oil fluid sensor and a vibration sensor;

[0079] A data fusion optimization module, used to optimize the fusion weights of the oil fluid and vibration data by using the firefly algorithm;

[0080] A feature extraction and subset selection module, used to extract and optimize the features of the fused oil fluid parameters and vibration signals;

[0081] A fault diagnosis module, used to input the optimized feature vector into the fault diagnosis model;

[0082] A fault early warning generation module, used to generate a fault early warning signal in real time according to the output of the fault diagnosis model;

[0083] A closed-loop update module, used to form a closed-loop optimization-diagnosis-feedback iteration mechanism;

[0084] The present invention constructs a full - process modular system including oil - fluid and vibration signal acquisition, fusion weight optimization, feature extraction, fault diagnosis, and closed - loop feedback update. The firefly algorithm is used to dynamically optimize the fusion weights of oil - fluid and vibration parameters, improving the adaptability of multi - source feature fusion. By performing feature subset selection on the fused data, the input dimension redundancy is reduced, enhancing the robustness and diagnostic accuracy. The key parameters output by the fault diagnosis module participate in subsequent early - warning scoring and feedback update, realizing closed - loop control between optimization, diagnosis, and early - warning, and ensuring that the system has the ability of continuous learning and adaptive optimization under different operating states.

[0085] In this embodiment, the modules are implemented through the following methods:

[0086] S1. The oil - fluid parameters of the fan gearbox are collected in real - time through an oil - fluid sensor, and the vibration signals of the fan gearbox are collected in real - time through a vibration sensor, and the collected oil - fluid parameters and vibration signals are pre - processed.

[0087] S2. The firefly algorithm is used to optimize the fusion weights of the oil - fluid parameters and vibration signals, adjusting the influence of data sources on the final diagnosis result.

[0088] S3. The bat algorithm is used to extract and optimize the features of the fused oil - fluid parameters and vibration signals, and select the optimal feature subset.

[0089] S4. The feature vectors of the oil - fluid and vibration data optimized by the firefly algorithm and the bat algorithm are input into the fault diagnosis model to generate optimized feature vectors and identify the fault types.

[0090] S5. According to the output of the fault diagnosis model, a fault early - warning signal is generated in real - time, and fault information and maintenance suggestions are provided to the operator.

[0091] S6. Continuously monitor the oil - fluid and vibration data of the fan gearbox, generate real - time monitoring data, and dynamically adjust the early - warning threshold according to the fault early - warning signal.

[0092] S7. According to the changes in the real - time monitoring data, adjust the optimization strategies of the firefly algorithm and the bat algorithm.

[0093] The present invention adopts an intelligent diagnosis mechanism combining fusion optimization and feature screening. The firefly algorithm is used to dynamically adjust the fusion weights of oil - fluid and vibration data, enhancing the collaborative expression ability of multi - source information for fault features. The bat algorithm is used to extract the optimal feature subset, constructing an efficient and low - redundancy diagnostic input vector, improving the accuracy and response speed of fault identification. The system continuously monitors the operating data and dynamically adjusts the early - warning threshold and optimization strategy in combination with the early - warning results, realizing closed - loop control of fault diagnosis - early - warning - feedback, and ensuring that the system has stability, self - adaptability, and long - term learning ability under complex working conditions.

[0094] In this embodiment, the oil fluid parameters include the oil fluid temperature, pressure, pollutant concentration, and viscosity of the fan gearbox, and the vibration signals specifically include vibration frequency, vibration amplitude, and vibration acceleration.

[0095] The present invention constructs a multi-parameter fusion feature space by collecting multi-dimensional oil fluid parameters such as the oil fluid temperature, pressure, pollutant concentration, and viscosity of the fan gearbox, and combining vibration signals such as vibration frequency, vibration amplitude, and acceleration. By comprehensively reflecting mechanical wear and operating dynamic states, it realizes a comprehensive perception of the health status of the gearbox and improves the accuracy of fault diagnosis and early warning ability.

