A Smart Inspection Method for Gas Turbine Leakage Based on Multi-Sensor Data Fusion

Abstract

This invention belongs to the field of intelligent technology and discloses a multi-sensor data fusion-based intelligent inspection method for gas turbine leaks. The method includes: deploying M sensors in each of T ranges, using a multi-level regression model to perform cross-sensor data verification, obtaining verification results, and automatically triggering an automatic sensor calibration mechanism based on the verification results; using an SVR model to predict leakage risk data for each range; for ranges with potential leaks, using an infrared thermal imaging camera for inspection and acquiring abnormal areas; for abnormal areas, triggering a multispectral camera to inspect the abnormal area, using a dynamic fusion method to obtain band-comprehensive data, and determining whether the abnormal area is leaking based on the band-comprehensive data; if a leak is detected, an early warning is immediately triggered; if no leak is detected, triggering the infrared thermal imaging camera and multispectral camera to perform high-frequency inspection of the abnormal area; this method can improve the accuracy and response speed of gas turbine leak inspection.

Description

Technical Field

[0001] This invention relates to the field of intelligent technology, specifically to an intelligent inspection method for gas turbine leaks based on multi-sensor data fusion. Background Technology

[0002] Existing intelligent gas turbine leak inspection systems based on multi-sensor data fusion have many shortcomings:

[0003] First, the gas turbine operating environment is harsh, with high temperature, vibration, electromagnetic interference or dust, which can cause sensor performance to degrade or data to be abnormal. The existing system lacks dynamic automatic calibration of the sensors, resulting in false alarms or missed alarms.

[0004] Then, the existing gas turbine leak inspection uses sensors to inspect and fuse the sensor data to determine whether there is an anomaly. It only performs a global inspection, that is, it collects data in various areas of the gas turbine to determine whether there is a leak or anomaly. However, this method is prone to ignoring local anomalies, especially small-scale anomalies or micro-leaks that are difficult to detect with global data. Because it only relies on global data, the existing method often cannot achieve detailed inspection of the abnormal area.

[0005] Furthermore, existing methods cannot effectively distinguish between anomalies caused by gas turbine leaks and those caused by equipment malfunctions or other environmental factors when detecting anomalies. They simply assume that the anomalies are caused by leaks, which leads to incorrect judgments and fails to provide sufficiently accurate leak information.

[0006] In view of this, the present invention proposes a multi-sensor data fusion-based intelligent inspection method for gas turbine leaks to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a multi-sensor data fusion-based intelligent inspection method for gas turbine leaks, comprising:

[0008] Step SS1: Deploy M sensors in each of the T ranges. Based on the data collected from the M sensors, use a multi-level regression model to perform cross-sensor data mutual verification and obtain the verification results. If the sensor data deviates from the verification results, the automatic calibration mechanism of that sensor will be automatically triggered for calibration.

[0009] Step SS2: In each range, using the calibration data collected by the automatically calibrated sensors, the SVR model is used to predict the leakage risk data for each range, and a leakage threshold is set. If the leakage risk data is greater than or equal to the leakage threshold, then there is a leakage risk in that range; if the leakage risk data is less than the leakage threshold, then the range is safe.

[0010] Step SS3: For areas with leakage risks, use an infrared thermal imaging camera to conduct inspections and use the dynamic error connectivity marking method to identify abnormal areas.

[0011] Step SS4: For abnormal areas, trigger the multispectral camera to inspect the abnormal area, use the dynamic fusion method to obtain the band composite data, and set the leakage judgment threshold. If the band composite data of the abnormal area is greater than the leakage judgment threshold, it means that the abnormality in the area is caused by gas turbine leakage; if the band composite data of the abnormal area is less than the leakage judgment threshold, it means that the abnormality in the area is caused by the equipment itself.

[0012] Step SS5: If the regional anomaly is caused by a gas turbine leak, an early warning will be triggered immediately; if the regional anomaly is caused by the equipment itself, the infrared thermal imaging camera and the multispectral camera will be triggered to conduct high-frequency inspections of the abnormal area.

[0013] Furthermore, the specific method for deploying M sensors in each of the T ranges includes:

[0014] Fiber optic methane sensors, temperature sensors, and pressure sensors are installed within the gas turbine valve area, fuel pipe expansion joint area, gas turbine fuel ring pipe area, combustion chamber area, combustion chamber bottom area, combustion chamber top area, and air outlet area inside the gas turbine casing.

[0015] Furthermore, the specific methods for cross-sensor data mutual verification using a multi-level regression model based on data collected from M sensors, obtaining verification results, and automatically triggering the automatic calibration mechanism of the sensor if the sensor data deviates from the verification results, include:

[0016] Based on data collected from M sensors, a multilevel regression model was used to model methane concentration data, temperature data, and pressure data. Temperature data and pressure data were input into the multilevel regression model to obtain validated methane concentration data; methane concentration data and pressure data were input into the multilevel regression model to obtain validated temperature data; methane concentration data and temperature data were input into the multilevel regression model to obtain validated pressure data.

