A bearing fault diagnosis method based on position constraint-adaptive dictionary learning in the complex field under variable speed conditions
By employing an adaptive dictionary learning method in the complex domain, the accuracy and robustness issues of bearing fault diagnosis under variable speed conditions were addressed. This method achieved unified representation of signal amplitude and phase information and adaptive optimization of parameters, thereby improving the accuracy and computational efficiency of fault feature extraction.
Patent Information
- Application Number
- CN202611146807.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-11-17
AI Technical Summary
Existing bearing fault diagnosis methods struggle to accurately extract fault features under variable speed conditions, suffer from insufficient utilization of phase information, poor noise immunity, high computational complexity, and reliance on manually set parameters, resulting in inadequate diagnostic accuracy and robustness.
A position-constrained adaptive dictionary learning method for the complex domain is adopted. A complex domain signal model is constructed through Hilbert transform. The Teager energy operator and the local orthogonal matching pursuit algorithm are combined, and complex sparse coefficients and Rényi entropy evaluation index are introduced to achieve a unified representation of the amplitude and phase information of the signal, and adaptive optimization of dictionary parameters is performed.
It improves the accuracy and robustness of bearing fault diagnosis under variable speed conditions, reduces computational complexity, enhances the ability to characterize non-stationary impact features, and improves the accuracy and adaptability of fault identification.
Smart Images

Figure CN123116046A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of mechanical fault diagnosis and signal processing technology, and more specifically, to a bearing fault diagnosis method based on position constraint-based complex domain adaptive dictionary learning under variable speed conditions. Background Technology
[0002] In rail transit equipment and rotating machinery systems, bearings are key components, and their operating status directly affects the safety and reliability of the system. Traditional methods for bearing fault diagnosis primarily rely on time-domain, frequency-domain, or time-frequency-domain analysis of vibration signals, extracting amplitude or energy characteristics to identify faults. However, in actual operation, especially during high-speed train starts, stops, acceleration, and deceleration, bearings operate at variable speeds, exhibiting significant non-stationary characteristics in their vibration signals. These characteristics include: (1) the fault impact interval dynamically changes with the rotational speed; (2) the impact response exhibits amplitude and frequency modulation phenomena; and (3) the impact signal is accompanied by significant phase drift. These characteristics make it difficult for traditional analysis methods based on the stationary assumption to accurately extract fault features. In recent years, dictionary learning and sparse representation methods have been widely applied in bearing fault diagnosis due to their ability to adaptively extract transient impact structures. While existing dictionary learning methods based on the real number domain can achieve sparse representation of impact signals to a certain extent, they still have the following shortcomings: (1) They only utilize amplitude information and ignore phase information, making it difficult to characterize the complete features of the impact structure; (2) They search in the entire time domain, resulting in high computational complexity; (3) They are susceptible to noise interference, leading to atomic mismatches; and (4) They have difficulty in impact alignment under variable speed conditions. Therefore, it is urgent to propose a novel fault feature extraction method that can simultaneously utilize amplitude and phase information, effectively constrain the impact search region, and has the ability to adaptively optimize parameters, so as to improve the accuracy and robustness of bearing fault diagnosis under variable speed conditions. Summary of the Invention
[0003] The present invention provides a bearing fault diagnosis method based on position constraint-based complex domain adaptive dictionary learning under variable speed conditions. This method can solve the problems of unstable feature extraction, insufficient utilization of phase information, poor noise resistance, and parameter dependence under variable speed conditions in existing bearing fault diagnosis methods. This improves the accuracy, robustness, and computational efficiency of bearing fault impact modeling under variable speed conditions.