[0096] In this embodiment, S2 includes the following specific steps:

[0097] S21. Optimize the fusion weights of the oil fluid parameters and vibration signals through the firefly algorithm, calculate the relative contribution degrees of the oil fluid parameters and vibration signals in fault diagnosis, and ensure that the sum of all weights is 1 and each weight value is within the range of [0, 1] during the optimization process;

[0098] S22. Initialize the weights and positions of the firefly individuals, set the initial brightness and attractiveness. The brightness represents the fitness of the individual, and the attractiveness is related to the distance from the optimization target and the fitness evaluation. Define the fitness function of the brightness as evaluating the weighted eigenvalue of the oil fluid parameters and vibration signals, and calculate the fitness of each individual:

[0099]

[0100] Among them, F i is the brightness of the i-th firefly, W i is the fusion weight of the i-th oil fluid parameter or vibration signal, x i is the normalized eigenvalue of the i-th oil fluid parameter or vibration signal, α i is the adjustment factor of the i-th oil fluid parameter or vibration signal, β is the balance adjustment factor, n is the number of oil fluid parameters and vibration signals, is used to measure the error of the weight and normalization process, i is the index number of the oil fluid parameters and vibration signals, (W i ·x i ) 2 represents the square operation on W i and x i ;

[0101] S23. According to the update rule of the firefly algorithm, perform iterative optimization. In each iteration, update the weights and positions of the oil fluid and vibration signals;

[0102] S24. Through multiple iterations, the firefly algorithm will update the fusion weights of the oil fluid and vibration data until the maximum number of iterations N is reachedmax , generate optimized fusion weights, which can achieve an optimal balance between oil parameters and vibration signals;

[0103] S25. After each iteration, monitor the convergence of the algorithm by evaluating the change of the objective function, and adjust the learning rate and weight update step size to ensure the stability and convergence speed of the optimization process.

[0104] The present invention constructs a multi-weight fusion optimization mechanism with the firefly algorithm as the core, and jointly models and adaptively optimizes the contribution degrees of oil parameters and vibration signals. The brightness function is used to comprehensively evaluate the fitness of each data source, and the weight vector and distance factor are combined to guide the individual to update its position, realizing the dynamic fusion of oil and vibration characteristics. By introducing a cooperative update mechanism of brightness difference, normalized distance and global optimal fitness, it is ensured that the individuals tend to the optimal solution during the optimization process. Set the maximum number of iterations, and dynamically adjust the update step size and weight parameters in combination with the change trend of the objective function, effectively improving the fusion accuracy and convergence efficiency, and enhancing the stability and sensitivity of the feature expression in the early stage of fault identification.

[0105] In this embodiment, S3 includes the following specific steps:

[0106] S31. Extract features from the fused oil parameter and vibration signal data, optimize the features using the bat algorithm, initialize the feature positions and velocities of bat individuals, and set the initial brightness and frequency. The brightness represents the fitness of the bat, and the frequency is used to control the flight speed and search step size of the bat. The feature positions and velocities are used to update the search process of the features. Define a new fitness function as:

[0107]

[0108] Wherein, is the brightness of the i-th bat, i is the number index of the oil parameter and vibration signal, λ j is the adjustment factor of the j-th feature, W j is the weight coefficient of the j-th feature, x j is the normalized value of the j-th feature, p is the exponent controlling the influence of each feature, j is the number of the feature, γ is the balance factor, m is the total number of features, is used to measure the difference between the sum of all feature weights and 1, |W j ·x j | p is the absolute value of the weighted normalized eigenvalue W j ·x j and perform a power operation;

[0109] S33. Perform iterative optimization through the flight update rule of the bat algorithm. In each iteration, update the feature positions and velocities of the oil parameter and vibration signal;

[0110] S34. Update the feature set and feature weights through multiple iterations until the maximum number of iterations N is reached max , and generate the optimal feature subset and feature weights.

