Online monitoring system and method for last-stage blade of steam turbine

Abstract

The invention relates to the technical field of online monitoring, in particular to an online monitoring system and method for a last-stage blade of a steam turbine, and the system comprises a data collection module, a preprocessing module, a feature fusion module and an early warning module. The data acquisition module synchronously acquires vibration data, temperature data and stress data through a wet steam corrosion resistant broadband acceleration sensor array, a platinum resistor temperature sensor, a thermal infrared imager and a fiber grating sensor. The feature fusion module is divided into three stages of processing, wherein a harmonic distortion factor, a local overheating intensity coefficient and a rain flow damage cumulant are screened and extracted in the first stage; a second-level space-time convolutional neural network constructs a multi-physics field incidence matrix to generate a failure mode confidence score; and the third-level comparison finite element model outputs a physical verification mark. And the closed-loop self-correction mechanism feeds back the early warning result to update the characteristic parameters, so that an anti-interference enhanced data processing closed loop is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring, and in particular to an online monitoring system and method for a last-stage blade of a steam turbine. Background Art

[0002] As the final energy extraction component of the turbine expansion process, the last-stage blades of the turbine must be designed to adapt to the low-pressure and low-density state of steam. The steam flow capacity is improved by increasing the blade length and optimizing the profile, thereby effectively converting the remaining steam energy into shaft power. Since the kinetic energy of steam is significantly reduced, longer blades increase the airflow contact area and improve the overall efficiency of the turbine. In order to cope with the challenges of high centrifugal stress and wet steam erosion under high-speed rotation, the blade material is usually made of corrosion-resistant high-strength alloy, and precision processing technology is used in the manufacturing process to ensure structural strength and extend service life.

[0003] The key technical pain point faced by online monitoring of the last-stage blades of steam turbines is that under high centrifugal stress and wet steam erosion environments, the real-time and accuracy of sensor data acquisition and algorithm analysis are insufficient, making it difficult to reliably capture early subtle fault characteristics; for example, in vibration monitoring, the complex modal changes caused by the high-speed rotation of the blades may lead to limited resolution of the fast Fourier transform analysis, making it impossible to effectively identify the spectral offset of micron-level fatigue cracks, thereby delaying fault warning and increasing the risk of unplanned downtime. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides an online monitoring system and method for the last-stage blades of a steam turbine, which solves the problem of insufficient reliability in online capture of early characteristics of micro-damage to the last-stage blades under high centrifugal stress and wet steam environment.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0006] In a first aspect, the present invention provides an online monitoring system for a last-stage blade of a steam turbine, comprising:

[0007] Data acquisition module, used to obtain vibration data, temperature data and stress data of the last stage blades of the steam turbine;

[0008] A preprocessing module, configured to process the vibration data, temperature data, and stress data acquired by the data acquisition module, eliminate environmental interference, and output processed signals;

[0009] A feature fusion module is used to extract multi-physical field features from the processed signals output by the preprocessing module, perform fusion analysis, and generate micro-damage feature information;

[0010] An early warning module, configured to execute early warning decisions based on the micro-damage feature information generated by the feature fusion module and output an early warning signal;

[0011] The data acquisition module outputs vibration data, temperature data and stress data to the preprocessing module, the preprocessing module outputs the processed signal to the feature fusion module, and the feature fusion module outputs micro-damage feature information to the early warning module.

[0012] Furthermore, in the steam turbine last stage blade online monitoring system of the present invention, the data acquisition module includes:

[0013] The vibration monitoring unit uses circumferentially distributed acceleration sensors to obtain vibration data;

[0014] Temperature monitoring unit, which uses a platinum resistance temperature sensor and an infrared thermal imager to obtain temperature data;

[0015] Stress monitoring unit, which uses fiber Bragg grating sensors to obtain stress data;

[0016] The vibration monitoring unit outputs vibration data to the preprocessing module, the temperature monitoring unit outputs temperature data to the preprocessing module, and the stress monitoring unit outputs stress data to the preprocessing module.

