Multi-parameter fusion triggered steam turbine generator unit vibration data recording method

Abstract

The invention relates to the technical field of industrial equipment state monitoring and data processing, and discloses a multi-parameter fusion triggered steam turbine generator unit vibration data recording method, which comprises the following steps: automatically judging whether a unit operates in a working condition steady state or transient state mode according to a macroscopic parameter fluctuation ratio; in the steady-state mode, monitoring is performed by comparing the structural difference between the real-time reference correlation matrix and the dynamic reference correlation matrix; in the transient mode, monitoring is carried out by matching the morphological difference between the real-time parameter track and the standardized template, and data recording is triggered when any monitored difference value exceeds the limit. Therefore, the method can automatically adapt to baseline drift generated by normal aging of the unit so as to avoid false alarm, and at the same time, for multi-parameter cooperative relationship disintegration caused by early failure, high sensitivity is shown so as to avoid missing alarm.

Description

Technical Field

[0001] This invention relates to a method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion, belonging to the field of industrial equipment condition monitoring and data processing technology. Background Technology

[0002] Currently, a common technique used in the operation and maintenance of steam turbine generator sets is to preset fixed safety thresholds for one or more operating parameters, such as vibration and temperature. When the real-time monitored parameter value exceeds the threshold, data recording is automatically triggered. This method provides a necessary monitoring means to respond to sudden faults with drastic changes in parameter amplitude. During operation, the state reference of a steam turbine generator set will continuously and slowly drift due to factors such as wear and aging. When applying the above-mentioned fixed threshold-based monitoring method to such dynamic systems, there are technical constraints between monitoring sensitivity and data recording reliability. If the threshold is tightened to capture early and minor faults, parameter fluctuations generated during normal operating condition switching will trigger a large number of unnecessary records, interfering with subsequent operation and maintenance analysis. If the threshold is relaxed to avoid false alarms, it will lead to a delay in the identification of some early faults.

[0003] To address this issue, existing technologies also employ algorithmic models to define the normal operating range of the unit. These methods still rely on comparing the real-time state with a reference system built from historical data. However, the initial stages of some early faults are not characterized by deviations in a single parameter value, but rather by a deterioration in the physical correlation between multiple operating parameters. Existing methods have limited ability to identify such fault characteristics. Besides the traditional methods based on fixed thresholds for single parameters, some seemingly more advanced diagnostic logics in existing technologies still adhere to the framework of state deviation assessment, thus exhibiting inherent flaws. For example, the Chinese patent with publication number CN110553821A... The invention patent discloses a method and system for visual diagnosis of turbine generator set faults. The core logic of this method is to judge the fault by comparing whether the difference between the real-time vibration vector and the historical reference vibration vector exceeds the limit. However, the essence of this approach is still to evaluate the absolute state of a single physical quantity (vibration). Its monitoring focus is limited to the change of the vibration vector itself. It lacks effective means to identify the deterioration of the inherent synergistic relationship between multiple different types of operating parameters (such as vibration, temperature, and pressure) caused by early faults. When the initial characteristics of the fault are manifested as the disintegration of multi-parameter correlation rather than a deviation of a single vibration vector, this method also faces the risk of missed detection and fails to fundamentally solve the problem.

[0004] In this context, the following technical problems need to be addressed: 1. Existing triggering methods are insufficient in distinguishing between normal operating condition fluctuations and parameter correlation changes caused by early faults, easily leading to false alarms or missed alarms; 2. The monitoring system is not highly adaptable to normal drift of the unit's operating baseline, requiring periodic manual model adjustments; 3. The triggering logic is based solely on parameter values ​​and does not record information on changes in the relationships between parameters, resulting in the recorded data lacking contextual information for fault tracing. Therefore, how to provide a data recording triggering method that can make judgments based on changes in the relationship structure between multiple operating parameters, adapt to normal drift of the unit's operating baseline, and effectively identify the deterioration of the cooperative relationship between parameters caused by early faults, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a method for recording vibration data of steam turbine generator sets by multi-parameter fusion triggering. Its main purpose is to solve the problems of existing data recording triggering methods based on fixed historical models, which are insufficient in adapting to the drift of the unit's operating baseline and in identifying the deterioration of the coordination relationship between parameters caused by early faults.

[0006] To achieve the above objectives, the present invention provides a method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion, comprising the following steps: Step a, Operation mode determination: Continuously monitor at least one macroscopic operating parameter of the steam turbine generator set, and determine the current operation mode as steady-state mode when the volatility of the macroscopic operating parameter within the first preset time window is lower than the first preset threshold; otherwise, determine it as transient mode. Step b, steady-state relationship instability monitoring: if and only if the judgment result of step a is the steady-state mode of the working condition, real-time data of multiple operating parameters are acquired to generate a real-time correlation matrix characterizing their inherent synergistic relationship, and the real-time correlation matrix is ​​compared with a benchmark correlation matrix dynamically updated based on a historical correlation matrix generated under recent historical steady state to determine the first drift value characterizing the structural difference between the two. Step c, transient behavior distortion monitoring: if and only if the judgment result of step a is a transient mode, the event type of the transient mode is identified based on the macroscopic state of the start and end of the transient mode, the preset standardized transient event trajectory template corresponding to the event type is obtained, and the real-time evolution trajectory of multiple operating parameters under the transient mode is matched with the trajectory template in an overall shape to determine the second mismatch value that characterizes the difference in trajectory shape. Step d: Unified data recording trigger. In steady-state operating mode, when the first drift value exceeds a preset first trigger threshold, the vibration data of the turbine generator set is triggered to be recorded; and in transient mode, when the second mismatch value exceeds a preset second trigger threshold, the vibration data of the turbine generator set is triggered to be recorded.

