A planetary gearbox compound fault self-adaptive decoupling diagnosis method based on fault feature orientation

Abstract

The present application relates to the technical field of rotating machinery fault diagnosis, and particularly relates to a planetary gearbox compound fault self-adaptive decoupling diagnosis method based on fault feature orientation. The present application constructs a complete planetary gearbox compound fault self-adaptive decoupling diagnosis process, from the acquisition of fault vibration signals, to the optimization of filter parameters by using a variable weight particle swarm algorithm, then the modal decomposition is carried out by using a blind deconvolution theory, and finally the diagnosis result is output by combining envelope spectrum analysis. The whole process is oriented to fault features, realizes the self-adaptive decoupling of compound faults, integrates the advantages of various advanced algorithms, can effectively process complex fault signals, improves the systematicness and accuracy of compound fault diagnosis, and lays a whole framework foundation for the accurate implementation of subsequent steps.

Description

Technical Field

[0001] This invention relates to the field of rotating machinery fault diagnosis technology, specifically an adaptive decoupling diagnosis method for complex faults in planetary gearboxes based on fault feature guidance. Background Technology

[0002] Planetary gearboxes, as core components of power transmission in heavy industry (such as port machinery and wind power equipment), endure extreme conditions such as high loads and low speeds for extended periods. Their critical components, such as gears and bearings, are prone to fatigue damage including tooth surface spalling and cracking. Particularly concerning is that, as a tightly coupled system, planetary gearboxes exhibit strong correlations between component failures, with compound failures (such as concurrent gear-bearing failures) occurring at a significantly higher probability than single failures. These compound failures accelerate equipment failure, ranging from reduced production efficiency to serious safety accidents. Therefore, there is an urgent need to develop efficient compound failure isolation and diagnosis technologies.

[0003] Decomposition algorithms are an effective means of separating different components in a signal. Scholars both domestically and internationally have proposed decomposition algorithms such as EMD, VMD, empirical wavelet transform, and symplectic geometric mode decomposition, achieving good application results in the field of rotating machinery fault diagnosis. Although these decomposition algorithms are widely used for signal decoupling, they lack a fault feature-guided mechanism during decomposition. Faced with complex components in planetary gearbox vibration signals, such as random impacts, strong background noise, and periodic harmonics, the decomposition results often exhibit mode aliasing, information loss, or redundant components, making it difficult to accurately separate the impact components related to the fault. Therefore, blind deconvolution (such as eigenmode decomposition) is commonly used for processing. Although blind deconvolution has made progress in fault feature extraction by designing FIR filters and optimizing kurtosis and correlation kurtosis, two key problems remain: 1. Parameters rely on experience: Core decomposition parameters (such as filter length, number of iterations, etc.) heavily depend on manual presets and lack adaptive optimization capabilities. Inappropriate parameter settings under different operating conditions or fault types will directly lead to decomposition failure, resulting in poor robustness.

[0004] 2. Deficiencies in the objective function: Existing indicators (such as correlation kurtosis) show a significant decrease in their ability to characterize periodic impact features under strong noise and dense harmonic interference. They struggle to distinguish between real fault impacts and interference pulses, resulting in low signal-to-noise ratios and periodic ambiguity in the decomposed "fault components," severely impacting the accuracy of subsequent diagnostics.

[0005] In summary, current technologies have not yet solved the core challenge of "how to adaptively and accurately decouple the complex fault characteristics of planetary gearboxes in complex noise environments." There is an urgent need for a diagnostic method that can simultaneously overcome the bottlenecks of both adaptive parameter optimization and fault characteristic enhancement. Summary of the Invention

[0006] To avoid and overcome the technical problems existing in the prior art, this invention provides an adaptive decoupling diagnosis method for complex faults in planetary gearboxes based on fault feature guidance. This invention acquires vibration signals from the planetary gearbox, adaptively optimizes candidate solutions for filtering parameters using a variable-weight particle swarm optimization algorithm, selects optimal mode decomposition parameters based on blind deconvolution theory and applies them to the decomposition of new signals, and finally achieves high-precision decoupling diagnosis of complex faults through envelope spectrum analysis.