[0004] According to the present invention, a bearing fault diagnosis method based on position constraint-based complex domain adaptive dictionary learning under variable speed conditions includes the following steps: Step 1: Perform Hilbert transform on the collected bearing vibration signal to construct an analytic signal, realizing the conversion of the signal from the real domain to the complex domain; Step 2: Calculate the instantaneous energy of the signal using the Teager energy operator, extract candidate impact points through piecewise adaptive thresholding, and expand the time window of the candidate impact points to construct a binary candidate mask function. Step 3: Construct a complex domain dictionary learning model under the constraint of the candidate mask function, and represent the signal as a complex convolution sparse form. The dictionary atoms are initialized with an exponentially decaying complex exponential function, and the impact intensity and phase shift are described simultaneously by complex sparse coefficients. Step 4: Use the local orthogonal matching pursuit algorithm to perform sparse encoding within the constrained region of the candidate mask function, and iteratively update the dictionary atoms by minimizing the reconstruction error to achieve adaptive learning of the complex field dictionary parameters; Step 5: Construct a complex domain Rényi entropy evaluation index within the atomic length and sparsity parameter space, and combine it with structural stability constraints based on entropy change rate to achieve adaptive optimization selection of dictionary parameters; Step 6: Reconstruct the signal using the optimized dictionary model, and extract the instantaneous fault characteristic frequency through time-frequency analysis to achieve accurate identification of bearing faults under variable speed conditions.
[0005] Preferably, in step 1, the analyzed signal is represented as: (1-1) Preferably, in step 2, for the signal The discrete form of the Teager energy operator is defined as shown in equation (1-2): (1-2) Under variable speed conditions, the signal energy level varies with time. A regional threshold is used to divide the signal into several time periods. Within each period, a threshold is adaptively constructed based on the local statistical characteristics of the TEO (Transient Energy Occurrence). The TEO sequence is divided into multiple time periods. The threshold corresponding to the energy sequence within a time period As shown in equation (1-3): (1-3) in, The mean of the energy sequence; The standard deviation of the energy sequence; The coefficient; when satisfying It's time, the corresponding moment. Selected as candidate impact points, thus obtaining a discrete candidate point set. ; After setting candidate points, based on the candidate point set It is further constructed into a binary mask function M(n) as shown in equation (1-4): (1-4) Simultaneously, a time window expansion operation is applied to the candidate points: for each candidate point Expand forward and backward The sampling points allow the mask to cover the entire impact response range.
[0006] As a preferred option, in step 3, the complex field dictionary learning model is: (1-5) in: Signals indicating reconstruction; For length is The complex number of atoms; The sparse coefficients are complex numbers. It is a sparse set of activation locations; This indicates the translation of dictionary atoms along the time axis; No. Complex coefficients Represented as: (1-6) In the formula: Indicates impact strength. This parameter characterizes the local phase shift of the impact response relative to the dictionary atoms, and the phase inconsistency of transient impacts can be adjusted by this parameter; Initialize dictionary atoms to complex exponential forms with exponential decay modulation, complex atoms for: (1-7) in, This is the equivalent damping factor; It is the structural resonance angular frequency.
[0007] As a preferred option, combine candidate impact mask functions By introducing positional prior constraints into the complex field dictionary structure, sparse coefficients are restricted to activation only within candidate impact regions: (1-8) The sparse coefficient matrix X corresponding to the complex field dictionary is as follows: (1-9) The optimization function for adaptive structure dictionary learning in complex fields is: (1-10) in: Represents analytic signals in the complex field; Represents the basis functions in a complex number dictionary; Describes the Euclidean norm of the complex field; Indicates the number of non-zero elements; The target sparsity corresponds to the number of impulse responses in the signal.
[0008] As a preferred option, in the dictionary structure, a dictionary of complex fields... and sparse coefficient matrix It is non-convex, and an alternating optimization strategy is used to solve it, which is divided into two parts: Step 1: Fix the dictionary of complex fields Fixed within the mask constraint, the sparse coefficient matrix is solved and updated. ; Step 2: Fix the sparse coefficient matrix Update the complex field dictionary atoms .