[0017] The real-time methane concentration, temperature, and pressure data are compared with the methane concentration, temperature, and pressure data validated by the multi-level regression model. If the data measured by the sensor deviates from the validation results (i.e., the validated methane concentration, temperature, and pressure data), it indicates that the sensor has a deviation, and the automatic calibration mechanism of the sensor is automatically triggered for calibration.

[0018] Furthermore, the specific method for predicting leakage risk data for each range using calibration data acquired by automatically calibrated sensors and an SVR model includes:

[0019] T databases are established for T ranges. Data collected by automatically calibrated sensors in each range is stored in the database. Each database contains M sets of samples. Each set of samples consists of methane concentration data, temperature data, pressure data, and leakage risk data. The M sets of samples are used as input data into the SVR model to obtain the predicted leakage risk data for each range.

[0020] Furthermore, the specific methods for using infrared thermal imaging cameras to inspect areas with leakage risks and for using dynamic error connectivity marking to identify abnormal areas include:

[0021] An infrared thermal imaging camera is used to conduct real-time inspections of areas with potential leakage risks and obtain thermal images. The inspection frequency of the infrared thermal imaging camera is u.

[0022] The median filtering method is used to reduce the noise in the thermal image to obtain a denoised thermal image.

[0023] Using the sliding window method, the average value of each pixel in the thermal image over n historical time periods is calculated to obtain the average thermal image.

[0024] Calculate the error of each pixel in the average thermal image and the denoised thermal image to obtain the error thermal image;

[0025] For the error thermal image, calculate the mean and standard deviation of each pixel in the error thermal image, and set a dynamic error threshold based on the mean and standard deviation;

[0026] In the error thermal imaging map, pixels with errors greater than the dynamic error threshold are marked in blue. Each blue pixel in the error thermal imaging map is assigned a unique label 0. The traversal method is used to detect whether the labels of the blue pixels with label 0 are the same on the top, bottom, left, right and the four diagonals. The connected component labeling method is used to identify the blue areas formed in the error thermal imaging map. The blue areas are abnormal areas.

[0027] Furthermore, for abnormal areas, the multispectral camera is triggered to inspect the abnormal areas, and the specific methods for obtaining band composite data using dynamic fusion include:

[0028] Once an abnormal area is identified, the multispectral camera is triggered. The multispectral camera's inspection frequency is f, and it uses the multispectral camera to collect N bands of each pixel in the abnormal area.

[0029] Each band includes data on radiation intensity, absorption bandwidth, spectral reflectance, and normalized absorption index;

[0030] For each pixel's N bands, a dynamic adaptive method is used to obtain D bands;

[0031] For D bands, a weighted average method is used to obtain the band composite data of each pixel. The band composite data of each pixel is then weighted and fused to obtain the band composite data of the abnormal area.

[0032] Furthermore, the specific method for obtaining D bands using the dynamic adaptive method for N bands of each pixel includes:

[0033] When the multispectral camera is triggered, it acquires N band data for each pixel in the abnormal area and uses the spectral angle matching analysis correction method to obtain the response intensity data of N bands of each pixel to the abnormal area.

[0034] The response intensities of N bands to the abnormal region are sorted in descending order. A response intensity threshold is set using expert experience. Bands with response intensities greater than or equal to the threshold are retained, while bands with response intensities less than the threshold are removed. Thus, D bands are retained for each pixel.

[0035] Furthermore, the specific method for obtaining the response intensity data of the N bands of each pixel to the abnormal region using the spectral angle matching analysis correction method includes:

[0036] The radiation intensity, absorption bandwidth, spectral reflectance, and normalized absorption index data of N bands of the pixel are used to construct N spectral vectors.

[0037] The spectral angle between the spectral vector of each band and the spectral eigenvector of the anomalous region is calculated using the following formula: Among them, S λb S represents the spectral vector of the b-th band. a θ represents the spectral data vector of the anomalous region. b The spectral angle between the spectral vector of the b-th band and the spectral data vector of the anomalous region is represented. The smaller the spectral angle, the more closely the band matches the anomalous region.

[0038] For each band, the Pearson correlation method is used to calculate the linear relationship between the spectral vector of each band and the spectral feature vector of the anomalous region, and the correlation value is obtained.

[0039] The correlation value is used to correct the spectral angle between the spectral vector of each band and the spectral data vector of the anomalous region, thus obtaining the correlation intensity data for each band. The response intensity data is then calculated from the correlation intensity data using the following formula: Where, rb R represents the correlation strength data of the b-th band, max(r) represents the maximum correlation strength data among the N bands, and R b This represents the response intensity data for the b-th band.

[0040] Furthermore, the spectral feature vector of the abnormal region is obtained by means of:

[0041] Given that the abnormal region consists of H pixels, N bands of spectral data are extracted for each pixel, and the spectral data are used to construct a spectral vector. The spectral vectors of the H pixels are then weighted using a weighted average method to obtain the spectral feature vector of the abnormal region.

