A method for gas turbine rotor health assessment

Abstract

The application discloses a kind of gas turbine rotor health state evaluation methods, it is related to gas turbine state monitoring technical field.The method includes: in the process of starting or stopping or variable load of gas turbine, rotation speed signal and rotor radial vibration displacement signal are collected, and the excitation-response pair corresponding to rotation speed change event is intercepted;Transfer delay spectrum is constructed to excitation-response pair;Multiple data under the health state of rotor are obtained to establish health benchmark transfer delay spectrum, and main peak time delay, main peak width and sub-peak quantity are extracted as health benchmark features;According to the distribution difference of current transfer delay spectrum and health benchmark and feature deviation, construct time delay spectrum distortion index;According to distortion index, the health state of rotor is graded and early warning.The application does not depend on fault sample, and realizes the detection of structural integrity degradation caused by early micro-crack or creep of rotor using starting and stopping transient data.

Description

Technical Field

[0001] This invention relates to the field of gas turbine condition monitoring technology, and more specifically, to a method for assessing the health status of a gas turbine rotor. Background Technology

[0002] Gas turbines frequently participate in peak shaving in new power systems, resulting in a significant increase in start-stop cycles. The rotor is subjected to alternating thermo-mechanical loads, making it susceptible to early microscopic damage such as microcracks or creep. As the core power component of the gas turbine, the rotor's structural integrity directly affects the safe and reliable operation of the unit; therefore, accurate assessment of the rotor's health status is of great importance.

[0003] Currently, there are two main approaches to rotor health status assessment. One approach requires fault samples for support. Chinese patent CN116861231B discloses a rotor fault diagnosis model generation and diagnosis method based on multi-transform domain. This method collects fault vibration signals from a simulated gas turbine rotor to construct training and testing datasets, and then uses short-time Fourier transform, wavelet transform, and complex network recursion. Figure 3 One approach extracts features from the transform domain and uses a convolutional neural network for fault detection. This method belongs to the supervised learning paradigm and requires labeled fault samples to train the model. Another approach requires the support of an accurate physical model. Chinese patent CN120160824A discloses a fault diagnosis method for major rotating components and bearings of industrial gas turbines. This method establishes engine performance and degradation models and performs fault diagnosis under steady-state and transient conditions. This approach relies on accurate engine performance and degradation models.

[0004] In real-world industrial environments, fault samples of gas turbine rotors are extremely scarce, and labeled data on early damage such as microcracks and creep are difficult to obtain, limiting the applicability of the first type of solution. The boundary conditions of gas turbine rotor systems are complex, and nonlinear factors are difficult to model accurately, making the diagnostic results of the second type of solution highly susceptible to model accuracy. Furthermore, the vibration amplitude changes caused by early microcracks or creep damage in the rotor are not significant, making reliable detection difficult in the early stages of degradation for both types of solutions. Therefore, a method for assessing the health status of a gas turbine rotor is proposed to address the above-mentioned problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for assessing the health status of a gas turbine rotor. This method addresses the problem that the prior art, under frequent start-stop peak shaving conditions, cannot detect and assess the structural integrity degradation of the rotor caused by early microcracks or creep without relying on fault samples or precise physical models, but only using the unit's own sensors during the start-stop transient process based on rotational speed and vibration displacement data.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for assessing the health status of a gas turbine rotor is provided to detect the structural integrity degradation of the rotor caused by early microcracks or creep under frequent start-stop peak shaving conditions, without relying on fault samples or precise physical models.

[0007] This method treats the rotor system as an information transmission channel, with the rotational speed signal as the excitation input and the vibration displacement signal as the response output. It measures the degradation of rotor structural integrity by analyzing the distortion degree of the information transmission delay structure.

[0008] The method includes the following steps: S1: During the start-up, shutdown, or load change of the gas turbine, the speed signal and rotor radial vibration displacement signal are collected, and the speed signal segment and vibration displacement signal segment corresponding to the speed change event are extracted to form an excitation-response pair.

[0009] The start-up, shutdown, or load change process of the gas turbine itself constitutes a transient excitation to the rotor system. The excitation-response pair obtained reflects the input-output relationship of the rotor under transient excitation.

[0010] Furthermore, the interception process is as follows: The moment when the rate of change of rotational speed exceeds a preset threshold is taken as the starting point of the transient event. The endpoint is defined as the moment when the detection speed re-enters the steady-state fluctuation zone and remains stable for a preset duration. The rotational speed signal and vibration displacement signal between the starting point and the ending point are extracted as the excitation-response pair. The extracted excitation-response pair contains the complete transient decay transition process, preserving the complete response information of the rotor from being disturbed to returning to steady state.

