Fan shaft fault diagnosis method based on hierarchical multi-objective optimization algorithm

Abstract

The present invention discloses a fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm. The method comprises the following steps: S1: forming a comprehensive dataset of fan shaft multimodal signals; S2: classifying the fan shaft fault modes according to a hierarchical strategy based on the comprehensive dataset of fan shaft multimodal signals; S3: forming a feature dataset for each layer; S4: forming an optimization model adapted to different operating conditions; S5: outputting an optimized diagnostic solution; S6: generating a comprehensive diagnostic result for the fan shaft; and S7: generating fault warning information and a health status assessment report for the fan shaft based on the comprehensive diagnostic result, while also providing fault location, fault type, and maintenance recommendations. The present invention addresses the common local optimality trap problem in traditional optimization methods and effectively reduces computational costs under complex fault modes.

Description

Technical Field

[0001] The present invention relates to the technical field of fans, and in particular to a fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm. Background Art

[0002] In recent years, with the continuous expansion of wind power generation, the wind turbine shaft system, as the core component of the wind turbine, its operating stability is crucial to the overall performance and safety of the wind power system. Wind turbines operate for a long time under complex climatic conditions, and the shaft system structure is susceptible to multiple influences of vibration, stress, temperature and external environment. If the shaft system fault information cannot be obtained in time, it may cause large-scale damage and shutdown of the equipment, resulting in economic losses. In order to reduce maintenance costs and ensure power generation efficiency, the wind power industry has gradually paid attention to the prevention and detection of shaft system faults. At present, some conventional diagnostic technologies based on vibration analysis, oil analysis or temperature monitoring have been widely used in actual wind turbine maintenance.

[0003] In the existing technology, many fan shaft fault diagnosis methods mainly rely on signal analysis of a single data source, such as only conducting time domain or frequency domain research on vibration signals or only monitoring changes in oil particle size. Although single-mode or finite-mode detection methods can detect some fault signs, they often lack comprehensiveness when facing complex coupled faults. When the fan operates at high speed, variable speed loads and interacts with multiple environmental parameters, the types of shaft faults are diverse and the occurrence mechanisms are complex. The existing technology is still insufficient in multi-modal signal fusion and adaptation to dynamic working conditions.

[0004] On the other hand, with the increase in wind turbine monitoring methods, the scale of collected operating data has grown exponentially. Existing technologies usually simply aggregate the data and use fixed models or threshold methods to make fault judgments, lacking comprehensive consideration of multiple objectives. Some systems that aim at real-time diagnosis often simplify the feature extraction process to ensure speed, resulting in a decrease in diagnostic accuracy. Those methods oriented towards high accuracy often require longer computing time or professional prior knowledge, and it is difficult to achieve ideal results under the complex and frequently changing operating conditions of wind turbines.

[0005] In summary, the existing wind turbine shaft fault diagnosis technology has obvious deficiencies in multimodal signal fusion, real-time accuracy balance, and dynamic operating condition adaptation. The diagnosis process lacks hierarchical refinement, making it difficult to balance the precise positioning of complex coupled faults with real-time identification under high load and variable speed conditions. The defects of existing technologies are particularly prominent in the context of the changing wind turbine operating environment, huge data volumes, and diverse fault modes, making it difficult to meet the modern wind power industry's requirements for high efficiency and reliability in shaft fault diagnosis. Summary of the Invention

[0006] One purpose of the present invention is to propose a fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm, which solves the common local optimal trap problem in traditional optimization methods and effectively reduces the computational cost under complex fault modes.

[0007] A fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm according to an embodiment of the present invention includes the following steps:

[0008] S1. Acquire multimodal signal data during the operation of the fan shaft system, and perform signal denoising, normalization processing, and time series segmentation on the multimodal signal data to form a comprehensive data set of multimodal signals of the fan shaft system;

[0009] S2. Based on the comprehensive dataset of wind turbine shaft multimodal signals, classify the wind turbine shaft failure modes according to a hierarchical strategy, and divide the failure modes into a basic layer, an intermediate layer, and a top layer;

[0010] S3. In each layer of the fault mode, the multimodal signal data is subjected to feature extraction and feature dimensionality reduction using signal processing techniques to form a feature data set for each layer;

[0011] S4. Construct a multi-objective optimization model based on the feature dataset extracted from each layer. The multi-objective optimization model uses fault diagnosis accuracy, real-time fault diagnosis, and adaptability to multimodal signal data as optimization objectives. Dynamically adjust the objective weights to form an optimization model that adapts to different operating conditions.

[0012] S5. Applying a gold panning optimization algorithm to solve the multi-objective optimization model, wherein the exploration phase samples feature data globally within the multi-objective optimization model to generate an initial candidate solution, and the mining phase performs local optimization near the initial candidate solution to output an optimized diagnostic solution;

[0013] S6. Based on the optimized diagnostic solution, perform fault diagnosis of the fan shaft system layer by layer in a hierarchical strategy, output the diagnostic results of each layer, and fuse the diagnostic results of each layer to generate a comprehensive diagnostic result of the fan shaft system;

[0014] S7. Generate fault warning information and health status assessment report of the fan shaft system based on the comprehensive diagnosis results, and provide fault location, fault type and maintenance suggestions.

[0015] Optionally, the S1 includes the following steps:

[0016] S11. Acquire multimodal signal data during the operation of the fan shaft system, wherein the multimodal signal data includes:

[0017] Vibration signal, indicating the vibration amplitude measured during the operation of the fan shaft system;

[0018] Speed ​​signal, indicating the speed change measured during the operation of the fan;

[0019] Temperature signal, indicating the temperature values ​​of key components of the shafting at different times;

[0020] Environmental parameter signals, including ambient temperature, humidity and air pressure parameters.