[0017] Furthermore, in the online monitoring system for the last-stage blades of a steam turbine according to the present invention, the preprocessing module performs integrated empirical mode decomposition and adaptive window length wavelet packet transform on the vibration data, performs an inverse problem regularization algorithm on the temperature data, and performs strain-stress transfer function conversion on the stress data;

[0018] Integrated empirical mode decomposition to separate high-frequency resonance components, and adaptive window length wavelet packet transform to filter basic vibration noise;

[0019] The inverse problem regularization algorithm reconstructs the temperature field, and the strain-stress transfer function is converted to lock the speed phase;

[0020] The preprocessing module outputs the processed signal to the feature fusion module for feature extraction.

[0021] Furthermore, in the steam turbine last-stage blade online monitoring system of the present invention, the feature fusion module is configured as follows:

[0022] The first level of feature screening extracts harmonic distortion factors and modal damping ratio attenuation from vibration data, calculates local overheat intensity coefficients from temperature data, and marks rain flow counting damage accumulation from stress data;

[0023] The second-level spatiotemporal correlation analysis uses a spatiotemporal convolutional neural network to analyze the spatiotemporal coupling relationship between the harmonic distortion factor, local overheating intensity coefficient, and rainflow counting damage accumulation to generate a failure mode confidence score.

[0024] The third level of physical verification compares the failure mode confidence score with the finite element model results and outputs the physical mechanism verification mark;

[0025] Among them, the first-level feature screening outputs the screening results to the second-level spatiotemporal correlation analysis, the second-level outputs the score to the third-level physical verification, and the third-level outputs the verification mark to the early warning module.

[0026] Furthermore, in the online monitoring system for the last-stage blades of a steam turbine according to the present invention, the early warning module includes:

[0027] Passive monitoring mode uses the hidden Markov model to establish a variable working condition baseline library, detects microcrack characteristics or wet steam erosion characteristics from micro-damage feature information, and triggers a four-level early warning signal;

[0028] Active diagnosis mode uses transfer function coherence analysis to inject virtual excitation signals, detect structural stiffness degradation, and output latent damage warnings;

[0029] Among them, the passive monitoring mode outputs early warning signals, and the active diagnosis mode outputs damage warnings, collaboratively responding to different operating states.

[0030] Furthermore, in the online monitoring system for the last-stage blades of the steam turbine according to the present invention, the adaptive window length wavelet packet transform in the preprocessing module and the transfer function coherence analysis in the early warning module cooperate to:

[0031] Coherence analysis of the transfer function of vibration data input filtered by wavelet packet transform to improve signal resolution accuracy;

[0032] The output results of the transfer function coherence analysis are fed back to the hidden Markov model to correct the variable operating condition baseline library; the data flows from the preprocessing module to the early warning module to enhance resistance to wet steam environment interference.

[0033] Furthermore, the steam turbine last stage blade online monitoring system of the present invention has a second-level spatiotemporal correlation analysis that identifies the location of the overheating zone from the spatiotemporal coupling relationship, and a third-level physical verification that uses a verification mark to activate an infrared thermal imager for re-inspection;

[0034] The spatiotemporal correlation analysis outputs the coordinates of the overheating zone to physical verification, and the physical verification outputs a flag that triggers special data collection by the temperature monitoring unit;

[0035] Data flows in a closed loop within the feature fusion module to verify reliability.

[0036] Furthermore, the steam turbine last stage blade online monitoring system of the present invention further includes a closed-loop self-correction mechanism:

[0037] The warning signal output by the warning module is input into the feature fusion module to drive the update of micro-damage feature information; the updated micro-damage feature information is fed back to the warning module to adjust the warning decision;

[0038] Data flows from the early warning module back to the feature fusion module to form an adaptive monitoring loop.

[0039] Furthermore, the steam turbine last stage blade online monitoring system of the present invention further includes a memory and a processor;

[0040] The memory stores program code;

[0041] The processor reads the program code in the memory and executes the functions of the data acquisition module, preprocessing module, feature fusion module and early warning module.

