Gas generator set fault detection method and device, equipment and storage medium

Abstract

The invention provides a gas generator set fault detection method and device, equipment and a storage medium, and belongs to the technical field of gas power generation, and the method comprises the steps: obtaining target detection data collected by different types of sensors of a gas generator set, and calculating a correlation coefficient between the target detection data and a target parameter; adding the target detection data of which the correlation coefficient is greater than a set threshold value into a fusion candidate set, and calculating a target weight corresponding to each target detection data in the fusion candidate set; based on each piece of target detection data in the fusion candidate set and the target weight corresponding to each piece of target detection data, obtaining comprehensive detection data; performing feature extraction on the comprehensive detection data to obtain time domain features and frequency domain features corresponding to the comprehensive detection data; and determining the fault type of the fault of the gas generator set based on the time domain feature and the frequency domain feature. According to the gas generator set fault detection method and device, the equipment and the storage medium provided by the invention, the accuracy of fault detection can be improved.

Description

Technical Field

[0001] This application belongs to the field of gas-fired power generation technology, and more specifically, relates to a method, device, equipment, and storage medium for fault detection of gas-fired generator sets. Background Technology

[0002] Gas generator sets, as important energy conversion equipment, play a crucial role in industrial production and energy supply. Their operational stability directly affects the safety and efficiency of the entire system. However, due to their complex structure and variable operating conditions, gas generator sets are prone to various faults during long-term operation. Failure to detect and accurately determine the type of fault in a timely manner can lead to accelerated equipment damage and even serious safety accidents, resulting in significant economic losses.

[0003] Currently, traditional fault detection methods for gas generator sets mostly rely on data from a single sensor. However, the information obtained by a single sensor is limited and cannot fully and accurately reflect the complex operating status of the unit. This can easily lead to misjudgments and missed judgments, which greatly limits the accuracy of fault detection. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, equipment, and storage medium for fault detection of gas generator sets, so as to improve the accuracy of fault detection.

[0005] A first aspect of this application provides a method for detecting faults in a gas generator set, comprising: Acquire target detection data collected by different types of sensors in the gas generator set. For each sensor's target detection data, calculate the correlation coefficient between the target detection data and the target parameter. The target parameter is a key parameter reflecting the operating status of the gas generator set. The target detection data is the preprocessed data of the raw detection data from different types of sensors. Add the target detection data with a correlation coefficient greater than a set threshold to the fusion candidate set, and calculate the target weight corresponding to each target detection data in the fusion candidate set; Based on the detection data of each target in the fusion candidate set and the target weight corresponding to each target detection data, comprehensive detection data is obtained; Feature extraction is performed on the comprehensive detection data to obtain the time-domain and frequency-domain features corresponding to the comprehensive detection data; The fault type of the gas generator set is determined based on time-domain and frequency-domain characteristics.

[0006] A second aspect of this application provides a gas generator set fault detection device, comprising: The correlation coefficient calculation module is used to acquire target detection data collected by different types of sensors in the gas generator set. For each sensor's target detection data, the correlation coefficient between the target detection data and the target parameter is calculated. The target parameter is a key parameter reflecting the operating status of the gas generator set. The target detection data is the preprocessed data of the raw detection data from different types of sensors. The weight calculation module is used to add target detection data with a correlation coefficient greater than a set threshold to the fusion candidate set, and to calculate the target weight corresponding to each target detection data in the fusion candidate set. The data fusion module is used to obtain comprehensive detection data based on the detection data of each target in the fusion candidate set and the target weight corresponding to each target detection data. The feature extraction module is used to extract features from the comprehensive detection data to obtain the time-domain and frequency-domain features corresponding to the comprehensive detection data. The fault detection module is used to determine the fault type of the gas generator set based on time-domain and frequency-domain characteristics.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described gas generator set fault detection method.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described gas generator set fault detection method.

[0009] The beneficial effects of the gas generator set fault detection method, device, equipment, and storage medium provided in this application are as follows: This application acquires preprocessed target detection data from different types of sensors, calculates the correlation coefficient with the target parameters, and filters out data with correlation coefficients greater than a set threshold to add to the fusion candidate set. This eliminates interfering data and focuses on key information. By calculating the target weights of the data in the fusion candidate set and then fusing the data based on these weights to obtain comprehensive detection data, the role of each effective data point can be fully utilized, comprehensively and accurately reflecting the unit's status. Extracting time-domain and frequency-domain features from the comprehensive detection data allows for the discovery of richer fault information. Based on these features, the fault type can be determined, improving the accuracy of fault detection, effectively avoiding equipment damage and safety accidents, and reducing economic losses. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating a gas generator set fault detection method provided in an embodiment of this application. Figure 2 This is a structural block diagram of a gas generator set fault detection device provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0014] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a gas generator set fault detection method provided in an embodiment of this application. The method can be executed by electronic equipment and may include: S101: Acquire target detection data collected by different types of sensors of the gas generator set. For the target detection data collected by each sensor, calculate the correlation coefficient between the target detection data and the target parameter. The target parameter is a key parameter reflecting the operating status of the gas generator set. The target detection data is the data after preprocessing the original detection data of different types of sensors.

[0015] In this embodiment, the gas generator set is a power device that uses gas as fuel to convert the chemical energy of gas into electrical energy.

[0016] Different types of sensors are detection devices installed in key parts of gas generator sets (such as engine block, turbine blades, gas pipelines, etc.) to collect physical quantity data such as temperature, pressure, vibration, and flow rate.

[0017] Target parameters are key indicators that reflect the core operating status or fault characteristics of a gas generator set (such as combustion efficiency, turbine blade vibration intensity, and gas flow stability), and are parameters used to measure whether a gas generator set is operating normally.

[0018] The correlation coefficient is a statistical indicator that measures the degree of linear correlation between two variables. In the fault detection of gas generator sets, by calculating the correlation coefficient between the target detection data and the target parameters, the ability of each sensor data to characterize the unit's operating status can be evaluated.

[0019] In this embodiment, raw detection data collected by different types of sensors from the gas generator set are acquired; the raw detection data is preprocessed to obtain target detection data. The preprocessing of the raw detection data includes: Timestamp alignment is performed on the raw detection data collected by different types of sensors to ensure that the time dimension of the data from each sensor is consistent. Adopt 3 The criteria identify outliers in the raw detection data and perform interpolation to repair the outliers; The original detection data after repair is standardized and mapped to the [0,1] interval; The target detection data is obtained by filtering noise from the standardized data using a wavelet threshold denoising algorithm.