[0111] The present invention uses the bat algorithm to optimize the features of the fused oil and vibration data, constructs a comprehensive fitness function based on brightness, frequency and feature normalization difference, and dynamically guides individuals to perform iterative search in the feature space. By jointly considering feature weights, position update and speed adjustment, the sensitivity of the feature extraction process to key fault information is improved. Set the maximum number of iterations and the local optimum judgment criterion to ensure that the feature set achieves a balance between global search and local refinement, effectively screen out a feature subset with strong representativeness and low redundancy, and improve the stability and recognition accuracy of the subsequent fault diagnosis model.

[0112] In this embodiment, S4 includes the following specific steps:

[0113] S41. Input the oil parameter and vibration data feature vectors optimized by the firefly algorithm and the bat algorithm into the fault diagnosis model, use the decision tree algorithm to identify the fault type, and construct a classifier model according to the training set to minimize the objective function:

[0114]

[0115] Wherein, is the parameter obtained by solving through the optimization algorithm, θ is the parameter set to be optimized, n is the number of oil parameters and vibration signals, m1 is the number of fault types, j is the index of the fault type, i is the number index of the oil parameter and vibration signal, W j is the weight of the jth type of fault, X i is the feature vector of the ith sample, μ j is the mean vector of the jth type of fault, is the brightness of the ith bat, F i is the brightness of the ith firefly, is the regularization factor, (X i -μ j ) 2 is the square difference between the feature vector X i of the ith sample and the mean vector μ j of the fault type j, is the brightness in the bat algorithm and the square difference between the brightness F i in the firefly algorithm;

[0116] S42. According to the optimized feature vectors, each feature is input into the fault diagnosis model, and the conditional probability of the fault type is calculated. The input feature vectors will be processed according to the parameters obtained during the training process to evaluate the probability of each fault type occurring, evaluate the contribution of each feature to the fault type, calculate the probability value for each fault type, and the output result is the posterior probability distribution of the fault type under the given features, and the most likely fault type is determined.

[0117] Based on the optimized oil fluid and vibration feature vectors, the present invention constructs an objective function that fuses the brightness differences of the bat algorithm and the glowworm algorithm, uses a decision tree model for fault type identification, optimizes the parameters to minimize the feature differences and the algorithm perception error, and improves the stability of the diagnosis model and the clarity of the classification boundary. By introducing the joint constraint of weight assignment and spatial distance, the discriminant ability for different fault features is strengthened. The output result is the posterior probability distribution of each type of fault, assisting the system to identify the most likely fault type and realizing the refined fault classification driven by multi-parameters and high-dimensional features.

[0118] In this embodiment, S5 includes the following specific steps:

[0119] S51. According to the parameters output by the fault diagnosis model, a fault warning score is generated in real time. If Ψ j >T threshold , it is determined that the warning of the j-th type of fault is triggered:

[0120]

[0121] Among them, Ψ j is the comprehensive warning score of the j-th type of fault, j is the index of the fault type, n is the total number of features, i is the index of the i-th input feature, ω ij is the weighting coefficient of feature i for fault type j, θ i is the optimal parameter vector the parameter weight corresponding to the i-th feature, is the brightness of the i-th bat, and F i is the brightness of the i-th glowworm;

[0122] S52. Conduct a confidence evaluation on the comprehensive warning score Ψ j of the j-th type of fault. Based on the parameters participating in the scoring, comprehensively analyze the credibility of the j-th type of fault score. If the θ i corresponding to most of the high-weight features is significantly higher than the remaining features and the brightness difference is stable, it indicates that the recognition result of the model for the fault has high consistency and reliability, and a confidence factor is generated;

[0123] S53. Automatically set the severity level of the fault according to the warning score and confidence factor, and determine the response time window and recommended maintenance priority according to the level;

[0124] S54. Generate a fault diagnosis report according to the scoring result. The report includes the fault type number, comprehensive score, confidence factor, severity level, and recommended disposal plan, and push the report to the wind turbine remote monitoring platform.