[0042] In a second aspect, the present invention provides a method for online monitoring of a last-stage blade of a steam turbine, which is applied to the above-mentioned last-stage blade online monitoring system of a steam turbine, comprising:

[0043] Step 1: Acquire vibration data, temperature data, and stress data of the last-stage blade of the steam turbine;

[0044] Step 2: Processing the vibration data, temperature data, and stress data acquired by the data acquisition module to eliminate environmental interference and output the processed signals;

[0045] Step 3, extracting multi-physical field features from the processed signal output by the preprocessing module, performing fusion analysis, and generating micro-damage feature information;

[0046] Step 4: Execute warning decision based on the micro-damage feature information generated by the feature fusion module and output a warning signal.

[0047] Beneficial effects of the present invention:

[0048] The beneficial effects of the present invention are reflected in the synergistic mechanism of anti-interference data acquisition, layered signal optimization and multi-level physical constraint fusion, which improves the reliability of micro-damage capture under high centrifugal stress and wet steam environments: the titanium alloy sealed package of the broadband acceleration sensor array resists wet steam corrosion, the root temperature measurement of the platinum resistor and the surface scanning of the infrared thermal imager construct a temperature gradient verification mechanism, the pre-buried fiber Bragg grating sensor captures dynamic strain, and the multi-source data synchronization eliminates environmental interference; the integrated empirical mode decomposition of the preprocessing module separates crack-sensitive high-frequency sub-modes, the adaptive window length wavelet packet transform filters out basic vibration noise, and the strain-stress transfer function conversion peels off the centrifugal stress component. The inverse problem regularization algorithm reconstructs the temperature field to ensure the basic signal quality of feature extraction; the feature fusion module optimizes the signal in three levels: the first level screens and extracts the harmonic distortion factor and the local overheating intensity coefficient; the second level uses a spatiotemporal convolutional network to construct a multi-physical field correlation matrix to identify spatial coupling characteristics; the third level uses physical verification to compare the finite element model output verification mark, breaking through the resolution limitations of traditional single-modal analysis; in dual-mode early warning, passive monitoring captures phase offset characteristics and broadband energy growth, and active diagnosis detects stiffness degradation trends. The early warning results drive the infrared thermal imager for directional re-inspection and feedback to update the feature parameters, significantly reducing the missed reporting rate of micron-level fatigue cracks and the false alarm rate in wet steam environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on the drawings without paying any creative labor.

[0050] Figure 1 A flow chart of a method for online monitoring of last-stage blades of a steam turbine provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings. In order to better understand the purpose of the present invention, the present invention is further described in detail below.

[0052] The present invention provides an online monitoring system for a last-stage blade of a steam turbine, comprising:

[0053] Data acquisition module, used to obtain vibration data, temperature data and stress data of the last stage blades of the steam turbine;

[0054] A preprocessing module, configured to process the vibration data, temperature data, and stress data acquired by the data acquisition module, eliminate environmental interference, and output processed signals;

[0055] A feature fusion module is used to extract multi-physical field features from the processed signals output by the preprocessing module, perform fusion analysis, and generate micro-damage feature information;

[0056] An early warning module, configured to execute early warning decisions based on the micro-damage feature information generated by the feature fusion module and output an early warning signal;

[0057] The data acquisition module outputs vibration data, temperature data and stress data to the preprocessing module, the preprocessing module outputs the processed signal to the feature fusion module, and the feature fusion module outputs micro-damage feature information to the early warning module.

[0058] The data acquisition module acquires vibration data, temperature data, and stress data of the last-stage blades of the steam turbine through an array of accelerometers resistant to wet steam corrosion, platinum resistance temperature sensors, infrared thermal imagers, and fiber Bragg grating sensors. Accelerometers are circumferentially distributed around the impeller rim to capture broadband vibration waveforms, platinum resistance sensors are implanted in blind holes at the blade root to monitor substrate temperature, infrared thermal imagers scan the surface temperature distribution of the blade through a pressure-resistant window, and fiber Bragg grating sensors are embedded in the high-stress area of ​​the blade steam inlet edge to track dynamic strain signals in real time. Multi-source data collaborates to overcome the interference of the wet steam environment on a single sensor and provide raw input for subsequent processing.

