Stator vibration detection-oriented strip noise removal method in any direction

Abstract

The invention provides a stator vibration detection-oriented strip noise removal method in any direction. The method comprises the following steps of: 1, deploying stator vibration detection equipment and acquiring data; step 2, constructing a variational stripe noise removal model; step 3, solving based on a split Bregman iteration model; and step 4, outputting the denoised stator vibration image u. According to the method, stripe noise in multiple directions can be removed at the same time, and the application range of an existing model is effectively expanded.

Description

Technical Field

[0001] This invention relates to the field of generator set condition monitoring and image processing technology, specifically to a method for removing arbitrary direction strip noise in stator vibration detection images, applicable to noise processing of stator vibration images acquired by laser non-contact measurement or contact vibration measuring instruments. Background Technology

[0002] The stator is the core component of a generator set, responsible for converting mechanical energy into electrical energy through electromagnetic induction. Its characteristics include complex structure, harsh operating environment, and potential vibration hazards. The stator consists of multiple components such as the iron core, windings, and insulation layers, weighing over a thousand tons and exceeding 20 meters in diameter. It withstands high temperatures (temperature rises up to 60°C) and mechanical stresses (such as rotor torque transmission) for extended periods. During operation, the stator may vibrate; abnormal vibration can lead to winding displacement, insulation wear, and even generator shutdown. Therefore, the electrical performance and insulation condition of the stator are crucial. Regular monitoring and maintenance can promptly detect potential faults, ensuring the safe and stable operation of the generator set and protecting the safety of personnel and equipment.

[0003] To address the safety concerns caused by stator frame vibration, a combination of laser non-contact measurement and contact vibration measurement will be used to simultaneously collect and compare data. Dual-modal data will be acquired at a single point in both the vertical and horizontal directions of the stator, thereby achieving the following functions: real-time monitoring of stator vibration parameters (frequency, amplitude, waveform distortion), distinguishing between intrinsic vibration and external interference (electromagnetic interference, noise), providing fault warning thresholds, and supporting optimized operation and maintenance decisions.

[0004] When performing target recognition and scene segmentation for stator vibration detection, it is necessary to remove these noises from the image beforehand; otherwise, it will significantly affect subsequent target detection, recognition, and size measurement. Stripe noise is a periodic or non-periodic bright and dark stripe interference phenomenon in remote sensing imaging systems, caused by a combination of factors, including differences in the response characteristics of multiple detectors (including inconsistencies in gain, offset, and sensitivity), non-uniformity of channel response due to sensor aging or insufficient calibration, abnormal radiation distribution caused by atmospheric scattering and changes in solar altitude angle, and scan line misalignment caused by satellite platform vibration or orbital offset. This noise significantly reduces the radiometric accuracy and geometric fidelity of the image, severely impacting subsequent applications such as quantitative inversion and ground feature classification.

[0005] Strip noise removal methods can be mainly divided into four categories: filtering-based methods, statistical methods, PDE-based optimization methods, and deep learning-based methods. Among them, filtering-based methods suppress strip noise by designing specific filters in the transform domain (such as the Fourier domain, wavelet domain, etc.). The core idea is to use the separability of strip noise in the transform domain for frequency domain filtering (CHEN JS, Destriping CMODIS Data by PowerFiltering[J]. IEEE Transactions on Geoscience and Remote Sensing, 2003, 41(9): 2119-2124). Although these methods are computationally efficient and simple to implement, their performance is heavily dependent on the choice of transform domain and the prior assumption of the periodic distribution of stripes. Moreover, the filtering process often leads to the loss of high-frequency information in the image, resulting in problems such as blurred edges and degraded texture details in the restored image.

[0006] The PDE-based optimization method models the destriping problem as an ill-posed inverse problem and constructs an optimization model by introducing regularization constraints. BOUALI and LADJAL first proposed a MODIS image destriping method based on a Unidirectional Total Variation (UTV) model (BOUALI M, LADJAL S. Toward Optimal Destriping of MODIS Data Using a Unidirectional Variational Model[J]. IEEE Transactions on Geoscience and Remote Sensing, 2011, 49(8): 2924-2935). This method utilizes the directional characteristics of striped noise and achieves noise removal by minimizing the energy function containing the UTV regularization term within a variational framework. Compared with filtering methods, the PDE-based optimization method can better preserve image edge and detail information, but its solution process uses the gradient descent algorithm, which has drawbacks such as sensitivity to initial values ​​and slow convergence speed, usually requiring a large number of iterations to reach a stable solution. Furthermore, existing UTV models can only handle stripe noise in a single direction (horizontal or vertical), and still lack effective solutions for multi-directional mixed stripe noise that may exist in real remote sensing images. These limitations have prompted researchers to further explore more efficient and robust destriping methods. Summary of the Invention

[0007] This invention aims to overcome the shortcomings of existing strip noise removal methods, such as inability to handle multi-directional noise, low solution efficiency, and easy loss of image details. It provides an arbitrary-directional strip noise removal method for stator vibration detection, which achieves efficient removal of multi-directional strip noise while preserving edge and texture information in the stator vibration image, providing reliable data for subsequent vibration parameter analysis.