[0021] S12. performing signal denoising processing on the acquired multimodal signal data, using a bandpass filter to eliminate high-frequency noise and low-frequency interference;

[0022] S13. performing normalization processing on the denoised multimodal signal data;

[0023] S14. Perform time series segmentation on the normalized multimodal signal data, setting the sliding window size to w and the step size to s, and segmenting the signal data into continuous time segments;

[0024] S15. The segmented vibration signal dataset D v , speed signal dataset D r , temperature signal dataset D T and environmental parameter signal dataset D e Summarize and form a comprehensive dataset of multimodal signals of the fan shaft system:

[0025] D={D v ,D r ,D T ,D e}.

[0026] Optionally, step S2 specifically includes:

[0027] S21. Perform fault information analysis on each record based on the comprehensive dataset D of the fan shaft multimodal signal, and define a fault information metric function to guide the fault information analysis of each record:

[0028] G(k)=ω v ·Φ(D v (k))+ω r ·Φ(D r (k))+ω T ·Φ(D T (k))+ω e ·Φ(D e (k));

[0029] Where k is the index of the record in the data set, G(k) is the fault information metric, ω v ,ω r ,ω T ,ω eare weighting coefficients for vibration signal data, speed signal data, temperature signal data, and environmental parameter signal data, respectively; Φ(·) is a function for calculating the abnormality degree of the input signal;

[0030] S22. Compare the fault information metric G(k) with the hierarchical threshold and define a hierarchical classification function:

[0031]

[0032] Where D(k) is the kth record in the data set, θ b and θ m To distinguish the base layer L b , middle layer L m With top floor L t The stratification threshold of

[0033] S23. In the base layer L b In

[15] , based on the univariate abnormal trends in the vibration signal dataset and the temperature signal dataset, the primary fault univariate detection formula is defined as follows:

[0034] A b (k) = max{|D v (k)-μ v |,|D T (k)-μ T |};

[0035] Among them, A b (k) is the abnormality measure of a single type of fault, μ v is the reference mean value of the vibration signal, μ T is the reference mean value of the temperature signal, if A b (k) If the set threshold is exceeded, it is determined to be a primary failure mode;

[0036] S24. In the middle layer L m In , complex coupling failure modes are detected and the coupling anomaly measurement formula is defined:

[0037] A m (k) = δ vr ·Ψ(D v (k),D r (k))+δ vT ·Ψ(D v (k),D T (k))+δ rT ·

[0038] Ψ(D r (k),D T (k));

[0039] Among them, A m(k) is the coupling fault anomaly metric, δ vr ,δ vT ,δ rT is the weight coefficient of different signal combinations, Ψ(·,·) is the function for evaluating the correlation of multiple signals, if A m (k) If the coupling fault threshold is exceeded, it is determined to be a complex coupling fault mode;

[0040] S25. On the top floor L t In the process, comprehensive failure modes under high load or special working conditions are identified and the comprehensive abnormality measurement formula is defined:

[0041] A t (k) = ζ e ·(D e (k)+1)×ln[1+D v (k)];

[0042] Among them, A t (k) is the comprehensive fault anomaly metric, ζ e D is the amplification factor of the environmental parameter’s response to the vibration signal. e (k) is the environmental parameter signal value corresponding to the kth record, D v (k) is the vibration signal value, if A t (k) If the comprehensive fault threshold is exceeded, it is determined to be a high-complexity comprehensive fault mode;

[0043] S26. According to the hierarchical classification function f L The output of (D(k)) is combined with A b (k), A m (k) and A t (k) is used to classify each record in the dataset D into the base layer L b , middle layer L m With top floor L t .

[0044] Optionally, step S3 specifically includes:

[0045] S31. In the base layer L b In this paper, time domain features are extracted for a single type of primary fault mode, and the characteristic values ​​of the vibration signal dataset and the temperature signal dataset are calculated, including the mean, variance and crest factor;

[0046] S32. In the middle layer L m In the paper, the frequency domain features are extracted for the complex coupled fault mode, the frequency distribution features of the vibration signal data set, the speed signal data set and the temperature signal data set are calculated, and the spectrum energy distribution is obtained by using the fast Fourier transform;

[0047] S33. On the top floor Lt In this paper, the time-frequency domain features are extracted for high-complexity comprehensive fault modes. The wavelet transform method is used to obtain multi-resolution signal features, and the characteristic energy of wavelet decomposition is defined as:

[0048]

[0049] Among them, W j,n is the nth coefficient of the signal in the jth layer of wavelet decomposition, J is the total number of layers of wavelet decomposition, E w is the characteristic energy in the time-frequency domain, which is used for the characteristic analysis of the comprehensive fault mode. N represents the total number of points of the signal in a certain decomposition layer.

[0050] S34. The time domain features, frequency domain features and time-frequency domain features extracted from the base layer, middle layer and top layer are respectively used to form feature data sets F b 、F m 、F t , define the comprehensive feature dataset as:

[0051] F={F b ,F m ,F t}.

[0052] Optionally, step S4 specifically includes:

[0053] S41. Based on the comprehensive feature dataset F, a dynamic weight adaptive allocation mechanism is introduced to construct a multi-objective optimization model that integrates fault diagnosis accuracy, real-time performance, and multimodal signal adaptability:

[0054]

[0055] Among them, α i (t) represents the priority weight of the i-th layer at time t, which is dynamically updated to adapt to different operating conditions. Denote the fault diagnosis accuracy, real-time and signal adaptability measurement functions of the base layer, middle layer and top layer respectively, λ A ,λ R ,λ S is the adjustment factor of the global objective function;

[0056] S42. Define a dynamic weight adaptive allocation mechanism, combining the complexity of failure modes and operating conditions to evaluate and calculate the weight allocation of each layer in real time:

[0057]

[0058] Among them, C i (t) is the fault complexity of layer i at time t, γ i is the initial weight of the i-th layer, β iis the dynamic adjustment coefficient of the weight of the i-th layer;

[0059] S43. Introduce a multi-objective co-evolution mechanism during the optimization process, use independent sub-optimization modules to optimize the feature data set of each layer, and define the optimization objective function of each layer as:

[0060]

[0061] Among them, ξ i Represents the optimization parameters of the i-th layer. Each sub-optimization module runs in parallel and shares intermediate results with each other;

[0062] S44. Define a global control strategy for model optimization, and perform unified optimization based on multi-objective co-evolution and the global objective function:

[0063]

[0064] Among them, κ is the global regularization coefficient, which is used to balance the model complexity and optimization objectives, and Φ(ξ) is the regularization function, which limits the search space of optimization parameters.

