A gas turbine rub fault diagnosis method based on multi-feature fusion

Abstract

The application discloses a kind of based on multi-feature fusion's gas turbine rub-impact fault diagnosis method, step one, the running data of gas turbine is collected in real time by multiple sensors;Step two, feature extraction is carried out to vibration data, and the key time-frequency features related to rub-impact are calculated;Step three, for speed data, the rub-impact energy index is calculated according to slight collision energy model;Step four, feature extraction is carried out to temperature data, and whether temperature anomaly appears is judged according to temperature feature condition;Step five, the weight of each feature is adjusted, and the weighted algorithm is used to fuse multiple features;Step six, the fault decision threshold is determined using ROC curve, and whether the fault occurs is judged;The application can improve the precision of gas turbine rub-impact fault diagnosis, monitor the running state of potential machine in real time, give early warning to abnormal condition, and adjust the operating condition of gas turbine according to the diagnosis result, so as to reduce the probability of rub-impact fault.

Description

Technical Field

[0001] This invention relates to the field of gas turbine fault diagnosis technology, and specifically to a gas turbine rubbing fault diagnosis method based on multi-feature fusion. Background Technology

[0002] As gas turbines evolve towards lighter, heavier, and higher-speed operation, their structural designs become more compact, requiring smaller rotor clearances. This makes rotor rubbing failure a major technical bottleneck. To address this issue, patent number 202411062505.0, entitled "A Control Method for Preventing Dynamic-Static Rubbing During Cold Start of a 9E Gas Turbine," indirectly adjusts the coaxiality between the turbine exhaust cylinder center and the generator center by adjusting the expansion of the turbine exhaust end bearing support leg. However, this adjustment method lacks flexibility and makes it difficult to predict fault occurrence in advance. Patent number 202410566546.7, entitled "A Method, Electronic Equipment, and Medium for Extracting Rotor Rubbing Fault Features in Rotating Machinery," uses the NS-TEMD method to decompose rotor vibration signals, effectively separating the power frequency component and transient impact component of early dynamic-static rubbing signals. However, the algorithm has low interpretability and unstable diagnostic results. Therefore, it is of great significance to invent a flexible and efficient method for diagnosing gas turbine rubbing faults. However, commonly used rubbing fault diagnosis methods often rely on vibration monitoring, but the early slight rubbing characteristics are not obvious and are easily drowned out by noise. Moreover, the vibration characteristics are easily confused with other faults. It is necessary to use indicators that can better reflect the essential characteristics of the fault as supplementary criteria. To solve these problems, the inventors proposed a rubbing fault diagnosis method based on multi-feature fusion. Summary of the Invention

[0003] The purpose of this invention is to provide a gas turbine rubbing fault diagnosis method based on multi-feature fusion. This method collects vibration, speed, and temperature data from multiple sensors, calculates time-frequency signals related to fluctuations based on the vibration data, establishes a minor collision energy model based on the speed data, and calculates the rubbing capability index. Feature indicators are extracted from temperature data, and a weighted fusion algorithm is used to fuse multiple features to obtain a comprehensive score, which is then compared with the fault threshold to determine whether a fault has occurred.

[0004] This invention provides a method for diagnosing gas turbine rubbing faults based on multi-feature fusion, the method being as follows:

[0005] Step 1: Real-time acquisition of gas turbine operating data using multiple sensors such as vibration sensors, speed sensors, and temperature sensors;

[0006] Step 2: Extract features from the vibration data and calculate the key time-frequency features related to rubbing.

[0007] Step 3: Calculate the collision energy index η based on the collision energy model using the rotational speed data;

[0008] Step 4: Extract features from the temperature data and determine whether there is a temperature anomaly based on the temperature features.

[0009] Step 5: Adjust the weights of each feature and use a weighted algorithm to fuse multiple features;

[0010] Step 6: Use the ROC curve to determine the fault decision threshold, and then compare the fused score with the fault decision threshold to determine whether a fault has occurred.

[0011] Furthermore, the feature set selection and processing are as follows:

[0012] (1) Divide the data panes and calculate four types of features that are highly correlated with signal fluctuations: root mean square amplitude, kurtosis, waveform factor and autocorrelation function as vibration signal features.

[0013] Root mean square amplitude: ;

[0014] kurtosis: ;

[0015] In the formula The mean, The standard deviation is denoted as .