[0125] Based on the optimal parameter output of the fault diagnosis model, the present invention constructs a comprehensive warning scoring mechanism that combines the feature weighted coefficient and the brightness difference of the double algorithm to realize the accurate trigger determination of different types of faults. By estimating the confidence of the scoring result, the concentration of the feature score and the consistency of the response are evaluated to ensure the interpretability and stability of the warning result. The system can automatically divide the fault severity level according to the score value and confidence factor, dynamically generate the fault level label, and output the graphic warning report, realizing the closed-loop intelligent warning management from diagnosis, evaluation to level classification and remote push.

[0126] In this embodiment, S6 includes the following specific steps:

[0127] S61. Continuously collect the oil fluid parameters and vibration signals during the operation of the wind turbine gearbox, construct the time series data containing multiple-dimensional features, and construct the monitoring matrix that changes with time. Each column is a feature vector at a moment;

[0128] S62. Input the real-time monitoring matrix into the fault diagnosis model, and conduct a comprehensive analysis of the current operating state in combination with the optimal parameter vector, the optimized feature vector, and the comprehensive warning score of the fault;

[0129] S63. Calculate the average fluctuation degree of all features in time as the current operating state fluctuation coefficient according to the change amplitude between the optimized feature vector and the feature vector of the previous moment;

[0130] S64. Dynamically adjust the warning threshold corresponding to the current fault type according to the current operating state fluctuation coefficient and the comprehensive warning score of the fault in the previous cycle. When the operating state fluctuates greatly, automatically increase the warning sensitivity and correspondingly lower the warning threshold. When the operating state fluctuates slightly, correspondingly increase the threshold;

[0131] S65. If it is detected that the operating state fluctuation continues to increase in multiple consecutive time periods and the fault score always exceeds the current warning threshold, automatically trigger the high-priority warning mechanism and fix the current warning threshold of the fault as the minimum acceptable value;

[0132] S66. Store the current optimized feature vector, the comprehensive early warning score of the fault, the optimized feature vector and the comprehensive early warning score of the fault, and the early warning threshold in the database within each sampling period.

[0133] Based on the multi-source characteristic data of the oil and vibration of the fan gearbox, the present invention constructs a time series monitoring matrix, integrates the current characteristic fluctuations and historical change trends, forms a state scoring system, and captures fault symptoms in real time. The system dynamically evaluates abnormal trends by calculating the characteristic fluctuation amplitude and the average change rate; combines the relationship between the score and the threshold, and adaptively adjusts the early warning sensitivity and grade standard. If the continuous abnormal fluctuations exceed the threshold, the priority early warning mechanism is automatically triggered and the response window of the current fault type is locked, effectively improving the response speed and controllability of sudden faults, and realizing accurate and safe predictive operation and maintenance management.

[0134] In this embodiment, S7 includes the following specific steps:

[0135] S71. During operation, based on the optimal parameter vector, the optimized feature vector, the comprehensive early warning score of the fault, the brightness of fireflies, and the brightness of bats collected at the current moment, construct a continuous monitoring data window to track and evaluate the performance in different time periods;

[0136] S72. When the change amplitude of the fusion weight and the brightness difference is detected to be lower than the set convergence determination value in multiple consecutive periods, it is determined that the current optimization process has tended to be stable, and then switch to the low-frequency update mode;

[0137] S73. If it is detected that the characteristic value fluctuation increases significantly, the early warning score continuously exceeds the early warning threshold, or the output parameters of the fault diagnosis model show a divergent trend, the optimization strategy reconstruction mechanism is automatically triggered to dynamically adjust the core control parameters of the firefly algorithm and the bat algorithm;

[0138] S74. For the firefly algorithm, according to the average fluctuation amplitude of the oil parameters and the vibration data feature vector optimized by the firefly algorithm and the bat algorithm in the current sampling period, adjust the attraction parameter and the individual movement step size. When it is detected that the characteristic value changes significantly, automatically increase the attraction parameter, so that the individuals with higher brightness in the algorithm have stronger attraction to the remaining individuals;

[0139] S75. For the bat algorithm, dynamically adjust the frequency adjustment parameter and the local perturbation amplitude according to the confidence factor. When the confidence factor is high, it indicates that the recognition tends to converge, appropriately reduce the frequency adjustment range of the bat individuals, and narrow the search perturbation amplitude. If the confidence factor decreases, operate in the reverse direction and expand the search frequency range;

[0140] After the dynamic update of the firefly algorithm and the bat algorithm is completed, restart the fusion weight optimization and feature extraction process to generate a new round of fusion feature vectors and input them into the fault diagnosis model to form a closed-loop optimization-diagnosis-feedback iteration mechanism.