[0042] Furthermore, when the anomaly in the area is caused by the device itself, the specific methods for triggering the infrared thermal imaging camera and the multispectral camera to perform high-frequency inspection of the anomaly area include:

[0043] For abnormal areas caused by the equipment itself, high-frequency inspection is triggered. The inspection frequency of the infrared thermal imaging camera is set to 2u, and the inspection frequency of the multispectral camera is set to 2f.

[0044] The technical effects and advantages of the intelligent inspection method for gas turbine leaks based on multi-sensor data fusion of this invention are as follows:

[0045] This invention utilizes multi-sensor fusion technology, combining methane concentration data, temperature data, and pressure data, to accurately identify gas turbine leakage risks and equipment malfunctions, predicting and triggering early warnings in advance. Through a multi-level regression model, cross-sensor data is mutually verified, and based on the verification results, the sensor's automatic calibration mechanism is automatically triggered for calibration, ensuring high accuracy and reliability of the data and effectively reducing the possibility of missed or false alarms.

[0046] Based on different ranges, the system combines the SVR model to predict the leakage risk data of that range and intelligently triggers targeted inspection measures to conduct global inspections, local inspections, and detailed inspections, thereby improving the accuracy and efficiency of inspections and ensuring that areas with leakage risks can be detected in a timely manner.

[0047] For areas with leakage risks, infrared thermal imaging cameras can be used to inspect these areas, accurately identifying regions with abnormal temperatures and helping to quickly find potential leak sources. Infrared thermal imaging cameras provide a non-contact detection method, avoiding the risks of manual operation, and ensuring the safety of inspection personnel, especially when inspecting high-temperature, high-pressure, or hazardous areas.

[0048] By collecting band data of the abnormal area using a multispectral camera, more detailed physical characteristic information can be obtained, such as radiation intensity, absorption bandwidth, and spectral reflectance. This multidimensional data can more accurately distinguish between anomalies caused by gas turbine leaks and those caused by equipment failures or other factors, providing a more scientific and detailed leak detection method.

[0049] As a non-contact monitoring tool, multispectral cameras can accurately monitor abnormal areas without interfering with the normal operation of equipment. This method can effectively avoid human interference and potential safety risks, especially when dealing with equipment in high temperature, high pressure or dangerous environments, which can greatly enhance safety.

[0050] By conducting precise inspections of abnormal areas, potential problems caused by the equipment itself can be identified in a timely manner. This not only reduces equipment downtime caused by equipment failure, but also helps technicians implement more precise maintenance or replacement strategies, further improving the overall operational reliability and stability of the equipment.

[0051] By distinguishing between anomalies caused by leaks and anomalies caused by equipment malfunctions, and by taking corresponding early warning and high-frequency inspection measures, the source of the problem can be located more accurately, reducing false alarms and missed alarms, and ensuring timely and effective intervention. This improves the system's response speed and accuracy, optimizes equipment operation and maintenance, and enhances safety assurance and operational efficiency. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of a multi-sensor data fusion intelligent inspection method for gas turbine leaks according to the present invention;

[0053] Figure 2 This is a schematic diagram of the fine inspection method of the present invention;

[0054] Figure 3 This is a schematic diagram of a gas turbine leak intelligent inspection system based on multi-sensor data fusion according to the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1

[0057] Please see Figure 1 and Figure 2As shown in the figure, this embodiment of a multi-sensor data fusion-based intelligent inspection method for gas turbine leaks includes:

[0058] Step SS1: Deploy M sensors in each of the T ranges. Based on the data collected from the M sensors, use a multi-level regression model to perform cross-sensor data mutual verification and obtain the verification results. If the sensor data deviates from the verification results, the automatic calibration mechanism of that sensor will be automatically triggered for calibration.

[0059] Step SS2: In each range, using the calibration data collected by the automatically calibrated sensors, the SVR model is used to predict the leakage risk data for each range, and a leakage threshold is set. If the leakage risk data is greater than or equal to the leakage threshold, then there is a leakage risk in that range; if the leakage risk data is less than the leakage threshold, then the range is safe.

[0060] Step SS3: For areas with leakage risks, use an infrared thermal imaging camera to conduct inspections and use the dynamic error connectivity marking method to identify abnormal areas.

[0061] Step SS4: For abnormal areas, trigger the multispectral camera to inspect the abnormal area, use the dynamic fusion method to obtain the band composite data, and set the leakage judgment threshold. If the band composite data of the abnormal area is greater than the leakage judgment threshold, it means that the abnormality in the area is caused by gas turbine leakage; if the band composite data of the abnormal area is less than the leakage judgment threshold, it means that the abnormality in the area is caused by the equipment itself.

[0062] Step SS5: If the regional anomaly is caused by a gas turbine leak, an early warning will be triggered immediately; if the regional anomaly is caused by the equipment itself, the infrared thermal imaging camera and the multispectral camera will be triggered to conduct high-frequency inspections of the abnormal area.