[0011] S2: Construct a transmission delay spectrum for the excitation-response pair, wherein the transmission delay spectrum represents the contribution intensity of the causal information of the current rotational speed to the vibration displacement at each delay time.

[0012] In a healthy rotor structure with simple constraints, information is primarily transmitted from rotational speed to vibration via a single dominant path, resulting in a transmission delay spectrum characterized by narrow peaks concentrated around a specific delay time. However, when microcracks or creep damage occur within the rotor, the introduced nonlinear constraints generate additional parasitic transmission paths, leading to distortions in the transmission delay spectrum, such as broadening of the main peak, generation of secondary peaks, or shifting of the main peak.

[0013] Furthermore, the process of constructing the transmission delay spectrum is as follows: Phase space reconstruction is performed on the vibration displacement signal in the excitation-response pair to obtain the historical state vector of the vibration displacement signal. The dynamic state information of the rotor system is recovered from the one-dimensional vibration displacement signal through phase space reconstruction. Within the preset time delay scanning range, for each time delay point, the transfer entropy of the current rotation speed to the vibration displacement after the time delay point is calculated under the condition that the historical state vector is known. The transfer entropy eliminates the influence of the historical information of the vibration displacement signal itself and extracts the unique causal information contribution of the current rotation speed to the vibration displacement. The propagation entropy at each delay point is normalized to form the propagation delay spectrum.

[0014] S3: Obtain data on multiple start-stop or load change processes under rotor health conditions, obtain the health reference transfer delay spectrum according to steps S1 to S2, and extract the main peak delay, main peak width and number of secondary peaks of the health reference transfer delay spectrum as health reference features.

[0015] The health baseline feature characterizes the information transmission delay structure of the rotor in a structurally intact state. The health state refers to the state in which the rotor is confirmed to be free of cracks and creep damage by borehole inspection. The health baseline feature is extracted by statistically averaging the transmission delay spectrum obtained under multiple health states, eliminating random fluctuations of single events and preserving stable transmission structure characteristics.

[0016] Furthermore, the method for extracting health benchmark features is as follows: The delay corresponding to the peak value in the health benchmark transmission delay spectrum is taken as the main peak delay, and the main peak delay reflects the delay time of the dominant transmission path. The width of half-peak height is taken as the width of the main peak, and the width of the main peak reflects the concentration of the dominant transmission path; The number of independent peaks whose height exceeds a preset proportion of the main peak height is taken as the number of secondary peaks. The number of secondary peaks reflects whether there are other identifiable secondary transmission paths besides the dominant transmission path.

[0017] Before step S4, a micro-delay entropy spectrum of the propagation delay spectrum is also constructed. The construction method is as follows: for each delay point of the propagation delay spectrum, the sample entropy is calculated from the propagation entropy sequence within the local neighborhood, which is used as the micro-delay entropy of that delay point, forming the micro-delay entropy spectrum. The micro-delay entropy spectrum captures the local random fluctuation characteristics of the propagation delay spectrum at a micro-delay scale.

[0018] The information transmission of a healthy rotor is highly deterministic, and the micro-delay entropy at each delay point is generally at a low level; the random behavior introduced by damage increases the micro-delay entropy of the relevant delay domain.

[0019] Furthermore, the statistical upper bound of the micro-delay entropy spectrum under healthy conditions is obtained multiple times, serving as the upper bound envelope of the healthy baseline micro-delay entropy spectrum. Exceeding this envelope indicates the occurrence of anomalous random propagation behavior not seen under healthy conditions.

[0020] S4: Construct a time delay spectrum distortion index based on the distribution difference between the current transmission delay spectrum and the health benchmark transmission delay spectrum, the main peak delay of the current transmission delay spectrum, the offset of the main peak width and the number of secondary peaks relative to the health benchmark characteristics, and the cumulative area of ​​the current micro-delay entropy spectrum exceeding the upper bound of the health benchmark micro-delay entropy spectrum.

[0021] The distortion index quantifies the degree of distortion in the information transmission delay structure, thereby measuring the degradation of rotor structural integrity. The cumulative area reflects the total amount of random transmissions exceeding the healthy baseline at the micro-delay scale.

