Fault early warning and positioning method for hot component of gas turbine

Abstract

The invention discloses a gas turbine hot component fault early warning and positioning method. The existing monitoring system is a protection system instead of an early warning system essentially. In practical application, when a monitoring system gives an alarm, hot parts of the gas turbine are often seriously damaged. The method comprises the following steps: acquiring data of gas turbine exhaust temperature measuring points and gas turbine loads from a database, and eliminating incomplete samples and interference items to obtain an effective sample set; traversing each sample in the effective sample set, sorting all exhaust temperature measuring point values in the samples, and calculating an average value and deviation to obtain a static deviation threshold model containing independent non-alarm value ranges of all measuring points; establishing a dynamic deflection angle model; performing abnormal state primary judgment; performing multi-rule alarm logic judgment according to the obtained abnormal state condition; and the system starts a fault positioning function and sends out an operation abnormity real-time early warning signal to prompt abnormity. The method is used for gas turbine hot component fault early warning and positioning.

Description

Technical Field

[0001] This invention relates to a method for early warning and location of faults in thermal components of a gas turbine. Background Technology

[0002] Gas turbines are one of the core pieces of equipment in power plants, and their hot-roller components are among the most important and expensive parts of the entire unit. These components operate under harsh conditions of high temperature and high pressure for extended periods, and their operational safety has a significant impact on the safe operation of the entire power plant. Statistics show that over 50% of gas turbine failures are related to the high-temperature hot-roller components, resulting in economic losses of tens of millions of yuan annually. Therefore, early warning systems for abnormal operation of gas turbine hot-roller components are particularly important.

[0003] Currently, power plants mainly rely on the following methods to monitor the thermal components of gas turbines: 1) Shutdown Inspection: The condition of critical components such as combustion chamber nozzles often needs to be inspected after the gas turbine is shut down, through disassembly or borehole probing. This method cannot achieve online real-time monitoring, and by the time faults such as nozzle erosion are discovered, severe thermal shock and secondary damage may have already occurred to downstream components; 2) Manufacturer protection systems: Power plants primarily rely on combustion protection systems provided by gas turbine manufacturers for monitoring. These systems typically determine combustion status by monitoring exhaust temperature distribution.

[0004] The existing technology has the following main problems and drawbacks: 1) Lack of early warning: Existing monitoring systems are essentially protection systems rather than early warning systems. In practical applications, by the time the monitoring system issues an alarm, the hot components of the gas turbine are often already severely damaged. 2) Forced shutdown: The triggering of alarms often means that the gas turbine can only operate at reduced load or shut down directly, which cannot be prevented in advance and leads to significant operational losses.

[0005] 3) Low sensitivity and high false alarm rate: The fundamental reason why existing systems struggle to provide early warnings is that parameters such as gas turbine exhaust temperature are highly susceptible to numerous interfering factors, including changes in operating load, environmental conditions, and manufacturing and installation errors. Under normal operating conditions, the range of changes in exhaust dispersion caused by these factors may even be greater than the changes caused by minor early-stage faults in hot components. Therefore, simply lowering the alarm threshold to improve sensitivity will result in an excessively high false alarm rate; while maintaining a high threshold will lead to a high missed alarm rate, failing to detect early anomalies. Summary of the Invention

[0006] The purpose of this invention is to address the problems of existing gas turbine thermal component monitoring systems, such as alarm delay, inability to achieve early warning, susceptibility to operating condition interference, and high false alarm and false alarm rates, by providing a method for early warning and location of gas turbine thermal component faults.