[0016] Waveform factor: ;

[0017] In the formula This is the absolute average amplitude.

[0018] Autocorrelation function: ;

[0019] (2) To eliminate the differences in the dimensions between different variables, the features are normalized.

[0020] Furthermore, the calculation process for the energy index of the minor collision energy model is as follows:

[0021] (1) The Jeffcott rotor model is used to make assumptions about the parameters of the system. The Coulomb friction model is used to describe the friction force, and higher-order nonlinear factors are ignored. The first-order response of the rotor is mainly analyzed. An initial gap between the rotor and the stator is considered. When the rotor radial displacement Exceed If so, it is considered that a collision has occurred.

[0022] (2) Based on d'Alembert's principle, the dynamic equations of the rotor system are established. The equation of motion of the rotor center when rubbing occurs is: ;

[0023] in denoted as ε, where ε is the rotor radial displacement and e is the rotor eccentricity.

[0024] Equation without friction: ;

[0025] in At that time, the stator does not participate in the action.

[0026] (3) To facilitate analysis, the variables are treated as dimensionless: ;

[0027] in For speed ratio, This is the damping ratio for attenuation.

[0028] After simplification, the rubbing time is obtained. When not touched The corresponding nonlinear differential equations can be used to study the system at different speed ratios. With damping ratio The corresponding patterns below.

[0029] (4) When the system is about to rub against each other, the kinetic energy of the rotor in the radial direction is defined as the collision energy. :

[0030] Where A is the rotor vibration amplitude, when This indicates that a collision is about to occur. This indicates the intensity of rubbing contact.

[0031] Substituting the analytical solution obtained from the steady-state response of amplitude A under non-collision conditions into the following: ;

[0032] After substituting, we get: ;

[0033] (5) To make the results relative and comparable, a relative collision energy index is introduced. Defined as: ;

[0034] in This represents the relative rubbing energy level under a unit system parameter. This indicates that the system is operating safely. There is a risk of collision. The larger the value, the higher the severity and probability of the collision.

[0035] (6) By changing the damping ratio Speed ​​ratio ,draw The response curve is used to analyze the sensitivity of the parameters.

[0036] when hour, Follow Monotonically increasing;

[0037] when hour, It rises first and then falls, with a peak value.

[0038] (7) In Drawing on a plane The joint distribution map revealed sensitive areas with a high risk of collisions. This diagram can be used to quantitatively assess the severity and probability of rubbing failures in gas turbines or rotor systems.

[0039] Furthermore, the temperature data acquisition and feature calculation process is as follows:

[0040] (1) At the upper and lower parts of the compressor exhaust cylinder, ≥3 temperature measuring points are arranged on the outer shell sections of the first and second stage turbine blades, symmetrically distributed to form a cylinder temperature distribution matrix.

[0041] (2) A temperature sensor with a sampling frequency of 1-5Hz is used to connect the signal to the DCS online monitoring system to obtain the temperature gradient change during the startup process in real time.

[0042] (3) Extract key temperature characteristic parameters:

[0043] Average temperature: ;

[0044] Temperature difference index: ;

[0045] Temperature rise rate: ;

[0046] Temperature asymmetry: ;

[0047] in, The upper cylinder temperature, This refers to the temperature of the lower cylinder.

[0048] (4) If in the cold start stage or If so, it is determined that there is a significant risk of thermal deviation.

[0049] Furthermore, the feature weights are set based on the following criteria:

[0050] (1) Select the fault signal with the greatest difference from the pre-experiment health signal, standardize the various feature data, homogenize the heterogeneous indicators, and use different algorithms to standardize the positive and negative indicators.

[0051] Positive indicators: ;

[0052] Negative indicators: ;

[0053] For ease of representation, the normalized data Still recorded as .

[0054] (2) Calculate the first The first item under the indicator The proportion of each sample value to the indicator : ;

[0055] (3) Calculate the first Entropy value of the item indicator: ;

[0056] in, ,satisfy .

[0057] (4) Calculate information entropy redundancy: ;

[0058] (5) Calculate the weights of each indicator: ;

[0059] (6) Calculate the comprehensive score using a weighted fusion algorithm. : ;

[0060] in .