[0007] To achieve the above objectives, the present invention provides the following technical solution: An adaptive decoupling diagnosis method for complex faults in planetary gearboxes based on fault characteristics includes the following diagnostic steps: S1. Acquire fault vibration signals during the operation of the planetary gearbox; S2. Use the variable weighted particle swarm optimization algorithm to optimize the candidate solutions for filtering parameters of the fault vibration signal; S3. Use blind deconvolution theory to perform mode decomposition on each set of candidate solutions for filtering parameters obtained in the optimization process, and obtain the optimal mode decomposition parameters. The process of obtaining the optimal mode decomposition parameters is as follows: S31. The finite impulse response filter bank is initialized using a Gaussian window function with a 50% overlap segmentation. S32. Combining blind deconvolution theory, modal decomposition is performed based on fault feature guidance and represented as a constraint problem; S33. Solve the constraint problem in step S32 using the eigenvector method to obtain the mode decomposition matrix of the vibration signal; S34. Calculate each modal component in the modal decomposition matrix. The correlation between modal components is analyzed, and the fitness function values ​​corresponding to the two modal components with the highest correlation are calculated. Modal components with fitness function values ​​less than a set threshold are removed. Through continuous iterative selection, the preset modal components are finally selected. The segment signal is reduced to the preset target number of decompositions. The optimal mode decomposition parameters are obtained. S4. Perform mode decomposition of the fault vibration signal according to the optimal mode decomposition parameters in step S3 to obtain the specific values ​​of the optimal mode decomposition parameters. S5. Perform envelope spectrum analysis on the specific values ​​to obtain the fault diagnosis results corresponding to the fault vibration signal.

[0008] As a further technical solution of the present invention, the optimization process is as follows: S21. Define the optimization space dimension of the variable-weight particle swarm optimization algorithm. The values ​​of the filtering parameters in the optimization space, including the filter length. The number of finite impulse response filters and expected types of failure modes , ; S22. Initialize the population in the optimization space composed of candidate solutions for each set of filter parameters. , Indicates the first Candidate solutions for group filter parameters Indicates the total number of candidate solutions for the filter parameters; the first Candidate solutions for group filter parameters in The position vector in the dimensional optimization space is Its velocity vector is , Indicates the first The first group of candidate solutions for filter parameters The values ​​of each filter parameter, Indicates the first The first group of candidate solutions for filter parameters The speed of each filter parameter; S23. Construct a fault impact index and use it as the fitness function of the variable weight particle swarm algorithm. S24. Update the position and velocity of candidate solutions for filtering parameters based on the fitness function; S25. After the set stopping condition is met, stop the iteration and output the candidate solutions of each set of filter parameters during the iteration process.

[0009] As a further technical solution of the present invention, the fitness function is expressed as follows: ; In the formula, Represents the fitness function. Indicates the first Types of failure modes; express The corresponding failure impact indicators.

[0010] As a further technical solution of the present invention, the update formula for the candidate solution of the updated filter parameters is expressed as follows: ; ; In the formula, express In the The value obtained during round iteration; express In the The value obtained during round iteration; express In the Positions of candidate solutions for filter parameters during round iteration; express In the The value obtained during round iteration; and All are learning factors; and All are uniformly random numbers; Indicates up to the number The historical optimal position of candidate solutions for filter parameters during round iteration; Indicates up to the number The global optimal position of the candidate solutions for the filter parameters in the population during round iteration; Indicates the inertia weight; express In the The value obtained during round iteration; Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value; Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value; express In the The value obtained during round iteration; Indicates up to the number During round iteration The average value.

[0011] As a further technical solution of the present invention, the segmentation method is as follows: ; In the formula, Indicates the sampling frequency of the vibration signal; , They represent the first Finite impulse response filter The upper and lower cutoff frequencies, .

