Steam turbine regulating valve state evaluation method and system

Abstract

The invention discloses a steam turbine regulating valve state evaluation method, and belongs to the technical field of thermal power plant steam turbine valve state monitoring, and the method comprises the steps: collecting the video stream data and operation data of a to-be-evaluated steam turbine regulating valve rod, recognizing the contour of the valve rod based on a YOLOv5 model according to the video stream data, and outputting the feature points of the valve rod; tracking the two-dimensional coordinates of the feature points of the valve rod to calculate the vibration parameters of the valve rod; the method comprises the following steps: calculating a Pearson correlation coefficient of operation data, screening vibration parameters related to a jam fault to generate a time domain displacement sequence, performing time alignment with the operation data, performing fast Fourier transform to obtain a frequency spectrum, and identifying a dominant frequency component, a frequency multiplication component and a sideband frequency; calculating a frequency deviation and an energy distribution difference based on a preset reference spectrum library; and judging the jam grade according to the frequency deviation and the energy distribution difference. The technical problem that the jamming fault level cannot be accurately quantified by an existing turbine regulating valve state monitoring method is solved.

Description

Technical Field

[0001] The invention relates to a method and system for evaluating the state of a steam turbine regulating valve, and belongs to the technical field of steam turbine valve state monitoring in thermal power plants. Background Art

[0002] Current steam turbine control valve condition monitoring technologies primarily rely on vibration sensor data, temperature monitoring, and manual inspections. Traditional methods identify jamming faults by analyzing the time-domain statistical characteristics or frequency-domain spectral characteristics of valve stem vibration signals. Some studies combine these with operating parameters such as steam pressure and valve opening for comprehensive analysis. For example, alarms are triggered when the peak value of vibration acceleration exceeds a threshold, or the presence of specific frequency harmonics in the spectrum is used to qualitatively assess jamming risk. Furthermore, image monitoring technology is increasingly being applied to equipment condition assessment, but it is limited to rough measurement of valve stem displacement or visual inspection of surface wear, and has yet to be deeply integrated with vibration parameters.

[0003] The core defect of the existing technology is that it is impossible to achieve quantitative graded assessment of jamming faults: traditional vibration analysis relies only on sensor signals, is easily disturbed by operating condition fluctuations, and fails to integrate image data of the actual motion trajectory of the valve stem, resulting in incomplete extraction of jamming features; manual inspections or image monitoring rely on experience and judgment, lack objective quantitative standards, and it is difficult to define the critical state of initial jamming and moderate jamming; existing methods can only trigger alarms through threshold exceeding, and cannot establish a continuous jamming grade classification system such as initial, moderate, and severe based on frequency deviation amplitude and energy distribution differences, which often causes maintenance decision lags or excessive maintenance; at the same time, because the dynamic changes of jamming characteristics under different load conditions are not taken into account, the use of fixed thresholds is prone to misjudgment or missed judgment, further limiting the accuracy of fault diagnosis. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for evaluating the status of a steam turbine control valve. By fusing the vibration characteristics of the video stream with the operating parameters for spectral analysis, the present invention solves the technical problem that the existing method for monitoring the status of a steam turbine control valve cannot accurately quantify the level of jamming faults, thereby realizing a graded assessment of the degree of jamming.

[0005] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions: In a first aspect, the present invention provides a method for evaluating the status of a steam turbine regulating valve, comprising: Collecting video stream data and operating data of a valve stem of a steam turbine control valve to be evaluated, wherein the operating data includes steam pressure data, temperature data, and valve opening data; According to the video stream data, identifying the valve stem contour based on a pre-trained YOLOv5 model and outputting feature points of the valve stem; The optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters. By calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault are screened and the time domain displacement sequence is generated. After time-aligning the time-domain displacement sequence with the operating data, a fast Fourier transform is performed to obtain the spectrum and identify the main frequency component, multiplier components, and sideband frequencies; Based on a preset reference spectrum library, the frequency deviation and energy distribution difference between the main frequency component and the reference main frequency component, the double frequency component and the reference double frequency component, and the sideband frequency and the reference sideband frequency are calculated; The jam level is determined based on the frequency deviation and energy distribution difference.