[0020] In this embodiment, the sensors of the gas generator set (such as vibration, temperature, and pressure sensors) may have asynchronous timestamps in the collected data due to differences in hardware sampling frequency or transmission delay. For example, if the vibration sensor collects data 1000 times per second, while the temperature sensor collects data 100 times per second, direct analysis will result in a misalignment of the time dimension.

[0021] Therefore, the timestamps of all sensor data can be calibrated using the unit's main controller clock as a reference. Missing data can be supplemented through linear interpolation, or the sampling frequency can be unified by downsampling (e.g., adjusting all to 100 times per second) to ensure that each set of data from different sensors corresponds to the equipment status at the same moment. This provides a time-synchronized data foundation for subsequent calculations of the correlation coefficient between sensor data and target parameters, avoiding distortion in correlation analysis due to time misalignment.

[0022] In this embodiment, the original detection data may also show abnormal values ​​due to sensor failure (such as momentary circuit break) or electromagnetic interference (such as vibration data suddenly jumping to 0 or values ​​far exceeding the normal range).

[0023] Therefore, it can be based on 3 The criteria (data values ​​exceeding the mean ± 3 standard deviations are considered outliers) are used to iterate through the original data sequence and mark outliers. For marked outliers, linear interpolation of nearby data (e.g., using the mean of the five normal data points before and after the outlier) or a sliding window mean method is used to correct them, avoiding interference from outliers in subsequent analysis. Noise points in the data are removed to ensure the continuity and reliability of the data sequence.

[0024] In this embodiment, the data dimensions and numerical ranges of different types of sensors vary greatly (e.g., temperature is measured in °C, with a value range of 20-100; vibration is measured in g, with a value range of 0.01-0.5). Direct fusion would lead to the parameters with larger values ​​dominating the analysis results.

[0025] Therefore, the min-max normalization formula is used to map all sensor data to the [0,1] interval. The min-max normalization formula is expressed as:

[0026] in, This represents the standardized sensor data. This represents the minimum value of the sensor data. This represents the maximum value of the sensor data. This represents the raw test data.

[0027] Standardized sensor data can eliminate dimensional differences, enabling different types of data such as temperature, pressure, and vibration to be compared and fused on the same order of magnitude, ensuring the fairness of correlation coefficient calculation and weight allocation.

[0028] Finally, a wavelet thresholding denoising algorithm is employed. The db4 wavelet basis is selected to perform a three-level decomposition on the standardized sensor data. The high-frequency coefficients obtained from the decomposition (mainly containing noise) are processed using a soft thresholding function, and then the data is reconstructed through inverse wavelet transform, filtering out high-frequency noise. This preserves the effective signals reflecting the equipment status in the data (such as low-frequency vibrations or trend changes in fault characteristics), improving the accuracy of subsequent feature extraction, for example, more clearly capturing the subtle vibration changes in early bearing wear.

[0029] In this embodiment, after the aforementioned timestamp alignment, outlier identification and repair, standardization processing, and noise filtering, the original detection data is transformed into time-synchronized, anomaly-free, dimensionless, and low-noise target detection data. Simultaneously, target parameters reflecting the core operating status of the unit (such as combustion efficiency, turbine speed stability, blade vibration intensity, and other key indicators) are identified. For the target detection data corresponding to each sensor, the Pearson correlation coefficient statistical method can be used to calculate the linear correlation between the data and the target parameter, obtaining the correlation coefficient. This identifies sensor data closely related to the operating status of the gas generator set, providing a basis for subsequent data fusion.

[0030] S102: Add the target detection data with a correlation coefficient greater than the set threshold to the fusion candidate set, and calculate the target weight corresponding to each target detection data in the fusion candidate set.

[0031] In this embodiment, the fusion candidate set is a collection of target detection data whose correlation coefficient exceeds a set threshold. These data are considered to be closely related to the target parameters and have the value to participate in data fusion.

[0032] The target weight is the weight value of each target detection data in the fusion candidate set, which is used to measure the importance of the data in the fusion process.

[0033] In this embodiment, target detection data with correlation coefficients exceeding a set threshold (e.g., absolute correlation coefficient ≥ 0.6) can be filtered out to form a fusion candidate set. These data are considered to be of significant value in reflecting target parameters. Based on the correlation coefficients between each target detection data and the target parameters in the candidate set, the target weight corresponding to each data can be calculated through normalization processing (e.g., the percentage of absolute correlation coefficients). Target detection data with higher correlation strength receives higher weights, ensuring that their impact on device status is more prominent in subsequent fusion.

[0034] S103: Based on the detection data of each target in the fusion candidate set and the target weight corresponding to each target detection data, the comprehensive detection data is obtained.

[0035] In this embodiment, the comprehensive detection data is the comprehensive data obtained by weighted fusion of target detection data in the fusion candidate set. It integrates the effective information from multiple sensors and can more comprehensively reflect the status of the gas generator set.

[0036] In this embodiment, a weighted fusion algorithm can be used to combine and calculate (e.g., weighted summation) all target detection data in the fusion candidate set according to their respective target weights to obtain comprehensive detection data. By integrating information from multiple highly correlated sensors, this embodiment can overcome the limitations of single sensor data, forming more comprehensive and reliable device status characterization data, and providing high-quality input for subsequent feature extraction.

[0037] S104: Extract features from the comprehensive detection data to obtain the time-domain and frequency-domain features corresponding to the comprehensive detection data.

[0038] In this embodiment, time-domain features are statistical features (such as mean, variance, and peak value) extracted from the time series of comprehensive detection data, which can reflect the numerical patterns and overall trends of data changes over time.

[0039] Frequency domain features are features extracted by converting time-domain data to the frequency domain using Fourier transform (such as dominant frequency and spectral energy distribution), which can reflect the energy distribution patterns of different frequency components in the data.