[0007] Preferably, both the real-time correlation matrix and the benchmark correlation matrix are matrices generated by calculating the Pearson correlation coefficients between pairs of multiple operating parameters; and the first drift value is determined by calculating the Frobenius norm between the real-time correlation matrix and the benchmark correlation matrix.

[0008] Preferably, the unified data recording triggering step further includes: if the trigger is due to the first drift value exceeding a preset first triggering threshold, then the real-time correlation matrix, the benchmark correlation matrix, and the first drift value are used as first context information and associated with the recorded vibration data for storage; if the trigger is due to the second mismatch value exceeding a preset second triggering threshold, then the event type standardized transient event trajectory template and the second mismatch value are used as second context information and associated with the recorded vibration data for storage.

[0009] Preferably, in the transient behavior distortion monitoring step, the overall morphological matching is calculated by using a dynamic time warping algorithm to determine the morphological distance between the real-time evolution trajectory and the trajectory template. The morphological distance is the second mismatch value.

[0010] Preferably, the method further includes a long-term baseline degradation early warning step: pre-storing a genesis correlation matrix calibrated when the turbine generator set is in a baseline healthy state and corresponding to a specific operating condition steady state; comparing the dynamically updated baseline correlation matrix with the genesis correlation matrix at a frequency lower than that used to generate the real-time correlation matrix to determine a third cumulative drift value characterizing long-term cumulative baseline degradation; and triggering the recording of vibration data of the turbine generator set when the third cumulative drift value exceeds a preset third trigger threshold.

[0011] Preferably, the method further includes a trigger threshold dynamic compensation step: after generating the real-time correlation matrix, a fingerprint structure entropy value representing the current signal-noise level is calculated based on the internal numerical distribution of the real-time correlation matrix; and the preset first trigger threshold is dynamically adjusted according to the fingerprint structure entropy value, the dynamic adjustment following a rule. ,in, The adjusted first trigger threshold, The preset first trigger threshold is the baseline value before compensation. The fingerprint structure entropy value, This is the preset positive gain coefficient.

[0012] Preferably, in the operation mode determination step, at least one macroscopic operation parameter is selected from at least one of active power and main steam pressure.

[0013] Preferably, the steady-state relationship instability monitoring step and the transient behavior distortion monitoring step include multiple operating parameters, including vibration parameters, temperature parameters and pressure parameters.

[0014] Preferably, the dynamic update of the benchmark correlation matrix is ​​generated by performing an exponentially weighted moving average calculation based on historical correlation matrices generated under one or more historical steady-state operating conditions before the real-time correlation matrix is ​​generated.

[0015] Preferably, in the transient behavior distortion monitoring step, the standardized transient event trajectory template includes a standardized historical data snapshot of multiple operating parameters changing over time or with a dominant macroscopic parameter under healthy conditions.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention discloses a multi-parameter fusion-triggered vibration data recording method for steam turbine generator sets. By monitoring the fluctuation rate of parameters such as active power, a steady-state window for analysis is identified. Within this window, instead of evaluating the values ​​of each operating parameter in isolation, the correlation coefficients between multiple parameters are calculated to generate a real-time steady-state fingerprint characterizing their inherent cooperative relationship structure. This fingerprint is then structurally compared with a benchmark steady-state fingerprint dynamically updated from recent historical steady-state fingerprints. This process shifts the core of the monitoring logic from judging whether absolute values ​​deviate from a fixed historical model to assessing whether the stability of the system's internal relationship network has experienced abnormal disturbances in the short term. Therefore, when normal wear or aging of the unit causes a slow drift in its operating baseline, the dynamically updated benchmark fingerprint smoothly adapts, avoiding false alarms caused by baseline changes. Furthermore, for the disintegration of the cooperative relationship between multiple parameters caused by early faults, even if the absolute values ​​of each parameter are still within the historical range, this method can identify it through abrupt changes in the fingerprint structure, avoiding missed alarms. By using the parallel transient process monitoring steps, the main scheme utilizes the logic used to determine the steady state of the operating conditions to identify the start and end points of transient events such as unit startup and load increase. During this period, the monitoring focus no longer focuses on the static relationship between parameters, but rather on matching the real-time evolution trajectory of multi-dimensional parameters during the transient process with the pre-stored standardized trajectory template representing healthy behavior. In this way, the main scheme focuses on the instability of the relational structure under steady state, while this supplementary step specifically monitors the distortion of behavioral trajectory under transient state. The two complement each other and together constitute a monitoring system covering both steady and transient states of the unit, enabling early faults that damage the system's dynamic response capability and prevent it from entering a steady state to be effectively recorded.