[0006] Furthermore, when the valve opening is greater than or equal to a preset threshold, video stream data and operation data of the valve stem of the steam turbine control valve to be evaluated are collected. The preset threshold is dynamically adjusted through a fuzzy control algorithm based on the unit load characteristics of the steam turbine to be evaluated.

[0007] Furthermore, the method further includes, after collecting the video stream data of the steam turbine regulating valve to be evaluated, preprocessing the video stream data: Performing ROI (Region of Interest) extraction on the video stream data to obtain valve stem area video stream data; After denoising the video stream data of the valve stem area by using the non-local mean based on adjacent frames, the valve stem edge feature video stream data is extracted by histogram equalization.

[0008] Furthermore, before training the YOLOv5 model, the YOLOv5 model is also improved. The improvement steps include: A feature point detection branch is added after the backbone network of the YOLOv5 model. The feature point detection branch includes two 3×3 convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. A 1×1 convolutional layer is used to identify the valve stem contour and output a heat map of the valve stem's feature points. The original bounding box prediction head of the YOLOv5 model is set as a multi-task detection head that simultaneously outputs 4 bounding box coordinate parameters and N feature point coordinate parameters. The multi-task detection head is used to jointly optimize bounding box regression and feature point detection through a combined loss function to output the feature points of the valve stem.

[0009] Furthermore, the optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters, including: In the first frame of the video stream data, the two-dimensional coordinates of the feature points output by the YOLOv5 model are loaded as the initial tracking point set; Apply the optical flow method frame by frame to calculate the displacement vector of the feature point and perform error checking to obtain the feature point displacement data; Filtering the low-confidence feature point displacement data according to a preset feature point existence probability threshold to obtain filtered feature point displacement data; Based on the filtered characteristic point displacement data, the vibration parameters of the valve stem are calculated, including: The displacement amplitude of the valve stem is calculated by synthesizing the displacement components of each feature point in the x-axis direction and the y-axis direction; Based on the displacement amplitude of the valve stem, the instantaneous velocity is calculated by displacement difference to obtain the vibration velocity of the valve stem; Based on the vibration velocity of the valve stem, the vibration acceleration of the valve stem is obtained by differential calculation.

[0010] Furthermore, by calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault were screened to generate a time domain displacement sequence, including: Perform time alignment processing based on the timestamp of the vibration parameters and the collection time of the operating data; Normalize the valve stem vibration parameters and operating data and map them to the same dimension range; The Pearson correlation coefficient between each vibration parameter and each operating data is calculated. According to the preset correlation coefficient threshold, the vibration parameters that are highly correlated with the jamming fault are screened out to generate a time domain displacement sequence.

[0011] Furthermore, after time alignment of the time domain displacement sequence and the operating data, a fast Fourier transform is performed to obtain the spectrum, and the main frequency component, multiplier components, and sideband frequencies are identified, including: Interpolate and resample the time domain displacement series and operating data according to the timestamp to synchronize the time axis; Apply anti-aliasing filter to pre-process the time domain displacement sequence after interpolation and resampling; Apply the Hanning window function to the preprocessed time domain displacement sequence and perform fast Fourier transform to obtain the spectrum; Locate the spectrum peak point and select the frequency components with amplitude greater than 0.1 times the maximum amplitude as candidate main frequency components; Taking the main frequency component as a reference, detect the frequency components at the integer multiple frequencies of the main frequency component, and take the frequency components at the integer multiple frequencies whose amplitudes are greater than 0.3 times the amplitude of the main frequency component as the frequency multiple components; With the main frequency component as the center, the sideband frequency modulated by the operating parameters near the main frequency component is detected, and the sideband frequency component with an amplitude greater than 0.2 times the amplitude of the main frequency component is taken as the sideband frequency.

[0012] Furthermore, the method for constructing the reference spectrum library includes: Collect historical video stream data and historical operating data of the turbine control valve under four status categories: normal state, initial jamming, moderate jamming, and severe jamming; According to the historical video stream data, identifying the valve stem contour based on a pre-trained YOLOv5 model and outputting feature points of the valve stem; The optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters. By calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault are screened to generate a historical time domain displacement sequence; After the historical time-domain displacement sequence is time-aligned with the historical operating data, a fast Fourier transform is performed to obtain the spectrum, and the reference main frequency component, reference multiplier component, and reference sideband frequency are identified; Based on the benchmark main frequency component, benchmark harmonic component and benchmark sideband frequency, a spectrum feature set is established according to four state categories: normal state, initial jamming, moderate jamming and severe jamming. The spectrum feature boundaries between the normal state and each level of jamming are determined through cluster analysis, and a benchmark spectrum library is constructed that includes the main frequency benchmark range, harmonic energy threshold and sideband modulation index.