[0040] In this embodiment, multi-dimensional feature extraction can be performed on the comprehensive detection data: from the time domain perspective, statistical quantities such as mean, variance, peak value, and kurtosis are calculated to reflect the overall trend and numerical distribution characteristics of the data over time; from the frequency domain perspective, the time domain data is converted into the frequency domain through Fourier transform to extract features such as the main frequency and spectral energy distribution, revealing the hidden periodic patterns and frequency components in the data.

[0041] S105: Determine the fault type of the gas generator set based on time-domain and frequency-domain characteristics.

[0042] In this embodiment, the fault type refers to the specific types of faults that may occur in the gas generator set (such as gas leakage, bearing wear, ignition system failure, and turbine imbalance), and is the final judgment result of fault detection.

[0043] In this embodiment, the extracted time-domain and frequency-domain features are input into the fault detection model (such as a machine learning model or a deep learning model). The fault detection model can judge the current equipment status by learning the mapping relationship between features and fault types, and finally output the specific fault type (such as gas leakage, bearing wear, ignition system abnormality, etc.). This step realizes the transformation from data features to fault diagnosis results, and completes the identification of faults in gas generator sets.

[0044] For example, a gas-fired power plant performs fault detection on the bearing system of a 1.5MW gas generator set. The specific process is as follows: Vibration sensors (to collect bearing vibration signals), temperature sensors (to monitor bearing temperature), and oil pressure sensors (to collect lubrication system pressure) are deployed, with raw data sampling frequencies of 1000Hz, 100Hz, and 100Hz, respectively. In the preprocessing stage, using the unit's main controller clock as a reference, the vibration data is downsampled to 100Hz to achieve timestamp alignment; 3 The criteria identify instantaneous pulse anomalies in vibration signals and repair them by interpolation of the mean of 5 points before and after; min-max normalization is used to map the three types of data to the [0,1] interval; high-frequency noise is filtered by db4 wavelet 3-level decomposition to obtain target detection data.

[0045] The bearing vibration intensity was selected as the target parameter. The correlation coefficients between the vibration, temperature, and pressure data and the target parameter were calculated to be 0.82, 0.65, and 0.41, respectively. A threshold of 0.6 was set, and the vibration and temperature data were included in the fusion candidate set. The weights were calculated based on the percentage of the absolute value of the correlation coefficients: vibration data 0.82 / (0.82+0.65)=0.56, temperature data 0.44. The weighted fusion yielded the comprehensive detection data.

[0046] During the feature extraction stage, time-domain features showed that the mean vibration increased from 0.05g to 0.12g, and the kurtosis increased from 3.2 to 6.8. Frequency-domain features, through Fourier transform, revealed that the energy proportion of the 120Hz main frequency increased from 15% to 42% (corresponding to the characteristic frequency of bearing rolling element failure). Inputting these features into the BiLSTM-Attention model, the model outputs the bearing wear failure type, which is consistent with the disassembly inspection results, verifying the effectiveness of the method.

[0047] As can be seen from the above, this embodiment acquires preprocessed target detection data from different types of sensors, calculates the correlation coefficient with the target parameters, and filters out data with correlation coefficients greater than a set threshold to add to the fusion candidate set. This eliminates interfering data and focuses on key information. By calculating the target weights of the data in the fusion candidate set and then fusing the data based on these weights to obtain comprehensive detection data, the role of each effective data point can be fully utilized to comprehensively and accurately reflect the unit's status. Extracting time-domain and frequency-domain features from the comprehensive detection data can uncover richer fault information. Based on these features, the fault type can be determined, improving the accuracy of fault detection, effectively avoiding equipment damage and safety accidents, and reducing economic losses.

[0048] In one embodiment of this application, calculating the correlation coefficient between target detection data and target parameters includes: The correlation coefficient between target detection data and target parameters is calculated based on the Pearson correlation coefficient. The Pearson correlation coefficient is expressed as:

[0049] in, This represents the correlation coefficient between the detection data of the i-th target and the target parameter T; This represents the specific value of the i-th target detection data at time t; This represents the average value of the i-th target detection data over a preset time period; This represents the specific value of the target parameter T at time t; This represents the average value of the target parameter T over a preset time period; n represents the total number of time points.

[0050] In this embodiment, the Pearson correlation coefficient is used to calculate the correlation coefficient between target detection data and target parameters. By measuring the degree of linear correlation between target detection data (such as feature data collected by sensors, monitoring indicators, etc.) and preset target parameters (such as standard thresholds, performance indicators, targets to be optimized, etc.), the strength and direction of the association between the two are determined, providing a basis for subsequent data analysis, decision-making, or model optimization.

[0051] In this embodiment, This represents the sum of the products of the deviations between the target detection data and the target parameters. This represents the deviation of the i-th detection data from its own mean at time t, reflecting the degree to which the data deviates from the overall level at that time. This represents the deviation of the target parameter from its mean at time t. and If the product of the two is positive, it indicates that at that moment, the detected data and the target parameter are simultaneously higher or lower than their respective means (same-direction deviation); if it is negative, it indicates opposite-direction deviation. The sign and magnitude of the sum reflect the overall coordinated change trend of the two.

[0052] In this embodiment, The standard deviation of the i-th test data (i.e., the square root of the variance, not divided by n-1 or n, but without affecting the relative magnitude of the correlation) reflects the dispersion of the test data itself. It represents the unstandardized standard deviation of the target parameter T, reflecting the degree of dispersion of the target parameter. The product of standard deviations serves to standardize the numerator, eliminating the influence of data dimensions (such as units and orders of magnitude) and making the data more uniform. It can uniformly measure correlation within the range of [-1,1].

[0053] As can be seen from the above, this embodiment, by using the Pearson correlation coefficient to calculate the correlation between target detection data and target parameters, can accurately quantify the degree of linear correlation between the two. Mean correction eliminates differences in data benchmarks, and standard deviation normalization avoids the influence of dimensions, making the results more comparable and reliable, and improving the effectiveness of the detection data application.

[0054] In one embodiment of this application, calculating the target weight corresponding to each target detection data in the fusion candidate set includes: Based on each target detection data in the fusion candidate set, the correlation coefficient between the target detection data and the target parameters is normalized to obtain the initial weight corresponding to each target detection data. Based on the detection data of each target in the fusion candidate set, the failure rate and calibration overdue rate of the sensor corresponding to each target detection data are obtained, and based on the failure rate and calibration overdue rate of the sensor corresponding to each target detection data, the health index of the sensor corresponding to each target detection data is obtained. The initial weights corresponding to each target detection data are corrected based on the health index to obtain the target weights corresponding to each target detection data.