[0009] Preferably, in step 4, during the sparse coding stage, a local orthogonal matching pursuit algorithm is used to search for the optimal atom and translation position within the candidate impact region; and during the dictionary update stage, the complex atoms are updated by minimizing the reconstruction error, and constraints are applied to each atom, as shown in the following expression: (1-11) Preferably, in step 5, the complex field Rényi entropy is introduced as a parameter evaluation index, assuming a given atomic length and sparsity parameters Under the condition of sparse coefficient matrix of complex field structure dictionary , No. The coefficients are represented as According to equation (1-6), it is normalized to a complex probability distribution p. k : (1-12) And satisfy: (1-13) In the formula, Represents the complex conjugate operation; Rényi entropy of complex field of order Defined as follows: (1-14) The Rényi entropy in the complex field is a measure of uncertainty within the framework of complex probability; when When the complex field Rényi entropy degenerates into the complex field Shannon entropy; when When the complex field Rényi entropy degenerates into the complex field Collision entropy, it focuses more on the energy concentration of the probability amplitude distribution, highlighting the dominant component; when When the complex field Rényi entropy degenerates into the complex field Hartley entropy, it reflects the number of possible states of the probability distribution and emphasizes sparsity or the size of the state space.
[0010] As a preferred approach, based on the Rényi entropy in the complex field, a structural stability index based on the rate of entropy change is introduced to characterize the stability of the sparse representation structure as the sparsity parameter changes; with a fixed atomic length... Under the condition, adjacent sparsity parameters Rényi entropy change rate in the complex field for: (1-15) in, Indicates fixed Time Rényi entropy value for complex fields with sparsity of _ ... Indicates fixed Time Rényi entropy values for complex fields with sparsity; and construct structural stability indices. : (1-16) in, The maximum entropy change rate at the corresponding atomic length is used for normalization; when The closer the value is to 1, the more stable the sparse representation structure becomes as the sparsity changes. When the structural stability index meets: (1-17) in, The sparse representation structure has entered a stable region at a preset threshold; therefore, the sparsity parameter that first satisfies the above equation is taken as the corresponding atomic length. optimal sparsity ; Obtaining the length of each atom Corresponding stable sparsity Subsequently, under stable structural conditions, the complex field Rényi entropy is used as a global evaluation index for the atomic length parameter; that is, in the parameter... Calculate the Rényi entropy value for the corresponding complex field and select the atom length with the smallest entropy value. : (1-18) Finally, the joint optimal parameter combination is obtained. .
[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) By constructing a complex domain analytical signal model, this invention integrates the amplitude and phase information of vibration signals into the analysis framework. Compared with the traditional method that only models based on the amplitude in the real domain, it can more completely characterize the fault impact response characteristics, thereby improving the impact alignment accuracy and weak fault identification capability.
[0012] (2) This invention introduces a candidate impulse pre-selection mechanism based on the Teager energy operator and constructs a position constraint mask by extending the time window, so that dictionary atoms are matched only in high-confidence impulse regions. Compared with the traditional full-time domain search method, it can effectively reduce computational complexity, reduce pseudo-impulse interference caused by noise, and improve the accuracy and robustness of sparse representation.
[0013] (3) This invention achieves adaptive optimization of dictionary atom length and sparsity parameters by constructing a joint evaluation mechanism of complex domain Rényi entropy and structural stability. Compared with traditional methods that rely on empirical parameter settings, this invention can automatically select the optimal parameters based on signal structural characteristics, thereby improving the stability and adaptability of fault feature extraction.
[0014] (4) This invention achieves stable extraction and adaptive evolution of fault impact structures under variable speed conditions through the synergistic effect of position prior constraints, sparse modeling in the complex domain and structural entropy optimization, which can effectively enhance the characterization ability of non-stationary impact features under complex noise background. Attached Figure Description
[0015] Figure 1 This is a flowchart of a bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex field under variable speed conditions, as described in one embodiment. Detailed Implementation
[0016] To further understand the content of this invention, a detailed description of the invention will be provided in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.
[0017] Example like Figure 1 As shown, this embodiment provides a bearing fault diagnosis method based on position constraint-based complex domain adaptive dictionary learning under variable speed conditions, which includes the following steps: Step 1: Perform Hilbert transform on the collected bearing vibration signal to construct an analytic signal, realizing the conversion of the signal from the real domain to the complex domain, thereby simultaneously representing the amplitude and phase information of the signal within a unified framework.