[0013] Furthermore, judging the jam level according to the frequency deviation and energy distribution difference includes: When the frequency deviation is less than 0.5 Hz and the energy distribution difference is less than 10%, it is judged to be in a normal state, indicating smooth movement without lag; When the frequency deviation is greater than or equal to 0.5 Hz and less than 2 Hz, and the energy distribution difference is greater than or equal to 10% and less than 30%, it is judged as early stage jamming, indicating occasional jamming and concentrated energy; When the frequency deviation is greater than or equal to 2Hz and less than 5Hz and the energy distribution difference is greater than or equal to 30% and less than 60%, it is judged as moderate jamming, indicating periodic jamming and energy is significantly concentrated in the harmonic frequency band; When the frequency deviation is greater than or equal to 5Hz and the energy distribution difference is greater than or equal to 60%, it is judged as severe jamming, indicating continuous jamming and abnormal energy accumulation leading to spectrum distortion.

[0014] In a second aspect, the present invention provides a steam turbine control valve status assessment system, characterized by being used to implement the steps of the steam turbine control valve status assessment method described in the first aspect.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention uses the YOLOv5 model to identify valve stem feature points and calculates vibration parameters based on the optical flow method. It screens key fault features through the Pearson correlation coefficient and obtains spectral features through fast Fourier transform. Finally, by comparing the frequency deviation and energy distribution differences with the benchmark spectrum library, it can accurately classify the jam level. This breaks through the limitations of traditional technologies that can only make qualitative judgments or single threshold alarms, provides a quantifiable decision-making basis for equipment maintenance, significantly improves the accuracy and predictive ability of status assessment, and realizes quantitative graded assessment of jamming faults.

[0016] 2. The present invention improves the YOLOv5 model and utilizes the feature point detection branch and multi-task detection head to synchronously output the valve stem contour bounding box and the coordinates of key feature points, thereby reducing the feature point positioning error. At the same time, combined with the optical flow displacement tracking and confidence filtering mechanism, the vibration parameters of the valve stem can be accurately calculated. It can still maintain a high feature tracking stability under complex steam disturbance conditions, providing a reliable data source for fault feature extraction.

[0017] 3. This invention utilizes spectrum analysis techniques based on time alignment and anti-aliasing filtering, combined with cluster analysis to construct a reference spectrum library. This approach enables the first dual-dimensional quantitative assessment of frequency deviation and energy distribution differences. By jointly analyzing the main frequency component, harmonic components, and sideband frequencies, it reduces the error in the classification of jamming levels and establishes a direct mapping relationship with maintenance strategies, reducing equipment operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a method for evaluating the status of a steam turbine control valve provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0020] Example 1

[0021] like Figure 1 As shown, this embodiment introduces a method for evaluating the status of a steam turbine control valve, including: Step 1: Collect video stream data and operating data of the valve stem of the steam turbine control valve to be evaluated.

[0022] In the present invention, the operating data includes steam pressure data, temperature data and valve opening data.

[0023] This invention dynamically collects valve stem motion video streams and operational data such as steam pressure, temperature, and valve opening to build a foundation for multi-source heterogeneous data fusion and analysis. The valve opening threshold utilizes a fuzzy control algorithm to dynamically adjust the operating condition range for triggering acquisition based on the unit's load characteristics, avoiding redundant, invalid data. The video stream data captures the dynamic displacement of the valve stem at a high frame rate, while the operational data reflects the external excitation of the valve stem motion by the operating conditions. This coupled data acquisition mechanism synchronously captures mechanical motion characteristics and operating condition parameters, providing complete contextual information for subsequent vibration parameter calculation and fault feature extraction. It is particularly suitable for capturing early signs of jamming under variable load conditions.

[0024] Step 2: Based on the video stream data, identify the valve stem contour based on the pre-trained YOLOv5 model and output the feature points of the valve stem.