[0022] Furthermore, the time delay spectrum distortion index is obtained by weighting the first distortion component, the second distortion component, and the third distortion component: The first distortion component is obtained by mapping the distribution divergence between the current transmission delay spectrum and the healthy baseline transmission delay spectrum, reflecting the distortion at the overall distribution level of the transmission delay spectrum.

[0023] The second distortion component is obtained by normalizing the cumulative area and reflects the degree of enhancement of the randomness of information transmission at the micro-delay scale.

[0024] The third distortion component is obtained by weighting the main peak delay offset, the main peak width broadening, and the change in the number of secondary peaks in the current transmission delay spectrum, and reflects the direct offset of the transmission path structure characteristics.

[0025] Furthermore, the calculation method for each distortion component is as follows: the distribution divergence is the symmetric KL divergence between the current propagation delay spectrum and the healthy baseline propagation delay spectrum. The mapping is: , in This is the first distortion component. Let KL divergence be the symmetric KL divergence. As a preset positive coefficient, when When deviating from the normal range of health status, rise.

[0026] The normalization is as follows: , in This is the second distortion component. The cumulative area, This is the normalization coefficient.

[0027] The third distortion component: , in , , The main peak delay, main peak width, and number of secondary peaks in the current propagation delay spectrum are given. , , For the aforementioned health benchmark characteristics, , , These are the weighting coefficients.

[0028] When microcracks appear in the rotor, the periodic opening and closing of the crack surface introduces nonlinear time-varying stiffness, leading to delayed drift and broadening of the main peak, and an increase in the first and second terms of the third distortion component. When creep damage occurs in the rotor, material relaxation changes the constraint state, resulting in the creation of additional parasitic paths, an increase in the number of secondary peaks, and an increase in the third term of the third distortion component.

[0029] S5: The rotor health status is graded according to the time delay spectrum distortion index. When the distortion index is lower than the first threshold, it is determined to be healthy. When the distortion index exceeds the first threshold but is lower than the second threshold, it is determined to have micro-damage. When the distortion index exceeds the second threshold, an early warning is triggered.

[0030] Furthermore, step S5 is followed by a trend warning process: the distortion index is recorded as a sequence of multiple events; if the sequence shows a significant monotonically increasing trend after trend testing, a progressive degradation warning is issued even if the current distortion index does not exceed the first threshold. The progressive degradation warning is for situations where the distortion index has not yet exceeded the threshold but has shown a continuous deterioration trend, indicating that the rotor may have slowly developing creep or microcrack accumulation.

[0031] Further, the rotational speed signal mentioned in step S1 is acquired by a gas turbine key phase sensor, and the vibration displacement signal is acquired by a gas turbine eddy current displacement sensor. The preset time delay scanning range mentioned in step S2 is 0.5ms to 500ms, with the lower limit corresponding to the shortest delay of direct mechanical force transmission and the upper limit covering the delay of thermoelastic coupling and fluid excitation coupling transmission paths.

[0032] The technical effects and advantages of this invention are as follows: (I) This invention treats the rotor system as an information transmission channel, using the rotational speed signal as the excitation input and the vibration displacement signal as the response output. It characterizes the delay structure of information transmission by constructing a transmission delay spectrum. A healthy rotor structure has simple constraints, and its transmission delay spectrum exhibits a concentrated narrow peak. Nonlinear constraints introduced by microcracks or creep generate additional parasitic transmission paths, causing the transmission delay spectrum to broaden its main peak, increase secondary peaks, or shift its main peak. By establishing a healthy baseline transmission delay spectrum and extracting baseline features, the distribution difference and feature shift between the current transmission delay spectrum and the baseline are quantified. A transmission delay spectrum distortion index is constructed, and graded assessment and trend warning are performed based on the distortion index. This method only requires healthy state data to establish a baseline, without relying on fault samples or precise physical models. By analyzing changes in the information transmission delay structure, it measures structural integrity degradation, thereby detecting signs of degradation in the early stages before significant changes in vibration amplitude.