[0065] Optionally, step S5 specifically includes:

[0066] S51. In the multi-objective optimization model, the comprehensive feature dataset F of the wind turbine shaft system is hierarchically mapped to the optimization parameter space Θ. Global sampling is performed using the exploration phase of the gold panning optimization algorithm:

[0067]

[0068] Among them, Θ (0) is the initial candidate solution set for global sampling, is the number of sampling points in the base layer, middle layer and top layer, Θ i is the optimized parameter subspace corresponding to the i-th layer, Uniform represents a uniform sampling distribution, which is used to cover the global range of the parameter space Θ;

[0069] S52. In conjunction with the topic of fan shaft fault diagnosis, calculate the multi-objective optimization function value for each candidate solution and define the multi-objective fault diagnosis optimization function as:

[0070]

[0071] in, They represent the fault diagnosis accuracy, real-time and signal adaptability objective function values ​​of the i-th layer diagnosis solution respectively;

[0072] S53. In the initial candidate solution set, according to the optimization function value F opt (θ k ) sorting selects the top N fdiagnostic solutions, and define the diagnostic solution set as:

[0073] Θ (f) ={θ k ∣F opt (θ k )≤δ f ,k=1,2,…,N f};

[0074] Among them, δ f is the optimal threshold, and the solution with the best optimization function value is selected, Θ (f) is the diagnostic solution set;

[0075] S54. During the excavation phase, combined with the specific fault mode characteristics of the fan shaft system, the diagnostic solution set Θ (f) Perform local optimization and define dynamic gradient mining based on the fan shaft characteristics as follows:

[0076]

[0077] in, is the feature regularization term based on the multimodal signal of the fan shaft system, ζ is the regularization weight, which is used to limit the parameter update range, It is the gradient calculation, used to guide the optimization direction;

[0078] S55. After global exploration and local mining are completed, the global and local optimization results are combined to output the final diagnostic solution set:

[0079]

[0080] Among them, Θ opt is the final diagnostic solution set, which includes the optimal parameters for fan shaft fault diagnosis, N o is the final number of solutions.

[0081] Optionally, step S6 specifically includes:

[0082] S61. Based on the diagnostic solution set, the diagnostic solution is input into the fault diagnosis models of the base layer, the middle layer, and the top layer according to the hierarchical strategy, and the hierarchical diagnosis process is started. Each layer model matches a specific fault mode according to the input solution;

[0083] S62. In the base layer, by analyzing the characteristics of a single type of primary fault, determine whether a single type of primary fault exists, and output the diagnosis results of the base layer, including the specific fault type and the preliminary diagnosis probability;

[0084] S63. In the middle layer, the cross-signal features of multiple signals are integrated to diagnose complex fault modes, the correlation trends between signals are dynamically analyzed, and the diagnosis results of the middle layer are output, including the complex coupling fault type and the diagnosis confidence level;

[0085] S64. At the top level, the system combines the comprehensive performance of environmental and vibration signals, focusing on fault diagnosis under high load or special operating conditions. Based on the multimodal signal fusion characteristics, it identifies comprehensive fault modes in complex operating environments and outputs the top-level diagnostic results.

[0086] S65. Fuse the diagnostic results of the base layer, middle layer, and top layer, assign dynamic weights based on the confidence and importance of the diagnostic results of each layer, and generate a comprehensive diagnostic result of the fan shaft system, including comprehensive fault types, overall diagnostic confidence, and recommended maintenance measures.

[0087] The beneficial effects of the present invention are:

[0088] (1) The present invention introduces a hierarchical strategy in the diagnosis process, divides the fault characteristics of the fan shaft system into a basic layer, an intermediate layer and a top layer, and adopts time domain, frequency domain and time-frequency domain feature extraction technologies for different levels of fault modes. The hierarchical feature processing ensures the full utilization of multimodal signals, and can perform targeted analysis on various fault types such as bearings, gears and shaft center offsets. Accuracy is improved as the primary goal through a multi-objective optimization model with dynamic weight adjustment.

[0089] (2) The present invention processes the optimization task in stages through the gold panning optimization algorithm. First, in the exploration stage, global uniform sampling is used to cover the parameter space. Then, in the mining stage, local search and gradient enhancement optimization are performed on the optimal solution. By introducing the dynamic gradient regularization term, the local optimal trap problem common in traditional optimization methods is solved, and the computational cost under complex fault modes is effectively reduced.