[0017] Furthermore, to avoid the adaptive mechanism of the main scheme losing sensitivity due to the synchronous drift of the reference fingerprint when facing extremely slow-developing cumulative faults, the method claimed in this invention introduces a dual-timescale comparison logic. While retaining the high-frequency comparison between the real-time fingerprint and the recent reference fingerprint to capture sudden anomalies, it adds a low-frequency step of comparing the currently dynamically updated reference steady-state fingerprint with a fixed genesis steady-state fingerprint representing the initial health state of the unit's lifecycle. This design enables the system to simultaneously diagnose short-term sudden relationship instability and provide early warning of long-term cumulative baseline degradation, offering both short-term and long-term dimensions for equipment health assessment. The decision-making basis solves the technical problem that chronic failures may be overlooked due to reference drift in adaptive monitoring. When triggering vibration data recording, the real-time steady-state fingerprint of the reference steady-state fingerprint before triggering and the drift value between the two are stored as context information associated with the vibration data. This information encapsulation method means that the post-event analysts no longer obtain isolated vibration waveforms, but a structured data package containing a chain of evidence of relationship instability. Analysts can directly know which parameters' synergistic relationship was first disrupted. This changes the previous diagnostic method that relied on expert experience for data archaeology and provides a direct logical basis for rapid and accurate fault tracing. Attached Figure Description

[0018] Figure 1 This is the overall flowchart of the dual-mode monitoring and unified data recording triggering of the present invention; Figure 2 This is a diagram showing the logical interaction relationships between the core functional modules of this invention; Figure 3 This is a timing diagram of the interactions between the components in the operation mode determination step of this invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The proposed multi-parameter fusion-triggered vibration data recording method for steam turbine generator sets, in one specific embodiment, is configured as a data processing flow that executes sequentially or in parallel a process consisting of an operating mode judgment step, a steady-state relationship instability monitoring step, a transient behavior distortion monitoring step, and a unified data recording triggering step. The operating mode judgment step guides the data processing flow to either the steady-state relationship instability monitoring step or the transient behavior distortion monitoring step based on the stability of the unit's operating conditions. These two monitoring steps evaluate the system's operating state from two dimensions: the static correlation structure between parameters and the dynamic behavior trajectory, respectively. The unified data recording triggering step then determines the system's operating state based on the evaluation results of any one of the monitoring steps. The process involves recording vibration data and storing relevant contextual information. The identification of transient pattern event types during transient behavior distortion monitoring is accomplished through an offline configured event rule base and an online matching engine. Specifically, for a particular transient event, such as a load increase process from 50% to 75% of rated power, the rules are defined as a logical set including initial trigger conditions, process state constraints, and termination confirmation conditions. The initial trigger condition is set as follows: the unit's active power remains stable within a range of 2% above and below 50% of rated power for 10 consecutive minutes, and the power fluctuation rate first exceeds 0.005 within a subsequent 180-second time window. The process state... The constraints require that the active power increases monotonically, while the termination confirmation condition is that the active power first enters the range of 75% of the rated power plus or minus 2%, and operates stably within this range for more than 5 minutes, i.e., the power fluctuation rate remains below 0.005. When the trajectory of the macroscopic operating parameters monitored online completely satisfies a certain rule in the preset rule base, the system determines that the event type of the current transient process corresponds to that rule. The multiple operating parameters used to generate the correlation matrix in the steady-state relationship instability monitoring step are not randomly combined, but selected according to the preset parameter analysis group. Each parameter analysis group is oriented towards a specific physical subsystem or potential fault mechanism, and its construction follows two basic principles. The first principle is the principle of physical proximity and causal correlation, which means that multiple measuring points on the same physical component, such as the vertical vibration, horizontal vibration, bearing oil temperature, and return oil temperature of the same bearing, are classified into the same analysis group to monitor the degradation of the physical correlation within the component. The second principle is the principle of functional flow correlation, which means that along the flow path of the process medium, the operating parameters of key upstream and downstream sections, such as main steam pressure, high-pressure cylinder exhaust pressure, and reheat steam temperature, are classified into another analysis group to monitor abnormal changes in the overall thermodynamic performance of the unit. During operation, the system can calculate the real-time correlation matrix corresponding to multiple such parameter analysis groups in parallel, thereby realizing distributed monitoring of different functional units of the unit.

[0021] During implementation, step a, the operation mode determination step, is executed first. This step continuously monitors a parameter that characterizes the overall operating condition of the turbine generator set at a sampling frequency of not less than 1Hz; here, active power is selected. Within a first preset time window of 180 seconds, the system collects continuous active power data points and calculates the ratio of the standard deviation to the mean of the data sequence, using this as the quantified value of volatility. The first preset threshold for distinguishing operating modes is determined through an offline calibration process. This process includes recording the active power volatility of the unit under specific transient processes (such as load increases) and multiple typical steady-state operating conditions, and selecting a value that can effectively distinguish the volatility distribution of these two types of operating conditions, for example, 0.005, as the first preset threshold. During online monitoring, if the current time window... If the calculated volatility is below 0.005, the system determines that the unit's current operating mode is a steady-state operating mode; otherwise, it determines it is a transient operating mode. If and only if the result of step a is a steady-state operating mode, the system executes step b, the steady-state relationship instability monitoring step. This step is used to monitor for early signs of failure where individual parameter values ​​are not abnormal, but the coordination relationship between parameters changes. The system simultaneously acquires multiple operating parameters, specifically including the vertical vibration of bearing number one, bearing number one oil temperature, and main steam pressure, using real-time data sequences over the past 60 seconds, and calculates the Pearson correlation coefficients between each pair of these parameters, thus forming a real-time correlation matrix. Simultaneously, the system maintains a benchmark correlation matrix representing the recent health status. This benchmark correlation matrix is ​​dynamically updated using an exponentially weighted moving average (EWMA) algorithm, specifically: Whenever a new real-time correlation matrix is ​​generated, according to the formula Update the baseline matrix, where, This is the updated baseline correlation matrix. For the newly generated real-time correlation matrix, This is the baseline correlation matrix before the update. A smoothing coefficient, ranging from 0 to 1 (e.g., 0.1), is used to update the baseline correlation matrix, allowing it to follow the baseline drift caused by normal aging. To quantify the deviation of the real-time state from the recent baseline, the system calculates the Frobenius norm between the real-time correlation matrix and the baseline correlation matrix. This norm is used as the first drift value, calculated by taking the square root of the sum of the squares of the differences between all corresponding elements of the two matrices. , and The real-time correlation matrix and the benchmark correlation matrix are respectively in the th... line, number The elements of the column are summed by traversing all corresponding elements of the two matrices to obtain a scalar value that represents the difference in the overall structure.