[0063] The specific methods for deploying M sensors in each of the T ranges include:

[0064] Fiber optic methane sensors, temperature sensors, and pressure sensors are installed within the gas turbine valve area, fuel pipe expansion joint area, gas turbine fuel ring pipe area, combustion chamber area, combustion chamber bottom area, combustion chamber top area, and air outlet area inside the gas turbine casing.

[0065] Existing sensors are calibrated periodically at set time intervals. For example, every certain number of days, months, or hours of use, the sensor will start the calibration equipment for standardization adjustment. The advantage of this method is that it is simple to operate, but the disadvantage is that the calibration time may not be consistent with the actual time when the sensor's performance deviation occurs, so it may miss the real-time changes in the sensor's performance deviation.

[0066] Secondly, calibration is performed based on standard references. For gas sensors (such as fiber optic methane sensors), a common practice is to use a standard gas of known concentration for calibration, compare the standard gas data with the data collected by the sensor, and adjust the sensor output to match the standard gas concentration. Although this method is relatively accurate, it requires manual operation and has high requirements for the experimental environment and gas concentration.

[0067] Based on data collected from M sensors, a multi-level regression model is used to perform cross-sensor data verification and obtain verification results. If the sensor data deviates from the verification results, the automatic calibration mechanism of that sensor is automatically triggered for calibration. The specific methods include:

[0068] Based on data collected from M sensors, a multilevel regression model was used to model methane concentration data, temperature data, and pressure data. Temperature data and pressure data were input into the multilevel regression model to obtain validated methane concentration data; methane concentration data and pressure data were input into the multilevel regression model to obtain validated temperature data; methane concentration data and temperature data were input into the multilevel regression model to obtain validated pressure data.

[0069] The real-time methane concentration, temperature, and pressure data are compared with the methane concentration, temperature, and pressure data validated by the multi-level regression model. If the data measured by the sensor deviates from the validation results, which are the validated methane concentration, temperature, and pressure data, it indicates that the sensor has a deviation, and the automatic calibration mechanism of the sensor is automatically triggered for calibration.

[0070] The calculation formula for the multilevel regression model is as follows: in, T pre and P pre This indicates the verified methane concentration data, temperature data, and pressure data. and X represents the intercept of methane concentration, temperature, and pressure data. T , and X P This indicates the actual temperature data, methane gas concentration data, and pressure data measured by the sensor. and The regression coefficients represent the methane concentration data, temperature data, and pressure data. The intercept and regression coefficients are obtained using the least squares method.

[0071] This invention collects data from multiple sensors in real time (such as methane concentration data, temperature data, and pressure data), and uses a multi-level regression model to cross-validate the sensor data. This allows for the timely detection of sensor performance deviations and the automatic triggering of a calibration mechanism when deviations occur, without the need for manual intervention. This method eliminates the limitations of existing periodic calibration and avoids the risks of missing calibration opportunities or human error.

[0072] Secondly, the multi-level regression model used analyzes the relationship between different sensors and realizes cross-sensor data mutual verification. This not only improves the accuracy of the data, but also minimizes errors caused by single sensor failure or deviation. This model-based calibration method is more scientific and systematic than the existing methods that rely on human experience or periodic calibration, and has high adaptability, enabling it to operate stably in complex environments.

[0073] Within each range, the specific methods for using SVR models to predict leakage risk data for each range, based on calibration data acquired using automatically calibrated sensors, include:

[0074] T databases are established for T ranges. Data collected by automatically calibrated sensors in each range is stored in the database. Each database contains M sets of samples. Each set of samples consists of methane concentration data, temperature data, pressure data, and leakage risk data. The M sets of samples are used as input data into the SVR model to obtain the predicted leakage risk data for each range.

[0075] Existing methods typically lack a mechanism for real-time prediction of leakage risk for each area; they usually determine directly whether a leakage has occurred.

[0076] By establishing databases across multiple ranges and using SVR models to model methane concentration, temperature, and pressure data within each range, leakage risks can be accurately predicted in each range.

[0077] Existing methods for leak detection using multi-sensor fusion typically rely on integrating data from multiple sensors to identify whether a leak has occurred. However, this approach has significant drawbacks when detecting minute leaks. First, while multi-sensor fusion can improve accuracy by combining data from different sensors, these methods often fail to sensitively capture minute changes when dealing with minute leaks. Minor leaks may only cause extremely subtle fluctuations in some sensors, while existing data fusion techniques are often designed to identify leaks over a larger area. This makes it easy for the signal of a minute leak to be drowned out by background noise, leading to missed detections.

[0078] Secondly, differences in the response characteristics, accuracy, and sensitivity of sensors may lead to distortion of the fusion results. Especially when the leakage signal is weak, the deviation or failure of a single sensor can have a significant impact on the overall judgment, thereby affecting the accurate detection of minute leaks.

[0079] In summary, while multi-sensor fusion is effective in detecting larger leaks, it suffers from insufficient sensitivity, poor adaptability, and susceptibility to sensor errors when judging small leaks, which limits its application in fine monitoring and early leak detection.