[0090] (3) The present invention generates comprehensive diagnostic results by hierarchically fusing the diagnostic results of the base layer, the middle layer, and the top layer, and assigning dynamic weights according to the diagnostic confidence and importance of each layer. This mechanism significantly enhances the fault diagnosis capability of the system in a complex operating environment. In the case of conflict or noise interference in multimodal signals, the hierarchical fusion significantly improves the robustness and adaptability of the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0092] Figure 1 This is a flow chart of a fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm proposed by the present invention. DETAILED DESCRIPTION

[0093] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0094] refer to Figure 1 A fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm includes the following steps:

[0095] S1. Acquire multimodal signal data during the operation of the fan shaft system, perform signal denoising, normalization, and time series segmentation on the multimodal signal data, and form a comprehensive dataset of the fan shaft system multimodal signals;

[0096] S2. Based on the comprehensive dataset of wind turbine shaft multimodal signals, classify the fault modes of the wind turbine shaft according to a hierarchical strategy, and divide the fault modes into the basic layer, the middle layer, and the top layer;

[0097] S3. In each fault mode layer, use signal processing technology to extract features and reduce the dimension of the multimodal signal data to form a feature dataset for each layer;

[0098] S4. Construct a multi-objective optimization model based on the feature dataset extracted from each layer. The multi-objective optimization model uses fault diagnosis accuracy, real-time fault diagnosis, and adaptability to multimodal signal data as optimization objectives. Dynamically adjust the objective weights to form an optimization model that adapts to different operating conditions.

[0099] S5. Apply the gold panning optimization algorithm to solve the multi-objective optimization model. In the exploration phase, feature data is sampled globally within the multi-objective optimization model to generate initial candidate solutions. In the mining phase, local optimization is performed near the initial candidate solutions, and the optimized diagnostic solution is output.

[0100] S6. Based on the optimized diagnostic solution, perform fault diagnosis of the fan shaft system layer by layer in a hierarchical strategy, output the diagnostic results of each layer, and fuse the diagnostic results of each layer to generate a comprehensive diagnostic result of the fan shaft system;

[0101] S7. Generate fault warning information and health status assessment report of the fan shaft system based on the comprehensive diagnosis results, and provide fault location, fault type and maintenance suggestions.

[0102] In this embodiment, S1 includes the following steps:

[0103] S11. Acquire multimodal signal data during the operation of the fan shaft system. The multimodal signal data includes:

[0104] Vibration signal, indicating the vibration amplitude measured during the operation of the fan shaft system;

[0105] Speed ​​signal, indicating the speed change measured during the operation of the fan;

[0106] Temperature signal, indicating the temperature values ​​of key components of the shafting at different times;

[0107] Environmental parameter signals, including ambient temperature, humidity and air pressure parameters.

[0108] S12. performing signal denoising processing on the acquired multimodal signal data, using a bandpass filter to eliminate high-frequency noise and low-frequency interference;

[0109] S13. performing normalization processing on the denoised multimodal signal data;

[0110] S14. Perform time series segmentation on the normalized multimodal signal data, setting the sliding window size to w and the step size to s, and segmenting the signal data into continuous time segments;

[0111] S15. The segmented vibration signal dataset D v , speed signal dataset D r , temperature signal dataset D T and environmental parameter signal dataset D e Summarize and form a comprehensive dataset of multimodal signals of the fan shaft system:

[0112] D={D v ,D r ,D T ,D e}.

[0113] In this embodiment, step S2 specifically includes:

[0114] S21. Perform fault information analysis on each record based on the comprehensive dataset D of the fan shaft multimodal signal, and define a fault information metric function to guide the fault information analysis of each record:

[0115] G(k)=ω v ·Φ(D v (k))+ω r ·Φ(D r (k))+ω T ·Φ(D T (k))+ω e ·Φ(D e (k));

[0116] Where k is the index of the record in the data set, G(k) is the fault information metric, ω v ,ω r ,ω T ,ω eare weighting coefficients for vibration signal data, speed signal data, temperature signal data, and environmental parameter signal data, respectively; Φ(·) is a function for calculating the abnormality degree of the input signal;

[0117] S22. Compare the fault information metric G(k) with the hierarchical threshold and define a hierarchical classification function:

[0118]

[0119] Where D(k) is the kth record in the data set, θ b and θ m To distinguish the base layer L b , middle layer L m With top floor L t The stratification threshold of

[0120] S23. In the base layer L b In

[15] , based on the univariate abnormal trends in the vibration signal dataset and the temperature signal dataset, the primary fault univariate detection formula is defined as follows:

[0121] A b (k) = max{|D v (k)-μ v |,|D T (k)-μ T |};

[0122] Among them, A b (k) is the abnormality measure of a single type of fault, μ v is the reference mean value of the vibration signal, μ T is the reference mean value of the temperature signal, if A b (k) If the set threshold is exceeded, it is determined to be a primary failure mode;

[0123] S24. In the middle layer L m In , complex coupling failure modes are detected and the coupling anomaly measurement formula is defined:

[0124] A m (k) = δ vr ·Ψ(D v (k),D r (k))+δ vT ·Ψ(D v (k),D T (k))+δ rT ·

[0125] Ψ(D r (k),D T (k));

[0126] Among them, A m(k) is the coupling fault anomaly metric, δ vr ,δ vT ,δ rT is the weight coefficient of different signal combinations, Ψ(·,·) is the function for evaluating the correlation of multiple signals, if A m (k) If the coupling fault threshold is exceeded, it is determined to be a complex coupling fault mode;

[0127] S25. On the top floor L t In the process, comprehensive failure modes under high load or special working conditions are identified and the comprehensive abnormality measurement formula is defined:

[0128] A t (k) = ζ e ·(D e (k)+1)×ln[1+D v (k)];

[0129] Among them, A t (k) is the comprehensive fault anomaly metric, ζ e D is the amplification factor of the environmental parameter’s response to the vibration signal. e (k) is the environmental parameter signal value corresponding to the kth record, D v (k) is the vibration signal value, if A t (k) If the comprehensive fault threshold is exceeded, it is determined to be a high-complexity comprehensive fault mode;

[0130] S26. According to the hierarchical classification function f L The output of (D(k)) is combined with A b (k), A m (k) and A t (k) is used to classify each record in the dataset D into the base layer L b , middle layer L m With top floor L t .