[0061] Furthermore, the methods for calculating the fault threshold and determining the fault are as follows:

[0062] (1) Integrate the score Sort by size from largest to smallest, and calculate each candidate threshold. Yoden Index : ;

[0063] Where TPR is the true positive rate and FPR is the false positive rate.

[0064] (2) All Connect the dots to form a curve, i.e., the ROC curve, which serves as an optimization tool for the threshold T.

[0065] (3) Maximum Oden Index Corresponding threshold This is the optimal threshold, while the area under the curve (AUC) is used to measure overall performance.

[0066] (4) If If so, it indicates that the machine has malfunctioned.

[0067] The design concept of this invention is as follows:

[0068] This invention collects vibration, rotational speed, and temperature data using multiple sensors, calculates time-frequency signals related to fluctuations based on the vibration data, establishes a collision energy model based on the rotational speed data, and calculates the impact resistance index. This invention extracts feature indicators from temperature data and then uses a weighted fusion algorithm to fuse multiple features to obtain a comprehensive score, which is compared with a fault threshold to determine whether a fault has occurred. This invention improves the accuracy of gas turbine rubbing fault diagnosis, while also enabling real-time monitoring of the submersible's operating status, providing early warnings of abnormal situations, and adjusting the gas turbine's operating conditions based on the diagnostic results, thereby reducing the probability of rubbing faults.

[0069] This invention has the following advantages:

[0070] (1) Strong interpretability: The slight collision energy model explains the physical meaning of collision energy. The model combines parameters such as rotational speed and damping ratio to calculate the collision energy index. This index can be used to assess the severity and probability of collision failure of the system. The definition of sensitive area can also help operators adjust operating conditions in a timely manner to avoid entering high-risk areas.

[0071] (2) High accuracy: This invention integrates the multi-source characteristics of vibration, rotational speed and temperature with nonlinear energy indicators, which improves the reliability of diagnosis. It also uses the ROC threshold method to quantify the risk level in order to assess the severity of the gas turbine rubbing failure.

[0072] (3) Low monitoring cost: The present invention can monitor the equipment during operation without stopping the machine, ensuring the working efficiency of the gas turbine, and without changing the structure and materials of the gas turbine components, resulting in low monitoring cost. At the same time, the present invention can be adapted to different models of gas turbines and rotor machinery, and has strong versatility. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of a multi-feature fusion method for diagnosing gas turbine rubbing faults.

[0074] Figure 2 This is a mechanical model diagram of the Jeffcott rotor system;

[0075] Figure 3 This is a diagram illustrating the friction force generated by touching and rubbing.

[0076] Figure 4 This is a schematic diagram of a slight collision energy model;

[0077] Figure 5 This is a diagram illustrating the contact / rubbing situation;

[0078] Figure 6 It is the relative rubbing energy index that follows Joint distribution map

[0079] Figure 7 This is a schematic diagram of the weighted fusion algorithm;

[0080] Figure 8 This is a diagram for fault diagnosis. Detailed Implementation

[0081] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings. These are merely exemplary embodiments of the present invention; however, it should be understood that the present invention can be implemented in various forms and is not limited to the embodiments described herein. These embodiments are provided to enable those skilled in the art to understand the present invention more clearly and thoroughly.

[0082] A PG9171(E) gas turbine at a power plant experienced cylinder deformation due to frequent start-stop cycles over a long period. During a cold start under load, dynamic and static contact occurred, causing vibrations to continuously increase and eventually leading to a turbine trip. Traditional monitoring methods rely on vibration threshold alarms: an alarm is triggered when the vibration exceeds 0.5 in / s, and a trip is initiated when the vibration exceeds 1 in / s. This alarm method relies on empirical thresholds, and it is difficult to trigger an alarm if the vibration amplitude does not exceed the limit, thus missing early, subtle faults. Therefore, this invention integrates multiple features to comprehensively screen and judge fault conditions.

[0083] Figure 1 shows a multi-feature fusion method for diagnosing gas turbine rubbing faults. The specific steps are as follows:

[0084] Step 1: Real-time acquisition of gas turbine operating data using multiple sensors such as vibration sensors, speed sensors, and temperature sensors;

[0085] Step 2: Extract features from the vibration data and calculate the key time-frequency features related to the impact and friction.

[0086] Step 3: Calculate the collision energy index (η) based on the rotational speed data and the collision energy model.