[0080] If there is a leak in the gas turbine, the leaked gas will cause temperature changes in the surrounding area. Leaks are particularly likely to occur at pipe connections, valves and seals. Infrared thermal imaging cameras can detect these local temperature changes.

[0081] Infrared thermal imaging cameras are installed around the gas turbine to ensure coverage of critical areas of the gas turbine equipment (such as gas pipelines, sealing joints, valves, etc.), especially areas prone to leakage, so as to conduct a thorough inspection of the gas turbine.

[0082] For areas with potential leakage risks, the specific methods for using infrared thermal imaging cameras for inspection and employing dynamic error connectivity marking to identify abnormal areas include:

[0083] An infrared thermal imaging camera is used to conduct real-time inspections of areas with potential leakage risks and obtain thermal images. The inspection frequency of the infrared thermal imaging camera is u.

[0084] The median filtering method is used to reduce the noise in the thermal image to obtain a denoised thermal image.

[0085] Using the sliding window method, the average value of each pixel in the thermal image over n historical time periods is calculated to obtain the average thermal image.

[0086] Calculate the error of each pixel in the average thermal image and the denoised thermal image to obtain the error thermal image;

[0087] For the error thermal image, calculate the mean and standard deviation of each pixel in the error thermal image, and set a dynamic error threshold based on the mean and standard deviation;

[0088] In the error thermal imaging map, pixels with errors greater than the dynamic error threshold are marked in blue. Each blue pixel in the error thermal imaging map is assigned a unique label 0. The traversal method is used to detect whether the labels of the blue pixels with label 0 are the same on the top, bottom, left, right and the four diagonals. The connected component labeling method is used to identify the blue areas formed in the error thermal imaging map. The blue areas are abnormal areas.

[0089] By using an infrared thermal imaging camera, areas with abnormal temperatures can be quickly identified. This method can accurately locate areas where leaks may occur, and is especially suitable for detecting temperature changes caused by leaks of methane, gas, etc. Infrared thermal imaging cameras have high spatial resolution, which can quickly scan a large area and find areas with abnormal temperatures or that do not meet normal environmental conditions. This provides clear target areas for subsequent detailed inspections, thereby improving inspection efficiency and accuracy.

[0090] Furthermore, for abnormal areas, the multispectral camera is triggered to inspect the abnormal areas, and the specific methods for obtaining band composite data using dynamic fusion include:

[0091] Once an abnormal area is identified, the multispectral camera is triggered. The multispectral camera's inspection frequency is f, and it uses the multispectral camera to collect N bands of each pixel in the abnormal area.

[0092] Each band includes data on radiation intensity, absorption bandwidth, spectral reflectance, and normalized absorption index;

[0093] For each pixel's N bands, a dynamic adaptive method is used to obtain D bands;

[0094] For D bands, a weighted average method is used to obtain the band composite data of each pixel. The band composite data of each pixel is then weighted and fused to obtain the band composite data of the abnormal area.

[0095] Existing methods typically use all bands or select a few common fixed bands for analysis, which has several significant drawbacks:

[0096] First, using all bands results in a huge amount of data, increasing the burden of processing and storage. At the same time, it also introduces a lot of background information that is irrelevant to target detection or may even cause interference, reducing the accuracy and efficiency of anomaly detection.

[0097] Secondly, while using a few fixed common bands can simplify the processing flow and reduce the demand for computing resources, this method ignores the differences in spectral characteristics under different scenarios and device conditions, and cannot flexibly cope with complex actual situations, resulting in inaccurate detection results or insufficient sensitivity.

[0098] Furthermore, the selection of fixed bands is often based on general experience and historical data, which makes it difficult to adapt to special situations or new changes in specific environments and easily misses important abnormal signals. Overall, these existing methods lack specificity and flexibility and cannot effectively distinguish the truly useful bands, thus limiting their performance and reliability in application scenarios such as intelligent inspection of gas turbine leaks in complex and dynamically changing environments.

[0099] For each pixel's N bands, the specific methods used to obtain D bands through dynamic adaptation include:

[0100] When the multispectral camera is triggered, it acquires N band data for each pixel in the abnormal area and uses the spectral angle matching analysis correction method to obtain the response intensity data of N bands of each pixel to the abnormal area.

[0101] The response intensities of N bands to the anomalous region are sorted in descending order. A response intensity threshold is set using expert experience. Bands with response intensities greater than or equal to the threshold are retained, while bands with response intensities less than the threshold are removed. Thus, each pixel retains D bands. The D bands retained by each pixel are not the same. For example, if the y-th pixel retains 3 bands, then D = 3; if the j-th pixel retains 6 bands, then D = 6.

[0102] This dynamic adaptive method acquires N band data for each pixel in the abnormal region using a multispectral camera, calculates the response intensity of the N bands of each pixel to the abnormal region using the spectral angle matching analysis correction method, sorts these response intensities in descending order, and selects the D bands that respond most strongly to the abnormal region based on a threshold set by expert experience, so that each pixel retains a different number of bands.