[0055] In this embodiment, normalization is the process of converting correlation coefficients of different magnitudes or ranges to a unified interval (such as 0-1) to ensure that the weights of each data are comparable and additive, and to avoid imbalance in weight allocation due to differences in the absolute values ​​of the coefficients.

[0056] The failure rate is the ratio of the number of times a sensor fails within a preset time to the total operating time. It reflects the stability of the sensor, and a higher failure rate indicates that the reliability of the sensor's output data is lower.

[0057] The calibration overdue rate is the proportion of a sensor's actual usage time exceeding the calibration cycle. Calibration overdue will lead to deviations in the detection data, and a higher overdue rate indicates a greater risk to the accuracy of the data.

[0058] The health index is a quantitative indicator derived from the combined sensor failure rate and calibration overdue rate, used to assess the overall reliability of the sensor.

[0059] The target weight is the final weight obtained after correcting the initial weight. It takes into account the correlation between the detection data and the target parameters as well as the reliability of the sensor itself, and is used for weighted calculation during subsequent data fusion.

[0060] In this embodiment, after obtaining the correlation coefficient between the target detection data and the target parameter, the correlation coefficients of all candidate data are normalized (e.g., the coefficients are converted into weight values ​​with a sum of 1 proportionally) to obtain the initial weight of each data, ensuring that the detection data with higher correlation has a larger initial weight.

[0061] For each target detection data corresponding to the sensor, its failure rate (the probability of failure occurring within a certain period of time) and calibration overdue rate (the proportion of time not calibrated on time) are collected. These two indicators are combined into a sensor health index through a preset algorithm (such as weighted summation, threshold scoring, etc.). The higher the health index, the more reliable the sensor status.

[0062] The initial weights are corrected using the sensor health index (if the health index is high, the weight of the corresponding data is increased, and vice versa), and finally the target weights that take into account both data relevance and sensor reliability are obtained, making the fusion results more accurate and reliable.

[0063] For example, suppose a gas generator set fusion candidate set includes vibration sensors ( ) and temperature sensor ( The target detection data was used. The correlation coefficient between the vibration sensor and the target parameter (bearing vibration intensity) was calculated to be 0.8, and the correlation coefficient between the temperature sensor and the target parameter (bearing vibration intensity) was 0.6. Both were normalized: initial weights... =0.8 / (0.8+0.6)=0.57, =0.6 / (0.8+0.6)=0.43.

[0064] Collect sensor status data. The system has been running for 720 hours in the past 30 days, with 3 failures (all of which were momentary communication interruptions), resulting in a failure rate of 3 / 720 = 0.0042 failures per hour. The calibration cycle is 90 days, and the system has been running for 30 days so the calibration overdue rate is 0 (not overdue).

[0065] The system has been running for 720 hours in the past 30 days, with 10 failures (including 5 data drifts), resulting in a failure rate of 10 / 720 = 0.0139 failures / hour. The calibration cycle is 60 days, and the system has been running for 78 days, exceeding the calibration deadline by 18 days, resulting in a calibration overdue rate of 18 / 60 = 0.3.

[0066] The weighted formula is: Health Index = 1 - (0.6 × Normalized Failure Rate + 0.4 × Calibration Overdue Rate).

[0067] Failure rate normalization: =0.0042 / (0.0042+0.0139)=0.23, =0.0139 / (0.0042+0.0139)=0.77.

[0068] Health Index: =1-(0.6×0.23+0.4×0)=0.86, =1-(0.6×0.77+0.4×0.3)=0.44.

[0069] Corrected formula: Target weight = (Initial weight × Health index) / Sum of (Initial weight × Health index) of all candidate sensors, resulting in: Target weight = (0.57 × 0.86) / (0.57 × 0.86 + 0.43 × 0.44) = 0.71 Target weight = 0.29.

[0070] In the final target weights, Due to the increased proportion of high correlation and high reliability, The high failure rate and the decrease in the proportion of serious overdue data make the fused data more consistent with the actual condition of the equipment.

[0071] As can be seen from the above, this embodiment employs a two-step weighting mechanism—obtaining initial weights through correlation coefficient normalization and correcting for sensor health indices—to ensure that data with stronger correlation to the target parameters receives higher weights. Furthermore, it quantifies sensor reliability through failure rate and calibration overdue rate, dynamically adjusting the weights. The final target weights balance data correlation and sensor status, reducing interference from unreliable data and improving the accuracy and robustness of the fusion results.

[0072] In one embodiment of this application, comprehensive detection data is obtained based on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data, including: We perform weighted fusion on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data to obtain the initial fused data; The deviation between the real-time fluctuation characteristics of the target parameter and the historical normal fluctuation characteristics is calculated. The historical normal fluctuation characteristics are obtained based on historical data statistics under normal operating conditions of the gas generator set. Based on the deviation, the corresponding dynamic calibration coefficient is obtained by querying the preset fluctuation correction coefficient table. The initial fusion data is coupled with the dynamic calibration coefficients to obtain comprehensive detection data.

[0073] In this embodiment, the initial fusion data is fusion data obtained by weighting by the target weight, which reflects the basic correlation characteristics of multi-sensor data, but does not consider the fluctuation of the real-time operating status of the device. Real-time fluctuation characteristics are the fluctuation features of target parameters within the current detection period, including variance, peak fluctuation frequency, trend change rate, etc., reflecting the dynamic changes in equipment operation. Historical normal fluctuation characteristics are target parameter fluctuation benchmarks obtained from the statistical analysis of historical data of long-term normal operation of gas generator sets. These include normal fluctuation range, frequency distribution pattern, trend stability, etc., and serve as a reference standard for judging whether the current fluctuation is abnormal. Deviation is an indicator that measures the difference between real-time fluctuation characteristics and historical normal fluctuation characteristics. The larger the value, the more the current fluctuation deviates from the normal state. It is often used to identify changes in operating conditions or potential faults. The fluctuation correction coefficient table is a preset table of correspondence between deviation and calibration coefficient, which is determined through training with historical data. For example, a deviation of 0.3 corresponds to a coefficient of 1.1, which is used to convert the degree of fluctuation into a quantifiable calibration parameter. The dynamic calibration coefficient is a coefficient obtained from the correction coefficient table based on the real-time deviation. It is used to dynamically adjust the initial fusion data so that the comprehensive test data is more in line with the current operating status of the equipment. Coupled computation is the process of performing mathematical operations (such as multiplication and weighted summation) on the initial fused data and dynamic calibration coefficients to combine the basic fused data with real-time state correction, and finally generate comprehensive detection data.