[0007] The above objectives are achieved through the following technical solutions: A method for early warning and location of faults in thermal components of a gas turbine, the method comprising the following steps: (1) By reviewing the operation log, determine the fault-free operation period of the gas turbine during this overhaul cycle, obtain the data of gas turbine exhaust temperature measurement point and gas turbine load from the database, and eliminate incomplete samples and interference items to obtain an effective sample set; (2) Traverse each sample in the effective sample set, sort all the exhaust temperature measurement points in the sample, calculate the average value and deviation to obtain a static deviation threshold model that includes the independent non-alarm value range of all measurement points. (3) Using the effective sample set obtained in step (1), establish a dynamic deflection angle model; (4) Read the real-time data of the gas turbine load and all exhaust temperature measurement points at the current moment. If the gas turbine load is lower than the rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. If the load meets the conditions, the calculation is exactly the same as step (2). (5) For each measuring point, compare its real-time deviation with the corresponding threshold range stored in the static deviation threshold model to make an initial judgment on the abnormal state; (6) Perform multi-rule alarm logic judgment based on the obtained abnormal status; (7) After the obtained multi-rule alarm logic triggers the formal early warning, the system starts the fault location function and issues a real-time early warning signal of abnormal operation to indicate the abnormality.

[0008] The method for early warning and location of faults in hot components of a gas turbine, wherein step (1) includes the following process: (11) By reviewing the operation log, determine the fault-free operating periods of the gas turbine during this overhaul cycle; (12) Extract data of all gas turbine exhaust temperature measurement points and gas turbine load during the period from the historical database at sampling intervals of 1 minute or 30 seconds; (13) Check the data integrity and delete incomplete samples that are missing measurement points or load data; (14) To eliminate interference from the start-up and shutdown process, the sample data of the gas turbine load being 40% lower than the rated load were deleted from the sample set to obtain the effective sample set.

[0009] The method for early warning and location of faults in hot components of a gas turbine, wherein step (2) includes the following process: (21) Traverse each sample in the valid sample set and sort all the exhaust temperature measurement points in the sample; (22) Delete the largest and smallest measuring point values ​​at a rate of 10%; (23) Calculate the average value of the remaining exhaust temperature measurement points as the average exhaust temperature of this sample; (24) Subtract the average exhaust temperature from all the original exhaust temperature measurement points in this sample to obtain the exhaust temperature deviation of each measurement point in this sample; (25) For each exhaust temperature measurement point, find the maximum and minimum deviation values ​​that have appeared in all samples; (26) For each measuring point, calculate its average deviation and deviation range; (27) Select a threshold coefficient k to amplify the deviation range; (28) Finally, the upper deviation alarm threshold and lower deviation alarm threshold of the measuring point are calculated, and a static deviation threshold model containing the independent non-alarm value range of all measuring points is established.

[0010] The method for early warning and location of faults in hot components of a gas turbine, wherein step (3) includes the following process: (31) Using the effective sample set screened in step (1), analyze the relationship between the gas turbine load and the highest temperature point of the gas turbine hot components; (32) By fitting the data, a functional relationship model describing the change of the gas turbine deflection angle with the power of the gas turbine generator is obtained, which is the dynamic deflection angle model. This functional model is stored together for real-time positioning.

[0011] The method for early warning and location of faults in hot components of a gas turbine, wherein step (4) includes the following process: (41) Read the real-time data of the gas turbine load and all exhaust temperature measuring points at the current moment; (42) If the gas turbine load is lower than the rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. (43) If the load meets the conditions, perform the same calculation as in step (2).

[0012] The method for early warning and location of faults in hot components of a gas turbine, wherein step (5) includes the following process: (51) For each measurement point, compare its real-time deviation with the corresponding threshold range stored in the static deviation threshold model; (52) If the real-time deviation exceeds the range, mark the status of the measuring point as abnormal and record the measuring point number corresponding to the maximum deviation and the measuring point number corresponding to the minimum deviation.

[0013] The method for early warning and location of faults in hot components of a gas turbine, wherein step (6) includes the following process: performing multi-rule alarm logic judgment based on the obtained abnormal state, with the following three rules: (61) Rule A: Combustion component malfunction; (62) Rule B: Persistent abnormality; (63) Rule C: Adjacent multiple points are abnormal; In step (5), if the system detects that two or more consecutively numbered measurement points are simultaneously marked as abnormal, the system will issue an alarm signal directly without waiting for the time filtering of rule B.