[0012] As a further technical solution of the present invention, the constraint problem is expressed as follows: ; In the formula, This indicates the selection of the modal component that best characterizes the fault. express In length The modal component values ​​at time, , Indicates the total length of the vibration signal; express Its length is Frequency of time , Indicates the filter length; This indicates that the vibration signal has a length of The value at time.

[0013] As a further technical solution of the present invention, the characteristic equation established by solving the constraint problem is as follows: ; ; ; ; In the formula, This indicates that the length gradually increases from 1 to [a certain value] by incrementing by 1 step. of indivual The matrix formed; express The values ​​of the modal components when the length is 1; express In length The modal component values ​​at time; Represents the vibration signal matrix; This represents the value of the vibration signal when the length is 1. This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; Indicates the first The frequency matrix of a finite impulse response filter; express The frequency when its length is 1; express Its length is The frequency of time.

[0014] As a further technical solution of the present invention, the calculation formula for the fault impact index is as follows: ; In the formula, express Periodic weighted kurtosis; express The linear kurtosis.

[0015] As a further technical solution of the present invention, the correlation index between the modal components is the cosine similarity, and the specific expression is as follows: ; In the formula, Indicates the first Types of failure modes; express In length The modal component values ​​at time; express and Cosine similarity between them.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs a complete adaptive decoupling diagnostic process for complex faults in planetary gearboxes. From acquiring fault vibration signals to optimizing filter parameters using a variable-weight particle swarm optimization algorithm, then performing mode decomposition through blind deconvolution theory, and finally combining envelope spectrum analysis to output diagnostic results, the entire process is fault-characteristic-oriented, achieving adaptive decoupling of complex faults. It integrates the advantages of multiple advanced algorithms, effectively handles complex fault signals, improves the systematicness and accuracy of complex fault diagnosis, and lays the overall framework foundation for the precise implementation of subsequent steps.

[0017] 2. By defining an optimization space that includes filter length, number, and desired fault mode types, a population composed of multiple candidate solutions is initialized, and a fitness function is constructed based on the fault impact index to guide iterative updates. This design can specifically search for optimal filter parameters, the population iteration mechanism ensures the comprehensiveness of the optimization, and the variable weight characteristic enhances the algorithm's adaptability to different fault scenarios. It can efficiently screen out candidate solutions that meet the fault characteristics, providing a high-quality parameter foundation for subsequent mode decomposition.

[0018] 3. By averaging the fault impact indices of multiple fault modes, the overall performance of different fault characteristics can be comprehensively reflected, avoiding the one-sidedness of single-mode evaluation. The design of taking the reciprocal makes the smaller the fitness function value, the more significant the corresponding fault characteristics, thereby accurately guiding the algorithm to optimize in the direction of enhancing fault characteristics, improving the pertinence and effectiveness of parameter optimization.

[0019] 4. The parameter update formula considers both the individual historical optimal position and the global optimal position of the candidate solution, and combines the learning factor and random numbers to balance the exploration and utilization capabilities during the optimization process. Furthermore, the adaptive adjustment strategy of the inertia weight (dynamically changing based on the comparison between the current fitness value and the average value) can maintain a larger inertia weight to expand the search range when the fitness is poor, and reduce the inertia weight to focus on local optimization when the fitness is good. This effectively avoids premature convergence of the algorithm and improves the efficiency, stability, and accuracy of the optimization process.

[0020] 5. The process of obtaining the optimal mode decomposition parameters combines Gaussian window function initialization, blind deconvolution theory, and eigenvector method, and enhances signal continuity through 50% overlap segmentation. Transforming mode decomposition into a constraint problem and solving it ensures that the decomposition process closely adheres to fault characteristics. Furthermore, by calculating the correlation of modal components and removing low-fitness components, redundant and interfering modes can be effectively removed, decomposing the signal to a preset target number, achieving precise decoupling of fault modes, and significantly improving the relevance and purity of mode decomposition.