[0022] When the judgment result of step a is transient mode, the system executes step c, namely the transient behavior distortion monitoring step, which aims to evaluate whether the unit's response behavior in the dynamic process deviates from the preset health mode. This step identifies the event type based on the start and end states of the transient mode (e.g., active power changes from 0MW to 500MW) and retrieves the corresponding standardized transient event trajectory template from the preset template library. This template stores standardized historical data of multiple operating parameters (such as speed and exhaust temperature of each cylinder) changing with time or with active power when the unit executes the same type of transient event in a healthy state. To compare the shape of the real-time trajectory with the template trajectory, the system uses the Dynamic Time Warping (DTW) algorithm to calculate the morphological distance between the two time series. This distance value is used as the second mismatch value. The DTW algorithm can match two time series with local differences in rate, thereby allowing for normal rhythm deviations in the actual execution of transient operations, while focusing on the distortion of the overall trajectory shape.

[0023] Finally, step d, the unified data recording triggering step, makes a decision based on the aforementioned monitoring results. In steady-state operating mode, if the calculated first drift value exceeds a preset first trigger threshold, vibration data recording of the turbine generator set is triggered. Simultaneously, the relevant context information for this recording, including the real-time correlation matrix at the moment of triggering, the baseline correlation matrix before triggering, and the first drift value, is associated and stored with the recorded vibration data. In transient mode, if the calculated second mismatch value exceeds a preset second trigger threshold, vibration data recording is also triggered, and the corresponding event type, the matched standardized transient event trajectory template, and the second mismatch value are stored as context information. To further enhance… In this embodiment, the method may further include two additional steps: First, a long-term baseline degradation early warning step, used to identify slowly accumulating performance degradation. This step pre-calibrates and stores a genesis correlation matrix corresponding to a specific operating condition as a health reference throughout the entire lifecycle, under the initial healthy state of the unit. At a low frequency (e.g., once daily), it compares the currently dynamically updated baseline correlation matrix with this genesis correlation matrix to calculate a third cumulative drift value. When this value exceeds a preset third trigger threshold, an early warning data record is triggered. Second, a trigger threshold dynamic compensation step, used to suppress false triggers caused by signal noise fluctuations. This step calculates the fingerprint structure entropy value of the matrix after generating the real-time correlation matrix. This value characterizes the degree of disorder in the numerical distribution within the matrix, and is adjusted according to dynamic rules. Compensation is applied to the first trigger threshold, wherein, The adjusted first trigger threshold, The first trigger threshold is the baseline value before compensation. The fingerprint structure entropy value, The rule is based on a preset positive gain coefficient, which ensures that when the signal noise increases (i.e., ... When the threshold is raised, the trigger value is increased. The corresponding increase will be made.

[0024] Example 1: In a 600MW steam turbine generator unit that has been operating continuously, the unit is being upgraded from a 450MW load point to a target load point of 500MW according to dispatch instructions. Under this condition, the oil film of the No. 2 bearing of the unit has deteriorated prematurely due to long-term operation, and its dynamic characteristics have changed. However, the absolute values ​​of the bearing vibration and temperature measured by the sensors do not exceed the normal range of historical data, nor do they reach the fixed alarm limit. After the active power of the unit begins to change, if the operation mode judgment step of the specific implementation method is executed, the system will judge the current operation mode as transient mode because it detects that the fluctuation rate of active power within a 180-second time window exceeds the first preset threshold of 0.005.