[0008] To achieve the above-mentioned technical features, the objective of this invention is as follows: a method for removing strip noise in any direction for stator vibration detection, comprising: Step 1, Stator vibration testing equipment deployment and data acquisition: Vibration detection equipment, including a laser Doppler vibrometer, a contact vibration meter, a reflector, a control box, and a storage device, is deployed at key points on the stator base to collect stator vibration image data. One laser Doppler vibrometer is deployed in the vertical and one in the horizontal directions of the stator, and the non-contact measurement distance is ≤1.8m. Step 2, construct a variational destriping noise model: Core model construction: Defined with noisy images and the denoised image The core variational model comprises two key components: a data fidelity term, which is achieved through... Constrain the similarity between the denoised image and the original image; Regularization term: introduced , directional gradient norm , And through penalty factors The suppression intensity in both directions is balanced to meet the requirements for removing strip noise in any direction; Model transformation: To facilitate the solution, auxiliary variables are introduced using the variable splitting method. , This transforms the unconstrained problem into a constrained problem; further, by introducing penalty terms and Bregman variables, it transforms the problem into an unconstrained problem that can be solved iteratively and alternately, laying the foundation for efficient solution. Step 3, Solve the model based on the split Bregman iteration: Initialization: Setting Auxiliary variables , The initial value of the Bregman variable, as well as the number of iterations, the maximum number of iterations, and the iteration stopping error parameter; Iterative solution process: While meeting the iteration conditions (maximum number of iterations not reached and error not less than the threshold), perform the following operations cyclically: Solve Subproblem: Obtain numerical solutions using Gauss-Seidel iteration; renew , Subproblem: Updating auxiliary variables using the shrink operator; Update the Bregman variable to maintain iteration stability; Dynamically adjust regularization parameters Based on the current pixel's 8-neighborhood , The mean directional gradient is updated in real time, enhancing the model's adaptability to local image features; Step 4: Output the denoised stator vibration image u .

[0009] Preferably, the vibration detection equipment further includes a multi-channel noise monitor with a frequency response range of 20Hz-20kHz and an accuracy of ±0.5dB; all equipment is encased in a metal shell and is kept at the same potential as the outer wall of the collector.

[0010] Preferably, in step 1, the control box and storage device are arranged next to the stator base; the storage device adopts solid-state drive or network-attached storage to ensure that it can store vibration data for a long time and has the ability to quickly read and back up. The wiring method needs to be reasonably planned to avoid interference and ensure the stability of data transmission.

[0011] Preferably, step 2 specifically includes: The variational destriping noise model is defined as follows: ; In the formula, g This represents an image with striped noise. u Represents a clear image after stripe noise removal; symbol Indicates the image f calculate x Gradient of direction; Indicates the image f calculate y Gradient of direction, Indicates matrix computation L Norm 1, which is used to compute a matrix X The sum of the absolute values ​​of all elements; Indicates matrix computation L 2-norm, i.e., the matrix norm. X Sum of squares of all elements; This is a regularization parameter used to balance the relationship between the image and the strip noise.

[0012] Preferably, step 2 further includes: Using the variable splitting method, auxiliary variables are introduced. , , The unconstrained problem is split into the following constrained problem: ; Introducing a penalty term transforms the constrained problem in step 2 into the following unconstrained problem: ; In the formula, , , These are penalty factors, all of which are constants greater than 0 and less than 1; Introducing Bregman variables Regarding step 3 The minimization problem is transformed into the following, respectively regarding Alternating Iterative Minimization Problem: ; ; ; ; In the formula, Representing variables The k The result of the second iteration Representing variables The k The result of the next iteration.

[0013] Preferably, step 2 further includes: Using Gauss-Seidel iterations u Subproblem solving, for u When minimizing the subproblem, it satisfies the following about u partial derivatives The following equation is obtained: ; Its Gauss-Seidel numerical iterative solution is: ; right d The sub-problem update method is as follows: ; Update using the shrink operator : ; ; in, ; According to the current pixel ( i , j The position is used to calculate the average gradient of its 8-neighborhood for dynamic parameter updates. : ; in, Gx and Gy They are x direction and y Mean of 8-neighbor gradient in the direction, e It is a natural constant. This is the error.