[0018] In step 1, the analytic signal is represented as: (1-1) Step 2: Calculate the instantaneous energy of the signal using the Teager energy operator, and extract candidate impact points by segmented adaptive thresholding. Extend the time window of the candidate impact points to construct a binary candidate mask function, thereby limiting the possible region of the impact response and narrowing the sparse search space.
[0019] For signals The discrete form of the Teager energy operator is defined as shown in equation (1-2): (1-2) Under variable speed conditions, the signal energy level changes with time. A regional threshold is used to divide the signal into several time periods. Within each period, a threshold is adaptively constructed based on the local statistical characteristics of the TEO (Transient Electron Oscillator), thereby achieving local candidate point detection. The local statistical characteristics are constructed using segmented mean and standard deviation, effectively avoiding false negatives and false negatives caused by speed variations. The TEO sequence is divided into multiple time periods... The threshold corresponding to the energy sequence within a time period As shown in equation (1-3): (1-3) in, The mean of the energy sequence; The standard deviation of the energy sequence; The coefficient; when satisfying It's time, the corresponding moment. Selected as candidate impact points, thus obtaining a discrete candidate point set. .
[0020] After setting candidate points, based on the candidate point set It is further constructed into a binary mask function M(n) as shown in equation (1-4): (1-4) Meanwhile, considering that actual impact responses typically have a finite duration, including a rise phase and a decay phase, this embodiment applies a time window expansion operation to the candidate points: for each candidate point Expand forward and backward The sampling points allow the mask to cover the entire impact response range.
[0021] Step 3: Construct a complex domain dictionary learning model under the constraint of the candidate mask function, and represent the signal as a complex convolution sparse form. The dictionary atoms are initialized with an exponentially decaying complex exponential function to characterize the damped oscillation characteristics of the bearing impact. The impact intensity and phase shift are described simultaneously through complex sparse coefficients to achieve alignment and accurate modeling of the impact response.
[0022] An adaptive dictionary structure for the complex domain is constructed within the framework of analytic signals. This dictionary structure is based on complex domain dictionary atoms and achieves a unified representation of the amplitude and phase changes of fault signals by modeling atoms in the complex domain.
[0023] Signals can be represented in complex form. In this embodiment, a complex convolutional sparse structure atom is constructed, and the complex domain dictionary learning model is as follows: (1-5) in: Signals indicating reconstruction; For length is The complex number of atoms; The sparse coefficients are complex numbers. It is a sparse set of activation locations; This indicates the translation of dictionary atoms along the time axis; No. Complex coefficients Represented as: (1-6) In the formula: Indicates impact strength. This parameter characterizes the local phase shift of the impact response relative to the dictionary atoms. The inconsistency of transient impact phases can be adjusted by this parameter, thus avoiding the problem of pulse alignment difficulties in the real domain model. Initialize dictionary atoms to complex exponential forms with exponential decay modulation, complex atoms for: (1-7) in, This is the equivalent damping factor; It is the structural resonance angular frequency.
[0024] Combined with candidate shock mask function By introducing positional prior constraints into the complex field dictionary structure, sparse coefficients are restricted to activation only within candidate impact regions: (1-8) The sparse coefficient matrix X corresponding to the complex field dictionary is as follows: (1-9) The optimization function for adaptive structure dictionary learning in complex fields is: (1-10) in: Represents analytic signals in the complex field; Represents the basis functions in a complex number dictionary; Describes the Euclidean norm of the complex field; Indicates the number of non-zero elements; The target sparsity corresponds to the number of impulse responses in the signal.
[0025] In dictionary structure, complex field dictionary and sparse coefficient matrix It is non-convex, and an alternating optimization strategy is used to solve it, which is divided into two parts: Step 1: Fix the dictionary of complex fields Fixed within the mask constraint, the sparse coefficient matrix is solved and updated. ; Step 2: Fix the sparse coefficient matrix Update the complex field dictionary atoms .
[0026] Step 4: Use the local orthogonal matching pursuit algorithm to perform sparse encoding within the constrained region of the candidate mask function, and iteratively update the dictionary atoms by minimizing the reconstruction error to achieve adaptive learning of the complex field dictionary parameters.