[0025] During this period, the steady-state relationship instability monitoring step is not activated, thus avoiding unnecessary data recording that may occur during the normal fluctuation of the load adjustment parameter. When the unit's active power stabilizes at 500MW, the operation mode judgment step, upon detecting that its volatility has fallen below 0.005, determines the current operation mode as a steady-state mode and activates the steady-state relationship instability monitoring step. This step collects real-time data of multiple operating parameters, including the vertical vibration of bearing No. 2 and oil temperature, and calculates and generates a real-time correlation matrix. Due to the early deterioration of the oil film and the weakening of the physical correlation between vibration and temperature signals, this change is reflected in the changes of the corresponding element values ​​in the real-time correlation matrix. The system then uses this real-time correlation matrix and a benchmark, dynamically updated using an exponentially weighted moving average algorithm, representing the unit's recent health status. The correlation matrices were compared, and the first drift value obtained by calculating the Frobenius norm between the two exceeded the preset first trigger threshold due to the weakening of the correlation. The system triggered the recording of the vibration data of the turbine generator set and stored the real-time correlation matrix at the trigger time, the baseline correlation matrix before the trigger, and the first drift value as context information associated with the recorded vibration data. When the operators analyzed the record, they knew from the context information that the reason for the data recording was that the synergistic relationship between the vibration and temperature parameters of the No. 2 bearing had deviated, even though the absolute values ​​of these two parameters themselves did not exceed the limits. Based on this information, the maintenance personnel pointed the inspection direction to the physical components affecting the correlation between vibration and temperature, namely the state of the bearing oil film, and finally confirmed and dealt with the early fault, stopping the further development of the fault.

[0026] Example 2: To verify the effectiveness of the method of the present invention, this embodiment constructs a digital simulation test platform for steam turbine generator sets driven by historical operating data. This platform can reproduce the multi-dimensional parameter dynamics of a 600MW-class unit under different operating conditions and allows the injection of controlled fault information. The parameter types and sampling rate specifications of its data output are consistent with those of the data acquisition system in the industrial field. The test sets up a control group and an experimental group. The control group adopts a single-parameter fixed threshold alarm method, and the alarm threshold for the vertical vibration of the No. 2 bearing is set to 100μm. The experimental group adopts the steady-state relationship instability monitoring method of the present invention, and the first trigger threshold for calculating the first drift value is set to 0.25. The test includes two scenarios. Scenario 1 simulates early faults caused by oil film deterioration. A disturbance that reduces the correlation between the vertical vibration of the No. 2 bearing and the oil temperature over time is injected into the simulation model, and the absolute values ​​of the two parameters are controlled not to exceed the alarm threshold of the control group. Scenario 2 simulates external operating condition disturbances. A disturbance simulating the fluctuation of circulating water temperature is injected into the simulation model. The disturbance caused synchronous fluctuations in the absolute values ​​of multiple parameters, with the peak values ​​of some parameters momentarily exceeding the threshold of the control group, but the physical correlation between the parameters remained stable. During the experiment, the data streams of the two scenarios were input into the control group and the experimental group for processing, respectively. In scenario one, as the injection fault developed, the Pearson correlation coefficient between the vertical vibration of bearing No. 2 and its oil temperature gradually decreased from the initial 0.92. When the coefficient dropped to 0.75, the first drift value calculated by the experimental group rose to 0.26, exceeding the first trigger threshold of 0.25 and triggering data recording. During this process, the control group did not generate an alarm. In scenario two, the fluctuation of the circulating water temperature caused a brief peak in the vertical vibration of bearing No. 2, reaching 101 μm, which exceeded the threshold of 100 μm of the control group and triggered an alarm. However, since the synergistic relationship between the parameters was not disrupted during the fluctuation, the first drift value calculated by the experimental group remained below 0.15 and did not trigger data recording. Table 1 shows the monitoring data at key time points in the two scenarios.

[0027] Table 1: Comparison of the responses of the two monitoring methods in different scenarios

[0028] The experimental results show that, in scenario one, the method used by the experimental group triggered data recording due to changes in the relationship structure between parameters when the absolute value of the parameters did not exceed the limit. However, in scenario two, data recording was not triggered for operating condition disturbances that only caused fluctuations in the absolute value of the parameters.

[0029] To further verify the beneficial effects of the method claimed in this invention compared to the conventional technology in the background section, the following comparative example 1 is established.

[0030] Comparative Example 1: This comparative example aims to simulate the single-parameter fixed threshold data recording method commonly used in the background art and compare its monitoring effect with the embodiment of the present invention. The test platform adopts the same steam turbine generator digital simulation test platform as Example 2. This platform can reproduce the operating dynamics of a 600MW-class unit and allows the injection of controlled fault or disturbance information. The only difference between the technical solution adopted in this comparative example and the test group of Example 2 of the present invention is the triggering logic of data recording: This comparative example does not adopt the monitoring method of structural instability based on multi-parameter relationship of the present invention, but adopts the conventional single-parameter fixed threshold monitoring method. That is, for the key parameter of vertical vibration of bearing No. 2, a fixed alarm threshold of 100μm is preset. An alarm is triggered and data is recorded only when the real-time measured value of the parameter exceeds this threshold. The test also includes two scenarios: Scenario 1: Early fault simulation. In order to verify the ability to identify weak faults, the simulation model is subjected to... Scenario 1: A fault message simulating early deterioration of the oil film in bearing No. 2 was injected into the simulation model. This fault message was characterized by disrupting the physical correlation between parameters such as vibration and temperature, leading to a continuous decrease in their correlation coefficient. However, the absolute amplitude of each parameter was controlled to prevent sudden changes in the short term, especially ensuring that the absolute value of the vertical vibration of bearing No. 2 did not reach the alarm threshold of 100μm. Scenario 2: Normal operating condition disturbance simulation. To verify the adaptability to non-fault fluctuations, an external operating condition disturbance simulating seasonal fluctuations in circulating water temperature was injected into the simulation model. This disturbance was characterized by causing synchronous and brief fluctuations in the absolute values ​​of multiple parameters, including vibration, but the intrinsic physical correlation between the parameters remained stable. During the disturbance, the peak value of the vertical vibration of bearing No. 2 was set to momentarily and slightly exceed the alarm threshold of 100μm. During the test, the data streams of the two scenarios were input into a monitoring system using conventional technology for processing. The test results are recorded in Table 2.