[0025] This paper introduces a feature point detection branch using an improved YOLOv5 model. Two 3×3 convolutional layers and a batch normalization-ReLU module are added to the backbone network, ultimately generating a feature point heatmap via a 1×1 convolutional layer. This design enables the model to simultaneously detect objects and locate key points, accurately outputting the valve stem outline bounding box and key feature point coordinates. The multi-task detection head uses a combined loss function to jointly optimize the bounding box regression and feature point detection tasks, significantly improving feature point location accuracy and providing a high-confidence initial point set for subsequent optical flow tracking.

[0026] Step 3: Use the optical flow method to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the vibration parameters of the valve stem.

[0027] This method uses optical flow to capture minute valve stem vibration displacements by calculating the displacement vector field of feature points frame by frame. The coordinates of the feature points in the first frame serve as the initial tracking benchmark, and the Lucas-Kanade algorithm is applied frame by frame to calculate the displacement vector. Tracking failure points are then eliminated through error checking. After filtering low-confidence data, the displacement components of the feature points in the x- and y-axis directions are synthesized, and the valve stem vibration velocity and acceleration parameters are calculated through differential calculation. This process enables non-invasive three-dimensional vibration measurement without the need for contact sensors, making it suitable for high-precision analysis of valve stem dynamic characteristics in high-temperature, high-pressure steam environments.

[0028] Step 4: By calculating the Pearson correlation coefficient between the vibration parameters of the valve stem and the operating data, the vibration parameters related to the jamming fault are screened to generate a time domain displacement sequence.

[0029] This method quantifies the correlation between vibration parameters and operating data using the Pearson correlation coefficient, enabling automatic screening of fault signatures. First, the video stream and operating data are timestamp-aligned and normalized to eliminate dimensional differences. The correlation coefficients between the vibration parameters and operating data, such as steam pressure and valve opening, are then calculated. Parameters with correlation coefficients exceeding a threshold, such as 0.75, are retained to generate a time-domain displacement sequence. This eliminates interference from irrelevant parameters, focusing on vibration signatures strongly associated with jamming, significantly improving the signal-to-noise ratio for subsequent spectrum analysis.

[0030] Step 5: After time-aligning the time-domain displacement sequence with the operating data, perform a fast Fourier transform to obtain the spectrum and identify the main frequency component, harmonic components, and sideband frequencies.

[0031] The present invention applies a Hanning window function to the time-domain displacement sequence and then performs an FFT transform, converting the time-domain vibration signal to the frequency domain. Anti-aliasing filter preprocessing is used to prevent spectrum leakage, locate spectrum peaks, and screen frequency components with amplitudes ≥0.1 times the maximum amplitude as candidate main frequency components. Using the main frequency component as a reference, frequency components at integer multiples of its frequency are detected, such as those with amplitudes ≥0.3 times the main frequency, and sideband frequencies modulated by operating parameters, such as those with amplitudes ≥0.2 times the main frequency. This allows analysis of the periodic impact characteristics and modulation phenomena in valve stem vibration, providing a quantitative basis for the frequency-domain characterization of jamming faults.

[0032] Step 6: Based on the preset reference spectrum library, calculate the frequency deviation and energy distribution difference between the main frequency component and the reference main frequency component, the double frequency component and the reference double frequency component, and the sideband frequency and the reference sideband frequency.

[0033] Based on a pre-defined benchmark spectrum library, this method quantitatively assesses the degree of jamming by comparing the frequency deviation and energy distribution differences of the current spectrum with those of normal and faulty benchmarks. The main frequency deviation reflects the degree of abnormality in the valve stem motion cycle, the energy difference of the harmonic distortion is revealed, and the sideband frequency modulation index characterizes the coupling characteristics of the operating conditions. This method transforms frequency domain characteristics into measurable fault indicators, breaking through the limitations of traditional qualitative judgment and providing a data-driven decision-making basis for jam classification.

[0034] Step 7: Determine the jam level based on the frequency deviation and energy distribution difference.

[0035] This invention uses dual-dimensional thresholds to categorize jam failures into four levels: normal, initial jam, moderate jam, and severe jam. For example, severe jam indicates persistent valve stem sticking and abnormal energy accumulation, leading to spectral distortion. This grading system is directly mapped to maintenance strategies, achieving a closed-loop process from condition monitoring to operation and maintenance decision-making, significantly improving the scientific and economic efficiency of equipment management.