[0014] The method for early warning and location of faults in hot components of a gas turbine, wherein step (7) includes the following process: (71) When rule A+B or rule C triggers a formal warning, the system starts the fault location function; the system reads the current gas turbine load; (72) Based on the established dynamic deflection angle model, the current gas turbine operating deflection angle is calculated in real time; (73) Obtain the measurement point number that triggered the alarm, and use the positioning algorithm to calculate the actual combustion component number where the abnormality occurred; (74) Finally, the system issues a real-time warning signal for abnormal operation of the hot component. This signal not only indicates the abnormality, but also clearly gives the number of the combustion hot component that is abnormal. Beneficial effects

[0015] 1. This invention employs a rigorous multi-rule alarm logic in the real-time early warning phase. Through adjacent point verification rules, it effectively filters out instantaneous interference or sensor noise at individual measuring points; through duration judgment rules, it further eliminates non-continuous, sporadic system fluctuations; and through adjacent multi-point anomaly immediate alarm rules, it ensures that no sudden, serious faults are missed due to time filtering, balancing sensitivity and safety. Only anomalies confirmed both spatially and temporally will trigger an early warning, greatly improving the reliability of the early warning signal and significantly reducing the false alarm rate.

[0016] 2. This invention establishes a deflection angle model offline and applies a positioning algorithm in real time during early warning. It can reverse the abnormal measurement point number detected to deduce the actual combustion component number where the abnormality occurred, enabling operators and maintenance personnel to quickly identify the source of the fault and shorten the fault investigation and downtime maintenance time. Attached Figure Description

[0017] AppendixFigure 1 This is the overall framework diagram of the present invention. Detailed Implementation

[0018] Reference Figure 1 This invention relates to a method for early warning and location of faults in gas turbine thermal components. The aim is to establish a model that can effectively encompass the effects of various operating conditions and environmental changes, accurately distinguishing between normal operating data and early abnormal data. This allows for early warning of operational anomalies in gas turbine thermal components, detecting abnormalities before serious damage occurs or before the protection system is triggered (load reduction or shutdown). The goal is to improve the monitoring level of the operating status of gas turbine thermal components, reducing the number of unplanned shutdowns and high maintenance costs caused by thermal component failures. The method includes the following steps: (1) By reviewing the operation log, determine the fault-free operation period of the gas turbine during this overhaul cycle, obtain the data of gas turbine exhaust temperature measurement point and gas turbine load from the database, and eliminate incomplete samples and interference items to obtain an effective sample set; Step (1) includes the following process: (11) By reviewing the operation log, determine the fault-free operating periods of the gas turbine during this overhaul cycle; (12) Extract data of all gas turbine exhaust temperature measurement points and gas turbine load during the period from the historical database at sampling intervals of 1 minute or 30 seconds; (13) Check the data integrity and delete incomplete samples that are missing measurement points or load data; (14) To eliminate interference from the start-up and shutdown process, the sample data of the gas turbine load being 40% lower than the rated load were deleted from the sample set to obtain the effective sample set.

[0019] (2) Traverse each sample in the effective sample set, sort all the exhaust temperature measurement points in the sample, calculate the average value and deviation to obtain a static deviation threshold model that includes the independent non-alarm value range of all measurement points. (21) Traverse each sample in the valid sample set and sort all the exhaust temperature measurement points in the sample; (22) Delete the largest and smallest measuring point values ​​at a rate of 10%; (23) Calculate the average value of the remaining exhaust temperature measurement points as the average exhaust temperature tp_avg of this sample; (24) Subtract the average exhaust temperature tp_avg from all the original exhaust temperature measurement points in this sample to obtain the exhaust temperature deviation tp_err for each measurement point in this sample; (25) For each exhaust temperature measurement point, find the maximum deviation tp_max and the minimum deviation tp_min that have appeared in all samples; (26) For each measuring point, calculate its average deviation dev_avg = (tp_max + tp_min) / 2 and deviation range = tp_max - tp_min; (27) Select a threshold coefficient k = 1.4 to amplify the deviation range: krange = k * range; (28) Finally, the upper deviation alarm threshold of the measuring point is th_upper = dev_avg + 0.5 * krange, and the lower deviation alarm threshold is th_lower = dev_avg - 0.5 * krange. Based on this, a static deviation threshold model containing the independent non-alarm value range [th_lower, th_upper] of all measuring points can be obtained.