[0031] Table 2: Comparison of the Response of Conventional Monitoring Methods in Different Scenarios

[0032] The experimental results show that in scenario one, although the vibration-temperature correlation coefficient, which characterizes the health of the oil film, decreased from 0.92 to 0.75, clearly indicating the existence and development of early faults, the conventional monitoring method used in this comparative example failed to trigger any records throughout the entire process because the absolute value of the vibration parameter never exceeded the fixed threshold of 100 μm. This confirms that the method has a technical defect of underreporting for early relationship deterioration faults that are not accompanied by deviations in the absolute value of the parameter. In scenario two, the vibration value briefly exceeded the limit (101 μm) caused by fluctuations in normal operating conditions, causing the conventional monitoring method used in this comparative example to trigger a record. An alarm and data recording were issued. However, the stable high correlation coefficient (0.90) indicates that the coordination relationship between the parameters of the unit has not been disrupted, and the system is actually in a healthy state. This confirms that the method is insufficient in distinguishing between real faults and normal operating fluctuations, and is prone to generating a large amount of irrelevant data due to false alarms, interfering with the analysis work of operation and maintenance personnel. By comparing the results of this comparison with the experimental results of Embodiment 2 of the present invention, it can be seen that the method claimed in the present invention solves the inherent contradiction between monitoring sensitivity and data reliability of conventional technology by changing the evaluation of the absolute value of parameters to the evaluation of the stability of the internal relationship network of the system.

[0033] Example 3: This embodiment combines Figures 1 to 3 This document describes a method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion, as follows: Figure 1 As shown, the process begins with the acquisition of multi-dimensional operating parameters of the unit, including active power, vibration, temperature, and pressure. It then proceeds to the operating mode determination stage, classifying the unit's operating mode into a steady-state mode or a transient mode based on the volatility of macroscopic parameters. If the operating mode is determined to be steady-state, the process enters the steady-state relationship instability monitoring path. This path first dynamically compensates and adjusts the first trigger threshold based on the signal noise level, i.e., the fingerprint structure entropy value. Then, by comparing the structural differences between the real-time and dynamic benchmark correlation matrices, the first drift value is calculated. If the operating mode is determined to be transient, the process continues... The transient behavior distortion monitoring path calculates the second mismatch value by matching the morphological differences between the real-time parameter trajectory and the standardized template. The results of the above two monitoring paths, together with the output of a parallel long-term baseline degradation early warning module, are fed into a unified data recording triggering link. This early warning module identifies long-term cumulative degradation by comparing the dynamic reference matrix and the genesis matrix. The triggering link immediately triggers data recording when any monitored drift value or mismatch value exceeds the limit, and finally generates a structured data package containing vibration data and accompanying contextual information such as matrix and drift value.

[0034] like Figure 2As shown, the core interface for operations and maintenance analysts, as end users, to interact with the system is the analysis of structured data packets. Within the system, the module for determining the operating mode is the starting point for decision-making. Its output is directed to the module for monitoring steady-state relationship instability and the module for monitoring transient behavior distortion. The decision logic of the module for monitoring steady-state relationship instability is also affected in real time by a dynamic compensation trigger threshold module. The monitoring results of the three parallel functional modules—monitoring steady-state relationship instability, monitoring transient behavior distortion, and warning of long-term baseline degradation—are all uniformly fed into the unified data recording module, which ultimately performs the data recording action and generates structured data packets for operations and maintenance analysts to use.

[0035] like Figure 3 As shown, the process begins with a power sensor continuously collecting active power data at a frequency of 1 Hz. The data buffer maintains a 180-second sliding window of data. Every second, the mode decision-maker requests a calculation from the volatility calculator, which calculates the ratio of the standard deviation to the mean based on the data sequence in the buffer and returns the volatility value. The mode decision-maker compares this volatility with a threshold of 0.005. If the volatility is less than or equal to 0.005, it is determined to be a steady-state mode, and a confirmation is sent to the steady-state processing path to initiate steady-state relationship instability monitoring. Conversely, if the volatility is greater than 0.005, it is determined to be a transient mode, and a confirmation is sent to the transient processing path to initiate transient behavior distortion monitoring. At the same time, when switching modes, the system automatically stops monitoring the previous mode.

[0036] Example 4: This embodiment provides an offline calibration procedure for determining the first trigger threshold. After a steam turbine generator unit that has undergone major overhaul is put into operation, in order to deploy the monitoring method of this invention, it is necessary to set a first trigger threshold that matches the unit's operating characteristics. To determine this threshold, the unit's data acquisition system continuously records historical operating data for more than one month. This dataset contains active power parameters for determining the operating mode, as well as multiple operating parameters for calculating the correlation matrix, including the vibration, temperature, and pressure of each bearing. After acquiring this historical operating dataset, data processing is performed. First, the entire dataset is traversed, and then... The operation mode determination step involves filtering out all time segments in steady-state mode based on whether the fluctuation rate of active power within a 180-second time window is less than 0.005. Subsequently, for each selected steady-state time segment, the steady-state relationship instability monitoring step is performed, i.e., a real-time correlation matrix is ​​generated and compared with a dynamically updated benchmark correlation matrix to calculate a corresponding first drift value. After traversing all steady-state time segments, a sample set consisting of thousands of first drift values ​​is obtained. This set represents the normal fluctuation range of the multidimensional parameter relationship structure of the unit under healthy conditions.