[0074] In this embodiment, a weighted summation algorithm is used to perform fusion calculations based on the target detection data and its corresponding target weight in the fusion candidate set. For example, when the vibration data weight is 0.7 and the temperature data weight is 0.3, the initial fused data = vibration data × 0.7 + temperature data × 0.3. By assigning weights to highlight the contribution of high-value data, multi-source sensor information is initially integrated to form a basic characterization of the equipment status.

[0075] Obtain real-time fluctuation characteristics of the target parameters (such as the standard deviation of vibration amplitude and pressure fluctuation frequency within 10 minutes from a certain moment) and historical normal fluctuation characteristics (obtained based on historical data during normal unit operation, such as the mean of the standard deviation of vibration amplitude under normal operating conditions). The fluctuation frequency is concentrated in the range of 5-10Hz.

[0076] The deviation between the real-time fluctuation characteristics of the target parameter and the historical normal fluctuation characteristics is calculated based on the relative deviation formula. The relative deviation formula is: Deviation = in, The standard deviation of the real-time fluctuation of the target parameter. This represents the standard deviation of historical normal fluctuations. A result of 0.3 indicates that the current fluctuation range deviates from the normal fluctuation range by 30%. The deviation measure measures the degree of abnormality in the operating status of the gas generator set. The higher the deviation, the greater the difference between the current operating condition and the normal condition, and the more important it is to enhance the sensitivity of the data to potential faults.

[0077] Based on the calculated deviation, the preset fluctuation correction coefficient table is queried. The fluctuation correction coefficient table is generated by training with historical fault data, as shown in Table 1.

[0078]

[0079] When the deviation is 0.3, the calibration coefficient is found to be 1.2 according to the table. The initial fused data is then mathematically coupled with the dynamic calibration coefficient (e.g., multiplied) to obtain the comprehensive detection data: Comprehensive test data = Dynamic calibration coefficient For example, if the initial fusion data is 0.5 and the calibration coefficient is 1.2, then the overall detection data is 0.6.

[0080] When equipment fluctuations are abnormal (high deviation), the amplitude of key features is amplified by calibration coefficients to enhance the data's sensitivity to early faults; when fluctuations are normal, the coefficients maintain the baseline value to avoid noise amplification caused by overcorrection, thus ensuring the stability and accuracy of the data.

[0081] As can be seen from the above, this embodiment obtains an initial fusion result by weighted fusion of multi-source data, and then dynamically calibrates the fused data by combining the deviation between the target parameters in real time and historical normal fluctuations. This approach leverages the dominant role of high-weight data while adapting to changes in equipment operating conditions through dynamic coefficients, enhancing the data's sensitivity to abnormal fluctuations, reducing the limitations of single static fusion, and improving the accuracy and adaptability of comprehensive detection data.

[0082] In one embodiment of this application, determining the fault type of a gas generator set based on time-domain and frequency-domain characteristics includes: The importance of time-domain features and frequency-domain features is evaluated, and importance scores for the time-domain features and frequency-domain features are obtained respectively; Features that meet the first condition and have an importance score are used as key features; the first condition is time-domain and frequency-domain features with an importance score greater than an importance threshold. Key features are input into a pre-defined fault classification model to obtain the fault type of the gas generator set. The fault classification model is trained based on the time-domain features, frequency-domain features, and labeled fault types corresponding to historical fault data.

[0083] In this embodiment, importance assessment is a process of quantifying the contribution of features in fault identification. The algorithm calculates the degree of influence of each feature on the classification result. The higher the score, the better the feature can distinguish fault types.

[0084] Importance score is a value (e.g., in the range of 0-1) that assesses the importance of a feature. It reflects the contribution of a feature to fault identification and serves as the basis for selecting key features.

[0085] The first condition is the criterion for selecting key features, namely, the importance score must be greater than the preset importance threshold to ensure that the most critical features for fault identification are retained.

[0086] Key features are high-contribution features selected after importance assessment. They reflect the core feature patterns of faults and are the core data input into the classification model.

[0087] Fault classification models are models built based on machine learning or deep learning algorithms. They are trained using historical fault data and can output the corresponding fault type based on input features, thereby achieving automated fault identification.

[0088] Historical fault data consists of characteristic data (time domain and frequency domain features) and corresponding fault type labels recorded when gas generator sets experienced past faults. It is the basic data for training fault classification models.

[0089] In this embodiment, the importance of the extracted time-domain features (such as mean and peak value) and frequency-domain features (such as dominant frequency and spectral energy) is quantified, and a feature importance evaluation algorithm is used to calculate the score of each feature. For example, in bearing wear faults, the kurtosis (time domain) and the proportion of 120Hz frequency energy (frequency domain) of the vibration signal may score significantly higher than other features, reflecting their greater contribution to fault identification.

[0090] An importance threshold is set (e.g., the score of the top 20% of features), and features with scores exceeding the threshold are selected as key features. For example, if a kurtosis score of 0.8 and a 120Hz energy percentage score of 0.7 are both higher than the threshold of 0.5, they are retained; while a mean temperature score of 0.3 is discarded. This reduces interference from redundant features, lowers the computational load of the model, and retains the core features most sensitive to faults.

[0091] Key features are input into a pre-defined fault classification model. This model has been trained using historical fault data (containing features of various fault types and their corresponding labeling types) and can learn the mapping relationship between features and fault types. For example, inputting key features such as high kurtosis and high 120Hz energy proportion will cause the model to output the bearing wear fault type, thus completing the identification from features to faults.

[0092] For example, in the fault detection of a gas generator set, the extracted time-domain features include vibration mean (F1), kurtosis (F2), and temperature peak (F3), and the frequency-domain features include the energy proportion at 120Hz (F4) and the energy proportion at 50Hz (F5).