[0027] In the sparse coding stage, a local orthogonal matching pursuit algorithm is used to search for the optimal atom and translation position within the candidate impact region, thereby significantly reducing computational complexity. In the dictionary update stage, the complex atoms are updated by minimizing the reconstruction error, and constraints are imposed on each atom to avoid scale uncertainty issues. The expression is shown below: (1-11) Step 5: Construct a complex domain Rényi entropy evaluation index within the atomic length and sparsity parameter space, and combine it with structural stability constraints based on entropy change rate to achieve adaptive optimization selection of dictionary parameters.
[0028] This embodiment introduces complex-domain Rényi entropy as a parameter evaluation index. Unlike traditional entropy indices based on real-valued energy distribution, complex-domain Rényi entropy not only considers the amplitude clustering characteristics of sparse coefficients but also further incorporates phase information. This allows it to simultaneously characterize the consistency of fault features in both energy distribution and phase structure, making it more suitable for variable-speed bearing fault signals with significant frequency modulation and phase drift characteristics. Let's assume a given atomic length... and sparsity parameters Under the condition of sparse coefficient matrix of complex field structure dictionary , No. The coefficients are represented as According to equation (1-6), it is normalized to a complex probability distribution p. k : (1-12) And satisfy: (1-13) In the formula, Represents the complex conjugate operation; The entropy of the complex field of order Rényi is defined as follows: (1-14) The Rényi entropy in the complex field is a measure of uncertainty within the framework of complex probability; when When the complex field Rényi entropy degenerates into the complex field Shannon entropy; when When the complex field Rényi entropy degenerates into the complex field Collision entropy, it focuses more on the energy concentration of the probability amplitude distribution, highlighting the dominant component; when When the complex field Rényi entropy degenerates into the complex field Hartley entropy, it reflects the number of possible states of the probability distribution and emphasizes sparsity or the size of the state space.
[0029] From the perspective of sparse representation, the Rényi entropy in the complex domain can simultaneously reflect the concentration of the amplitude distribution and the consistency of the phase structure of the sparse coefficients. When the dictionary parameters are set appropriately, the sparse representation in the complex domain will exhibit a small number of atomic responses with high amplitudes and stable phase structures, at which point the Rényi entropy in the complex domain will take a smaller value; conversely, if the parameters are set improperly, the distribution of sparse coefficients tends to be scattered or the phase is disordered, and the corresponding entropy value will increase significantly.
[0030] Based on the above analysis, this embodiment uses the extension of the Rényi entropy in the complex field as the parameter combination. The evaluation metric achieves adaptive selection of dictionary atom length and sparsity parameters by searching for regions in the parameter space that minimize entropy and have stable structures.
[0031] However, the entropy minimization criterion is susceptible to local fluctuations and noise. To further characterize the stability of the sparse representation structure as the sparsity parameter changes, a structural stability index based on the rate of entropy change is introduced on top of the Rényi entropy in the complex field. This index is used to characterize the stability of the sparse representation structure as the sparsity parameter changes. This is done while maintaining a fixed atom length. Under the condition, adjacent sparsity parameters Rényi entropy change rate in the complex field for: (1-15) in, Indicates fixed Time Rényi entropy value for complex fields with sparsity of _ ... Indicates fixed Time Rényi entropy values for complex fields with sparsity; and construct structural stability indices. : (1-16) in, The maximum entropy change rate at the corresponding atomic length is used for normalization; when The closer the value is to 1, the more stable the sparse representation structure becomes as the sparsity changes. When the structural stability index meets: (1-17) in, A preset threshold (usually 0.95~0.99) is set, at which point the sparse representation structure has entered a stable region; therefore, the sparsity parameter that first satisfies the above equation is taken as the corresponding atomic length. optimal sparsity ; Obtaining the length of each atom Corresponding stable sparsity Subsequently, under stable structural conditions, the complex field Rényi entropy is used as a global evaluation index for the atomic length parameter; that is, in the parameter... Calculate the Rényi entropy value for the corresponding complex field, and select the atom length with the smallest entropy value: (1-18) Finally, the joint optimal parameter combination is obtained. .