[0037] Finally, statistical analysis is performed on the sample set of the first drift value to calculate its probability density distribution. The value corresponding to the 99.9th percentile of this distribution is selected as the first trigger threshold for the unit. In a specific calibration process, if the mean of the collected 10,000 drift value samples is 0.08 and the standard deviation is 0.05, then the value corresponding to its 99.9th percentile is approximately... The calculation result is 0.2345; based on this calculation, the first trigger threshold of the unit is set to 0.24; through this calibration procedure, a statistically based first trigger threshold can be set for different units.

[0038] Example 5: This embodiment provides an offline calibration procedure for generating standardized transient event trajectory templates and determining the second trigger threshold. After the monitoring system is deployed on a steam turbine generator unit, in order to enable the transient behavior distortion monitoring step, a template library containing multiple standardized transient event trajectory templates needs to be established for the unit. For the event type of increasing from 50% load to 75% load, the template generation procedure is as follows: First, confirm that the unit is in a healthy operating state. Then, within a predetermined time, repeat the standard load increase operation from the 50% load stabilization point to the 75% load stabilization point 5 to 10 times, and simultaneously record multiple operating parameters during each operation, including the main steam temperature and exhaust pressure of each cylinder, and the trajectory data of the changes over time. After the data collection is completed, the system aligns the multiple trajectory data with the active power as the reference time axis and calculates their average trajectory and standard deviation boundary, which together constitute a standardized transient event trajectory template for this event type and stores it in the library. This procedure can be repeated to generate corresponding trajectory templates for other transient event types such as unit startup or shutdown.

[0039] After the template library is constructed, a second trigger threshold is determined for the transient behavior distortion monitoring step. This determination process utilizes 5 to 10 transient event trajectory data collected during template generation under healthy conditions. Each trajectory is matched with the corresponding standardized transient event trajectory template generated for that event type using a dynamic time warping algorithm, thereby calculating a set of second mismatch value samples representing the normal fluctuation range. Subsequently, statistical analysis is performed on this set of second mismatch value samples to calculate its probability distribution, and the value corresponding to the 99.9% quantile of this distribution is selected as the second trigger threshold for that event type. Through this procedure, a statistically based second trigger threshold that matches the operating characteristics of the local unit can be set for each event type in the template library.

[0040] Example 6: This embodiment aims to address the key parameter in the dynamic compensation step of the trigger threshold, namely the positive gain coefficient. This provides a reproducible offline calibration procedure. In industrial environments with high background noise, to avoid false triggering of monitoring steps due to steady-state instability caused by sensor signal quality fluctuations, dynamic compensation rules are required. Positive gain coefficient in Calibration is performed; the goal of this calibration is to determine a... The value is set so that it can effectively suppress the fluctuation of the normal first drift value caused by the increase of signal noise from exceeding the first trigger threshold without reducing the sensitivity to real faults.

[0041] This calibration procedure utilizes historical operational data on unit health status spanning more than one month. First, by analyzing the high-frequency components of the original signal, the dataset is divided into a low-noise data subset and a high-noise data subset. Second, to quantify the impact of noise on the correlation matrix structure, steady-state data segments are extracted from both subsets, and their corresponding fingerprint structure entropy values ​​are calculated. The calculation includes: extracting the absolute values ​​of the off-diagonal elements of the correlation matrix, and normalizing these values ​​to form a probability distribution. Then apply the formula Calculate the entropy value. The calculation results show that the high-noise data subset corresponds to The values ​​are generally higher than those in the low-noise data subset; finally, to determine... The value is calculated using only a subset of high-noise data, where all data originates from a healthy state, and any value exceeding the first trigger threshold is considered valid. The first drift value is considered a false trigger, and the optimization process traverses a preset... Value range, such as from 0.1 to 5.0, and calculate in each... Apply dynamic threshold under the given value. Then, the number of false triggers in this high-noise data subset is analyzed. Finally, a subset is selected that can reduce the number of false triggers to an acceptable level, such as below 0.1% of the total sample size, while also being the smallest. The value is used as a calibration result; for example, if the calculation finds that... The false trigger rate decreased to 0.08%, while it further increased. If the value has no effect on reducing the false trigger rate, then the positive gain coefficient will be adjusted. The value of is determined to be 2.5.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion, characterized in that, Includes the following steps: Operating mode determination: The turbine generator unit continuously monitors at least one macroscopic operating parameter. When the volatility of the macroscopic operating parameter is lower than the first preset threshold within the first preset time window, the current operating mode is determined to be the steady-state operating mode; otherwise, it is determined to be the transient mode. Steady-state relationship instability monitoring: If and only if the operating mode is determined to be a steady-state operating mode, real-time data of multiple operating parameters are acquired. Based on the real-time data of multiple operating parameters acquired when the operating mode is determined to be a steady-state operating mode, a real-time correlation matrix is ​​generated. The real-time correlation matrix is ​​compared with a benchmark correlation matrix that is dynamically updated based on a historical correlation matrix generated under recent historical steady-state conditions to determine the first drift value that characterizes the structural difference between the two. Transient behavior distortion monitoring: If and only if the operating mode is determined to be a transient mode, the event type of the transient mode is identified based on the macroscopic state of the start and end of the transient mode, the preset standardized transient event trajectory template corresponding to the event type is obtained, and the real-time evolution trajectory of multiple operating parameters under the transient mode is matched with the trajectory template in an overall shape to determine the second mismatch value that represents the difference in trajectory shape. Unified data recording trigger: In steady-state operating mode, when the first drift value exceeds a preset first trigger threshold, the vibration data of the turbine generator set is triggered to be recorded; and in transient mode, when the second mismatch value exceeds a preset second trigger threshold, the vibration data of the turbine generator set is triggered to be recorded.

2. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, Both the real-time correlation matrix and the benchmark correlation matrix are generated by calculating the Pearson correlation coefficients between each pair of multiple operating parameters; the first drift value is determined by calculating the Frobenius norm between the real-time correlation matrix and the benchmark correlation matrix.

3. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, The unified data recording triggering step further includes: if the trigger is due to the first drift value exceeding the preset first triggering threshold, then the real-time correlation matrix, the benchmark correlation matrix, and the first drift value are used as first context information and associated with the recorded vibration data for storage; if the trigger is due to the second mismatch value exceeding the preset second triggering threshold, then the event type standardized transient event trajectory template and the second mismatch value are used as second context information and associated with the recorded vibration data for storage.

4. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, In the transient behavior distortion monitoring step, the overall morphological matching is calculated by using a dynamic time warping algorithm to determine the morphological distance between the real-time evolution trajectory and the trajectory template. The morphological distance is the second mismatch value.

5. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, It also includes long-term baseline degradation early warning steps: A genesis correlation matrix corresponding to a specific operating condition steady state is pre-stored when the steam turbine generator set is in a baseline healthy state; The dynamically updated baseline correlation matrix is ​​compared with the genesis correlation matrix at a frequency lower than that used to generate the real-time correlation matrix in order to determine the third cumulative drift value that characterizes long-term cumulative baseline degradation. When the third cumulative drift value exceeds a preset third trigger threshold, the vibration data of the steam turbine generator set is triggered to be recorded.

6. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, It also includes a step to trigger dynamic threshold compensation: After generating the real-time correlation matrix, a fingerprint structure entropy value representing the current signal-noise level is calculated based on the internal numerical distribution of the real-time correlation matrix. Based on the fingerprint structure entropy value, a preset first trigger threshold is dynamically adjusted, and the dynamic adjustment follows a set rule. ,in, The adjusted first trigger threshold, The preset first trigger threshold is the baseline value before compensation. The fingerprint structure entropy value, This is the preset positive gain coefficient.

7. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, In the operation mode determination step, at least one macroscopic operation parameter is selected from at least one of active power and main steam pressure.

8. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, The dynamic update of the benchmark correlation matrix is ​​generated by performing an exponentially weighted moving average calculation based on the historical correlation matrices generated under one or more historical steady-state operating conditions before the real-time correlation matrix is ​​generated.

9. The method for recording vibration data of a steam turbine generator set triggered by multi-parameter fusion according to claim 1, characterized in that, In the transient behavior distortion monitoring step, the standardized transient event trajectory template includes a standardized historical data snapshot of multiple operating parameters changing over time or with a dominant macroscopic parameter under healthy conditions.

10. A multi-parameter fusion-triggered vibration data recording system for steam turbine generator sets, characterized in that, include: The mode determination module is used to determine the operating mode: the turbine generator set continuously monitors at least one macroscopic operating parameter. When the volatility of the macroscopic operating parameter is lower than the first preset threshold within the first preset time window, the current operating mode is determined to be the steady-state operating mode; otherwise, it is determined to be the transient mode. The instability monitoring module is used for steady-state relationship instability monitoring: when the operating mode is determined to be a steady-state operating mode, it acquires real-time data of multiple operating parameters, generates a real-time correlation matrix based on the real-time data of multiple operating parameters acquired when the operating mode is determined to be a steady-state operating mode, and compares the real-time correlation matrix with a benchmark correlation matrix that is dynamically updated based on a historical correlation matrix generated under recent historical steady-state conditions to determine the first drift value that characterizes the structural difference between the two. The distortion monitoring module is used for transient behavior distortion monitoring: when the operating mode is determined to be a transient mode, the event type of the transient mode is identified based on the macroscopic state of the start and end of the transient mode, the preset standardized transient event trajectory template corresponding to the event type is obtained, and the real-time evolution trajectory of multiple operating parameters under the transient mode is matched with the trajectory template in an overall shape to determine the second mismatch value that represents the difference in trajectory shape. The recording trigger module is used to unify data recording triggers: in steady-state operating mode, when the first drift value exceeds a preset first trigger threshold, the vibration data of the turbine generator set is triggered to be recorded; and in transient mode, when the second mismatch value exceeds a preset second trigger threshold, the vibration data of the turbine generator set is triggered to be recorded.

Images