[0059] The preprocessing module receives the vibration data, temperature data and stress data output by the data acquisition module and performs targeted signal optimization. It uses integrated empirical mode decomposition to separate the high-frequency resonance components of the vibration data, and combines it with the adaptive window length wavelet packet transform to filter out the basic vibration noise of the unit. It applies the inverse problem regularization algorithm to the temperature data to reconstruct the three-dimensional temperature field, and introduces a dynamic compensation factor for the material thermal conductivity to correct the measurement deviation caused by wet steam. The stress data is converted through the strain-stress transfer function to achieve speed phase locking and remove the main frequency component of the centrifugal stress. The processed signal eliminates the influence of environmental interference, improves data credibility and is output to the feature fusion module.

[0060] The feature fusion module extracts multi-physical field features from the processed signal output by the preprocessing module and performs a three-level fusion analysis. The first-level feature screening extracts the harmonic distortion factor and modal damping ratio attenuation from the vibration data, calculates the local overheating intensity coefficient from the temperature data, and marks the rainflow counting damage accumulation from the stress data. The second-level spatiotemporal correlation analysis uses a spatiotemporal convolutional neural network to mine the spatiotemporal coupling relationship between the harmonic distortion factor, the local overheating intensity coefficient, and the rainflow counting damage accumulation to generate a failure mode confidence score. The third-level physical verification compares the failure mode confidence score with the finite element model simulation results, and outputs a physical mechanism verification mark when the spatial distance between the overheating zone and the vibration anomaly point is less than a threshold and the stress concentration factor exceeds the standard. The three-level analysis is progressive, fusing data features with physical constraints to generate micro-damage feature information, which is output to the early warning module.

[0061] The early warning module executes dual-mode early warning decisions based on the micro-damage feature information output by the feature fusion module. The passive monitoring mode uses a hidden Markov model to establish a variable operating condition baseline library. When the micro-crack feature or wet steam erosion feature exceeds the adaptive threshold, a fourth-level early warning signal is triggered. The active diagnosis mode periodically injects virtual excitation signals, detects the structural stiffness degradation trend through transfer function coherence analysis, and outputs a latent damage warning. The dual modes collaborate to cover different operating states, and the early warning signal drives the three-dimensional digital twin model to calibrate the damage location and activate special re-inspections. In addition, the closed-loop self-correction mechanism feeds the early warning signal output by the early warning module back to the feature fusion module to update the micro-damage feature information, and then flows back to the early warning module to adjust the decision parameters, forming an adaptive monitoring loop from early warning to feature fusion data, continuously improving the reliability of the system under variable operating conditions.

[0062] The vibration monitoring unit features an array of wide-band acceleration sensors, resistant to wet steam corrosion, distributed evenly around the rim of the turbine's final-stage impeller. These sensors are housed in a titanium alloy housing and a polyimide insulation coating, transmitting the original vibration waveform via redundant shielded cables. The temperature monitoring unit combines an embedded platinum resistance temperature sensor with a high-speed infrared thermal imager. The platinum resistance is implanted in a blind hole at the blade root to monitor the base temperature, while the infrared thermal imager scans the blade surface temperature distribution through a pressure-resistant window, establishing a base-to-surface temperature gradient calibration mechanism. The stress monitoring unit incorporates a special fiber Bragg grating sensor, micromachined by femtosecond laser, embedded in the high-stress area of ​​the blade steam inlet edge to capture dynamic strain coupled with centrifugal force and steam flow loads in real time. When these three units output data to the preprocessing module, the vibration monitoring unit transmits a sampling frequency signal of ≥20kHz, the temperature monitoring unit simultaneously outputs temperature field coordinate information, and the stress monitoring unit adds a speed phase marker.

[0063] The preprocessing module performs integrated empirical mode decomposition on the vibration data, decomposing the original waveform into several intrinsic mode function components and isolating the 0.5-2kHz high-frequency resonant submodes. The adaptive window length wavelet packet transform dynamically adjusts the window function length based on unit speed fluctuations, filtering out fundamental vibration noise components below 50Hz. An inverse problem regularization algorithm is applied to the temperature data, and a dynamic compensation factor for the blade material thermal conductivity is used to reconstruct the three-dimensional temperature field, eliminating measurement deviations caused by wet steam condensation. The stress data is converted through a strain-stress transfer function matrix, combined with speed phase markers to strip away the centrifugal stress main frequency component. In the processed signals, the vibration data retains high-frequency resonant characteristics, the temperature data outputs a three-dimensional field distribution diagram, and the stress data locks into the steam flow excitation response frequency band.