[0087] Step 4: Extract features from the temperature data and determine whether there is a temperature anomaly based on the temperature features.

[0088] Step 5: Adjust the weights of each feature and use a weighted algorithm to fuse multiple features;

[0089] Step 6: Use the ROC curve to determine the fault decision threshold, and then compare the fused score with the fault decision threshold to determine whether a fault has occurred.

[0090] In specific implementation, the feature set selection and processing in step two are as follows:

[0091] (1) Collect the radial acceleration signal of the bearing, with a sampling frequency of 10kHz and a data window length of 1s. Calculate the peak values ​​of four types of features with high correlation to signal fluctuation within the data window: root mean square amplitude, kurtosis, waveform factor and autocorrelation function as vibration signal features.

[0092] Root mean square amplitude: ;

[0093] kurtosis: ;

[0094] In the formula The mean, The standard deviation is denoted as .

[0095] Waveform factor: ;

[0096] In the formula This is the absolute average amplitude.

[0097] Autocorrelation function: ;

[0098] (2) In order to eliminate the difference in the dimensions between different variables, the features are normalized to obtain the feature vector.

[0099] In specific implementation, the calculation process for the energy index of the collision energy model in step three is as follows:

[0100] (1) Real-time acquisition of rotor speed signal. During the startup process, the speed increases from 0 to the rated speed of 3000 rpm. The Jeffcott rotor model is used to make assumptions about the parameters of the system, as shown in Figure 2. Rotor mass 1200kg, rotation speed Stator radial stiffness at 3000 rpm 1.5×10 7 N / m, rotor stiffness coefficient 1.2×10 7 N / m, rotor radial displacement is The radial and tangential friction forces are respectively and Considering the initial gap between the rotor and stator Approximately 0.5mm, when hour, ;

[0101] when hour, ;

[0102] When the rotor radial displacement u exceeds If so, it is considered that a collision has occurred.

[0103] (2) The friction force is described using the Coulomb friction model, and higher-order nonlinear factors are ignored. The first-order response of the rotor is mainly analyzed, as shown in Figure 3. The dynamic equation of the rotor system is established according to d'Alembert's principle. The motion equation of the rotor center when friction occurs is: ;

[0104] in denoted as , where is the rotor radial displacement and e is the rotor eccentricity.

[0105] Equation without friction: ;

[0106] in At that time, the stator does not participate in the action.

[0107] (3) To facilitate analysis, the variables are treated as dimensionless: ;

[0108] in For speed ratio, To reduce the damping ratio, Set it to 0.05.

[0109] After simplification, the rubbing time is obtained. When not touched The corresponding nonlinear differential equations can be used to study the system at different speed ratios. With damping ratio The corresponding patterns below.

[0110] (4) As shown in Figure 4, when the system is about to rub against each other, the kinetic energy of the rotor in the radial direction is defined as the collision energy. : ;

[0111] Where A is the rotor vibration amplitude, when This indicates that a collision is about to occur. This indicates the intensity of rubbing contact.

[0112] Substituting the analytical solution obtained from the steady-state response of amplitude A under non-collision conditions into the following: ;

[0113] After substituting, we get: ;

[0114] (5) To make the results relative and comparable, a relative collision energy index is introduced. Defined as: ;

[0115] in This represents the relative rubbing energy level under a unit system parameter. This indicates that the system is operating safely. There is a risk of collision. The larger the value, the more severe the impact and the higher the probability of impact. Possible impact failures are shown in Figure 5. Real-time rotational speed. The value is 2850 rpm. Substituting this value into a minor collision energy model, the relative rubbing energy index is calculated. The value is 0.85 (>0.5 indicates a risk of collision).

[0116] (6) By changing the damping ratio Speed ​​ratio ,draw The response curve is used to analyze the sensitivity of the parameters.

[0117] when hour, Follow Monotonically increasing;

[0118] when hour, It rises first and then falls, with a peak value.

[0119] (7) In Draw on a plane as shown in Figure 6 The joint distribution map revealed sensitive areas with a high risk of collisions. This diagram can be used to quantitatively assess the severity and probability of rubbing failures in gas turbines or rotor systems.

[0120] In specific implementation, the temperature data acquisition and feature calculation process in step four is as follows:

[0121] (1) At the upper and lower parts of the compressor exhaust cylinder, ≥3 temperature measuring points are arranged on the outer shell sections of the first and second stage turbine blades, symmetrically distributed to form a cylinder temperature distribution matrix.