[0020] (3) Using the effective sample set obtained in step (1), establish a dynamic deflection angle model; (31) Using the effective sample set screened in step (1), analyze the relationship between the gas turbine load and the highest temperature point of the gas turbine hot components; (32) By fitting the data, a function Gdeg = f(Pw) is obtained to describe the change of the gas turbine deflection angle Gdeg with the gas turbine generator power Pw. This is the dynamic deflection angle model. This function model is stored together for real-time positioning.

[0021] (4) Read the real-time data of the gas turbine load and all exhaust temperature measurement points at the current moment. If the gas turbine load is lower than the rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. If the load meets the conditions, the calculation is exactly the same as step (2). (41) Read the real-time data of the gas turbine load Pw and all exhaust temperature measuring points at the current moment; (42) If Pw is lower than the 40% rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. (43) If the load meets the conditions, perform the same calculation as step (2): sort all current real-time exhaust temperature measurement points; delete the maximum and minimum measurement points proportionally; calculate the average value of the remaining measurement points as the average exhaust temperature value cur_avg for this calculation; calculate the real-time deviation tp_dev between each exhaust temperature measurement point and cur_avg.

[0022] (5) For each measuring point, compare its real-time deviation with the corresponding threshold range stored in the static deviation threshold model to make an initial judgment on the abnormal state; (51) For each measurement point, compare its real-time deviation tp_dev with the corresponding threshold range [th_lower, th_upper] stored in the static deviation threshold model; (52) If tp_dev exceeds this range (tp_dev > th_upper or tp_dev < th_lower), mark the status of this measurement point as abnormal, and record the measurement point number NOmax corresponding to the maximum deviation and the measurement point number Nomin corresponding to the minimum deviation.

[0023] (6) Perform multi-rule alarm logic judgment based on the obtained abnormal status; Perform multi-rule alarm logic judgment based on the obtained abnormal status. There are three rules as follows: (61) Rule A: Combustion component is abnormal; Take the occurrence of the maximum deviation (NOmax) as an example: When the measurement point NOmax is marked as abnormal, the system does not immediately alarm, but additionally reads the real-time deviations of its adjacent measurement points (such as NOmax - 1 and NOmax + 1), and determines whether the real-time deviations of these two adjacent measurement points also exceed their respective upper deviation thresholds th_upper (for example, T(NOmax - 1) - cur_avg > th_upper_for_NOmax - 1). Only when the NOmax point is abnormal and both its left and right adjacent points are determined to be abnormal, the system initially judges that the combustion heat component is abnormal. The judgment logic for the minimum deviation NOmin is the same, comparing whether its adjacent points are all lower than th_lower.

[0024] (62) Rule B: Continuous abnormality; If the judgment result of Rule A is that the combustion heat component is initially judged to be abnormal, the system starts a timer. If the abnormal state of this combustion heat component lasts for a duration Th exceeding a preset threshold Ts = 30 seconds (60 seconds is also acceptable, with lower sensitivity requirements), the system officially issues a warning of abnormal operation of the heat component. If the duration does not exceed Ts, it is regarded as interference and no alarm is issued.

[0025] (63) Rule C: Adjacent multiple points are abnormal; In step (5), if the system detects that 2 or more consecutively numbered measurement points are simultaneously marked as abnormal, the system directly issues an alarm signal without waiting for the time filtering of Rule B.