[0064] The first-level feature screening in the feature fusion module extracts harmonic distortion factors from preprocessed vibration data to quantify waveform distortion and calculates modal damping ratio attenuation to assess structural energy dissipation characteristics. Local overheating intensity coefficients are calculated from temperature data to calibrate areas of abnormal temperature rise, and rainflow counting damage accumulation is marked from stress data to record cyclic loading history. The second-level spatiotemporal correlation analysis utilizes a spatiotemporal convolutional neural network to establish a spatiotemporal correlation matrix of harmonic distortion factors, local overheating intensity coefficients, and rainflow counting damage accumulation. A cross-domain failure mode confidence score is generated when the spatial distance between the abnormal vibration frequency point and the overheating zone is less than 5 mm and the stress concentration factor exceeds 1.8. The third-level physical verification compares this score with the finite element model simulation results and outputs a physical mechanism verification flag to trigger a targeted re-inspection.

[0065] The early warning module's passive monitoring mode utilizes a hidden Markov model to construct a baseline library for variable operating conditions. When microcracks exhibit a sudden increase in 1 / 3 octave energy with a phase shift exceeding 5°, or when wet steam erosion exhibits a sustained increase in broadband energy below 200 Hz, four warning levels (observe, caution, warning, and danger) are triggered. The active diagnostic mode periodically injects a virtual excitation signal with controllable amplitude, detects the blade structural stiffness degradation gradient through transfer function coherence analysis, and outputs a latent damage warning when no abnormal characteristics are present. The dual modes collaboratively respond to changes in operating status: passive monitoring responds to overt faults in real time, while active diagnosis periodically scans for latent damage.

[0066] Vibration data filtered by the adaptive window length wavelet packet transform in the preprocessing module is input into the transfer function coherence analysis unit of the early warning module, improving the response resolution of the virtual excitation signal. The structural stiffness degradation gradient output by the transfer function coherence analysis is fed back into the hidden Markov model to modify the stiffness threshold parameters in the variable operating condition baseline library. As data flows from the preprocessing module to the early warning module via the feature fusion module, the wet steam noise component suppressed by the wavelet packet transform reduces the transfer function analysis error, forming an enhanced anti-interference link.

[0067] The coordinates of the overheating area output by the second-level spatiotemporal correlation analysis are input into the third-level physical verification unit. When the physical mechanism verification flag is activated, the infrared thermal imager is driven to perform a high-frequency re-inspection of the coordinate area at 30 frames per second. The re-inspected temperature data is returned to the feature fusion module to update the local overheating intensity coefficient, and the verification results are synchronously marked to the failure mode confidence score matrix.

[0068] The four-level warning signals output by the early warning module are fed into the feature fusion module, triggering a mechanism to update micro-damage feature information. The passive warning signal drives the spatiotemporal correlation analysis unit to expand the sampling range of monitoring points, while the active damage warning signal adjusts the frequency band used to calculate the cumulative damage during the first-level feature screening. The updated micro-damage feature information is fed back to the early warning module to adjust the state transition probability parameters of the hidden Markov model.

[0069] The program code stored in the memory includes sensor control instructions for the data acquisition module, algorithm parameter configuration sets for the preprocessing module, a finite element model database for the feature fusion module, and a variable operating condition baseline library for the early warning module. When the processor reads this program code, it schedules the sampling frequency of the vibration monitoring unit, the window length adjustment logic of the preprocessing module, the neural network inference tasks of the feature fusion module, and the virtual excitation signal injection period of the early warning module according to millisecond timing, ensuring the synchronization of the parallel execution of multiple modules.