[0036] Example 2

[0037] Based on the same inventive concept as Example 1, this example introduces the implementation steps of a method for evaluating the status of a steam turbine regulating valve, including: Step 1: Collect video stream data and operating data of the valve stem of the steam turbine control valve to be evaluated.

[0038] In this embodiment, when the valve opening is greater than or equal to a preset threshold, video stream data and operating data of the valve stem of the regulating valve of the steam turbine to be evaluated are collected. The preset threshold is dynamically adjusted through a fuzzy control algorithm based on the unit load characteristics of the steam turbine to be evaluated. The operating data includes steam pressure data, temperature data and valve opening data.

[0039] After collecting the video stream data of the steam turbine regulating valve to be evaluated, this embodiment preprocesses the video stream data: Step 1.1: Extract the ROI (Region of Interest) of the video stream data to obtain the valve stem area video stream data.

[0040] Step 1.2: After denoising the video stream data of the valve stem region using the non-local mean based on adjacent frames, the valve stem edge feature video stream data is extracted through histogram equalization.

[0041] Step 2: Based on the video stream data, identify the valve stem contour based on the pre-trained YOLOv5 model and output the feature points of the valve stem.

[0042] Before training the YOLOv5 model, this embodiment further includes improving the YOLOv5 model. The improvement steps include: Step 2.1: Add a feature point detection branch after the backbone network of the YOLOv5 model. The feature point detection branch includes two 3×3 convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. The valve stem contour is identified through a 1×1 convolutional layer and the feature point heat map of the valve stem is output.

[0043] Step 2.2: Set the original bounding box prediction head of the YOLOv5 model to a multi-task detection head that simultaneously outputs 4 bounding box coordinate parameters and N feature point coordinate parameters. The multi-task detection head is used to jointly optimize bounding box regression and feature point detection through a combined loss function to output the feature points of the valve stem.

[0044] Step 3: Use the optical flow method to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the vibration parameters of the valve stem.

[0045] In this embodiment, the optical flow method is used to track the two-dimensional coordinates of the feature points of the valve stem frame by frame to calculate the vibration parameters of the valve stem, including: Step 3.1: In the first frame of the video stream data, load the two-dimensional coordinates of the feature points output by the YOLOv5 model as the initial tracking point set.

[0046] Step 3.2: Apply the optical flow method frame by frame to calculate the displacement vector of the feature point and perform error checking to obtain the feature point displacement data.

[0047] Step 3.3: Filter the feature point displacement data with low confidence according to the preset feature point existence probability threshold to obtain filtered feature point displacement data.

[0048] Step 3.4: Calculate the vibration parameters of the valve stem based on the filtered feature point displacement data, including: Step 3.4.1: Calculate the displacement amplitude of the valve stem by synthesizing the displacement components of each feature point in the x-axis direction and the y-axis direction.

[0049] Step 3.4.2: Based on the displacement amplitude of the valve stem, calculate the instantaneous velocity by displacement difference to obtain the vibration velocity of the valve stem.

[0050] Step 3.4.3: Based on the vibration velocity of the valve stem, obtain the vibration acceleration of the valve stem through differential calculation.

[0051] Step 4: By calculating the Pearson correlation coefficient between the vibration parameters of the valve stem and the operating data, the vibration parameters related to the jamming fault are screened to generate a time domain displacement sequence.

[0052] In this embodiment, the Pearson correlation coefficient between the vibration parameters of the valve stem and the operating data is calculated, and the vibration parameters related to the jamming fault are screened to generate a time domain displacement sequence, including: Step 4.1: Perform time alignment based on the timestamps of the vibration parameters and the acquisition time of the operating data.

[0053] Step 4.2: Normalize the valve stem vibration parameters and operating data and map them to the same dimension range.

[0054] Step 4.3: Calculate the Pearson correlation coefficient between each vibration parameter and each operating data. Based on the preset correlation coefficient threshold, select the vibration parameters that are highly correlated with the jamming fault and generate a time domain displacement sequence.

[0055] Step 5: After time-aligning the time-domain displacement sequence with the operating data, perform a fast Fourier transform to obtain the spectrum and identify the main frequency component, harmonic components, and sideband frequencies.