[0131] In this embodiment, step S3 specifically includes:

[0132] S31. In the base layer L b In this paper, time domain features are extracted for a single type of primary fault mode, and the characteristic values ​​of the vibration signal dataset and the temperature signal dataset are calculated, including the mean, variance and crest factor;

[0133] S32. In the middle layer L m In the paper, the frequency domain features are extracted for the complex coupled fault mode, the frequency distribution features of the vibration signal data set, the speed signal data set and the temperature signal data set are calculated, and the spectrum energy distribution is obtained by using the fast Fourier transform;

[0134] S33. On the top floor Lt In this paper, the time-frequency domain features are extracted for high-complexity comprehensive fault modes. The wavelet transform method is used to obtain multi-resolution signal features, and the characteristic energy of wavelet decomposition is defined as:

[0135]

[0136] Among them, W j,n is the nth coefficient of the signal in the jth layer of wavelet decomposition, J is the total number of layers of wavelet decomposition, E w is the characteristic energy in the time-frequency domain, which is used for the characteristic analysis of the comprehensive fault mode. N represents the total number of points of the signal in a certain decomposition layer.

[0137] S34. The time domain features, frequency domain features and time-frequency domain features extracted from the base layer, middle layer and top layer are respectively used to form feature data sets F b 、F m 、F t , define the comprehensive feature dataset as:

[0138] F={F b ,F m ,F t}.

[0139] In this embodiment, step S4 specifically includes:

[0140] S41. Based on the comprehensive feature dataset F, a dynamic weight adaptive allocation mechanism is introduced to construct a multi-objective optimization model that integrates fault diagnosis accuracy, real-time performance, and multimodal signal adaptability:

[0141]

[0142] Among them, α i (t) represents the priority weight of the i-th layer at time t, which is dynamically updated to adapt to different operating conditions. Denote the fault diagnosis accuracy, real-time and signal adaptability measurement functions of the base layer, middle layer and top layer respectively, λ A ,λ R ,λ S is the adjustment factor of the global objective function;

[0143] S42. Define a dynamic weight adaptive allocation mechanism, combining the complexity of failure modes and operating conditions to evaluate and calculate the weight allocation of each layer in real time:

[0144]

[0145] Among them, C i (t) is the fault complexity of layer i at time t, γ i is the initial weight of the i-th layer, β iis the dynamic adjustment coefficient of the weight of the i-th layer;

[0146] S43. Introduce a multi-objective co-evolution mechanism during the optimization process, use independent sub-optimization modules to optimize the feature data set of each layer, and define the optimization objective function of each layer as:

[0147]

[0148] Among them, ξ i Represents the optimization parameters of the i-th layer. Each sub-optimization module runs in parallel and shares intermediate results with each other;

[0149] S44. Define a global control strategy for model optimization, and perform unified optimization based on multi-objective co-evolution and the global objective function:

[0150]

[0151] Among them, κ is the global regularization coefficient, which is used to balance the model complexity and optimization objectives, and Φ(ξ) is the regularization function, which limits the search space of optimization parameters.

[0152] In this embodiment, step S5 specifically includes:

[0153] S51. In the multi-objective optimization model, the wind turbine shaft system comprehensive feature dataset F is hierarchically mapped to the optimization parameter space Θ, and global sampling is performed using the exploration phase of the gold panning optimization algorithm:

[0154]

[0155] Among them, Θ (0) is the initial candidate solution set for global sampling, is the number of sampling points in the base layer, middle layer and top layer, Θ i is the optimized parameter subspace corresponding to the i-th layer, Uniform represents a uniform sampling distribution, which is used to cover the global range of the parameter space Θ;

[0156] S52. In conjunction with the topic of fan shaft fault diagnosis, calculate the multi-objective optimization function value for each candidate solution and define the multi-objective fault diagnosis optimization function as:

[0157]

[0158] in, They represent the fault diagnosis accuracy, real-time and signal adaptability objective function values ​​of the i-th layer diagnosis solution respectively;

[0159] S53. In the initial candidate solution set, according to the optimization function value F opt (θ k ) sorting selects the top N fdiagnostic solutions, and define the diagnostic solution set as:

[0160] Θ (f) ={θ k ∣F opt (θ k )≤δ f ,k=1,2,…,N f};

[0161] Among them, δ f is the optimal threshold, and the solution with the best optimization function value is selected, Θ (f) is the diagnostic solution set;

[0162] S54. During the excavation phase, combined with the specific fault mode characteristics of the fan shaft system, the diagnostic solution set Θ (f) Perform local optimization and define dynamic gradient mining based on the fan shaft characteristics as follows:

[0163]

[0164] in, is the feature regularization term based on the multimodal signal of the fan shaft system, ζ is the regularization weight, which is used to limit the parameter update range, It is the gradient calculation, used to guide the optimization direction;

[0165] S55. After global exploration and local mining are completed, the global and local optimization results are combined to output the final diagnostic solution set:

[0166]

[0167] Among them, Θ opt is the final diagnostic solution set, which includes the optimal parameters for fan shaft fault diagnosis, N o is the final number of solutions.

[0168] In this embodiment, step S6 specifically includes:

[0169] S61. Based on the diagnostic solution set, the diagnostic solution is input into the fault diagnosis models of the base layer, the middle layer, and the top layer according to the hierarchical strategy, and the hierarchical diagnosis process is started. Each layer model matches a specific fault mode according to the input solution;

[0170] S62. In the base layer, by analyzing the characteristics of a single type of primary fault, determine whether a single type of primary fault exists, and output the diagnosis results of the base layer, including the specific fault type and the preliminary diagnosis probability;

[0171] S63. In the middle layer, the cross-signal features of multiple signals are integrated to diagnose complex fault modes, the correlation trends between signals are dynamically analyzed, and the diagnosis results of the middle layer are output, including the complex coupling fault type and the diagnosis confidence level;

[0172] S64. At the top level, the system combines the comprehensive performance of environmental and vibration signals, focusing on fault diagnosis under high load or special operating conditions. Based on the multimodal signal fusion characteristics, it identifies comprehensive fault modes in complex operating environments and outputs the top-level diagnostic results.