[0032] Step 6: Reconstruct the signal using the optimized dictionary model, and extract the instantaneous fault feature frequency through time-frequency analysis to achieve accurate identification of bearing faults under variable speed conditions. In this embodiment, short-time Fourier transform analysis is performed on the reconstructed signal to generate a time-frequency representation graph, and time-frequency ridges are extracted. Calculate the instantaneous fault characteristic frequency, compare it with the specific order of the bearing's theoretical fault frequency, and determine the bearing fault type.
[0033] The adaptive parameter selection strategy is as follows: Table 1. Adaptive dictionary parameter selection algorithm based on Rényi entropy and structural stability in the complex field.
[0034] This embodiment can effectively utilize the amplitude and phase information of the signal, and significantly improve the accuracy, robustness and computational efficiency of bearing fault feature extraction under variable speed conditions through position constraints and adaptive parameter optimization.
[0035] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the figures shown are only one embodiment of the present invention; the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.
Claims
1. A bearing fault diagnosis method based on position constraint in complex domain adaptive dictionary learning under variable speed conditions, characterized in that, Includes the following steps: Step 1: Perform Hilbert transform on the collected bearing vibration signal to construct an analytic signal, realizing the conversion of the signal from the real domain to the complex domain; Step 2: Calculate the instantaneous energy of the signal using the Teager energy operator, extract candidate impact points through piecewise adaptive thresholding, and expand the time window of the candidate impact points to construct a binary candidate mask function. Step 3: Construct a complex domain dictionary learning model under the constraint of the candidate mask function, and represent the signal as a complex convolution sparse form. The dictionary atoms are initialized with an exponentially decaying complex exponential function, and the impact intensity and phase shift are described simultaneously by complex sparse coefficients. Step 4: Use the local orthogonal matching pursuit algorithm to perform sparse encoding within the constrained region of the candidate mask function, and iteratively update the dictionary atoms by minimizing the reconstruction error to achieve adaptive learning of the complex field dictionary parameters; Step 5: Construct a complex domain Rényi entropy evaluation index within the atomic length and sparsity parameter space, and combine it with structural stability constraints based on entropy change rate to achieve adaptive optimization selection of dictionary parameters; Step 6: Reconstruct the signal using the optimized dictionary model, and extract the instantaneous fault characteristic frequency through time-frequency analysis to achieve accurate identification of bearing faults under variable speed conditions.
2. The bearing fault diagnosis method based on location-constrained complex domain adaptive dictionary learning under varying operating conditions according to claim 1, characterized in that, In step 1, the analytic signal is represented as: (1-1)。 3. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions as described in claim 2, characterized in that, In step 2, for the signal The definition of the Teager energy operator in discrete form is given by equation (1-2): (1-2); The signal energy level changes with time under variable speed conditions. The signal is divided into several time periods using a region threshold, and a threshold is adaptively constructed in each period according to the local statistical characteristics of the TEO. The TEO sequence is divided into multiple time periods, and the threshold corresponding to the energy sequence in each time period is as formula (1-3): (1-3) (1-3); wherein, is the mean of the energy sequence; is the standard deviation of the energy sequence; is a coefficient; when is satisfied, the corresponding time is selected as a candidate impact point, thereby obtaining a discrete candidate point set ; After the candidate points are set, the candidate point set is used to determine the final point set is further configured to a binary mask function M(n) as in equation (1-4): (1-4); At the same time, a time window expansion operation is applied to the candidate points: for each candidate point , the window is expanded sample points in both directions, so that the mask covers the complete impulse response interval.
4. The bearing fault diagnosis method based on location-constrained complex domain adaptive dictionary learning under varying operating conditions according to claim 3, characterized in that, In step 3, the complex field dictionary learning model is as follows: (1-5); wherein: represents the reconstructed signal; is a complex digital dictionary atom of length ; and is a complex sparse coefficient; is a sparse activation position set; represents a shift of the dictionary atom in the time axis; The first complex coefficient is represented as: (1-6); wherein: represents the impact strength, then the local phase shift of the impact response relative to the dictionary atom is characterized, and the transient impact phase inconsistency is adjusted by this parameter; The dictionary atoms are initialized to the form of exponentially decaying modulated complex exponentials, complex atoms are: (1-7); wherein is the equivalent damping factor; is the structural resonant angular frequency.
5. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions according to claim 4, characterized in that, Combined with candidate shock mask function By introducing positional prior constraints into the complex field dictionary structure, sparse coefficients are restricted to activation only within candidate impact regions: (1-8); The sparse coefficient matrix X corresponding to the complex field dictionary is as follows: (1-9); The optimization function for adaptive structure dictionary learning in complex fields is: (1-10); in: Analytical signals representing the complex field; Represents the basis functions in a complex number dictionary; Describes the Euclidean norm of the complex field; Indicates the number of non-zero elements; The target sparsity corresponds to the number of impulse responses in the signal.
6. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions as described in claim 5, characterized in that, In dictionary structure, complex field dictionary sparse coefficient matrix It is non-convex, and an alternating optimization strategy is used to solve it, which is divided into two parts: Step 1: Fix the dictionary of complex fields Fixed within the mask constraint, the sparse coefficient matrix is solved and updated. ; Step 2: Fix the sparse coefficient matrix Update the complex field dictionary atoms .
7. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions as described in claim 6, characterized in that, In step 4, during the sparse coding stage, a local orthogonal matching pursuit algorithm is used to search for the optimal atom and translation position within the candidate impact region; and during the dictionary update stage, the complex atoms are updated by minimizing the reconstruction error, and constraints are applied to each atom, as shown in the following expression: (1-11)。 8. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions as described in claim 7, characterized in that, In step 5, the Rényi entropy of the complex field is introduced as a parameter evaluation index, assuming a given atomic length... and sparsity parameters Under the condition of sparse coefficient matrix of complex field structure dictionary , No. The coefficients are represented as According to equation (1-6), it is normalized to a complex probability distribution p. k : (1-12); And satisfy: (1-13); In the formula, Represents the complex conjugate operation; Rényi entropy of complex field of order Defined as follows: (1-14); The Rényi entropy in the complex field is a measure of uncertainty within the framework of complex probability; when When the complex field Rényi entropy degenerates into the complex field Shannon entropy; when When the complex field Rényi entropy degenerates into the complex field Collision entropy, it focuses more on the energy concentration of the probability amplitude distribution, highlighting the dominant component; when When the complex field Rényi entropy degenerates into the complex field Hartley entropy, it reflects the number of possible states of the probability distribution and emphasizes sparsity or the size of the state space.
9. The bearing fault diagnosis method based on position constraint-based adaptive dictionary learning in the complex domain under variable speed conditions as described in claim 8, characterized in that, Based on the Rényi entropy in the complex field, a structural stability index based on the rate of entropy change is introduced to characterize the stability of sparse representation structures as the sparsity parameter changes; with a fixed atom length Under the condition, adjacent sparsity parameters Rényi entropy change rate in the complex field for: (1-15); in, Indicates fixed Time Rényi entropy value for complex fields with sparsity of _ ... Indicates fixed Time Rényi entropy values for complex fields with sparsity; and construct structural stability indices. : (1-16); in, The maximum entropy change rate at the corresponding atomic length is used for normalization; when The closer the value is to 1, the more stable the sparse representation structure becomes as the sparsity changes. When the structural stability index meets: (1-17); in, The sparse representation structure has entered a stable region at a preset threshold; therefore, the sparsity parameter that first satisfies the above equation is taken as the corresponding atomic length. optimal sparsity ; Obtaining the length of each atom Corresponding stable sparsity Subsequently, under stable structural conditions, the complex field Rényi entropy is used as a global evaluation index for the atomic length parameter; that is, in the parameter... Calculate the Rényi entropy value for the corresponding complex field and select the atom length with the smallest entropy value. : (1-18); Finally, the joint optimal parameter combination is obtained. .