[0070] When implementing an online monitoring system for the last-stage turbine blades, the data acquisition module includes an array of broadband acceleration sensors distributed uniformly around the rim of the last-stage impeller. These sensors are encapsulated in a titanium alloy housing and a polyimide insulating coating to resist wet steam corrosion. Platinum resistance temperature sensors are embedded in blind holes at the blade root to monitor the base temperature. Simultaneously, a high-speed infrared thermal imager is deployed through the cylinder pressure-resistant window to scan the blade surface, establishing a base-to-surface temperature gradient calibration mechanism. A fiber Bragg grating sensor is embedded in the high-stress area of ​​the blade's steam inlet edge. Its femtosecond laser-micromachined grating structure suppresses hydrogen-induced attenuation and captures the dynamic strain coupled with centrifugal force and steam flow in real time. Three streams of raw data are transmitted to the preprocessing module via redundant shielded cables. The vibration data is sampled at a frequency of ≥20 kHz, the temperature data is labeled with spatial coordinates, and the stress data is synchronized with a key-phase pulse signal.

[0071] The preprocessing module performs integrated empirical mode decomposition on the vibration data, isolating high-frequency resonant submodes between 0.5 and 2 kHz. Wavelet packet transform with an adaptive window length is then used to filter out fundamental vibration noise below 50 Hz. An inverse regularization algorithm is applied to the temperature data, incorporating a dynamic compensation factor for the thermal conductivity of the titanium alloy to reconstruct the three-dimensional temperature field and eliminate measurement deviations caused by wet steam condensation. The stress data is converted through a strain-stress transfer function matrix, combining it with the bond phase signal to remove the dominant frequency component of the centrifugal stress. The processed signal retains the crack-sensitive frequency band in the vibration data, outputs the spatial temperature field distribution in the temperature data, and locks the stress data to the steam flow excitation response band.

[0072] The first-level feature screening in the feature fusion module extracts harmonic distortion factors from processed vibration data to quantify waveform distortion and calculates modal damping ratio attenuation to assess the structural energy consumption state. Local overheating intensity coefficients are calculated from temperature data to identify areas of abnormal temperature rise, and rainflow count damage accumulation is output from stress data to record cyclic loading history. The second-level spatiotemporal correlation analysis utilizes a spatiotemporal convolutional neural network to construct a multi-physics field feature correlation matrix. When the spatial distance between the vibration anomaly and the overheating zone is less than 5 mm and the stress concentration factor exceeds 1.8, a cross-domain failure mode confidence score is generated. The third-level physical verification compares this score with the finite element model simulation results and outputs a physical mechanism verification flag to trigger a targeted re-inspection.

[0073] The early warning module's passive monitoring mode builds a variable-condition baseline library based on a hidden Markov model. A Level 4 warning signal is triggered when a sudden increase in 1 / 3-octave energy accompanied by a phase shift greater than 5°, or a sustained increase in broadband energy below 200 Hz, indicating wet steam erosion, is detected. The active diagnostic mode injects a virtual excitation signal with an amplitude of 0.5 N·m every 10 minutes, detects stiffness degradation gradients through transfer function coherence analysis, and outputs a latent damage warning. Vibration data processed with an adaptive window length wavelet packet transform is fed into the transfer function analysis unit to improve signal resolution. The analysis results are then fed back to modify the hidden Markov model threshold parameters, forming an enhanced anti-interference link.

[0074] When the physical verification flag is activated, the infrared thermal imager performs a 30-frame-per-second recheck of the overheated coordinate area. This recheck data is fed back to the feature fusion module to update the local overheat intensity coefficient. A warning signal is also fed into the feature fusion module to trigger parameter updates: passive warnings expand the sampling range of monitoring points, while active warnings adjust the frequency band for damage accumulation calculation. The updated micro-damage signature information is fed back to the warning module to optimize decision parameters. During system deployment, the memory stores sensor control instructions, a finite element model database, and a variable operating condition baseline library. The processor schedules each module's tasks on a millisecond-by-millisecond basis to ensure real-time capture of micron-level cracks in wet steam environments.

[0075] Second, see Figure 1 The method for online monitoring of the last-stage blades of a steam turbine provided by the present invention is applied to the above-mentioned online monitoring system for the last-stage blades of a steam turbine, comprising:

[0076] Step 1: Acquire vibration data, temperature data, and stress data of the last-stage blade of the steam turbine;

[0077] Step 2: Processing the vibration data, temperature data, and stress data acquired by the data acquisition module to eliminate environmental interference and output the processed signals;

[0078] Step 3, extracting multi-physical field features from the processed signal output by the preprocessing module, performing fusion analysis, and generating micro-damage feature information;

[0079] Step 4: Execute warning decision based on the micro-damage feature information generated by the feature fusion module and output a warning signal.