[0014] Preferably, the penalty factor The value range is 0-1, and it is used for dynamic equilibrium. x direction and y The intensity of strip noise suppression in the direction.

[0015] Preferably, the update formula for the Bregman variable is: ; ;

[0016] Preferably, the x Mean of directional gradient Gx and the mean gradient in the y-direction G y For the current pixel ( i , j All pixels within the 8-neighborhood of ) x The average of the absolute values ​​of the directional gradients y The average of the absolute values ​​of the directional gradients.

[0017] Preferably, in step 3, the following steps are performed iteratively until the stopping condition is met: and ; In the formula, This represents the maximum number of iterations.

[0018] The present invention has the following beneficial effects: 1. This invention breaks through the limitations of single-direction strip noise processing and achieves efficient removal of noise in any direction.

[0019] Existing PDE-based optimization methods (such as the UTV model) can only handle strip noise in a single direction, such as horizontal or vertical, and cannot cope with multi-directional mixed noise that may exist in stator vibration detection (such as oblique strips caused by differences in equipment installation angle and electromagnetic interference direction).

[0020] This invention introduces into the variational model x , y Weighted regularization term (penalty factor) of directional gradient (Dynamically balancing the suppression intensity in two directions) can simultaneously suppress strip noise in any direction, significantly expanding the applicable scope of noise processing, especially suitable for complex noise scenarios in stator vibration detection.

[0021] 2. This invention improves the solution efficiency and solves the problems of slow convergence and sensitivity to initial values ​​in traditional methods.

[0022] Existing PDE-based optimization methods mostly employ gradient descent algorithms for solving problems, which have drawbacks such as slow convergence speed (requiring numerous iterations) and sensitivity to initial values ​​(easily getting trapped in local optima), making it difficult to meet the timeliness requirements of real-time monitoring of stator vibration.

[0023] This invention employs the split Bregman iteration method to solve the model. By splitting variables, updating auxiliary variables, and adjusting Bregman variables, the complex optimization problem is decomposed into subproblems that can be iterated alternately, which greatly improves the convergence speed. At the same time, it avoids the dependence of gradient descent on initial values, resulting in stronger iterative stability and the ability to quickly output denoised results to meet real-time monitoring requirements.

[0024] 3. This invention reduces the loss of image details while balancing noise removal and feature preservation.

[0025] Existing filtering-based methods (such as Fourier domain filtering) are prone to losing high-frequency information in the image when suppressing noise, resulting in blurred edges (such as winding structure boundaries) and textures (such as small vibration waveforms) in the stator vibration image, which affects the accuracy of subsequent vibration parameter (frequency, amplitude) analysis.

[0026] This invention achieves detail retention through the following two points: the regularization term adopts... L 1 norm (not L The 2-norm can better preserve the sharpness of image edges; Dynamically adjust regularization parameters (Based on real-time updates of the gradient mean of the 8-neighborhood of pixels), the model adapts to local image features (such as preserving more details in areas of severe vibration and enhancing suppression in areas of dense noise), and retains key feature information of stator vibration to the maximum extent while removing strip noise.

[0027] 4. Adapts to the complex environment of stator vibration testing, improving testing reliability.

[0028] Stator vibration detection is conducted in harsh environments (high temperature, strong electromagnetic interference), and the acquired images have complex noise types. Existing methods are not sufficiently targeted to this scenario.

[0029] This invention, through an optimized noise removal model, can effectively handle strip noise caused by optical path interference of laser vibrometers, electromagnetic noise of equipment, etc. The output denoised image can more accurately reflect the actual vibration state of the stator, providing reliable data support for subsequent vibration parameter extraction (such as frequency and amplitude) and fault early warning, and indirectly improving the safety of generator set operation.

[0030] In summary, this invention comprehensively overcomes the shortcomings of existing strip noise removal methods through integrated optimization of multi-directional noise processing capabilities, efficient solution algorithms, detail preservation mechanisms, and scene adaptability, providing more accurate, efficient, and reliable technical support for stator vibration detection. Attached Figure Description

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] Figure 1 This is a MODIS striped noise image.

[0033] Figure 2 After 40 iterations Figure 1 The noise reduction results.

[0034] Figure 3 This is an infrared image.