[0033] (II) This invention acquires data based on the gas turbine's own key phase sensor and eddy current displacement sensor, requiring no additional hardware modifications. It utilizes transient data that would otherwise be discarded during start-up, shutdown, and load changes for evaluation, without increasing testing costs. This method transforms the evaluation object from the amplitude characteristics of vibration response to the delayed structural characteristics of information transmission, exhibiting fundamental sensitivity to nonlinear constraints introduced by early damage such as microcracks and creep. It provides a detection method for condition monitoring of peak-shaving gas turbine rotors that does not rely on prior failures. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the evaluation process of the method of the present invention; Figure 2 This is a schematic diagram illustrating the mechanism of transmission delay spectrum construction and distortion analysis in this invention; Figure 3 This is a flowchart of the health assessment decision-making process with trend warning of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Example 1 As attached Figures 1 to 3 The method for assessing the health status of a gas turbine rotor, as shown, addresses the structural integrity degradation of the rotor caused by microcracks or creep during frequent start-stop peak shaving of gas turbines, based on the following technical premises: When microcracks or creep damage appear inside the rotor, the periodic opening and closing of the crack surface or material relaxation introduces nonlinear time-varying constraints. These nonlinear constraints are equivalent to adding parasitic transmission paths with different delay times in addition to the original information transmission paths. As a result, the transmission time delay spectrum from rotational excitation to vibration response exhibits detectable distortions such as main peak shift, broadening, or secondary peak proliferation. By quantifying the degree of this distortion, the degree of rotor structural degradation can be assessed without relying on fault samples.

[0037] The specific implementation process of each step is explained below.

[0038] S1. Obtain stimulus-response pairs During start-up, shutdown, or load changes, the rotor speed of a gas turbine undergoes significant changes, which constitute a transient excitation to the rotor system. This step utilizes the unit's onboard sensors to collect transient process data, extracting the speed signal segments and vibration displacement signal segments corresponding to the speed change events to form excitation-response pairs.

[0039] The rotational speed signal is acquired by a key phase sensor. The key phase sensor is installed at the end of the rotor shaft. When a mark on the shaft passes the sensor, it outputs a pulse. The rotational speed sequence is calculated by measuring the time interval between adjacent pulses.

[0040] The vibration displacement signal is acquired by an eddy current displacement sensor. The sensor is mounted on the bearing housing at the bearing section, with a preset gap maintained between the probe and the rotor journal. The output voltage signal is calibrated and converted into the rotor radial vibration displacement value.

[0041] The sampling frequency of both signals is preset to 2kHz.

[0042] Perform a first-order backward difference on the speed sequence to obtain the speed change rate sequence. Set a speed change rate threshold. , The value is taken as three times the normal fluctuation amplitude of the unit's speed during steady-state operation. When the absolute value of the speed change rate first exceeds... When that moment arrives, record that moment as the starting point of the transient event.

[0043] The speed signal is continuously monitored from the starting point. The speed is stabilized once it re-enters the steady-state fluctuation zone and remains stable for a preset duration. The moment when steady state is reached is recorded as the endpoint. The steady-state fluctuation band is defined as the range within 0.5% above and below the stable speed value. The larger of three times the damping decay time constant of the rotor's first critical speed and 5s is selected to ensure that the extracted data segment contains the complete transient decay transition process.

[0044] The rotational speed signal and vibration displacement signal between the starting point and the ending point are extracted, and their respective mean values ​​are subtracted to obtain the standardized excitation-response pair.

[0045] In a single start-stop or load-changing operation, both the acceleration and deceleration phases contain stages of continuous speed change. The rate of speed change in each phase exceeds [a certain threshold]. Each sub-process is extracted as an independent excitation-response pair. For example, during a cold start, the acceleration phase from the turning gear speed to the rated speed can be extracted as one excitation-response pair, and the deceleration phase from the rated speed to the turning gear speed can be extracted as another. During rapid load changes, the transient speed fluctuations caused by changes in load commands can also be extracted as corresponding excitation-response pairs.

[0046] S2, Constructing the propagation delay spectrum After obtaining the excitation-response pair, a transmission delay spectrum from rotational speed to vibration is constructed to reflect the delay structure of information transmission between rotational speed excitation and vibration response.

[0047] First, phase space reconstruction is performed on the vibration displacement signal sequence to recover the dynamic state information of the system from the one-dimensional time series. Using the vibration displacement signal sequence as input, the mutual information function between the signal and its delayed version is calculated, and the delay corresponding to the first minimum point of the mutual information function is taken as the reconstruction delay time. .

[0048] Then, using the vibration displacement signal sequence and Using the spurious nearest neighbor method as input, the minimum embedding dimension is determined. Starting with dimension 1, the phase space is reconstructed incrementally. In each dimension, the nearest neighbor of each point is found, and the relative rate of change of the nearest neighbor distance after adding one dimension is calculated. When the proportion of points with a rate of change exceeding 10% is less than 5%, the current dimension is output as the starting dimension. .