[0080] This invention uses three technical approaches to collaboratively address the challenge of capturing micro-damage in high centrifugal stress and wet steam environments. The first involves establishing an anti-interference data acquisition system. A titanium alloy-sealed broadband acceleration sensor array is evenly distributed around the rim of the final-stage impeller to resist wet steam corrosion. Platinum resistance temperature sensors are embedded in blind holes at the blade root and scanned through the infrared thermal imager window to establish a temperature gradient calibration. A hydrogen-suppressed fiber Bragg grating sensor is embedded in the high-stress area of ​​the steam inlet. Multi-physics field raw data is transmitted through redundant shielding to eliminate environmental interference, providing high-fidelity input for micro-damage analysis.

[0081] The second stage implements layered signal optimization and physical constraint fusion. The preprocessing module performs integrated empirical mode decomposition on the vibration data to isolate crack-sensitive high-frequency submodes, and then filters out basic vibration noise through adaptive window length wavelet packet transform. A regularization algorithm with thermal conductivity compensation is applied to the temperature data to reconstruct the three-dimensional temperature field. The stress data is subjected to speed phase locking to separate the centrifugal stress component. The feature fusion module processes the optimized signal in three stages: the first stage screens and extracts harmonic distortion factors, overheat intensity coefficients, and rainflow damage accumulation; the second stage uses a spatiotemporal convolutional network to construct a multi-physics correlation matrix to identify spatial coupling features; and the third stage compares finite element models to output physical verification signatures. This algorithmic collaboration overcomes the resolution limitations of traditional single-modal analysis.

[0082] The third layer deploys a dual-mode early warning and closed-loop self-correction mechanism. The passive monitoring mode uses a Hidden Markov Model (HMM) variable-condition baseline library to capture microcrack phase shift characteristics and broadband energy growth from wet steam erosion, triggering a Level 4 early warning. The active diagnostic mode injects a virtual excitation signal to detect stiffness degradation trends. This early warning signal drives the infrared thermal imager to conduct a targeted re-inspection of the target area, while simultaneously feeding feedback to the feature fusion module to update damage calculation parameters.

Claims

1. An online monitoring system for the last stage blades of a steam turbine, characterized in that: include: Data acquisition module, used to obtain vibration data, temperature data and stress data of the last stage blades of the steam turbine; A preprocessing module, configured to process the vibration data, temperature data, and stress data acquired by the data acquisition module, eliminate environmental interference, and output processed signals; A feature fusion module is used to extract multi-physical field features from the processed signals output by the preprocessing module, perform fusion analysis, and generate micro-damage feature information; An early warning module, configured to execute early warning decisions based on the micro-damage feature information generated by the feature fusion module and output an early warning signal; The data acquisition module outputs vibration data, temperature data and stress data to the preprocessing module, the preprocessing module outputs the processed signal to the feature fusion module, and the feature fusion module outputs micro-damage feature information to the early warning module.

2. The on-line monitoring system for the last stage blades of a steam turbine according to claim 1, characterized in that: The data acquisition module includes: The vibration monitoring unit uses circumferentially distributed acceleration sensors to obtain vibration data; Temperature monitoring unit, which uses a platinum resistance temperature sensor and an infrared thermal imager to obtain temperature data; Stress monitoring unit, which uses fiber Bragg grating sensors to obtain stress data; The vibration monitoring unit outputs vibration data to the preprocessing module, the temperature monitoring unit outputs temperature data to the preprocessing module, and the stress monitoring unit outputs stress data to the preprocessing module.

3. The on-line monitoring system for the last stage blades of a steam turbine according to claim 2, characterized in that: The preprocessing module performs integrated empirical mode decomposition and adaptive window length wavelet packet transform on vibration data, performs inverse problem regularization algorithm on temperature data, and performs strain-stress transfer function conversion on stress data; Integrated empirical mode decomposition to separate high-frequency resonance components, and adaptive window length wavelet packet transform to filter basic vibration noise; The inverse problem regularization algorithm reconstructs the temperature field, and the strain-stress transfer function is converted to lock the speed phase; The preprocessing module outputs the processed signal to the feature fusion module for feature extraction.