[0021] 6. By calculating the upper and lower cutoff frequencies of the filters using explicit formulas, and based on the vibration signal sampling frequency and the number of filters, the frequency band division of the finite impulse response filter bank becomes more scientific. This segmentation method ensures that the frequency bands of each filter neither overlap excessively nor omit any, accurately covering the effective frequency range of the signal. Each filter can specifically capture fault information in a specific frequency band, providing a structurally sound and comprehensively covered frequency foundation for subsequent mode decomposition, and enhancing the reliability of fault feature extraction.

[0022] 7. Modal decomposition is transformed into a constrained optimization problem aimed at maximizing the fault impact index. The objective function directly targets the salience of fault characteristics, ensuring that the decomposition process focuses on enhancing fault information. The constraints clearly define the convolution relationship between the modal components and the original signal and filter, making the decomposition results mathematically traceable, ensuring the clear physical meaning of the modal components, and effectively extracting the modes that best characterize the fault features, thus improving the purposefulness and accuracy of the decomposition.

[0023] 8. By constructing characteristic equations in matrix form, modal components are represented as the product of the vibration signal matrix and the filter matrix, transforming the complex mode decomposition process into quantifiable mathematical operations. This matrix representation clarifies the relationship between modal components, the original signal, and the filter, facilitating efficient solutions using the eigenvector method, improving computational rigor and operability, and providing a concise mathematical model for computer implementation, ensuring the accuracy of the mode decomposition results.

[0024] 9. Periodically weighted kurtosis can highlight periodic fault impacts, while linear kurtosis enhances the detection sensitivity of impact signals. The product of the two can comprehensively capture the periodic and transient characteristics of fault impacts, and is especially suitable for identifying multiple impact features in planetary gearbox complex faults, significantly improving the comprehensiveness and sensitivity of fault feature characterization.

[0025] 10. Using cosine similarity as an index of modal component correlation, highly correlated redundant modes can be effectively identified by quantifying the similarity between different modal components. This method is simple to calculate and has clear physical meaning. It can accurately eliminate highly interfering modal components, retain independent modes that can reflect different fault characteristics, achieve effective decoupling of compound faults, reduce mutual interference between modes, and improve the accuracy of subsequent fault diagnosis. Attached Figure Description

[0026] Figure 1 This is a flowchart of the adaptive decoupling diagnosis method for complex faults in planetary gearboxes according to the present invention.

[0027] Figure 2 This is a schematic diagram of a planetary gearbox fault simulation test bench in an embodiment of the present invention.

[0028] Figure 3 This is a time-domain signal diagram of a combined fault of sun gear peeling and planetary gear cracking in an embodiment of the present invention.

[0029] Figure 4 This is a Fourier spectrum diagram of a combined fault of sun gear peeling and planetary gear cracking in an embodiment of the present invention.

[0030] Figure 5 This is the initialization curve of the FIR filter bank based on the Gaussian window function in an embodiment of the present invention.

[0031] Figure 6 This is an iterative curve diagram of the mode decomposition parameter optimization process using the variable weight particle swarm algorithm in an embodiment of the present invention.

[0032] Figure 7 This is a time-domain waveform diagram of mode #1 when using the fault diagnosis model to predict the planetary gear crack fault type in this invention.

[0033] Figure 8 This is the envelope spectrum of mode #1 when using the fault diagnosis model to predict the planetary gear crack fault type in this invention.

[0034] Figure 9 This is the time-domain waveform diagram of mode #2 when using the fault diagnosis model to predict the planetary gear crack fault type in this invention.

[0035] Figure 10This is the envelope spectrum of mode #2 when using the fault diagnosis model to predict the fault type of planetary gear cracks in this invention.

[0036] In the diagram: 1. Equipment mounting base; 2. Three-phase AC motor; 3. First coupling; 4. Planetary reduction gearbox; 5. Second coupling; 6. Magnetic powder brake; 7. Vibration damping base. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see Figure 1 In this embodiment of the invention, an adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault characteristics includes the following: I. Obtaining Fault Vibration Signals like Figure 2 As shown, the planetary gearbox fault simulation test bench built in this study mainly consists of core components such as equipment mounting base 1, three-phase AC motor 2, first coupling 3, planetary gearbox 4, second coupling 5, magnetic powder brake 6, and shock-absorbing base 7.