[0141] The present invention constructs a dynamic monitoring window for feature vectors and diagnostic scores, and performs real-time perception and intervention on the optimization based on the continuous data change trend. The system dynamically adjusts the core parameters of the firefly and bat algorithms, including the attractiveness, frequency range, and perturbation amplitude, according to the score change rate, brightness deviation, and confidence factor, to achieve the linkage control of the algorithm convergence speed and search depth. In the case of slow convergence or abnormal fluctuations, the system can actively execute strategy reconstruction and optimize the individual behavior model. After the update is completed, a new fusion feature vector is generated and fed back to the fault diagnosis model to construct an intelligent iterative closed-loop mechanism of "acquisition-optimization-identification-re-optimization".

[0142] Example 1:

[0143] To verify the feasibility of the present invention in implementation, the present invention is applied to the gearbox operation monitoring system of a 1.5MW doubly-fed wind turbine in a wind farm. The wind farm is located in a typical low-temperature and high-wind area, and the equipment is in an intermittent high-load fluctuation state for a long time. The failure rate of the gearbox components is relatively high, and common problems include lubricating oil deterioration, abnormal gear meshing, and bearing fatigue rupture. The traditional single-channel vibration monitoring system cannot meet the detection requirements of early hidden faults, and the fault identification is delayed and the false alarm rate is relatively high, which greatly affects the stability of equipment operation and the planning of operation and maintenance.

[0144] The wind farm installs and deploys the fan gearbox fault diagnosis and early warning system based on the fusion of oil and vibration parameters proposed by the present invention on the target unit. The system includes a high-precision oil sensor and a three-axis IEPE vibration sensor, and collects 4 types of parameters such as oil temperature, pressure, particle contamination degree, and viscosity, as well as 3 types of signals such as vibration acceleration, frequency, and amplitude. The sensor data sampling frequency is once every 10 seconds, and the data is transmitted to the edge computing terminal in real time for preliminary preprocessing, including operations such as filtering, normalization, and outlier removal.

[0145] The system applies the firefly algorithm to optimize the fusion weight of the oil and vibration signals, and automatically adjusts the contribution ratio of the two types of data to the final diagnosis decision. Within the first 10 days of operation, through the adaptive update of the brightness function and individual attractiveness, the fusion weight is stably distributed as follows: vibration parameters account for 61.7%, and oil parameters account for 38.3%. Subsequently, the system uses the bat algorithm to iteratively search the fused feature space, and finally selects a 5-dimensional optimal feature vector, including the oil temperature change rate, vibration acceleration peak value, frequency deviation degree, viscosity change slope, and pollution particle concentration density. This subset shows the highest fault classification discriminability in multiple rounds of tests.

[0146] The optimized feature vector is input into the decision tree model for fault identification and type prediction. Meanwhile, the system constructs a scoring mechanism to generate multi-dimensional fault warning scores. At 15:46 on the afternoon of August 7, 2024, the system first output a medium-level warning score for this unit and prompted that there was an abnormal lubrication state. At 16:20 on the same day, after the inspection personnel arrived at the scene to check, it was confirmed that the oil viscosity was low, the oil level dropped, and the particle concentration increased. It was initially judged that the insufficient lubrication was caused by the change of the main gear pair clearance. After replacing the oil and the gearbox filter element according to the suggestions, the warning score quickly dropped back to the normal range within the following week.

[0147] During the continuous monitoring period of 30 days, the system detected 5 abnormal events in total, including 4 minor wear warnings and 1 medium-level alarm, all of which were consistent with the on-site maintenance records and manual inspection information. The overall fault identification accuracy rate of the system reached 97.3%. Compared with the traditional method, the average early warning time was advanced by more than 41 minutes, and the average response time was only 6.5 seconds, and there were no false alarms or missed alarms.