4. The on-line monitoring system for the last stage blades of a steam turbine according to claim 3, characterized in that: The feature fusion module is configured as follows: The first level of feature screening extracts harmonic distortion factors and modal damping ratio attenuation from vibration data, calculates local overheat intensity coefficients from temperature data, and marks rain flow counting damage accumulation from stress data; The second-level spatiotemporal correlation analysis uses a spatiotemporal convolutional neural network to analyze the spatiotemporal coupling relationship between the harmonic distortion factor, local overheating intensity coefficient, and rainflow counting damage accumulation to generate a failure mode confidence score. The third level of physical verification compares the failure mode confidence score with the finite element model results and outputs the physical mechanism verification mark; Among them, the first-level feature screening outputs the screening results to the second-level spatiotemporal correlation analysis, the second-level outputs the score to the third-level physical verification, and the third-level outputs the verification mark to the early warning module.

5. The on-line monitoring system for the last stage blades of a steam turbine according to claim 4, characterized in that: The early warning module includes: Passive monitoring mode uses the hidden Markov model to establish a variable working condition baseline library, detects microcrack characteristics or wet steam erosion characteristics from micro-damage feature information, and triggers a four-level early warning signal; Active diagnosis mode uses transfer function coherence analysis to inject virtual excitation signals, detect structural stiffness degradation, and output latent damage warnings; Among them, the passive monitoring mode outputs early warning signals, and the active diagnosis mode outputs damage warnings, collaboratively responding to different operating states.

6. The on-line monitoring system for the last stage blades of a steam turbine according to claim 5, characterized in that: The adaptive window length wavelet packet transform in the preprocessing module works in conjunction with the transfer function coherence analysis in the early warning module: Coherence analysis of the transfer function of vibration data input filtered by wavelet packet transform to improve signal resolution accuracy; The output results of the transfer function coherence analysis are fed back to the hidden Markov model to correct the variable operating condition baseline library; the data flows from the preprocessing module to the early warning module to enhance resistance to wet steam environment interference.

7. The on-line monitoring system for the last stage blades of a steam turbine according to claim 6, characterized in that: The second-level spatiotemporal correlation analysis identifies the location of the overheating zone from the spatiotemporal coupling relationship, and the third-level physical verification uses the verification mark to activate the infrared thermal imager for re-inspection; The spatiotemporal correlation analysis outputs the coordinates of the overheating zone to physical verification, and the physical verification outputs a flag that triggers special data collection by the temperature monitoring unit; Data flows in a closed loop within the feature fusion module to verify reliability.

8. The on-line monitoring system for the last stage blades of a steam turbine according to claim 7, characterized in that: Also includes a closed-loop self-correcting mechanism: The warning signal output by the warning module is input into the feature fusion module to drive the update of micro-damage feature information; the updated micro-damage feature information is fed back to the warning module to adjust the warning decision; Data flows from the early warning module back to the feature fusion module to form an adaptive monitoring loop.

9. The on-line monitoring system for the last stage blades of a steam turbine according to claim 8, characterized in that: Also includes memory and processor; The memory stores program code; The processor reads the program code in the memory and executes the functions of the data acquisition module, preprocessing module, feature fusion module and early warning module.

10. A method for online monitoring of a last-stage blade of a steam turbine, applied to the online monitoring system for the last-stage blade of a steam turbine according to any one of claims 1 to 9, characterized in that: include: Step 1: Acquire vibration data, temperature data, and stress data of the last-stage blade of the steam turbine; Step 2: Processing the vibration data, temperature data, and stress data acquired by the data acquisition module to eliminate environmental interference and output the processed signals; Step 3, extracting multi-physical field features from the processed signal output by the preprocessing module, performing fusion analysis, and generating micro-damage feature information; Step 4: Execute warning decision based on the micro-damage feature information generated by the feature fusion module and output a warning signal.

Images