[0035] Figure 4 After 80 iterations Figure 3 The noise reduction results. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] Example 1: See Figure 1-4 A method for removing strip noise in any direction for stator vibration detection includes the following steps: Step 1: Deployment and monitoring of stator vibration detection equipment; To ensure the stator vibration testing equipment operates normally under field conditions and can collect complete vibration data, holes need to be drilled in the stator base to install a laser reflector and a contact vibration meter. Simultaneously, the non-contact measurement distance must be ≤1.8m. Power should be drawn from the wall, and measurement points must be selected from key locations on the stator. Based on a laser Doppler vibration meter (LDV), one point is deployed in the vertical and one in the horizontal directions of the stator, requiring a total of two LDV detectors and adjustable mounting brackets, ensuring the non-contact measurement distance is ≤1.8m. Through precise installation of the reflector (high reflectivity film) and optical path calibration, nanometer-level vibration resolution (displacement resolution at the PM level) is achieved.

[0038] To address the issue of refractive index fluctuations caused by temperature, humidity, air pressure changes, and noise affecting the optical path of non-contact vibration meters, thus interfering with the accuracy of nanometer-level vibration measurements, a laser Doppler vibration meter (resistant to electromagnetic interference and humidity ≥90%) was selected. A multi-channel noise monitor (accuracy ±0.5dB, frequency response range 20Hz-20kHz) was also deployed to ensure measurement accuracy in complex environments. The control box and storage system are located next to the stator base, using high-performance storage devices such as solid-state drives (SSDs) or network attached storage (NAS) to ensure the storage of long-term vibration data and the ability to quickly read and back up data. The cabling is carefully planned to avoid interference and ensure stable data transmission. Considering the impact of electromagnetic interference, all equipment located in the collector room is encased in a metal shell and kept at equipotential with the outer wall of the collector room to reduce the impact of external electromagnetic interference on equipment performance.

[0039] Step 2: Define a variational destriping noise model for the monitored objects and data; ; In the formula, g This represents an image with striped noise. u Represents a clear image after stripe noise removal; symbol Indicates the image f calculate x Gradient of direction; Indicates the image f calculate y Gradient of direction, Indicates matrix computation L Norm 1, which is used to compute a matrix X The sum of the absolute values ​​of all elements; Indicates matrix computation L 2-norm, i.e., the matrix norm. X Sum of squares of all elements; This is a regularization parameter used to balance the relationship between the image and the strip noise.

[0040] Based on this, the variable splitting method is used to introduce auxiliary variables. , , The unconstrained problem is split into the following constrained problem: ; Introducing a penalty term transforms the constrained problem in step 2 into the following unconstrained problem: ; Introducing Bregman variables Regarding step 3 The minimization problem is transformed into the following, respectively regarding Alternating Iterative Minimization Problem: ; ; ; ; In the formula, Representing variables The k The result of the second iteration Representing variables The k The result of the next iteration.

[0041] Using Gauss-Seidel iterations u Subproblem solving, for u When minimizing the subproblem, it satisfies the following about u partial derivatives The following equation is obtained: ; Its Gauss-Seidel numerical iterative solution is: ; right d The sub-problem update method is as follows: ; Update using the shrink operator : ; ; in, ; According to the current pixel ( i , j The position is used to calculate the average gradient of its 8-neighborhood for dynamic parameter updates. : ; in, Gx and Gy They are x direction and y Mean of 8-neighbor gradient in the direction, e It is a natural constant. This is the error.

[0042] Step 3: Based on Step 2, a model algorithm for arbitrary direction strip noise removal method for stator vibration detection is proposed.

[0043] initialization: , Number of iterations Maximum number of iterations Iteration stopping error .

[0044] While and

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] k := k +1; EndWhile The method of this invention can remove strip noise in multiple directions simultaneously, effectively expanding the application scope of existing models.

[0053] The UTV model also incorporates X , Y The direction-weighted regularization term is dynamically adjusted using the mean of the local 8-neighborhood gradient. X , Y Directional weights are used to suppress stripes in different directions, thereby achieving the goal of removing multi-directional stripe noise simultaneously.