[0026] (7) After the multi-rule alarm logic triggers a formal warning, the system starts the fault location function and issues a real-time warning signal of abnormal operation to indicate the abnormality.

[0027] (71) When rule A+B or rule C triggers a formal warning, the system activates the fault location function. The system reads the current gas turbine load Pw; (72) Based on the deflection angle model Gdeg = f(Pw) established in step (3), calculate the current gas turbine deflection angle Gdeg in real time; (73) Obtain the measurement point number that triggered the alarm, apply the positioning algorithm, and calculate the actual combustion component number NOr that caused the abnormality: NOr = NOmax - int(Gdeg / a) + b or NOr = NOmin - int(Gdeg / a) + b, where b is the total number of combustion heat components, a is 360 / b, and int is the rounding function; (74) Finally, the system issues a real-time warning signal for abnormal operation of the hot component. This signal not only indicates the abnormality, but also clearly gives the number NOr of the combustion hot component where the abnormality occurred.

[0028] In steps (2) and (4) of this invention, the method of calculating tp_avg and cur_avg by removing extreme values ​​and then averaging is used. The median can be used instead of the shortened average. The median is also robust to extreme outliers, is simple to calculate, and can achieve a similar effect in containing operating condition fluctuations.

[0029] In step (3) of this invention, a data fitting method is used to establish a continuous function model Gdeg = f(Pw). Alternatively, a discrete lookup table can be established instead of a continuous function. This table stores the fixed deflection angles corresponding to different load intervals. During real-time positioning, the deflection angle is obtained directly by querying the table based on the interval where the current load Pw is located.

[0030] This invention establishes a dynamically adaptive benchmark model: abandoning the traditional fixed threshold, it statistically models historical data of gas turbines over long periods and under all operating conditions. The core of this model is to calculate the normal deviation range of each exhaust temperature measurement point relative to a dynamically corrected average value under healthy conditions. This correction is achieved by removing extreme maximum and minimum values ​​before calculating the average value, thus enabling the model to adaptively accommodate normal operational fluctuations.

[0031] Multi-dimensional alarm rules are integrated: In the real-time warning phase, more complex judgment logic is introduced. It not only judges whether a single measuring point deviates from the baseline, but also introduces adjacent point verification logic. That is, when an anomaly occurs at one point, its adjacent measuring points must also show a coordinated deviation before it is confirmed as a combustion component anomaly. Simultaneously, by combining duration judgment and multi-point simultaneous anomaly rules, the accuracy of the warning is significantly improved, effectively distinguishing between transient interference and genuine faults.

[0032] Achieving precise fault location: After confirming the anomaly, this invention introduces an operating deflection angle model. This model analyzes historical data to fit a functional relationship between the power generation load and the deflection angle of the high-temperature point measurement location. During real-time early warning, this model and the current load are used to reverse-engineer the actual combustion component number where the anomaly occurred by analyzing the detected abnormal measurement point number, thus achieving a closed loop from anomaly detection to precise fault location.

Claims

1. A method for early warning and location of faults in thermal components of a gas turbine, characterized in that: The method includes the following steps: (1) By reviewing the operation log, determine the fault-free operation period of the gas turbine during this overhaul cycle, obtain the data of gas turbine exhaust temperature measurement point and gas turbine load from the database, and eliminate incomplete samples and interference items to obtain an effective sample set; (2) Traverse each sample in the effective sample set, sort all the exhaust temperature measurement points in the sample, calculate the average value and deviation to obtain a static deviation threshold model that includes the independent non-alarm value range of all measurement points. (3) Using the effective sample set obtained in step (1), establish a dynamic deflection angle model; (4) Read the real-time data of the gas turbine load and all exhaust temperature measurement points at the current moment. If the gas turbine load is lower than the rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. If the load meets the conditions, the calculation is exactly the same as step (2). (5) For each measuring point, compare its real-time deviation with the corresponding threshold range stored in the static deviation threshold model to make an initial judgment on the abnormal state; (6) Perform multi-rule alarm logic judgment based on the obtained abnormal status; (7) After the obtained multi-rule alarm logic triggers the formal early warning, the system starts the fault location function and issues a real-time early warning signal of abnormal operation to indicate the abnormality.

2. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (1) includes the following process: (11) By reviewing the operation log, determine the fault-free operating periods of the gas turbine during this overhaul cycle; (12) Extract data of all gas turbine exhaust temperature measurement points and gas turbine load during the period from the historical database at sampling intervals of 1 minute or 30 seconds; (13) Check the data integrity and delete incomplete samples that are missing measurement points or load data; (14) To eliminate interference from the start-up and shutdown process, the sample data of the gas turbine load being 40% lower than the rated load were deleted from the sample set to obtain the effective sample set.

3. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (2) includes the following process: (21) Traverse each sample in the valid sample set and sort all the exhaust temperature measurement points in the sample; (22) Delete the largest and smallest measuring point values ​​at a rate of 10%; (23) Calculate the average value of the remaining exhaust temperature measurement points as the average exhaust temperature of this sample; (24) Subtract the average exhaust temperature from all the original exhaust temperature measurement points in this sample to obtain the exhaust temperature deviation of each measurement point in this sample; (25) For each exhaust temperature measurement point, find the maximum and minimum deviation values ​​that have appeared in all samples; (26) For each measuring point, calculate its average deviation and deviation range; (27) Select a threshold coefficient k to amplify the deviation range; (28) Finally, the upper deviation alarm threshold and lower deviation alarm threshold of the measuring point are calculated, and a static deviation threshold model containing the independent non-alarm value range of all measuring points is established.

4. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (3) includes the following process: (31) Using the effective sample set screened in step (1), analyze the relationship between the gas turbine load and the highest temperature point of the gas turbine hot components; (32) By fitting the data, a functional relationship model describing the change of the gas turbine deflection angle with the power of the gas turbine generator is obtained, which is the dynamic deflection angle model. This functional model is stored together for real-time positioning.

5. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (4) includes the following process: (41) Read the real-time data of the gas turbine load and all exhaust temperature measuring points at the current moment; (42) If the gas turbine load is lower than the rated load threshold set in step (1), it is determined that the calculation conditions are not met, no warning judgment is made, and the next calculation cycle is waited for. (43) If the load meets the conditions, perform the same calculation as in step (2).

6. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (5) includes the following process: (51) For each measurement point, compare its real-time deviation with the corresponding threshold range stored in the static deviation threshold model; (52) If the real-time deviation exceeds the range, mark the status of the measuring point as abnormal and record the measuring point number corresponding to the maximum deviation and the measuring point number corresponding to the minimum deviation.

7. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (6) includes the following process: performing multi-rule alarm logic judgment based on the obtained abnormal status, with the following three rules: (61) Rule A: Combustion component malfunction; (62) Rule B: Persistent abnormality; (63) Rule C: Adjacent multiple points are abnormal; In step (5), if the system detects that two or more consecutively numbered measurement points are simultaneously marked as abnormal, the system will issue an alarm signal directly without waiting for the time filtering of rule B.

8. The method for early warning and location of faults in thermal components of a gas turbine according to claim 1, characterized in that: Step (7) includes the following process: (71) When rule A+B or rule C triggers a formal warning, the system starts the fault location function; the system reads the current gas turbine load; (72) Based on the established dynamic deflection angle model, the current gas turbine operating deflection angle is calculated in real time; (73) Obtain the measurement point number that triggered the alarm, apply the positioning algorithm, and calculate the actual combustion component number where the abnormality occurred; (74) Finally, the system issues a real-time warning signal for abnormal operation of the hot component. This signal not only indicates the abnormality, but also clearly gives the number of the combustion hot component that is abnormal.

Images