[0103] The advantage of this method lies in its ability to improve detection accuracy in a targeted manner, as it dynamically selects the most relevant band based on the actual response intensity, thereby effectively reducing noise interference. It is also highly flexible and can adapt to the complex and ever-changing scenarios in gas turbine leak detection, without the need to preset fixed bands, thus reducing the need for manual intervention. In addition, this method can significantly reduce the amount of redundant data, optimize the data load and computational efficiency of subsequent processing, and make the entire inspection process more efficient.

[0104] The specific methods for obtaining the response intensity data of N bands for the abnormal region for each pixel using the spectral angle matching analysis correction method include:

[0105] The radiation intensity, absorption bandwidth, spectral reflectance, and normalized absorption index data of N bands of the pixel are used to construct N spectral vectors.

[0106] The spectral angle between the spectral vector of each band and the spectral eigenvector of the anomalous region is calculated using the following formula: Among them, S λb S represents the spectral vector of the b-th band. a θ represents the spectral data vector of the anomalous region. b The spectral angle between the spectral vector of the b-th band and the spectral data vector of the anomalous region is represented. The smaller the spectral angle, the more closely the band matches the anomalous region.

[0107] For each band, the Pearson correlation method is used to calculate the linear relationship between the spectral vector of each band and the spectral feature vector of the anomalous region, and the correlation value is obtained.

[0108] The correlation value is used to correct the spectral angle between the spectral vector of each band and the spectral data vector of the anomalous region, thus obtaining the correlation intensity data for each band. The response intensity data is then calculated from the correlation intensity data using the following formula: Where, r b R represents the correlation strength data of the b-th band, max(r) represents the maximum correlation strength data among the N bands, and R b This represents the response intensity data for the b-th band;

[0109] Regarding the spectral angle matching analysis correction method: First, by constructing a spectral vector containing data on radiation intensity, absorption bandwidth, spectral reflectance, and normalized absorption index, the unique spectral characteristics of each band can be comprehensively captured, providing a rich information foundation for subsequent analysis. Next, spectral angle matching analysis can directly quantify the similarity between each band and the spectral characteristics of the abnormal region (a smaller spectral angle indicates a higher matching degree), thereby improving the accuracy and sensitivity of detection. Furthermore, combining Pearson correlation analysis for linear relationship analysis and spectral angle correction allows for more detailed adjustment of the importance weight of each band, ensuring accurate identification of abnormal signals even in complex and variable environments. Finally, the process of calculating response intensity data based on correlation intensity data allows for dynamic adjustment of the contribution of each band according to actual observations, avoiding the biases and limitations that may arise from traditional fixed-parameter methods. This enhances the robustness and reliability of the overall system, achieving efficient and accurate monitoring of abnormal phenomena such as gas turbine leaks. This spectral angle matching analysis correction method not only improves the accuracy of anomaly detection, enabling the identification of minute leaks, but also optimizes resource utilization efficiency, providing strong technical support for intelligent inspection.

[0110] Furthermore, the spectral feature vector of the abnormal region is obtained by means of:

[0111] Given that the abnormal region consists of H pixels, N bands of spectral data are extracted for each pixel, and the spectral data are used to construct a spectral vector. The spectral vectors of the H pixels are then weighted using a weighted average method to obtain the spectral feature vector of the abnormal region.

[0112] The advantages of this multispectral camera inspection method are that by acquiring data from multiple bands through a multispectral camera and processing this data using a dynamic adaptive method, the accuracy and sensitivity of leak detection can be significantly improved. Unlike existing single-band detection methods, the integrated data from multiple bands can provide more comprehensive information on abnormal areas, helping to identify minute leaks more accurately. By fusing data from different bands using a weighted average method, environmental interference and data noise can be effectively reduced, improving the response capability to weak leak signals.

[0113] In addition, the dynamic adaptive method can automatically adjust the band selection according to the actual situation, making the detection process more flexible and accurate. This method can flexibly remove irrelevant or interfering bands based on the response intensity data of each pixel, thereby improving the robustness and accuracy of the detection. Especially when facing anomalies caused by complex environments and the equipment itself, it can accurately distinguish between leakage sources and equipment failures.

[0114] This meticulous inspection mechanism makes the detection of minor leaks more reliable, significantly reduces the probability of missed and false alarms, and improves the efficiency and security of the entire monitoring system.

[0115] Furthermore, if the abnormal area is caused by the device itself, the specific methods for triggering the infrared thermal imaging camera and multispectral camera to perform high-frequency inspection of the abnormal area include:

[0116] For abnormal areas caused by the equipment itself, high-frequency inspection is triggered. The inspection frequency of the infrared thermal imaging camera is set to 2u, and the inspection frequency of the multispectral camera is set to 2f.