[0049] The same method is used for the speed signal and Reconstruction is performed. After reconstruction, the vibration displacement signal yields a state vector sequence. Among these, the historical state vector... From time Previous Composed of several delayed sampling points: The rotational speed signal is also reconstructed into a state vector sequence.

[0050] After phase space reconstruction, the propagation delay spectrum is calculated. The preset delay scan range is 0.5 ms to 500 ms. The lower limit of 0.5 ms corresponds to the shortest delay for direct mechanical force transmission, while the upper limit of 500 ms covers the delays of slower propagation paths such as thermoelastic coupling and fluid excitation coupling. Scan step size. It is fixed at 0.5ms, consistent with the sampling period.

[0051] For each time delay value within the scanning range Calculate the conditional transfer entropy . Measurement in the known history of the vibration signal itself Under these conditions, the current rotational speed Capable of prediction Vibration after time How much additional definitive information is provided. The calculation formula is: ; in, For the current moment The rotational speed value, for The rotor radial vibration displacement value at time t. This is the historical state vector of the vibration displacement signal.

[0052] When calculating, first perform statistics , and The joint probability distribution of the three factors. Then, statistical analysis of each factor is performed separately in the known... and under conditions The conditional probability, and in the case where only the known under conditions The conditional probability. When the rotational speed... For prediction When there is an additional contribution, the first conditional probability is greater than the second, and the logarithmic term is positive. It is positive.

[0053] When the speed When there is no additional contribution, the two conditional probabilities are equal. It is 0. The larger the value, the stronger the causal driving force of the rotational speed on the vibration during that delay time.

[0054] The joint probability distribution is estimated using the multidimensional histogram method: right Take the interval formed by the maximum and minimum values ​​within the data segment and divide it into equal parts. A range.

[0055] right Each component of the dimensionality is divided into intervals formed by its respective maximum and minimum values. A range.

[0056] right Divide the interval formed by its maximum and minimum values ​​into equal parts. A range.

[0057] , , The value of is 1 / 15 of the total number of data points.

[0058] The frequency of each data point falling into each joint interval is counted, and the ratio of this frequency to the total number of data points is used as the joint probability estimate. The conditional probability is obtained by the ratio of the joint probability to the marginal probability. For zero-frequency intervals, a smoothing constant of 0.001 is added before recalculating the probability to avoid undefined logarithms.

[0059] As one implementation method, when the data sample size is small or a more continuous and smooth probability estimate is required, the joint probability distribution can be estimated using kernel density estimation. A Gaussian kernel function centered at each data point is placed, and the kernel functions of all data points are superimposed to obtain the overall probability density estimate.

[0060] The kernel function bandwidth is determined according to Scott's rule, that is, the bandwidth is equal to the number of sample points. Power of the power multiplied by the standard deviation of the data Let's consider the dimension of the variables. Substitute the joint probability density and marginal probability density obtained from kernel density estimation into... The formula calculates the transfer entropy value through numerical integration. This method avoids the discrete errors introduced by partition boundaries in the histogram method.

[0061] After iterating through all scan delay values, we obtain Follow The changing curve. Normalize this curve, and then... Value divided by all point The sum of the values ​​yields the normalized propagation delay spectrum. After normalization, The vertical axis represents the proportion of information contribution at each delay time to the total contribution.

[0062] In a healthy rotor structure with simple constraints, information is primarily transmitted from rotational speed to vibration via a single dominant path. It appears as a clean, narrow peak concentrated around a specific delay time.

[0063] When microcracks or creep damage appear inside the rotor, nonlinear constraints introduce parasitic transmission paths. This can lead to broadening of the main peak, generation of secondary peaks, or shift in the position of the main peak. For example, microcracks cause the crack surface to open and close periodically, forming new delayed couplings outside the original propagation path. New secondary peaks appear. Creep leads to material relaxation and changes in constraint, causing the main peak to shift and broaden.

[0064] S3. Establish health benchmarks Establish a health baseline during the operational phase to confirm the rotor's health status. The health status is confirmed as follows: after a major overhaul, borehole inspection reveals no cracks or creep damage; or after a new machine is put into operation, the running-in period confirms that the operating parameters are normal and vibration monitoring shows no abnormalities.

[0065] During this health phase, data is collected from multiple start-stop or load change processes, typically no less than 10 times, so that the health baseline can fully cover the range of fluctuations under normal operating conditions.