[0039] The assembly relationship of the test bench is as follows: the shock-absorbing base 7 is placed on the workbench, and the equipment mounting base 1 is fixedly installed on its upper part; the three-phase AC motor 2, the planetary reduction gearbox 4, and the magnetic powder brake 6 are all mounted on the equipment mounting base 1, forming a stable layout structure. Among them, the output end of the three-phase AC motor 2 is connected to the input end of the planetary reduction gearbox 4 through the first coupling 3, and the output end of the planetary reduction gearbox 4 is connected to the input end of the magnetic powder brake 6 through the second coupling 5, thus forming a complete power transmission and loading system.

[0040] To simulate a combined fault in the planetary gearbox, an electrical discharge machining (EDM) and wire EDM process was used to create a combined fault of sun gear spalling and planetary gear cracking. To simulate low-speed, heavy-load operation in an industrial setting, the drive motor speed was set to 150 r / min and the load to 300 N·m. An accelerometer was mounted on the surface of the planetary gearbox housing 4 to collect vibration signals, with a sampling frequency of 12800 Hz and a sampling duration of 20 s. Under these experimental conditions, the parameters and characteristic frequencies of the planetary gearbox 4 are shown in Table 1.

[0041] Table 1. Parameters and characteristic frequencies of planetary reduction gearboxes

[0042] An accelerometer is installed on the surface of the planetary gearbox corresponding to the bearing or gear ring to collect vibration signals during the operation of the planetary gearbox. x ( n ),in, n =1,2,…, N , N Vibration signal x ( n The length of the acquired vibration signal. The time-domain and frequency-domain waveforms are as follows: Figure 3 and Figure 4 As shown, in the time domain, the gearbox vibration signal waveform is stable with no obvious periodic impact characteristics, except for random pulse interference impact characteristics between 10s and 15s. Therefore, the health status of the gearbox cannot be determined from the time domain waveform. From the spectrum, it can be seen that the amplitude of the planetary gearbox meshing frequency is low in the frequency domain. f m The sidebands are submerged by strong noise, making it impossible to extract effective fault feature information; therefore, simple time-frequency analysis cannot effectively identify the current health status of the planetary gearbox.

[0043] II. Finding the best solution

[0044] Initialize the variable-weight particle swarm optimization algorithm and construct the fitness function. F obj The variable-weight particle swarm optimization algorithm is used to adjust the mode decomposition parameters of the mode decomposition method, specifically the filter length. L Initial number of filters K and expected decomposition mode number K d Perform adaptive optimization.

[0045] The optimization process is as follows: S21. Since the optimization parameter is the filter length L Initial number of filters K Expected decomposition mode number K d Therefore, the optimization space dimension of the variable-weight particle swarm optimization algorithm is set to be... D =3; Filter length in the optimization space L The value range is [2, 200], and the initial number of filters is... K The value range is [5, 10], and the expected number of decomposed modes is... K d The value range is [1, 4].

[0046] S22. Initialize the population in the optimization space composed of candidate solutions for each set of filter parameters. , Indicates the first Candidate solutions for group filter parameters Indicates the total number of candidate solutions for the filter parameters; the first Candidate solutions for group filter parameters in The position vector in the dimensional optimization space is Its velocity vector is , Indicates the first The first group of candidate solutions for filter parameters The values ​​of each filter parameter, Indicates the first The first group of candidate solutions for filter parameters The velocity of each filter parameter; the position of each particle represents the filter length. L Initial number of filters K Expected decomposition mode number K d The candidate solutions for the three parameters are represented by the velocity of each particle, which indicates the search direction and step size of the candidate solution. This determines the filter length L and the initial number of filters in the next iteration step. K Expected decomposition mode number K d The update direction and distance of the three parameters in the optimization space.

[0047] S23. Construct a fault impact index and use it as the fitness function of the variable-weight particle swarm optimization algorithm. The fitness function is expressed as follows: ; In the formula, Represents the fitness function. Indicates the first Types of failure modes; express The corresponding failure impact indicators.