[0117] The main purpose of using high-frequency inspection is to increase the monitoring frequency of abnormal areas of equipment, thereby capturing potential problems or changes caused by the equipment itself more promptly. In the event of equipment malfunction, increasing the inspection frequency (such as doubling the inspection frequency of infrared thermal imaging cameras and multispectral cameras) can more accurately monitor the equipment status and avoid missing subtle changes. High-frequency inspection can provide more detection data in a short time, which helps to discover equipment faults or abnormalities more quickly and reduce the risk of equipment damage. In addition, rapid response can improve inspection efficiency and take timely measures in the early stages of equipment failure to avoid potentially greater damage or safety hazards, ensuring the normal operation of equipment and the stability of the system.

[0118] In this embodiment, by using multi-sensor fusion technology and combining methane concentration data, temperature data, and pressure data, the risk of gas turbine leakage and equipment failure can be accurately identified, and early warnings can be triggered in advance. By using a multi-level regression model to verify cross-sensor data, the automatic calibration mechanism of the sensors is automatically triggered based on the verification results, ensuring high accuracy and reliability of the data and effectively reducing the possibility of missed and false alarms.

[0119] Based on different ranges, the leak risk data of that range is predicted by combining the SVR model, and targeted inspection measures are intelligently triggered to carry out global inspections, which improves the accuracy and efficiency of inspections and ensures that areas with leak risks can be discovered in a timely manner.

[0120] For areas with leakage risks, infrared thermal imaging cameras can be used to inspect these areas, accurately identifying regions with abnormal temperatures and helping to quickly find potential leak sources. Infrared thermal imaging cameras provide a non-contact detection method, avoiding the risks of manual operation, and ensuring the safety of inspection personnel, especially when inspecting high-temperature, high-pressure, or hazardous areas.

[0121] By collecting band data of the abnormal area using a multispectral camera, more detailed physical characteristic information can be obtained, such as radiation intensity, absorption bandwidth, and spectral reflectance. This multidimensional data can more accurately distinguish between anomalies caused by gas turbine leaks and those caused by equipment failures or other factors, providing a more scientific and detailed leak detection method.

[0122] As a non-contact monitoring tool, multispectral cameras can accurately monitor abnormal areas without interfering with the normal operation of equipment. This method can effectively avoid human interference and potential safety risks, especially when dealing with equipment in high temperature, high pressure or dangerous environments, which can greatly enhance safety.

[0123] By conducting precise inspections of abnormal areas, potential problems caused by the equipment itself can be identified in a timely manner. This not only reduces equipment downtime caused by equipment failure, but also helps technicians implement more precise maintenance or replacement strategies, further improving the overall operational reliability and stability of the equipment.

[0124] By distinguishing between anomalies caused by leaks and anomalies caused by equipment malfunctions, and by taking corresponding early warning and high-frequency inspection measures, the source of the problem can be located more accurately, reducing false alarms and missed alarms, and ensuring timely and effective intervention. This improves the system's response speed and accuracy, optimizes equipment operation and maintenance, and enhances safety assurance and operational efficiency.

[0125] Example 2

[0126] Please see Figure 3 As shown, parts not described in detail in this embodiment are described in Embodiment 1. A multi-sensor data fusion-based intelligent inspection system for gas turbine leaks is provided, including:

[0127] Sensor dynamic calibration unit: M sensors are deployed in each of the T ranges. Based on the data collected from the M sensors, a multi-level regression model is used to perform cross-sensor data mutual verification and obtain the verification results. If the sensor data deviates from the verification results, the automatic calibration mechanism of that sensor is automatically triggered for calibration.

[0128] Global Inspection Unit: Within each range, calibration data collected by automatically calibrated sensors is used to predict the leakage risk data of each range using an SVR model, and a leakage threshold is set. If the leakage risk data is greater than or equal to the leakage threshold, then there is a leakage risk in that range; if the leakage risk data is less than the leakage threshold, then the range is safe.

[0129] Local inspection unit: For areas with leakage risks, infrared thermal imaging cameras are used for inspection, and dynamic error connectivity marking method is used to identify abnormal areas;

[0130] Fine-grained inspection unit: For abnormal areas, a multispectral camera is triggered to inspect the abnormal area, and a dynamic fusion method is used to obtain the comprehensive band data. A leakage judgment threshold is set. If the comprehensive band data of the abnormal area is greater than the leakage judgment threshold, it means that the abnormality in the area is caused by gas turbine leakage; if the comprehensive band data of the abnormal area is less than the leakage judgment threshold, it means that the abnormality in the area is caused by the equipment itself.

[0131] Early warning unit: If the regional anomaly is caused by a gas turbine leak, an early warning will be triggered immediately; if the regional anomaly is caused by the equipment itself, the infrared thermal imaging camera and multispectral camera will be triggered to conduct high-frequency inspections of the abnormal area.

[0132] Example 3

[0133] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the intelligent inspection method for gas turbine leakage based on multi-sensor data fusion provided above.

[0134] Since the electronic device described in this embodiment is the electronic device used to implement the intelligent gas turbine leak inspection method based on multi-sensor data fusion in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the intelligent gas turbine leak inspection method based on multi-sensor data fusion in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any electronic device used by those skilled in the art to implement the intelligent gas turbine leak inspection method based on multi-sensor data fusion in the embodiments of this application falls within the scope of protection of this application.