[0066] For each transient event, the corresponding propagation delay spectrum is obtained according to steps S1 and S2. Since the preset delay scan range and scan step size are fixed for each calculation, the delay point coordinates of all propagation delay spectra are completely consistent. The arithmetic mean of the propagation delay spectra under all healthy states is calculated point-by-point to obtain the healthy baseline propagation delay spectrum. By averaging out random fluctuations in individual events, stable transmission structure characteristics are preserved.

[0067] from Three health benchmark features were extracted from them.

[0068] Peak delay for The horizontal delay value corresponding to the maximum value of the vertical axis. If there are multiple peaks of equal height, the one with the smallest delay is taken as the main peak.

[0069] Main peak width The width of the peak is measured at half the height of the main peak, which is the time delay difference spanned from the left half-height position to the right half-height position of the peak.

[0070] Number of secondary peaks for The number of independent peaks whose mid-peak height exceeds 5% of the main peak height. The method for determining an independent peak is: [The text abruptly ends here, so the translation stops as well.] Curve detection identifies local maxima and local minima. A region with a single maximum between two adjacent minima is treated as an independent peak. Peak height conditions are checked for each peak, and the total number of independent peaks that meet the conditions is counted.

[0071] The aforementioned health baseline features were extracted by statistically averaging the transmission delay spectra obtained under multiple health states.

[0072] Prior to step S4, a micro-delay entropy spectrum is constructed to capture the local structural details of the propagation delay spectrum at the micro-delay scale.

[0073] For the propagation delay spectrum Each delay point in Take Centered on, with a width of The neighborhood window. Take 20 scan steps ,Right now Extract the propagation entropy values ​​corresponding to each delay point within the window, and arrange them in ascending order of delay to form a sequence. .

[0074] With sequence The sample entropy is calculated for the input, and the output value is used as the delay point. Local sample entropy The parameters for calculating sample entropy are: pattern dimension is 2, similarity tolerance... Get sequence 0.2 times the standard deviation. After traversing all delay points, the local sample entropies of each point are connected in delay order to obtain the micro-delay entropy spectrum. .

[0075] The transmission of information in a healthy rotor is highly deterministic, and the transmission entropy at each delay point varies within its local neighborhood, with the overall local sample entropy remaining at a low level.

[0076] When damage introduces random micro-opening or micro-impact behavior, the propagation entropy of certain delay domains exhibits irregular fluctuations in their local neighborhoods, and the local sample entropy increases detectably. For example, a microcrack causes a local abrupt change in propagation entropy at the moment of opening and closing, which manifests as a spike at the corresponding delay point in the micro-delay entropy spectrum.

[0077] The upper bound envelope of the health baseline micro-delay entropy spectrum is obtained as follows: For multiple data points under healthy conditions, the micro-delay entropy spectrum is calculated for each time, resulting in multiple healthy micro-delay entropy spectrum curves.

[0078] At each delay point Above, calculate the mean value of the entropy spectrum of all healthy micro-delays at that point. and standard deviation ,Pick This serves as the upper bound for the delay point.

[0079] The upper bounds of all delay points are concatenated in delay order to form the upper bound envelope of the healthy baseline micro-delay entropy spectrum. .

[0080] S4, Construction of the time delay spectrum distortion index The distortion index is obtained by weighted summation of three distortion components. These three components reflect the deviation of the propagation delay spectrum from the healthy baseline at different scales.

[0081] First distortion component This reflects the difference between the overall distribution of the propagation delay spectrum and the health baseline. Calculate the current propagation delay spectrum. Transmission delay spectrum with health benchmark Symmetric KL divergence .

[0082] Depend on and Add them together to get the result.

[0083] in , The summation range covers all delay points. For or A zero delay point is where the logarithmic term is 0.

[0084] get Then, press calculate . Take multiple times in a healthy state Standard deviation of values The reciprocal. This setting makes... In a typical healthy state The value is close to 0, when When it deviates from the normal range It rose rapidly, approaching 1.

[0085] Second distortion component This reflects the degree of increased randomness in information transmission at the micro-delay scale. For each delay point... Determine the current micro-delay entropy spectrum Is it greater than the upper bound of the health benchmark envelope? .

[0086] like Calculate the excess amount Multiply by step size Obtain the area micro-element of that point. Accumulate the area micro-element of all points outside the scanned area to obtain the cumulative area. .