[0093] The importance of time-domain and frequency-domain features was evaluated using the random forest Gini coefficient method: F1 score 0.2, F2 score 0.8, F3 score 0.3, F4 score 0.7, and F5 score 0.1. An importance threshold of 0.5 was set, and F2 (kurtosis) and F4 (120Hz energy percentage) were selected as key features.

[0094] F2 (measured value 6.8) and F4 (measured value 42%) were input into a convolutional neural network fault classification model trained on historical data. The model learns the feature patterns of historical bearing wear cases where kurtosis > 5 and 120Hz energy percentage > 30%, and outputs the fault type as bearing wear.

[0095] As can be seen from the above, this embodiment selects key features by evaluating feature importance, reducing redundant information interference and lowering the computational load of the model. Combined with a fault classification model trained on historical data, it accurately identifies fault types using key features, retaining the most critical features for fault identification while improving diagnostic accuracy through the mapping relationships learned by the model, thus achieving efficient and reliable fault type determination.

[0096] In one embodiment of this application, the gas generator set fault detection method further includes: Calculate the confidence score of the fault type output by the fault classification model; In response to the failure classification model outputting a failure type with a confidence level lower than a confidence threshold, target key features are obtained from the key features. The target key features are those key features arranged from high to low importance scores, satisfying a preset number of features. The preset number is less than the number of key features. The fault type of the gas generator set is determined by matching the target key features with the fault case library of gas generator sets based on the matching results.

[0097] In this embodiment, the confidence level is a quantitative value (e.g., in the range of 0-1) that quantifies the reliability of the output results of the fault classification model. It reflects the model's certainty regarding the matching relationship between the current feature and the fault type. The higher the value, the more reliable the judgment.

[0098] The confidence threshold is the critical value that triggers a secondary decision. When the model outputs a confidence level lower than this value, the result needs to be verified through other means to avoid misjudgment caused by low reliability.

[0099] Multiple key features are sorted from high to low based on their importance scores. The target key features are the top N target key features selected from the key features (N is a preset number). The target key features are the core features that are most representative of the fault and are used to improve the accuracy of case matching.

[0100] The preset number is the number of key features selected for the target, which is less than the total number of key features. The purpose is to focus on core features and reduce matching complexity.

[0101] The fault case database is a database that stores historical fault events of gas generator sets. Each record contains information such as the key characteristics of the fault, the fault type, and the handling solution, which serves as a reference for case matching.

[0102] In this embodiment, the confidence level of the fault classification model output (such as the probability value of the model predicting bearing wear) is first calculated. If the value is lower than the preset confidence threshold (such as 0.7), it indicates that the model's recognition reliability of the current feature is insufficient, triggering a secondary judgment mechanism.

[0103] The selected key features are sorted from highest to lowest importance score, and the top N features are selected as target key features (N is a preset number, such as the first 2 when there are 5 key features). For example, if the key features are kurtosis (score 0.8), 120Hz energy ratio (0.7), and temperature fluctuation (0.6), the first 2 are selected as target key features, focusing on the core features that are most distinguishable from faults.

[0104] The target's key features are matched with historical cases in the fault case database using similarity analysis (e.g., by calculating feature vector similarity using Euclidean distance). The fault type corresponding to the case with the highest matching degree is selected as the final result. For example, if the target's key feature "kurtosis 6.8 + 120Hz energy 42%" has a feature similarity of 92% with the "bearing wear" case in the case database, it is determined to be this fault type.

[0105] As can be seen from the above, this embodiment calculates the confidence level of the model output and triggers a secondary judgment mechanism when the confidence level is insufficient. Highly important target key features are selected and matched with a fault case library to supplement the model's shortcomings with historical experience. This approach achieves efficient initial judgment through the model and improves the diagnostic reliability of low-confidence scenarios by leveraging the case library, reducing misjudgments caused by feature ambiguity or novel faults, and enhancing the robustness and accuracy of fault type determination.

[0106] In one embodiment of this application, matching is performed between target key features and a fault case library of gas generator sets, and the fault type of the gas generator set fault is determined based on the matching result, including: Calculate the weighted similarity between the key features of the target and each historical failure case in the failure case library; The fault type corresponding to the historical fault case with the highest weighted similarity is selected as the candidate fault type; and the actual occurrence probability of the candidate fault type within a preset time window is obtained. If the actual probability of occurrence is greater than the probability threshold, the candidate fault type is determined as the fault type of the gas generator set.

[0107] In this embodiment, the weighted similarity is the similarity between the target key feature and the historical case features calculated by combining the feature importance weights. The higher the weight of the feature, the greater its impact on the similarity result, and the more accurately it reflects the matching degree of the core features.

[0108] Candidate fault types are the fault types corresponding to the historical fault cases with the highest weighted similarity to the target key features, and are the suspected fault types obtained from the initial matching.

[0109] The preset time window is the time range within which the actual probability of failure is statistically analyzed, ensuring that the probability data reflects recent equipment failure patterns and enhancing timeliness.

[0110] The actual occurrence probability is the proportion of the number of times a candidate fault type actually occurs within a preset time window to the total number of faults, quantifying the actual occurrence frequency of that fault type.

[0111] The probability threshold is a critical value for determining the validity of candidate fault types. It is used to filter fault types that have high similarity but rarely occur in reality, so as to avoid rare cases interfering with the judgment results.

[0112] In this embodiment, the key features of the target and the corresponding features of each historical fault case in the fault case library are weighted according to the feature importance score and the similarity is calculated. For example, if the key features of the target are kurtosis (weight 0.6) and 120Hz energy ratio (weight 0.4), and the cosine similarity between these two features of a certain historical case and the target features are 0.9 and 0.8 respectively, then the weighted similarity = 0.6 × 0.9 + 0.4 × 0.8 = 0.86, which measures the degree of fit between the feature patterns of the two.

[0113] The fault type corresponding to the historical fault case with the highest weighted similarity is selected as the candidate fault type (such as bearing wear). At the same time, the actual occurrence probability of this type within a preset time window (such as the past year) is calculated (such as 12 occurrences out of 30 faults, with a probability of 40%), reflecting the actual occurrence frequency of this fault type.