[0122] (2) A temperature sensor with a sampling frequency of 1-5Hz is used to connect the signal to the DCS online monitoring system to obtain the temperature gradient change during the startup process in real time.

[0123] (3) Extract key temperature characteristic parameters:

[0124] Average temperature: ;

[0125] Temperature difference index: ;

[0126] Temperature rise rate: ;

[0127] Temperature asymmetry: ;

[0128] in, The upper cylinder temperature, This refers to the temperature of the lower cylinder.

[0129] (4) Calculate the average temperature difference between the upper and lower parts of the cylinder block within 30 minutes after startup. for Temperature asymmetry for According to empirical thresholds, if the temperature difference and temperature rise rate exceed the safety threshold during the cold start-up phase ( or If the result is positive, it indicates a risk of thermal deviation.

[0130] ;

[0131] in The features are normalized. For feature weights, and .

[0132] In specific implementation, in step five, the feature weights are set based on the following criteria for determining the fault condition:

[0133] (1) Conduct preliminary experiments and collect health signals of the gas turbine under normal operating conditions as a reference benchmark.

[0134] (2) Select the fault signal with the greatest difference from the pre-experiment health signal, standardize the various feature data, homogenize the heterogeneous indicators, and use different algorithms to standardize the positive and negative indicators.

[0135] Positive indicators: ;

[0136] Negative indicators: ;

[0137] For ease of representation, the normalized data Still recorded as .

[0138] (3) Calculate the first The first item under the indicator The proportion of each sample value to the indicator : ;

[0139] (4) Calculate the first Entropy value of the indicator: ;

[0140] in, ,satisfy .

[0141] (5) Calculate information entropy redundancy: ;

[0142] (6) Calculate the weights of each indicator: ;

[0143] Based on historical health and fault samples, the vibration feature weight is 0.35, the rubbing energy index weight is 0.40, and the temperature feature weight is 0.25.

[0144] (7) As shown in Figure 7, the weighted fusion algorithm is used to calculate the comprehensive score. : ;

[0145] in The current state comprehensive score is obtained after weighted fusion. It is 0.72.

[0146] In specific implementation, step six involves the following methods for calculating the fault threshold and determining the fault:

[0147] (1) Integrate the score Sort by size from largest to smallest, and calculate the threshold for each candidate. Yoden Index :

[0148] Where TPR is the true positive rate and FPR is the false positive rate.

[0149] (2) All Connect the dots to form a curve, i.e., the ROC curve, which serves as an optimization tool for the threshold T.

[0150] (3) Maximum Oden Index Corresponding threshold This is the optimal threshold. The area under the curve (AUC) is used to measure overall performance. Based on historical data ROC curves, the optimal fault determination threshold is... It is 0.60.

[0151] (4) As shown in Figure 8, due to This indicates a machine malfunction, triggering an alarm in the system.

[0152] To demonstrate the advancements of this invention compared to existing technologies, the diagnostic results of this embodiment are compared with those of the traditional vibration threshold method, as shown in the table below: ;

[0153] The comparative results show that the method described in this invention, during the cold start-up of the PG9171(E) gas turbine, can issue an early warning of rubbing faults based on multi-source feature fusion indicators, before the vibration reaches the traditional alarm threshold. This improves the gas turbine system's ability to prevent early faults and allows valuable time for operational adjustments. Simultaneously, this invention provides interpretable risk quantification assessment indicators through rubbing energy and temperature characteristics, offering a reference direction for locating the specific cause of the fault. Furthermore, this method is applicable to a wider range of faults than traditional methods, adapting to complex scenarios such as cold starts, cylinder deformation, and vibration-induced cylinder tripping, demonstrating outstanding practical value.

[0154] The above description is merely a preferred embodiment of the present invention and does not constitute any limitation on the present invention. Any simple modifications, alterations, or equivalent structural changes made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for diagnosing gas turbine rubbing faults based on multi-feature fusion, characterized in that, Includes the following steps: Step 1: Real-time acquisition of gas turbine operating data using multiple sensors, including vibration sensors, speed sensors, and temperature sensors; Step 2: Extract features from the vibration data and filter out time-frequency features related to rubbing; Step 3: Calculate the collision energy index based on the rotational speed data and the collision energy model. ; Step 4: Extract features from the temperature data and determine whether there is a temperature anomaly based on the temperature features. Step 5: Adjust the weights of each feature and use a weighted algorithm to fuse multiple features; Step 6: Use the ROC curve to determine the fault decision threshold, and then compare the fused score with the fault decision threshold to determine whether a fault has occurred.

2. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, Step 1 is described in detail as follows: 2.1) Divide the data panes and calculate the four types of features that are most correlated with signal fluctuations as vibration signal features, including: root mean square amplitude, kurtosis, waveform factor and autocorrelation function; 2.2) Normalize the various features.

3. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, Step 3 is as follows: 3.1) Establish the Jeffcott rotor dynamics model. Assumptions are made about the system parameters, including the speed ratio, damping ratio, and the existence of an initial gap between the rotor and stator. When the rotor radial displacement Exceed At that time, it is considered that a collision has occurred; 3.2) Calculate radial collision energy Combined with rotor vibration amplitude With initial gap Based on the relationship between the radial collision energy and the steady-state response under no-collision conditions, a formula for calculating radial collision energy is obtained. 3.3) Introducing the relative collision energy index , This represents the relative rubbing energy level under a unit system parameter. This indicates that the system is operating safely. There is a risk of collision. The larger the value, the higher the severity and probability of the collision. 3.4) Drawing With speed ratio With damping ratio The joint response distribution will Between 0.6 and 1.6, The distribution area in the range of 0 to 0.3 is defined as the sensitive area.

4. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, Step 4 is as follows: 4.1) Temperature measuring points are arranged in the thermal stress-sensitive areas of the gas turbine, including the upper and lower parts of the compressor exhaust cylinder; 4.2) Connect the signal to the DCS online monitoring system to collect temperature signals in real time; 4.3) The temperature difference that reflects the spatial temperature gradient and the rate of temperature rise reflecting the temperature time gradient The index is used as a temperature characteristic; 4.4) If during the cold start phase or If so, it is determined that there is a significant risk of thermal deviation.

5. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 1, characterized in that, Step 5 is described in detail below: 5.1) Determine the weight of each feature based on the information entropy, and the sum of the weights of all features is 1; 5.2) For the sample Overall score It is calculated by summing the products of each normalized feature and its corresponding feature weight.

6. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 5, characterized in that, Step 6 is as follows: 6.1) Use ROC curves to analyze the true positive rate and false positive rate under different candidate thresholds; 6.2) Calculate the Youden index corresponding to each candidate threshold. ; 6.3) Maximum Gordian Index Corresponding threshold This is the optimal threshold; 6.4) When At that time, the system issued a collision warning.

7. The gas turbine rubbing fault diagnosis method based on multi-feature fusion as described in claim 5, characterized in that, The basis for setting the weights in 5.1) is as follows: 5.1.1) Collect characteristic signals of the gas turbine under normal operating conditions as health signals, where normal operating conditions refer to all parameters of the gas turbine being within the set range; 5.1.2) Select the fault signal with the greatest difference from the pre-experiment health signal, standardize various feature data, realize the homogenization of heterogeneous indicators, and use different algorithms to standardize positive and negative indicators. 5.1.3) Calculate the first The first item under the indicator The proportion of each sample value to the indicator ; 5.1.4) Calculate the first The entropy value e of the item index j ; 5.1.5) Calculate the information entropy redundancy d j ; 5.1.6) Calculate the weight ω of each indicator. j ; 5.1.7) Calculate the comprehensive score using a weighted fusion algorithm. .

8. The gas turbine rubbing fault diagnosis method based on multi-feature fusion according to claim 6, characterized in that, The calculation method for the fault threshold in section 6.1 is as follows: 6.1.1) Integrate the scoring Sort by size from largest to smallest, and calculate the threshold for each candidate. Yoden Index The Yoden index is obtained by subtracting the false positive rate from the true positive rate. 6.1.2) Plot the false positive rate and true positive rate corresponding to each candidate threshold as the horizontal and vertical axes to obtain the ROC curve; 6.1.3) Maximum Yorden Index Corresponding threshold That is, the optimal threshold, and the area under the curve (AUC) is used to measure the overall performance; 6.1.4) If If so, it indicates that the machine has malfunctioned.

Images