[0087] according to calculate . Take a healthy state Maximum observation The reciprocal of the number, making the healthy state Not exceeding 0.1 When significantly increased Approaching 1.

[0088] Third distortion component A direct offset reflecting the structural characteristics of the transmission path. From the current transmission delay spectrum. In the same manner as in step S3, the main peak delay is extracted. Main peak width and the number of secondary peaks Using health benchmark characteristics , , Calculate using the following formula : ; in , , These are weighting coefficients, and the sum of the three is 1. Their values ​​are determined based on the coefficient of variation of each feature under healthy conditions; the smaller the coefficient of variation, the more stable the feature, and the greater its weight.

[0089] When microcracks appear in the rotor, the periodic opening and closing of the crack surface introduces nonlinear time-varying stiffness, leading to delayed drift and broadening of the main peak, and an increase in the first and second terms in the equation. When creep damage occurs, material relaxation alters the constraint state, resulting in the creation of additional parasitic paths, increasing the number of secondary peaks. Increase, the third term increases. The denominator uses... This is to avoid the denominator being zero when the number of secondary peaks at the health baseline is zero.

[0090] The three distortion components are weighted. , , The distortion index is obtained by weighted summation. : ; in , , These are the weighting coefficients. Default settings , , It can also be adjusted according to the fluctuation range of each component under healthy conditions. When more sensitive detection of early damage is required, the value can be appropriately increased. This is because the micro-delay entropy spectrum is more sensitive to the micro-opening and closing behavior introduced by cracks.

[0091] The value range is [0, 1]. The closer it is to 0, the closer the transmission delay spectrum is to the health baseline, and the better the rotor structure integrity.

[0092] S5. Graded Assessment and Trend Early Warning Preset first threshold Second threshold . Take the distortion index multiple times under healthy conditions mean Add 3 standard deviations . Pick 2 times.

[0093] Calculated for each transient event Make a judgment: when At that time, the rotor is determined to be healthy and its structural transmission characteristics are normal.

[0094] when If micro-damage is detected, the monitoring frequency should be increased or the borehole inspection cycle should be shortened.

[0095] when When this occurs, an early warning is triggered, indicating that the structure has undergone significant degradation and that a shutdown inspection should be arranged as soon as possible.

[0096] For a slow-developing degradation process, record the corresponding events of multiple consecutive start-stop or load changes. The sequences are arranged chronologically. The Mann-Kendall trend test is used to analyze monotonic trends.

[0097] Given a sequence as input, compare the size relationships of data points pairwise and count the number of pairs where the later value is greater than the earlier value. The logarithm of the second value being less than the first value Calculate the test statistic . It is calculated from the sequence length and the number of nodes in it. A node is a data point in the sequence that takes the same value. When a node appears... The conclusion needs to be corrected.

[0098] Standardized statistics The calculation method is as follows: when hour ,when hour ,when hour .like and If the value is positive, then the sequence exhibits a significant monotonically increasing trend. At this point, even if the current value is positive... Not exceeding It also issued a warning of gradual degradation.

[0099] As another implementation method, when the sequence is short or the degradation rate needs to be obtained directly, a linear regression slope test can be used. The least-squares linear fit is performed using the sequence as input to obtain the slope. and coefficient of determination .like Exceeding the preset threshold and If a significant upward trend is observed, a warning of gradual degradation will be issued. The mean slope of the natural fluctuation series of the distortion index under healthy conditions is added to three times the standard deviation. This method can not only determine the presence or absence of a trend, but also... Directly quantify the degradation rate.

[0100] The above provides a detailed description of the gas turbine rotor health status assessment method provided by this invention. Any equivalent substitutions made by those skilled in the art to the specific algorithm selection and parameter determination methods in each step without departing from the technical concept of this invention should fall within the protection scope of this invention.