[0114] If the actual occurrence probability of a candidate fault type is greater than a preset probability threshold (e.g., 30%), it is confirmed as the final fault type; otherwise, further investigation is required. For example, the actual occurrence probability of bearing wear is 40% > 30%, so it is determined to be this fault type to avoid matching fault types with high similarity but very low occurrence.

[0115] In one embodiment of this application, in response to the actual occurrence probability being less than or equal to a probability threshold, the top M historical fault cases with the second highest weighted similarity are selected as candidate types, their actual occurrence probability within a preset time window is calculated, and types with probabilities exceeding the threshold are filtered out; then, the trend consistency (such as feature fluctuation period, amplitude change rate) between the target key features and the historical features of the candidate types is compared, and the type with the highest trend matching degree is retained; if there is still no type that meets the conditions, it is marked as a fault to be confirmed, triggering a manual review process, and the feature data is entered into the case library.

[0116] In this embodiment, three historical fault cases ranked 2nd to 4th in weighted similarity from the fault case library are selected as candidate types. The actual occurrence probability of these types within a preset time window is calculated, and cases with probabilities still ≤ a threshold are filtered out. For example, after the original candidate type bearing crack (probability 25% < 30%) is excluded, the next highest similarity types bearing wear (probability 40%), rotor imbalance (probability 35%), and insufficient lubrication (probability 20%) are selected.

[0117] For the selected candidate types, the dynamic trends of the target key features and corresponding historical case features (such as the rate of increase of vibration kurtosis and the fluctuation period of the 120Hz energy proportion) are compared, and the consistency is quantified by a trend matching algorithm (such as dynamic time warping). For example, the target feature kurtosis increased from 3.2 to 6.8 within 5 minutes, and the trend matching degree with the historical case of bearing wear reached 85%, which is significantly higher than the 50% of rotor imbalance. Therefore, bearing wear was retained.

[0118] If the trend matching degree of all candidate types fails to meet the preset standard (e.g., <60%), it is determined to be a fault pending confirmation, triggering a manual review process (e.g., engineers make a judgment based on equipment operation records and on-site inspection data). At the same time, the key features of this target and the final confirmation result are entered into the fault case library to update the data distribution of the case library and provide more comprehensive sample support for subsequent model training and case matching.

[0119] As can be seen from the above, this embodiment calculates the weighted similarity to match the most relevant historical cases and verifies it by combining the actual occurrence probability within a preset time window, ensuring that the fault type determination conforms to both the feature pattern and the actual occurrence pattern. It highlights the matching degree of core features through weighting and filters rare matching results with probability thresholds, reducing misjudgments and improving the accuracy and reliability of fault type determination.

[0120] Corresponding to the gas generator set fault detection method in the above embodiment, Figure 2 This is a structural block diagram of a gas generator set fault detection device provided according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The gas generator set fault detection device 20 includes: a correlation coefficient calculation module 21, a weight calculation module 22, a data fusion module 23, a feature extraction module 24, and a fault detection module 25.

[0121] The correlation coefficient calculation module 21 is used to acquire target detection data collected by different types of sensors of the gas generator set. For the target detection data collected by each sensor, the correlation coefficient between the target detection data and the target parameter is calculated. The target parameter is a key parameter reflecting the operating status of the gas generator set. The target detection data is the data after preprocessing the original detection data of different types of sensors. The weight calculation module 22 is used to add target detection data with a correlation coefficient greater than a set threshold to the fusion candidate set, and to calculate the target weight corresponding to each target detection data in the fusion candidate set. Data fusion module 23 is used to obtain comprehensive detection data based on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data; Feature extraction module 24 is used to extract features from the comprehensive detection data to obtain the time-domain features and frequency-domain features corresponding to the comprehensive detection data; The fault detection module 25 is used to determine the fault type of the gas generator set based on time domain characteristics and frequency domain characteristics.

[0122] In one embodiment of this application, the correlation coefficient calculation module 21 is specifically used for: The correlation coefficient between target detection data and target parameters is calculated based on the Pearson correlation coefficient. The Pearson correlation coefficient is expressed as:

[0123] in, This represents the correlation coefficient between the detection data of the i-th target and the target parameter T. This represents the specific value of the i-th target detection data at time t. This represents the average value of the i-th target detection data over a preset time period. This represents the specific value of the objective parameter T at time t. This represents the average value of the target parameter T over a preset time period, where n represents the total number of time points.

[0124] In one embodiment of this application, the weight calculation module 22 is specifically used for: Based on each target detection data in the fusion candidate set, the correlation coefficient between the target detection data and the target parameters is normalized to obtain the initial weight corresponding to each target detection data. Based on the detection data of each target in the fusion candidate set, the failure rate and calibration overdue rate of the sensor corresponding to each target detection data are obtained, and based on the failure rate and calibration overdue rate of the sensor corresponding to each target detection data, the health index of the sensor corresponding to each target detection data is obtained. The initial weights corresponding to each target detection data are corrected based on the health index to obtain the target weights corresponding to each target detection data.

[0125] In one embodiment of this application, the data fusion module 23 is specifically used to: perform weighted fusion on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data to obtain initial fusion data; The deviation between the real-time fluctuation characteristics of the target parameter and the historical normal fluctuation characteristics is calculated. The historical normal fluctuation characteristics are obtained based on historical data statistics under normal operating conditions of the gas generator set. Based on the deviation, the corresponding dynamic calibration coefficient is obtained by querying the preset fluctuation correction coefficient table. The initial fusion data is coupled with the dynamic calibration coefficients to obtain comprehensive detection data.

[0126] In one embodiment of this application, the fault detection module 25 is specifically used for: The importance of time-domain features and frequency-domain features is evaluated, and importance scores for the time-domain features and frequency-domain features are obtained respectively; Features that meet the first condition and have an importance score are used as key features; the first condition is time-domain and frequency-domain features with an importance score greater than an importance threshold. Key features are input into a pre-defined fault classification model to obtain the fault type of the gas generator set. The fault classification model is trained based on the time-domain features, frequency-domain features, and labeled fault types corresponding to historical fault data.

[0127] In one embodiment of this application, the fault detection module 25 is further configured to: Calculate the confidence score of the fault type output by the fault classification model; In response to the failure classification model outputting a failure type with a confidence level lower than a confidence threshold, target key features are obtained from the key features. The target key features are those key features arranged from high to low importance scores, satisfying a preset number of features. The preset number is less than the number of key features. The fault type of the gas generator set is determined by matching the target key features with the fault case library of gas generator sets based on the matching results.