[0048] The formula for calculating the fault impact index is as follows: ; In the formula, express Periodic weighted kurtosis; express The linear kurtosis.

[0049] S24. Update the position and velocity of candidate solutions for the filter parameters based on the fitness function. The update formula is as follows: ; ; In the formula, express In the The value obtained during round iteration; express In the The value obtained during round iteration; express In the Positions of candidate solutions for filter parameters during round iteration; express In the The value taken during round iteration. and All are learning factors, with values ​​ranging from [0, 4]. In this embodiment of the invention... c 1 and c Both are taken as 1.5. and All are uniformly random numbers, with values ​​ranging from [0, 1], and are randomly generated during each iteration. Indicates up to the number The historical optimal position of the candidate solution for the filter parameters during round iteration. Indicates up to the number The global optimal position of the candidate solutions for the filtering parameters in the population during round iteration. The inertia weight is represented, and its value range is typically [0.2, 1]. In this embodiment of the invention, ω The initial value is generated by a random number and is dynamically updated with each iteration step by the following formula, with a maximum value of 1 and a minimum value of 0.2. express In the The value obtained during round iteration; Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value. Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value; express In the The value obtained during round iteration; Indicates up to the number During round iteration The average value.

[0050] S25. Upon reaching the set maximum number of iterations... TAfter reaching 20, the iteration stops, and the candidate solutions for each set of filter parameters during the iteration process are output. The historical optimal position of the particle is... P best The optimal position of the population is G best .

[0051] III. Obtaining the Optimal Mode Decomposition Parameters

[0052] Based on the obtained global optimal mode decomposition parameters L , K and K d The acquired vibration signals were analyzed using the modal decomposition method. x ( n Modal decomposition was performed to obtain... K d One optimal modal component u i ( n ), i =1,2,…, K d The specific steps are as follows: S31. The finite impulse response filter bank is initialized using a Gaussian window function with a 50% overlap segmentation method; the segmentation method is shown in the following equation: ; In the formula, Indicates the sampling frequency of the vibration signal; , They represent the first Finite impulse response filter The upper and lower cutoff frequencies, .

[0053] The filter bank initialization segmentation results are as follows Figure 5 As shown, the filters overlap, which can better preserve the impulse response characteristics.

[0054] S32. Combining blind deconvolution theory, modal decomposition is performed based on fault feature guidance and represented as a constraint problem; the constraint problem is expressed as follows: ; In the formula, This indicates the selection of the modal component that best characterizes the fault. express In length The modal component values ​​at time, , Indicates the total length of the vibration signal; express Its length is Frequency of time , Indicates the filter length; This indicates that the vibration signal has a length of The value at time.

[0055] S33. Solve the constraint problem in step S32 using the eigenvector method to obtain the mode decomposition matrix of the vibration signal; The characteristic equation established by solving the constraint problem is as follows: ; ; ; ; In the formula, This indicates that the length gradually increases from 1 to [a certain value] by incrementing by 1 step. of indivual The matrix formed; express The values ​​of the modal components when the length is 1; express In length The modal component values ​​at time; Represents the vibration signal matrix; This represents the value of the vibration signal when the length is 1. This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; Indicates the first The frequency matrix of a finite impulse response filter; express The frequency when its length is 1; express Its length is The frequency of time.

[0056] S34. Calculate each modal component in the modal decomposition matrix. The correlation between modal components is analyzed, and the fitness function values ​​corresponding to the two modal components with the highest correlation are calculated. Modal components with fitness function values ​​less than a set threshold are removed. Through continuous iterative selection, the preset modal components are finally selected. The segment signal is reduced to the preset target number of decompositions. The optimal mode decomposition parameters are obtained.

[0057] The correlation index between modal components is cosine similarity, and its specific expression is as follows: ; In the formula, Indicates the first Types of failure modes; express In length The modal component values ​​at time; express and Cosine similarity between them.