Claims

1. A method for removing strip noise in any direction for stator vibration detection, characterized in that, include Step 1, Stator vibration testing equipment deployment and data acquisition: Vibration detection equipment, including a laser Doppler vibrometer, a contact vibration meter, a reflector, a control box, and a storage device, is deployed at key points on the stator base to collect stator vibration image data. One laser Doppler vibrometer is deployed in the vertical and one in the horizontal directions of the stator, and the non-contact measurement distance is ≤1.8m. Step 2, construct a variational destriping noise model: Core model construction: Defined with noisy images and the denoised image The core variational model comprises two key components: a data fidelity term, which is achieved through... Constrain the similarity between the denoised image and the original image; Regularization term: introduced , directional gradient norm , And through penalty factors The suppression intensity in both directions is balanced to meet the requirements for removing strip noise in any direction; Model transformation: To facilitate the solution, auxiliary variables are introduced using the variable splitting method. , This transforms the unconstrained problem into a constrained problem; further, by introducing penalty terms and Bregman variables, it transforms the problem into an unconstrained problem that can be solved iteratively and alternately, laying the foundation for efficient solution. Step 3, Solve the model based on the split Bregman iteration: Initialization: Setting Auxiliary variables , The initial value of the Bregman variable, as well as the number of iterations, the maximum number of iterations, and the iteration stopping error parameter; Iterative solution process: While meeting the iteration conditions (maximum number of iterations not reached and error not less than the threshold), perform the following operations cyclically: Solve Subproblem: Obtain numerical solutions using Gauss-Seidel iteration; renew , Subproblem: Updating auxiliary variables using the shrink operator; Update the Bregman variable to maintain iteration stability; Dynamically adjust regularization parameters Based on the current pixel's 8-neighborhood , The mean directional gradient is updated in real time, enhancing the model's adaptability to local image features; Step 4: Output the denoised stator vibration image u .

2. The method for removing strip noise in any direction for stator vibration detection according to claim 1, characterized in that, In step 1, the vibration detection equipment also includes a multi-channel noise monitor with a frequency response range of 20Hz-20kHz and an accuracy of ±0.5dB; all equipment is encased in a metal shell and is kept at the same potential as the outer wall of the collector.

3. The method for removing strip noise in any direction for stator vibration detection according to claim 1, characterized in that, In step 1, the control box and storage device are arranged next to the stator base; the storage device uses solid-state drives or network-attached storage to ensure that it can store long-term vibration data and has the ability to quickly read and back up. The wiring method needs to be planned reasonably to avoid interference and ensure the stability of data transmission.

4. The method for removing strip noise in any direction for stator vibration detection according to claim 1, characterized in that, Step 2 specifically includes: The variational destriping noise model is defined as follows: ； In the formula, g This represents an image with striped noise. u Represents a clear image after stripe noise removal; symbol Indicates the image f calculate x Gradient of direction; Indicates the image f calculate y Gradient of direction, Indicates matrix computation L Norm 1, which is used to compute a matrix X The sum of the absolute values ​​of all elements; Indicates matrix computation L 2-norm, i.e., the matrix norm. X Sum of squares of all elements; This is a regularization parameter used to balance the relationship between the image and the strip noise.

5. The method for removing strip noise in any direction for stator vibration detection according to claim 4, characterized in that, Step 2 also includes: Using the variable splitting method, auxiliary variables are introduced. , , The unconstrained problem is split into the following constrained problem: ； Introducing a penalty term transforms the constrained problem in step 2 into the following unconstrained problem: ； In the formula, , , These are penalty factors, all of which are constants greater than 0 and less than 1; Introducing Bregman variables Regarding step 3 The minimization problem is transformed into the following, respectively regarding Alternating Iterative Minimization Problem: ； ； ； ； In the formula, Representing variables The k The result of the second iteration Representing variables The k The result of the next iteration.

6. The method for removing strip noise in any direction for stator vibration detection according to claim 5, characterized in that, Step 2 also includes: Using Gauss-Seidel iterations u Subproblem solving, for u When minimizing the subproblem, it satisfies the following about u partial derivatives The following equation is obtained: ； Its Gauss-Seidel numerical iterative solution is: ； right d The sub-problem update method is as follows: ； Update using the shrink operator : ； ； in, ; According to the current pixel ( i , j The position is used to calculate the average gradient of its 8-neighborhood for dynamic parameter updates. : ； in, Gx and Gy They are x direction and y Mean of 8-neighbor gradient in the direction, e It is a natural constant. This is the error.

7. The method for removing strip noise in any direction for stator vibration detection according to claim 6, characterized in that, The penalty factor The value range is 0-1, and it is used for dynamic equilibrium. x direction and y The intensity of strip noise suppression in the direction.

8. The method for removing strip noise in any direction for stator vibration detection according to claim 6, characterized in that, The update formula for the Bregman variable is: ； 。 9. The method for removing strip noise in any direction for stator vibration detection according to claim 6, characterized in that, The x Mean of directional gradient Gx and the mean gradient in the y-direction G y For the current pixel ( i , j All pixels within the 8-neighborhood of ) x The average of the absolute values ​​of the directional gradients y The average of the absolute values ​​of the directional gradients.

10. The operation control method of the arbitrary direction strip noise removal method for stator vibration detection as described in claim 9, characterized in that, In step 3, the following steps are executed iteratively until the stopping condition is met: and ; In the formula, This represents the maximum number of iterations.

Images