Claims

1. A method for assessing the health status of a gas turbine rotor, characterized in that, Includes the following steps: S1: During the start-up, shutdown, or load change process of the gas turbine, the speed signal and rotor radial vibration displacement signal are collected, and the speed signal segment and vibration displacement signal segment corresponding to the speed change event are extracted to form an excitation-response pair; S2: Construct a transmission delay spectrum for the excitation-response pair, wherein the transmission delay spectrum represents the contribution intensity of the causal information of the current rotational speed to the vibration displacement at each delay time; S3: Obtain data on multiple start-stop or load change processes under rotor health status, obtain the health reference transfer delay spectrum according to steps S1 to S2, and extract the main peak delay, main peak width and number of secondary peaks of the health reference transfer delay spectrum as health reference features. S4: Based on the distribution difference between the current transmission delay spectrum and the health benchmark transmission delay spectrum, and the offset of the main peak delay, the main peak width, and the number of sub-peaks of the current transmission delay spectrum relative to the health benchmark characteristics, construct the delay spectrum distortion index; S5: The rotor health status is graded according to the time delay spectrum distortion index. When the distortion index is lower than the first threshold, it is determined to be healthy. When the distortion index exceeds the first threshold but is lower than the second threshold, it is determined to have micro-damage. When the distortion index exceeds the second threshold, an early warning is triggered.

2. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The interception mentioned in step S1 includes: The moment when the rate of change of rotational speed exceeds a preset threshold is taken as the starting point of the transient event. The endpoint is defined as the moment when the detection speed re-enters the steady-state fluctuation zone and remains stable for a preset duration. The rotational speed signal and vibration displacement signal between the starting point and the ending point are extracted as the excitation-response pair.

3. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The construction of the propagation delay spectrum in step S2 includes: The phase space of the vibration displacement signal in the excitation-response pair is reconstructed to obtain the historical state vector of the vibration displacement signal; Within the preset time delay scanning range, for each time delay point, calculate the transfer entropy of the current rotational speed to the vibration displacement after the time delay point, given the known historical state vector; The propagation entropy at each delay point is normalized to form the propagation delay spectrum.

4. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The extraction of health benchmark features in step S3 includes: The delay corresponding to the peak value in the health benchmark transmission delay spectrum is taken as the main peak delay, the width of half peak height is taken as the main peak width, and the number of independent peaks whose peak height exceeds a preset proportion of the main peak height is taken as the number of secondary peaks.

5. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The steps preceding step S4 also include: Construct the micro-delay entropy spectrum of the propagation delay spectrum, wherein, for each delay point of the propagation delay spectrum, the sample entropy is calculated by taking the propagation entropy sequence in the local neighborhood as the micro-delay entropy of that delay point, thus forming the micro-delay entropy spectrum; Obtain the statistical upper bound of the micro-delay entropy spectrum under the healthy state multiple times, and use it as the upper bound envelope of the healthy baseline micro-delay entropy spectrum; In step S4, when constructing the time delay spectrum distortion index, the cumulative area of ​​the current micro-delay entropy spectrum exceeding the upper bound of the healthy baseline micro-delay entropy spectrum is also considered.

6. The method for assessing the health status of a gas turbine rotor according to claim 5, characterized in that, The time delay spectrum distortion index mentioned in step S4 is obtained by weighting the first distortion component, the second distortion component, and the third distortion component; The first distortion component is obtained by mapping the distribution divergence between the current transmission delay spectrum and the health baseline transmission delay spectrum; The second distortion component is obtained by normalizing the cumulative area; The third distortion component is obtained by weighting the main peak delay offset, the main peak width broadening, and the change in the number of secondary peaks in the current propagation delay spectrum.

7. The method for assessing the health status of a gas turbine rotor according to claim 6, characterized in that, The distribution divergence is the symmetric KL divergence between the current transmission delay spectrum and the health baseline transmission delay spectrum; The mapping is ,in This is the first distortion component. Let KL divergence be the symmetric KL divergence. Preset positive coefficient; The normalization is ,in This is the second distortion component. The cumulative area, These are the normalization coefficients; The third distortion component ,in , , The main peak delay, main peak width, and number of secondary peaks in the current propagation delay spectrum are given. , , For the aforementioned health benchmark characteristics, , , These are the weighting coefficients.

8. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, Step S5 is followed by: The distortion indexes that record multiple events form a sequence; If the sequence shows a significant monotonically increasing trend after trend testing, a progressive degradation warning is issued even if the current distortion index does not exceed the first threshold.

9. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The health status mentioned in step S3 is the state in which the rotor is confirmed to be free of cracks and creep damage by borehole inspection; the health benchmark feature is extracted by statistically averaging the transmission delay spectrum obtained under multiple health statuses.

10. The method for assessing the health status of a gas turbine rotor according to claim 1, characterized in that, The speed signal mentioned in step S1 is acquired by the gas turbine key phase sensor, and the vibration displacement signal is acquired by the gas turbine eddy current displacement sensor. The preset time delay scan range in step S2 is 0.5ms to 500ms.

Images