[0128] In one embodiment of this application, the fault detection module 25 is further configured to: Calculate the weighted similarity between the key features of the target and each historical failure case in the failure case library; The fault type corresponding to the historical fault case with the highest weighted similarity is selected as the candidate fault type; and the actual occurrence probability of the candidate fault type within a preset time window is obtained. If the actual probability of occurrence is greater than the probability threshold, the candidate fault type is determined as the fault type of the gas generator set.

[0129] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the correlation coefficient calculation module 21, weight calculation module 22, data fusion module 23, feature extraction module 24, and fault detection module 25 are shown.

[0130] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0131] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0132] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store information such as fault types, fault case libraries, and fusion candidate sets.

[0133] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the gas generator set fault detection method provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0134] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0135] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0136] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces or units, or they may be electrical, mechanical, or other forms of connection.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0140] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0141] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for fault detection in a gas generator set, characterized in that, include: Acquire target detection data collected by different types of sensors in the gas generator set, and calculate the correlation coefficient between the target detection data and the target parameters for each sensor. The target parameters are key parameters that reflect the operating status of the gas generator set; The target detection data is the preprocessed data of the raw detection data from different types of sensors; Add the target detection data with a correlation coefficient greater than a set threshold to the fusion candidate set, and calculate the target weight corresponding to each target detection data in the fusion candidate set; Based on the detection data of each target in the fusion candidate set and the target weight corresponding to each detection data, comprehensive detection data is obtained; Feature extraction is performed on the comprehensive detection data to obtain the time-domain features and frequency-domain features corresponding to the comprehensive detection data; The fault type of the gas generator set is determined based on the time-domain characteristics and the frequency-domain characteristics.

2. The method as described in claim 1, characterized in that, The calculation of the correlation coefficient between the target detection data and the target parameters includes: The correlation coefficient between target detection data and target parameters is calculated based on the Pearson correlation coefficient. The Pearson correlation coefficient is expressed as follows: in, This represents the correlation coefficient between the detection data of the i-th target and the target parameter T. This represents the specific value of the i-th target detection data at time t. This represents the average value of the i-th target detection data over a preset time period. This represents the specific value of the objective parameter T at time t. This represents the average value of the target parameter T over a preset time period, where n represents the total number of time points.

3. The method as described in claim 1, characterized in that, The calculation of the target weight corresponding to each target detection data in the candidate set of fusion includes: Based on each target detection data in the fusion candidate set, the correlation coefficient between the target detection data and the target parameters is normalized to obtain the initial weight corresponding to each target detection data. Based on the detection data of each target in the fusion candidate set, the failure rate and calibration overdue rate of the sensor corresponding to each target detection data are obtained, and based on the failure rate and calibration overdue rate of the sensor corresponding to each target detection data, the health index of the sensor corresponding to each target detection data is obtained. The initial weights corresponding to each target detection data are corrected based on the health index to obtain the target weights corresponding to each target detection data.

4. The method as described in claim 1, characterized in that, The comprehensive detection data is obtained based on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data, including: The initial fused data is obtained by weighting and fusing each target detection data and the target weight corresponding to each target detection data in the fusion candidate set; The deviation between the real-time fluctuation characteristics of the target parameter and the historical normal fluctuation characteristics is calculated. The historical normal fluctuation characteristics are obtained based on historical data statistics under normal operating conditions of the gas generator set. The corresponding dynamic calibration coefficient is obtained by querying the preset fluctuation correction coefficient table based on the deviation. The initial fusion data is coupled with the dynamic calibration coefficient to obtain comprehensive detection data.

5. The method as described in claim 1, characterized in that, The method of determining the fault type of the gas generator set based on the time-domain features and the frequency-domain features includes: The importance of the time-domain features and the frequency-domain features is evaluated to obtain importance scores for the time-domain features and the frequency-domain features, respectively. The features corresponding to the importance scores that satisfy the first condition are taken as key features; the first condition is time-domain features and frequency-domain features whose importance scores are greater than an importance threshold. The key features are input into a preset fault classification model to obtain the fault type of the gas generator set; the fault classification model is trained based on the time domain features, frequency domain features and labeled fault types corresponding to historical fault data.

6. The method as described in claim 5, characterized in that, Also includes: Calculate the confidence level of the fault type output by the fault classification model; In response to the fault classification model outputting a confidence level for the fault type that is lower than a confidence threshold, target key features are obtained from the key features; the target key features are features arranged from high to low importance scores among the key features, satisfying a preset number of features; The preset quantity is less than the quantity of the key features; Based on the target's key features, a fault case database of gas generator sets is matched, and the fault type of the gas generator set is determined according to the matching results.

7. The method as described in claim 6, characterized in that, The process involves matching the target's key features with a fault case database of gas generator sets, and determining the fault type of the gas generator set based on the matching results, including: Calculate the weighted similarity between the key features of the target and each historical failure case in the failure case library; The fault type corresponding to the historical fault case with the highest weighted similarity is taken as the candidate fault type; and the actual occurrence probability of the candidate fault type within a preset time window is obtained. If the actual occurrence probability is greater than the probability threshold, the candidate fault type is determined as a fault type of the gas generator set.

8. A fault detection device for a gas generator set, characterized in that, include: The correlation coefficient calculation module is used to acquire target detection data collected by different types of sensors in the gas generator set. For the target detection data collected by each sensor, the correlation coefficient between the target detection data and the target parameter is calculated. The target parameters are key parameters that reflect the operating status of the gas generator set; The target detection data is the preprocessed data of the raw detection data from different types of sensors; The weight calculation module is used to add the target detection data with a correlation coefficient greater than a set threshold to the fusion candidate set, and to calculate the target weight corresponding to each target detection data in the fusion candidate set. The data fusion module is used to obtain comprehensive detection data based on each target detection data in the fusion candidate set and the target weight corresponding to each target detection data. The feature extraction module is used to extract features from the comprehensive detection data to obtain the time-domain features and frequency-domain features corresponding to the comprehensive detection data; The fault detection module is used to determine the fault type of the gas generator set based on the time-domain features and the frequency-domain features.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images