[0173] S65. Fuse the diagnostic results of the base layer, middle layer, and top layer, assign dynamic weights based on the confidence and importance of the diagnostic results of each layer, and generate a comprehensive diagnostic result of the fan shaft system, including comprehensive fault types, overall diagnostic confidence, and recommended maintenance measures.

[0174] Example 1:

[0175] At a coastal wind farm with 200 wind turbines and an average annual wind speed of 9.2 m / s, the climatic conditions are complex and changeable, and the wind turbines are under long-term high-load operation. One wind turbine (numbered WT-135) exhibited intermittent vibration anomalies during operation. Maintenance personnel observed through the monitoring system that the amplitude of its vibration signal fluctuated greatly under high-load conditions, accompanied by a gradual increase in temperature. Due to the wide distribution of the wind farm and the different operating conditions of each wind turbine, traditional manual regular maintenance methods are unable to detect such potential faults in a timely manner, which may lead to further deterioration of the fault, causing the entire machine to shut down or even more serious component damage.

[0176] In order to promptly detect and diagnose the fault of the fan shaft system, the operation and maintenance team decided to apply the fan shaft system fault diagnosis method based on the hierarchical multi-objective optimization algorithm of the present invention to comprehensively analyze and diagnose the operating status of the fan to verify the effectiveness and adaptability of this method.

[0177] Maintenance personnel first retrieved the operating data of the WT-135 wind turbine over the past three months, including vibration signals, speed signals, temperature signals, and environmental parameters (wind speed, humidity, and air pressure). This totaled 4,320 sets of data, each corresponding to one operating cycle (10 minutes). In data preprocessing, the original signals were denoised to remove high-frequency noise and low-frequency interference. All signals were normalized to the range of [0,1] and segmented into 216 data segments according to the time series, with each segment being 20 time points long.

[0178] Then, according to the hierarchical multi-objective optimization diagnosis method of the present invention, the pre-processed data is input into the basic layer, middle layer and top layer diagnosis models:

[0179] At the foundational level, a preliminary analysis of single signal characteristics was performed to identify possible primary bearing faults. The system detected an abnormal concentration of characteristic peak frequencies within the vibration signal within the frequency domain, and a continuous upward trend in the temperature signal, initially indicating possible bearing rolling element wear.

[0180] At the intermediate level, the coupling relationship between the vibration and speed signals was combined to further analyze complex coupled faults. The diagnostic model found that the instantaneous frequency variation in the speed signal was a multiple of the dominant frequency of the vibration signal, potentially related to shaft center offset. Furthermore, the increased correlation between the temperature and vibration signals further pointed to a coupled fault mode.

[0181] At the top level, multimodal signal fusion is used to identify complex faults under complex operating conditions. The diagnostic model, combined with environmental parameters, found that when wind speeds exceeded 12 meters per second, the dominant frequency of the vibration signal and fluctuations in environmental parameters were highly correlated, while the rate of temperature rise increased significantly. This combined analysis indicated the presence of a combined pattern of gear meshing anomalies and bearing failures under high-load operating conditions.

[0182] Ultimately, the diagnostic results from each layer were integrated through dynamic weighting to generate a comprehensive diagnosis: the WT-135 wind turbine exhibited bearing rolling element wear, accompanied by axial misalignment and gear meshing anomalies, with a 92% probability of overall fault diagnosis. The system recommended immediate shutdown for maintenance and offered maintenance recommendations, including bearing and gear replacement.

[0183] To verify the effectiveness of the proposed method, a comparative experiment was conducted with traditional diagnostic methods. 4,320 sets of historical data from the WT-135 and similar data from five other wind turbines (a total of 25,920 sets) were selected as training samples. The test set consisted of 1,080 sets of new data under different fault modes. The diagnostic accuracy, diagnosis time, and comprehensive performance indicators were compared.

[0184]

[0185] It can be seen from the experimental data that the method of the present invention is 16.3 percentage points higher than the traditional single signal analysis method in diagnostic accuracy, the diagnosis time is shortened by about 65%, and the comprehensive fault diagnosis rate is significantly improved, showing stronger adaptability and real-time performance under complex operating conditions.

[0186] Through the fault diagnosis case of the wind turbine WT-135, the method of the present invention demonstrates its high efficiency and accuracy under complex fault modes. Combined with comparative experimental data, it verifies the significant advantages of this method in multimodal signal fusion, multi-objective optimization and hierarchical diagnosis, providing a reliable solution for the intelligent fault diagnosis of the wind turbine shaft system.

[0187] The present invention introduces a hierarchical strategy in the diagnosis process, divides the fault characteristics of the fan shaft system into a basic layer, an intermediate layer and a top layer, and adopts time domain, frequency domain and time-frequency domain feature extraction technologies for fault modes at different levels. Through feature hierarchical processing, it ensures the full utilization of multimodal signals, and can perform targeted analysis on various fault types such as bearings, gears and shaft center offsets. Accuracy is improved as the primary goal through a multi-objective optimization model with dynamic weight adjustment.

[0188] The present invention processes the optimization task in stages through the gold panning optimization algorithm. First, in the exploration stage, global uniform sampling is used to cover the parameter space. Then, in the mining stage, local search and gradient enhancement optimization are performed on the optimal solution. By introducing the dynamic gradient regularization term, the local optimal trap problem common in traditional optimization methods is solved, and the computational cost under complex failure modes is effectively reduced.

[0189] The present invention generates comprehensive diagnostic results through hierarchical fusion of diagnostic results at the base layer, middle layer, and top layer, and assigns dynamic weights according to the diagnostic confidence and importance of each layer. The mechanism significantly enhances the system's fault diagnosis capability in complex operating environments, and significantly improves the robustness and adaptability of the results through hierarchical fusion in the presence of conflicts or noise interference in multimodal signals.