[0058] IV. Obtaining the specific values ​​of the optimal mode decomposition parameters

[0059] Based on the obtained optimal mode decomposition parameters, the newly acquired vibration signal is subjected to mode decomposition in step S3 to obtain the specific values ​​of the optimal mode decomposition parameters. The details are as follows: Repeat steps two and three until the maximum number of iterations is reached. T =20, and based on the comparison of the current optimal mode decomposition parameters generated in each iteration step. L , K and K d The corresponding fitness function F obj The value of . Update the fitness function. F obj The minimum value corresponds to the mode decomposition parameters L , K and K d For the vibration signal acquired in step one x ( n Global optimal mode decomposition parameters L , K and K d Regarding the vibration signal acquisition in step one x ( n The iterative curve of the mode decomposition parameter optimization process is as follows: Figure 6 As shown in the figure, when iterating to the fifth step, the globally optimal decomposition parameters are obtained, and the combination of decomposition parameters is: K =6, K d =2, L =86.

[0060] V. Obtaining Fault Diagnosis Results

[0061] Based on the obtained optimal mode decomposition parametersK =6, K d =2, L =86, proceed to step three to obtain the vibration signal acquired in step one. x ( n The optimal mode decomposition result is obtained, and envelope spectrum analysis is performed on the optimal mode decomposition result to obtain the fault diagnosis result.

[0062] Regarding the vibration signal acquired in step one x ( n The optimal decomposition mode results in the time domain and envelope spectrum are as follows: Figures 7 to 10 As shown, there are two optimal decomposition modes. This can be seen from the envelope spectrum. Figure 8 Mode #1 exhibits abundant sun gear fault-related features. Figure 10 The planetary gear fault-related features appeared in mode #2, so the diagnosis result was a combined fault of the sun gear and planetary gears, which is consistent with the experimental preset and verifies the effectiveness of the fault diagnosis method of the present invention in decoupling diagnosis of combined faults in planetary gearboxes.

[0063] In summary, this invention provides an adaptive decoupling diagnosis method for complex faults in planetary gearboxes based on fault feature guidance, achieving adaptive separation and extraction of complex faults in planetary gearboxes. This method constructs a composite fault impact index (FII) based on periodic weighted kurtosis and linear kurtosis to address the identification of the periodicity and impact of fault features in vibration signals. By combining cosine similarity and the multi-objective optimization rules of FII, the optimal filter length, number of filter banks, and expected modal components are obtained, achieving adaptive separation and extraction of complex fault features and effectively overcoming the blindness of manually selecting modal decomposition parameters. This method is highly immune to random impact interference, sensitive to cyclic impact features, and can achieve effective decoupling and separation of complex faults without prior fault knowledge.

[0064] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fault feature-guided adaptive decoupling diagnosis method for complex faults in planetary gearboxes, characterized in that, The diagnostic steps include the following: S1. Acquire fault vibration signals during the operation of the planetary gearbox; S2. Use the variable weighted particle swarm optimization algorithm to optimize the candidate solutions for filtering parameters of the fault vibration signal; S3. Use blind deconvolution theory to perform mode decomposition on each set of candidate solutions for filtering parameters obtained in the optimization process, and obtain the optimal mode decomposition parameters. The process of obtaining the optimal mode decomposition parameters is as follows: S31. The finite impulse response filter bank is initialized using a Gaussian window function with a 50% overlap segmentation. S32. Combining blind deconvolution theory, modal decomposition is performed based on fault feature guidance and represented as a constraint problem; S33. Solve the constraint problem in step S32 using the eigenvector method to obtain the mode decomposition matrix of the vibration signal; S34. Calculate each modal component in the modal decomposition matrix. The correlation between modal components is analyzed, and the fitness function values ​​corresponding to the two modal components with the highest correlation are calculated. Modal components with fitness function values ​​less than a set threshold are removed. Through continuous iterative selection, the preset modal components are finally selected. The segment signal is reduced to the preset target number of decompositions. The optimal mode decomposition parameters are obtained. S4. Perform mode decomposition of the fault vibration signal according to the optimal mode decomposition parameters in step S3 to obtain the specific values ​​of the optimal mode decomposition parameters. S5. Perform envelope spectrum analysis on the specific values ​​to obtain the fault diagnosis results corresponding to the fault vibration signal.

2. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance as described in claim 1, characterized in that, The optimization process is as follows: S21. Define the optimization space dimension of the variable-weight particle swarm optimization algorithm. The values ​​of the filtering parameters in the optimization space, including the filter length. The number of finite impulse response filters and expected fault modes , ; S22. Initialize the population in the optimization space composed of candidate solutions for each set of filter parameters. , Indicates the first Candidate solutions for group filter parameters Indicates the total number of candidate solutions for the filter parameters; the first Candidate solutions for group filter parameters in The position vector in the dimensional optimization space is Its velocity vector is , Indicates the first The first group of candidate solutions for filter parameters The values ​​of each filter parameter, Indicates the first The first group of candidate solutions for filter parameters The speed of each filter parameter; S23. Construct a fault impact index and use it as the fitness function of the variable weight particle swarm algorithm. S24. Update the position and velocity of candidate solutions for filtering parameters based on the fitness function; S25. After the set stopping condition is met, stop the iteration and output the candidate solutions of each set of filter parameters during the iteration process.

3. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance according to claim 2, characterized in that, The fitness function is expressed as follows: ； In the formula, Represents the fitness function. Indicates the first Types of failure modes; express The corresponding failure impact indicators.

4. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance according to claim 3, characterized in that, The update formula for the candidate solutions of the updated filter parameters is expressed as follows: ； ； In the formula, express In the The value obtained during round iteration; express In the The value obtained during round iteration; express In the Positions of candidate solutions for filter parameters during round iteration; express In the The value obtained during round iteration; and All are learning factors; and All are uniformly random numbers; Indicates up to the number The historical optimal position of candidate solutions for filter parameters during round iteration; Indicates up to the number The global optimal position of the candidate solutions for the filter parameters in the population during round iteration; Indicates the inertia weight; express In the The value obtained during round iteration; Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value; Indicates up to the number During round iteration The historical minimum value; Indicates up to the number During round iteration The historical maximum value; express In the The value obtained during round iteration; Indicates up to the number During round iteration The average value.

5. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance according to claim 1, characterized in that, The segmentation method is shown in the following formula: ； In the formula, Indicates the sampling frequency of the vibration signal; , They represent the first Finite impulse response filter The upper and lower cutoff frequencies, .

6. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance according to claim 5, characterized in that, The constraint problem is represented as follows: ； In the formula, This indicates the selection of the modal component that best characterizes the fault. express In length The modal component values ​​at time, , Indicates the total length of the vibration signal; express Its length is Frequency of time , Indicates the filter length; This indicates that the vibration signal has a length of The value at time.

7. The adaptive decoupling diagnosis method for composite faults in planetary gearboxes based on fault feature guidance according to claim 6, characterized in that, The characteristic equation established by solving the constraint problem is as follows: ； ； ； ； In the formula, This indicates that the length gradually increases from 1 to [a certain value] by incrementing by 1 step. of indivual The matrix formed; express The values ​​of the modal components when the length is 1; express In length The modal component values ​​at time; Represents the vibration signal matrix; This represents the value of the vibration signal when the length is 1. This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; This indicates that the vibration signal has a length of The value at time; Indicates the first The frequency matrix of a finite impulse response filter; express The frequency when its length is 1; express Its length is The frequency of time.

8. A fault-feature-oriented adaptive decoupling diagnosis method for composite faults in planetary gearboxes according to any one of claims 1-7, characterized in that, The formula for calculating the fault impact index is as follows: ； In the formula, express Periodic weighted kurtosis; express The linear kurtosis.

9. A fault feature-guided adaptive decoupling diagnosis method for composite faults in planetary gearboxes according to any one of claims 5-7, characterized in that, The correlation index between modal components is cosine similarity, and its specific expression is as follows: ； In the formula, Indicates the first Types of failure modes; express In length The modal component values ​​at time; express and Cosine similarity between them.

Images