[0056] In this embodiment, after the time domain displacement sequence is time-aligned with the operating data, a fast Fourier transform is performed to obtain a spectrum, and the main frequency component, the multiplier component, and the sideband frequency are identified, including: Step 5.1: Interpolate and resample the time domain displacement series and the running data according to the timestamp to synchronize the time axis.

[0057] Step 5.2: Apply an anti-aliasing filter to pre-process the time-domain displacement sequence after interpolation and resampling.

[0058] Step 5.3: Apply the Hanning window function to the preprocessed time-domain displacement sequence and perform fast Fourier transform to obtain the spectrum.

[0059] Step 5.4: Locate the spectrum peak points and select frequency components with amplitudes greater than 0.1 times the maximum amplitude as candidate main frequency components.

[0060] Step 5.5: Using the main frequency component as a reference, detect the frequency components at integer multiples of the main frequency component, and take the frequency components at integer multiples of the frequency component whose amplitude is greater than 0.3 times the amplitude of the main frequency component as the multiple frequency components.

[0061] Step 5.6: With the main frequency component as the center, detect the sideband frequency modulated by the operating parameters near the main frequency component, and take the sideband frequency component with an amplitude greater than 0.2 times the amplitude of the main frequency component as the sideband frequency.

[0062] Step 6: Based on the preset reference spectrum library, calculate the frequency deviation and energy distribution difference between the main frequency component and the reference main frequency component, the double frequency component and the reference double frequency component, and the sideband frequency and the reference sideband frequency.

[0063] In this embodiment, the method for constructing the reference spectrum library includes: Step 6.1: Collect historical video stream data of the turbine control valve in four status categories: normal state, initial jamming, moderate jamming, and severe jamming, and historical operating data of the corresponding working conditions.

[0064] Step 6.2: Based on the historical video stream data, identify the valve stem contour based on the pre-trained YOLOv5 model and output the feature points of the valve stem.

[0065] Step 6.3: Use the optical flow method to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters.

[0066] Step 6.4: By calculating the Pearson correlation coefficient between the vibration parameters of the valve stem and the operating data, the vibration parameters related to the jamming fault are screened to generate a historical time domain displacement series.

[0067] Step 6.5: After time-aligning the historical time-domain displacement sequence with the historical operating data, perform a fast Fourier transform to obtain the spectrum and identify the reference main frequency component, reference multiplier component, and reference sideband frequency.

[0068] Step 6.6: Based on the baseline main frequency component, baseline harmonic component, and baseline sideband frequency, establish a spectrum feature set according to the four state categories: normal state, initial jamming, moderate jamming, and severe jamming. Use cluster analysis to determine the spectrum feature boundaries between the normal state and each level of jamming. Build a baseline spectrum library that includes the main frequency baseline range, harmonic energy threshold, and sideband modulation index.

[0069] Step 7: Determine the jam level based on the frequency deviation and energy distribution difference.

[0070] In this embodiment, judging the jam level according to the frequency deviation and energy distribution difference includes: When the frequency deviation is less than 0.5 Hz and the energy distribution difference is less than 10%, it is judged to be in a normal state, indicating smooth movement without lag; When the frequency deviation is greater than or equal to 0.5 Hz and less than 2 Hz, and the energy distribution difference is greater than or equal to 10% and less than 30%, it is judged as early stage jamming, indicating occasional jamming and concentrated energy; When the frequency deviation is greater than or equal to 2Hz and less than 5Hz and the energy distribution difference is greater than or equal to 30% and less than 60%, it is judged as moderate jamming, indicating periodic jamming and energy is significantly concentrated in the harmonic frequency band; When the frequency deviation is greater than or equal to 5Hz and the energy distribution difference is greater than or equal to 60%, it is judged as severe jamming, indicating continuous jamming and abnormal energy accumulation leading to spectrum distortion.

[0071] Example 3

[0072] Based on the same inventive concept as the other embodiments, this embodiment provides a steam turbine control valve status assessment system, which is used to implement the steps of the steam turbine control valve status assessment method described in embodiment 1 or 2.

[0073] For specific functional implementation, please refer to the relevant content in the method of Example 1 and will not be repeated here.

[0074] Example 4

[0075] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the steps of the method of the above-mentioned embodiment 1 or 2 are implemented.

[0076] Example 5

[0077] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions. When the computer instructions are executed by a processor, the steps of the method in the above-mentioned embodiment 1 or 2 are implemented.