[0190] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A fan shaft fault diagnosis method based on a hierarchical multi-objective optimization algorithm, characterized in that: The steps include: S1. Acquire multimodal signal data during the operation of the fan shaft system, and perform signal denoising, normalization processing, and time series segmentation on the multimodal signal data to form a comprehensive data set of multimodal signals of the fan shaft system; S2. Based on the comprehensive dataset of wind turbine shaft multimodal signals, classify the wind turbine shaft failure modes according to a hierarchical strategy, and divide the failure modes into a basic layer, an intermediate layer, and a top layer; S3. In each layer of the fault mode, the multimodal signal data is subjected to feature extraction and feature dimensionality reduction using signal processing techniques to form a feature data set for each layer; S4. Construct a multi-objective optimization model based on the feature dataset extracted from each layer. The multi-objective optimization model uses fault diagnosis accuracy, real-time fault diagnosis, and adaptability to multimodal signal data as optimization objectives. Dynamically adjust the objective weights to form an optimization model that adapts to different operating conditions. S5. Applying a gold panning optimization algorithm to solve the multi-objective optimization model, wherein the exploration phase samples feature data globally within the multi-objective optimization model to generate an initial candidate solution, and the mining phase performs local optimization near the initial candidate solution to output an optimized diagnostic solution; The step S5 specifically includes: S51. In the multi-objective optimization model, the wind turbine shaft system comprehensive feature dataset F is hierarchically mapped to the optimization parameter space Θ, and global sampling is performed using the exploration phase of the gold panning optimization algorithm; S52. Combined with the topic of fan shaft fault diagnosis, calculate the multi-objective optimization function value for each candidate solution and define the multi-objective fault diagnosis optimization function as F opt (θ k ); S53. In the initial candidate solution set, according to the multi-objective fault diagnosis optimization function F opt (θ k ) sorting selects the top N f diagnostic solutions, and define the diagnostic solution set as: I (f) ={θ k ∣F opt (i k )≤δ f ,k=1,2,…,N f }; Among them, δ f As the threshold, select the solution with the best optimization function value, Θ (f) is the diagnostic solution set; S54. During the excavation phase, combined with the fault mode characteristics of the fan shaft system, the diagnostic solution set Θ (f) Perform local optimization and define dynamic gradient mining based on the fan shaft characteristics as follows: in, is the feature regularization term based on the multimodal signal of the fan shaft system, ζ is the regularization weight, which is used to limit the parameter update range, It is the gradient calculation, used to guide the optimization direction; S55. After global exploration and local mining are completed, the global and local optimization results are combined to output the final diagnostic solution set: Among them, Θ opt is the final diagnostic solution set, which includes the optimal parameters for fan shaft fault diagnosis, N o is the final solution number; S6. Based on the optimized diagnostic solution, perform fault diagnosis of the fan shaft system layer by layer in a hierarchical strategy, output the diagnostic results of each layer, and fuse the diagnostic results of each layer to generate a comprehensive diagnostic result of the fan shaft system; S7. Generate fault warning information and health status assessment report of the fan shaft system based on the comprehensive diagnosis results, and provide fault location, fault type and maintenance suggestions.

2. The fan shaft fault diagnosis method based on the hierarchical multi-objective optimization algorithm according to claim 1 is characterized in that: Said S1 comprises the following steps: S11. Acquire multimodal signal data during the operation of the fan shaft system, wherein the multimodal signal data includes: Vibration signal, indicating the vibration amplitude measured during the operation of the fan shaft system; Speed ​​signal, indicating the speed change measured during the operation of the fan; Temperature signal, indicating the temperature values ​​of key components of the shafting at different times; Environmental parameter signals, including ambient temperature, humidity and air pressure parameters; S12. performing signal denoising processing on the acquired multimodal signal data, using a bandpass filter to eliminate high-frequency noise and low-frequency interference; S13. performing normalization processing on the denoised multimodal signal data; S14. Perform time series segmentation on the normalized multimodal signal data, setting the sliding window size to w and the step size to s, and segmenting the signal data into continuous time segments; S15. The segmented vibration signal dataset D v , speed signal dataset D r , temperature signal dataset D T and environmental parameter signal dataset D e Summarize and form a comprehensive dataset of multimodal signals of the fan shaft system: D={D v ,D r ,D T ,D e }。 3. The fan shaft fault diagnosis method based on hierarchical multi-objective optimization algorithm according to claim 1 is characterized in that: The step S2 specifically includes: S21. Perform fault information analysis on each record based on the comprehensive dataset D of the fan shaft multimodal signal, and define a fault information metric function to guide the fault information analysis of each record: G(k)=ω v ·Φ(D v (k))+ω r ·Φ(D r (k))+ω T ·Φ(D T (k))+ω e ·Φ(D e (k)); Where k is the index of the record in the data set, G(k) is the fault information metric, ω v ,ω r ,ω T ,ω e are weighting coefficients for vibration signal data, speed signal data, temperature signal data, and environmental parameter signal data, respectively; Φ(·) is a function for calculating the abnormality degree of the input signal; S22. Compare the fault information metric G(k) with the hierarchical threshold and define a hierarchical classification function: Where D(k) is the kth record in the data set, θ b and θ m To distinguish the base layer L b , middle layer L m With top floor L t The stratification threshold of S23. In the base layer L b In [15], based on the univariate abnormal trends in the vibration signal dataset and the temperature signal dataset, the primary fault univariate detection formula is defined as follows: A b (k)=max{|D v (k)-m v |,|D T (k)-m T |}; Among them, A b (k) is the abnormality measure of a single type of fault, μ v is the reference mean value of the vibration signal, μ T is the reference mean value of the temperature signal, if A b (k) If the set threshold is exceeded, it is determined to be a primary failure mode; S24. In the middle layer L m In , complex coupling failure modes are detected and the coupling anomaly measurement formula is defined: A m (k)=δ vr ·Ψ(D v (k),D r (k))+δ vT ·Ψ(D v (k),D T (k))+δ rT ·Ψ(D r (k),D T (k)); Among them, A m (k) is the coupling fault anomaly metric, δ vr ,δ vT ,δ rT is the weight coefficient of different signal combinations, Ψ(·,·) is the function for evaluating the correlation of multiple signals, if A m (k) If the coupling fault threshold is exceeded, it is determined to be a complex coupling fault mode; S25. On the top floor L t In the process, comprehensive failure modes under high load or special working conditions are identified and the comprehensive abnormality measurement formula is defined: A t (k)=ζ e ·(D e (k)+1)×ln[1+D v (k)]; Among them, A t (k) is the comprehensive fault anomaly metric, ζ e D is the amplification factor of the environmental parameter’s response to the vibration signal. e (k) is the environmental parameter signal value corresponding to the kth record, D v (k) is the vibration signal value, if A t (k) If the comprehensive fault threshold is exceeded, it is determined to be a high-complexity comprehensive fault mode; S26. According to the hierarchical classification function f L The output of (D(k)) is combined with A b (k), A m (k) and A t (k) is used to classify each record in the dataset D into the base layer L b , middle layer L m With top floor L t .