[0148] Table 1 Summary Table of Data for the Fault Diagnosis and Warning System of Wind Turbine Gearboxes Based on the Fusion of Oil and Vibration Parameters

[0149]

[0150] Table 1 shows that the system proposed by the present invention has extremely high robustness and recognition ability in the actual operating environment. It not only realizes the deep fusion of oil and vibration data, but also realizes the closed-loop adaptive control of feature extraction, fault identification and warning response through the collaborative optimization of the firefly and bat algorithms, which is significantly better than the existing traditional methods. This system is applicable to a variety of wind turbine models and operating conditions, and has extremely high promotion and application value.

[0151] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters, characterized in that, Including: An oil parameter and vibration signal acquisition module, which is used to collect the oil parameters and vibration signals of the fan gearbox in real time through an oil sensor and a vibration sensor; A data fusion optimization module, which is used to optimize the fusion weights of oil and vibration data by using the firefly algorithm; A feature extraction and subset selection module, which is used to extract and optimize the features of the fused oil parameters and vibration signals; A fault diagnosis module, which is used to input the optimized feature vector into the fault diagnosis model; A fault warning generation module, which is used to generate a fault warning signal in real time according to the output of the fault diagnosis model; A closed-loop update module, which is used to form a closed-loop optimization-diagnosis-feedback iteration mechanism.

2. A fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters, characterized in that, The modules are implemented through the following methods: S1. Collect the oil parameters of the fan gearbox in real time through an oil sensor, collect the vibration signals of the fan gearbox in real time through a vibration sensor, and preprocess the collected oil parameters and vibration signals; S2. Use the firefly algorithm to optimize the fusion weights of the oil parameters and vibration signals, and adjust the influence of the data source on the final diagnosis result; S3. Use the bat algorithm to extract and optimize the features of the fused oil parameters and vibration signals, and select the optimal feature subset; S4. Input the feature vectors of the oil and vibration data optimized by the firefly algorithm and the bat algorithm into the fault diagnosis model, generate the optimized feature vector, and identify the fault type; S5. Generate a fault warning signal in real time according to the output of the fault diagnosis model, and provide fault information and maintenance suggestions to the operator; S6. Continuously monitor the oil and vibration data of the fan gearbox, generate real-time monitoring data, and dynamically adjust the warning threshold according to the fault warning signal; S7. Adjust the optimization strategies of the firefly algorithm and the bat algorithm according to the changes in the real-time monitoring data.

3. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, wherein The oil parameters specifically include the oil temperature, pressure, pollutant concentration and viscosity of the fan gearbox, and the vibration signals specifically include the vibration frequency, vibration amplitude and vibration acceleration.

4. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, wherein The specific content of S1 includes: S11. Preprocess the collected oil parameters, use median filtering to denoise the oil parameters, filter out the outliers caused by sensor noise, use Kalman filtering to smooth the oil temperature and pressure parameters, and use linear interpolation to fill in the missing oil parameters; S12. Preprocess the collected vibration signals, use a band-pass filter to filter the vibration signals, remove the low-frequency noise and high-frequency noise, retain the effective vibration information within the frequency range, use Hilbert transform to extract the instantaneous vibration amplitude, and perform detrending processing, and use the IQR-based outlier detection method to detect and remove abnormal vibration data points.

5. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, characterized in that, The specific content of S2 includes: S21. Optimize the fusion weights of the oil parameters and vibration signals through the firefly algorithm, calculate the relative contribution degrees of the oil parameters and vibration signals in fault diagnosis, and ensure that the sum of all weights is 1 and each weight value is within the range of [0,1] during the optimization process; S22. Initialize the weights and positions of firefly individuals, set the initial brightness and attractiveness. The brightness represents the fitness of an individual, and the attractiveness is related to the distance from the optimization goal and the fitness evaluation. Define the fitness function of brightness to evaluate the weighted eigenvalues of oil parameters and vibration signals, and calculate the fitness of each individual: Among them, F i is the brightness of the i-th firefly, W i is the fusion weight of the i-th oil parameter or vibration signal, x i is the normalized eigenvalue of the i-th oil parameter or vibration signal, α i is the adjustment factor of the i-th oil parameter or vibration signal, β is the balance adjustment factor, n is the number of oil parameters and vibration signals, is used to measure the error of the weighting and normalization processes, i is the index number of the oil parameters and vibration signals, (W i ·x i ) 2 represents the square operation on W i and x i ; S23. According to the update rules of the firefly algorithm, perform iterative optimization. In each iteration, update the weights and positions of the oil and vibration signals; S24. Through multiple iterations, the firefly algorithm updates the fusion weight of the oil fluid and vibration data until the maximum number of iterations N is reached max , generating an optimized fusion weight that can achieve an optimal balance between the oil fluid parameters and the vibration signal; S25. After each iteration, monitor the convergence of the algorithm by evaluating the change of the objective function, and adjust the learning rate and the weight update step size to ensure the stability and convergence speed of the optimization process.

6. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, characterized in that, The specific steps of S3 include: S31. Extract features from the fused oil parameter and vibration signal data, optimize the features using the bat algorithm, initialize the feature positions and velocities of bat individuals, and set the initial brightness and frequency. The brightness represents the fitness of a bat, and the frequency is used to control the flying speed and search step size of the bat. The feature positions and velocities are used to update the search process of the features. Define a new fitness function as: Among them, is the brightness of the i-th bat, where i is the index number of the oil fluid parameter and the vibration signal, and λ j is the adjustment factor of the j-th feature, and W j is the weight coefficient of the j-th feature, x j is the normalized value of the j-th feature, p is the exponent that controls the influence of each feature, j is the feature number, γ is the balance factor, and m is the total number of features. is used to measure the difference between the sum of all feature weights and 1, |W j ·x j | p is the absolute value of the weighted normalized eigenvalue W j ·x j followed by a power operation; S33. Perform iterative optimization through the flight update rules of the bat algorithm. In each iteration, update the feature positions and velocities of the oil parameters and vibration signals; S34. Update the feature set and feature weights through multiple iterations until the maximum number of iterations N is reached max , and generate the optimal feature subset and feature weights.

7. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, wherein The specific steps of S4 include: S41. Input the oil parameter and vibration data feature vectors optimized by the firefly algorithm and the bat algorithm into the fault diagnosis model, use the decision tree algorithm to identify the fault type, and construct a classifier model based on the training set to minimize the objective function: Among them, is the parameter obtained by solving through the optimization algorithm, θ is the parameter set to be optimized, n is the number of oil parameters and vibration signals, m1 is the number of fault types, is the index of the fault type, i is the number index of the oil parameter and the vibration signal, is the weight of the i i-th class of faults, X is the mean vector of the i-th class of faults, is the brightness of the i-th bat, F i is the brightness of the i-th firefly, is the regularization factor, is the feature vector X of the i-th sample i and the fault type mean vector between the square difference, is the brightness in the bat algorithm and the brightness F in the firefly algorithm i between the square difference; S42. According to the optimized feature vectors, input each feature into the fault diagnosis model, calculate the conditional probability of the fault type, process the input feature vectors according to the parameters obtained during the training process, evaluate the probability of each fault type occurring, evaluate the contribution of each feature to the fault type, and calculate the probability value for each fault type. The output result is the posterior probability distribution of the fault type under the given features, and determine the most likely fault type.

8. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, characterized in that, The specific steps of S5 include: S51. Generate a fault warning score in real time according to the parameters output by the fault diagnosis model. If then it is determined that the fault warning of the th type is triggered: Among them, is the comprehensive early warning score for the type of fault is the index of the fault type, is the total number of features, is the index of the th input feature, is the feature 's weighted coefficient for the fault type ; is the optimal parameter vector the parameter weight corresponding to the th feature in it, is the brightness of the th bat, is the brightness of the th firefly; S52. For the comprehensive warning score of the type fault, perform confidence evaluation, and comprehensively analyze the reliability of the type fault score based on the parameters participating in the scoring. If the values corresponding to most high-weight features are significantly higher than the remaining features and the brightness difference is stable, it indicates that the recognition result of the fault has high consistency and reliability, and a confidence factor is generated; S53. Automatically set the severity level of the fault according to the warning score and confidence factor, and determine the response time window and recommended maintenance priority according to the level; S54. Generate a fault diagnosis report according to the scoring result. The report includes the fault type number, comprehensive score, confidence factor, severity level, and recommended disposal plan, and push the report to the fan remote monitoring platform.

9. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, characterized in that, The specific steps of S6 include: S61. Continuously collect the oil parameters and vibration signals during the operation of the fan gearbox, construct time series data containing multi-dimensional features, and construct a monitoring matrix that changes with time. Each column is a feature vector at a certain moment; S62. Input the real-time monitoring matrix into the fault diagnosis model, and perform a comprehensive analysis of the current operating state in combination with the optimal parameter vector, optimized feature vector, and comprehensive warning score of the fault; S63. Calculate the average fluctuation degree of all features in time as the current operating state fluctuation coefficient according to the change amplitude between the optimized feature vector and the feature vector at the previous moment; S64. Dynamically adjust the warning threshold corresponding to the current fault type according to the comprehensive warning score of the current operating state fluctuation coefficient and the faults in the previous cycle. When the operating state fluctuates greatly, automatically increase the warning sensitivity and correspondingly lower the warning threshold. When the operating state fluctuates little, correspondingly increase the threshold; S65. If it is detected that the operating state fluctuations continuously increase in multiple consecutive time periods and the fault score always exceeds the current warning threshold, automatically trigger a high-priority warning mechanism and fix the current warning threshold of the fault to the minimum acceptable value; S66. In each sampling cycle, store the current optimized feature vector, the comprehensive warning score of the fault, the optimized feature vector and the comprehensive warning score of the fault, and the warning threshold into the database.

10. The fault diagnosis and early warning system for a fan gearbox based on the fusion of oil fluid and vibration parameters according to claim 1, wherein The specific steps of S7 are as follows: S71. During operation, based on the optimal parameter vector, optimized feature vector, comprehensive warning score of the fault, firefly brightness, and bat brightness collected at the current moment, construct a continuous monitoring data window to track and evaluate the performance in different time periods; S72. When the change amplitude of the fusion weight and brightness difference is detected to be lower than the set convergence determination value in multiple consecutive cycles, it is determined that the current optimization process has tended to be stable, and then switch to the low-frequency update mode; S73. If it is detected that the eigenvalue fluctuation increases significantly, the warning score continuously exceeds the warning threshold, or the output parameters of the fault diagnosis model show a divergent trend, automatically trigger an optimization strategy reconstruction mechanism to dynamically adjust the core control parameters of the firefly algorithm and the bat algorithm; S74. For the firefly algorithm, adjust the attractiveness parameter and the individual movement step size according to the average fluctuation amplitude of the oil fluid parameters and vibration data feature vector optimized by the firefly algorithm and the bat algorithm in the current sampling cycle. When it is detected that the eigenvalue change increases significantly, automatically increase the attractiveness parameter so that the individuals with higher brightness in the algorithm have stronger attraction to the remaining individuals; S75. For the bat algorithm, dynamically adjust the frequency adjustment parameter and the local perturbation amplitude according to the confidence factor. When the confidence factor is high, it indicates that the recognition tends to converge. Appropriately reduce the frequency adjustment range of the bat individuals and narrow the search perturbation amplitude. If the confidence factor decreases, operate in the reverse, expanding the search frequency range; S76. After completing the dynamic update of the firefly algorithm and the bat algorithm, restart the fusion weight optimization and feature extraction process to generate a new round of fusion feature vectors and input them into the fault diagnosis model to form a closed-loop optimization-diagnosis-feedback iteration mechanism.

Images