[0078] In summary, the present invention uses the YOLOv5 model to identify valve stem feature points and calculates vibration parameters based on the optical flow method, screens key fault features through the Pearson correlation coefficient, and then obtains spectral features through fast Fourier transform. Finally, by comparing the frequency deviation and energy distribution differences of the benchmark spectrum library, the jam level can be accurately divided, breaking through the limitations of traditional technology that can only make qualitative judgments or single threshold alarms, providing a quantifiable decision-making basis for equipment maintenance, significantly improving the accuracy and predictive ability of status assessment, and realizing quantitative graded assessment of jamming faults.

[0079] The present invention improves the YOLOv5 model and utilizes the feature point detection branch and the multi-task detection head to synchronously output the valve stem contour bounding box and the coordinates of the key feature points, thereby reducing the feature point positioning error. At the same time, combined with the optical flow displacement tracking and confidence filtering mechanism, the vibration parameters of the valve stem can be accurately calculated. It can still maintain a high feature tracking stability under complex steam disturbance conditions, providing a reliable data source for fault feature extraction.

[0080] This method, based on spectrum analysis techniques using time alignment and anti-aliasing filtering, combines cluster analysis with a baseline spectrum library to achieve the first dual-dimensional quantitative assessment of frequency deviation and energy distribution differences. By jointly analyzing the main frequency component, harmonic components, and sideband frequencies, it reduces the error in the classification of jamming levels and establishes a direct mapping relationship with maintenance strategies, reducing equipment operation and maintenance costs.

[0081] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0085] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

Claims

1. A method for evaluating the status of a steam turbine regulating valve, characterized in that: include: Collecting video stream data and operating data of a valve stem of a steam turbine control valve to be evaluated, wherein the operating data includes steam pressure data, temperature data, and valve opening data; According to the video stream data, identifying the valve stem contour based on a pre-trained YOLOv5 model and outputting feature points of the valve stem; The optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters. By calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault are screened and the time domain displacement sequence is generated. After time-aligning the time-domain displacement sequence with the operating data, a fast Fourier transform is performed to obtain the spectrum and identify the main frequency component, multiplier components, and sideband frequencies; Based on a preset reference spectrum library, the frequency deviation and energy distribution difference between the main frequency component and the reference main frequency component, the double frequency component and the reference double frequency component, and the sideband frequency and the reference sideband frequency are calculated; The jam level is determined based on the frequency deviation and energy distribution difference.

2. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: When the valve opening is greater than or equal to a preset threshold, video stream data and operation data of the valve stem of the steam turbine control valve to be evaluated are collected. The preset threshold is dynamically adjusted through a fuzzy control algorithm based on the unit load characteristics of the steam turbine to be evaluated.

3. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: The method further includes preprocessing the video stream data after collecting the video stream data of the steam turbine regulating valve to be evaluated: Performing ROI (Region of Interest) extraction on the video stream data to obtain valve stem area video stream data; After denoising the video stream data of the valve stem area by using the non-local mean based on adjacent frames, the valve stem edge feature video stream data is extracted by histogram equalization.

4. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: Before training the YOLOv5 model, the YOLOv5 model is also improved. The improvement steps include: A feature point detection branch is added after the backbone network of the YOLOv5 model. The feature point detection branch includes two 3×3 convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. A 1×1 convolutional layer is used to identify the valve stem contour and output a heat map of the valve stem's feature points. The original bounding box prediction head of the YOLOv5 model is set as a multi-task detection head that simultaneously outputs 4 bounding box coordinate parameters and N feature point coordinate parameters. The multi-task detection head is used to jointly optimize bounding box regression and feature point detection through a combined loss function to output the feature points of the valve stem.

5. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: The optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters, including: In the first frame of the video stream data, the two-dimensional coordinates of the feature points output by the YOLOv5 model are loaded as the initial tracking point set; Apply the optical flow method frame by frame to calculate the displacement vector of the feature point and perform error checking to obtain the feature point displacement data; Filtering the low-confidence feature point displacement data according to a preset feature point existence probability threshold to obtain filtered feature point displacement data; Based on the filtered characteristic point displacement data, the vibration parameters of the valve stem are calculated, including: The displacement amplitude of the valve stem is calculated by synthesizing the displacement components of each feature point in the x-axis direction and the y-axis direction; Based on the displacement amplitude of the valve stem, the instantaneous velocity is calculated by displacement difference to obtain the vibration velocity of the valve stem; Based on the vibration velocity of the valve stem, the vibration acceleration of the valve stem is obtained by differential calculation.