4. The fan shaft fault diagnosis method based on hierarchical multi-objective optimization algorithm according to claim 1 is characterized in that: The step S3 specifically includes: S31. In the base layer L b In this paper, time domain features are extracted for a single type of primary fault mode, and the characteristic values ​​of the vibration signal dataset and the temperature signal dataset are calculated, including the mean, variance and crest factor; S32. In the middle layer L m In the paper, the frequency domain features are extracted for the complex coupled fault mode, the frequency distribution features of the vibration signal data set, the speed signal data set and the temperature signal data set are calculated, and the spectrum energy distribution is obtained by using the fast Fourier transform; S33. On the top floor L t In this paper, the time-frequency domain features are extracted for high-complexity comprehensive fault modes. The wavelet transform method is used to obtain multi-resolution signal features, and the characteristic energy of wavelet decomposition is defined as: Among them, W j,n is the nth coefficient of the signal in the jth layer of wavelet decomposition, J is the total number of layers of wavelet decomposition, E w is the characteristic energy in the time-frequency domain, which is used for the characteristic analysis of the comprehensive fault mode. N represents the total number of points of the signal in a certain decomposition layer. S34. The time domain features, frequency domain features and time-frequency domain features extracted from the base layer, middle layer and top layer are respectively used to form feature data sets F b 、F m 、F t , define the comprehensive feature dataset as: F={F b ,F m ,F t }。 5. The fan shaft fault diagnosis method based on hierarchical multi-objective optimization algorithm according to claim 1 is characterized in that: The step S4 specifically includes: S41. Based on the comprehensive feature dataset F, a dynamic weight adaptive allocation mechanism is introduced to construct a multi-objective optimization model that integrates fault diagnosis accuracy, real-time performance, and multimodal signal adaptability: Among them, α i (t) represents the priority weight of the i-th layer at time t, which is dynamically updated to adapt to different operating conditions. Denote the fault diagnosis accuracy, real-time and signal adaptability measurement functions of the base layer, middle layer and top layer respectively, λ A ,λ R ,λ S is the adjustment factor of the global objective function; S42. Define a dynamic weight adaptive allocation mechanism, combining the complexity of failure modes and operating conditions to evaluate and calculate the weight allocation of each layer in real time: Among them, C i (t) is the fault complexity of layer i at time t, γ i is the initial weight of the i-th layer, β i is the dynamic adjustment coefficient of the weight of the i-th layer; S43. Introduce a multi-objective co-evolution mechanism during the optimization process, use independent sub-optimization modules to optimize the feature data set of each layer, and define the optimization objective function of each layer as: Among them, ξ i Represents the optimization parameters of the i-th layer. Each sub-optimization module runs in parallel and shares intermediate results with each other; S44. Define a global control strategy for model optimization, and perform unified optimization based on multi-objective co-evolution and the global objective function: Among them, κ is the global regularization coefficient, which is used to balance the model complexity and optimization objectives, and Φ(ξ) is the regularization function, which limits the search space of optimization parameters.

6. The fan shaft fault diagnosis method based on hierarchical multi-objective optimization algorithm according to claim 1 is characterized in that: The step S6 specifically includes: S61. Based on the diagnostic solution set, the diagnostic solution is input into the fault diagnosis models of the base layer, the middle layer, and the top layer according to the hierarchical strategy, and the hierarchical diagnosis process is started. The models of each layer match the fault mode according to the input solution; S62. In the base layer, by analyzing the characteristics of a single type of primary fault, determine whether a single type of primary fault exists, and output the diagnosis results of the base layer, including the specific fault type and the preliminary diagnosis probability; S63. In the middle layer, the cross-signal features of multiple signals are integrated to diagnose complex fault modes, the correlation trends between signals are dynamically analyzed, and the diagnosis results of the middle layer are output, including the complex coupling fault type and the diagnosis confidence level; S64. At the top level, the system combines the comprehensive performance of environmental and vibration signals, focusing on fault diagnosis under high load or special operating conditions. Based on the multimodal signal fusion characteristics, it identifies comprehensive fault modes in complex operating environments and outputs the top-level diagnostic results. S65. Fuse the diagnostic results of the base layer, middle layer, and top layer, assign dynamic weights based on the confidence and importance of the diagnostic results of each layer, and generate a comprehensive diagnostic result of the fan shaft system, including comprehensive fault types, overall diagnostic confidence, and recommended maintenance measures.

Images