6. The method for evaluating the status of a steam turbine regulating valve according to claim 5, wherein: By calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault are screened to generate a time domain displacement sequence, including: Perform time alignment processing based on the timestamp of the vibration parameters and the collection time of the operating data; Normalize the valve stem vibration parameters and operating data and map them to the same dimension range; The Pearson correlation coefficient between each vibration parameter and each operating data is calculated. According to the preset correlation coefficient threshold, the vibration parameters that are highly correlated with the jamming fault are screened out to generate a time domain displacement sequence.

7. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: After time-aligning the time-domain displacement sequence with the operating data, a fast Fourier transform is performed to obtain the spectrum, and the main frequency component, multiplier components, and sideband frequencies are identified, including: Interpolate and resample the time domain displacement series and operating data according to the timestamp to synchronize the time axis; Apply anti-aliasing filter to pre-process the time domain displacement sequence after interpolation and resampling; Apply the Hanning window function to the preprocessed time domain displacement sequence and perform fast Fourier transform to obtain the spectrum; Locate the spectrum peak point and select the frequency components with amplitude greater than 0.1 times the maximum amplitude as candidate main frequency components; Taking the main frequency component as a reference, detect the frequency components at the integer multiple frequencies of the main frequency component, and take the frequency components at the integer multiple frequencies whose amplitudes are greater than 0.3 times the amplitude of the main frequency component as the frequency multiple components; With the main frequency component as the center, the sideband frequency modulated by the operating parameters near the main frequency component is detected, and the sideband frequency component with an amplitude greater than 0.2 times the amplitude of the main frequency component is taken as the sideband frequency.

8. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: The method for constructing the reference spectrum library includes: Collect historical video stream data and historical operating data of the turbine control valve under four status categories: normal state, initial jamming, moderate jamming, and severe jamming; According to the historical video stream data, identifying the valve stem contour based on a pre-trained YOLOv5 model and outputting feature points of the valve stem; The optical flow method is used to track the two-dimensional coordinates of the valve stem's feature points frame by frame to calculate the valve stem's vibration parameters. By calculating the Pearson correlation coefficient between the valve stem vibration parameters and the operating data, the vibration parameters related to the jamming fault are screened to generate a historical time domain displacement sequence; After the historical time-domain displacement sequence is time-aligned with the historical operating data, a fast Fourier transform is performed to obtain the spectrum, and the reference main frequency component, reference multiplier component, and reference sideband frequency are identified; Based on the benchmark main frequency component, benchmark harmonic component and benchmark sideband frequency, a spectrum feature set is established according to four state categories: normal state, initial jamming, moderate jamming and severe jamming. The spectrum feature boundaries between the normal state and each level of jamming are determined through cluster analysis, and a benchmark spectrum library is constructed that includes the main frequency benchmark range, harmonic energy threshold and sideband modulation index.

9. The method for evaluating the status of a steam turbine regulating valve according to claim 1, wherein: The jam level is determined based on the frequency deviation and energy distribution differences, including: When the frequency deviation is less than 0.5 Hz and the energy distribution difference is less than 10%, it is judged to be in a normal state, indicating smooth movement without lag; When the frequency deviation is greater than or equal to 0.5 Hz and less than 2 Hz, and the energy distribution difference is greater than or equal to 10% and less than 30%, it is judged as early stage jamming, indicating occasional jamming and concentrated energy; When the frequency deviation is greater than or equal to 2Hz and less than 5Hz and the energy distribution difference is greater than or equal to 30% and less than 60%, it is judged as moderate jamming, indicating periodic jamming and energy is significantly concentrated in the harmonic frequency band; When the frequency deviation is greater than or equal to 5Hz and the energy distribution difference is greater than or equal to 60%, it is judged as severe jamming, indicating continuous jamming and abnormal energy accumulation leading to spectrum distortion.

10. A steam turbine regulating valve status assessment system, characterized in that: The method is used to implement the steps of the method for evaluating the status of a steam turbine control valve according to any one of claims 1 to 9.

Images