[0135] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0136] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A multi-sensor data fusion intelligent inspection method for gas turbine leakage, characterized in that, The method comprises the following steps: Step SS1, M sensors are arranged in each of T ranges, cross-sensor data mutual verification is performed by using a multi-level regression model based on data collected from the M sensors, and a verification result is obtained, if the data of a sensor deviates from the verification result, an automatic calibration mechanism of the sensor is automatically triggered for calibration; Step SS2, in each range, calibration data collected by the automatically calibrated sensor is used to predict leakage risk data of each range by using an SVR model, and a leakage threshold is set, if the leakage risk data is greater than or equal to the leakage threshold, the range has a leakage risk; if the leakage risk data is less than the leakage threshold, the range is safe; Step SS3, for the range with a leakage risk, an infrared thermal imaging camera is used for inspection, and a dynamic error connected marking method is used to obtain an abnormal area; The specific method comprises the following steps: Real-time inspection is performed on the range with a leakage risk by using an infrared thermal imaging camera to obtain a thermal imaging image, wherein the infrared thermal imaging camera inspection frequency is u; Median filtering is used to denoise the thermal imaging image to obtain a denoised thermal imaging image; A sliding window method is used to calculate the average value of each pixel point in the thermal imaging image in the last n time to obtain an average thermal imaging image; The error of each pixel point in the average thermal imaging image and the denoised thermal imaging image is calculated to obtain an error thermal imaging image; For the error thermal imaging image, the mean and standard deviation of each pixel point in the error thermal imaging image are calculated, and a dynamic error threshold is set based on the mean and standard deviation; In the error thermal imaging image, the pixel points with an error greater than the dynamic error threshold are marked in blue, each blue pixel point in the error thermal imaging image is assigned a unique label 0, and a traversal method is used to detect whether the labels of the upper, lower, left and right and four diagonal lines of the blue pixel point with the label 0 are the same, a connected region marking method is used to identify the blue region formed in the error thermal imaging image, and the blue region is an abnormal area; Step SS4, for the abnormal area, a multispectral camera is triggered to inspect the abnormal area, a dynamic fusion method is used to obtain band synthesis data, and a leakage judgment threshold is set, if the band synthesis data of the abnormal area is greater than the leakage judgment threshold, it indicates that the abnormal area is caused by gas turbine leakage; if the band synthesis data of the abnormal area is less than the leakage judgment threshold, it indicates that the abnormal area is caused by the equipment itself; 2. The method of claim 1, wherein, Step SS5, when the abnormal area is caused by gas turbine leakage, an alarm is immediately triggered; when the abnormal area is caused by the equipment itself, the infrared thermal imaging camera and the multispectral camera are triggered to perform high-frequency inspection on the abnormal area. The specific method of arranging M sensors in each of T ranges comprises: Optical fiber type methane sensors, temperature sensors and pressure sensors are installed in the gas turbine valve range, the fuel pipe expansion joint range, the gas turbine fuel ring pipe range, the combustion chamber range, the combustion chamber bottom range, the combustion chamber top range and the air outlet range in the gas turbine cover shell.

3. The method of claim 2, wherein, The specific way that the automatic calibration mechanism of the sensor is triggered for calibration when the data collected from the M sensors deviates from the verification result includes: Based on the data collected from the M sensors, the multi-level regression model is used to model the methane concentration data, temperature data and pressure data, and the temperature data and pressure data are input into the multi-level regression model to obtain the verified methane concentration data; the methane concentration data and pressure data are input into the multi-level regression model to obtain the verified temperature data, and the methane concentration data and temperature data are input into the multi-level regression model to obtain the verified pressure data. The real-time methane concentration data, temperature data and pressure data are compared with the methane concentration data, temperature data and pressure data verified by the multi-level regression model, and if the data measured by the sensor deviates from the verification result, the verification result, i.e. the verified methane concentration data, temperature data and pressure data, indicates that the sensor deviates, and the automatic calibration mechanism of the sensor is automatically triggered for calibration.

4. The method of claim 3, wherein, The specific way that the SVR model is used to predict the leakage risk data of each range using the calibration data collected by the automatically calibrated sensor in each range includes: T databases are established for T ranges, and the data collected by the automatically calibrated sensor in each range is stored in the database. The database includes M samples, each sample is composed of methane concentration data, temperature data, pressure data and leakage risk data, and the M samples are input into the SVR model as input data to obtain the predicted leakage risk data of each range.

5. The method of claim 4, wherein, The specific way that the multispectral camera is triggered to patrol the abnormal area and the dynamic fusion method is used to obtain the band comprehensive data includes: When the abnormal area is determined, the multispectral camera is triggered, the patrol frequency of the multispectral camera is f, and the multispectral camera collects N bands of each pixel point in the abnormal area; Each band includes radiation intensity, absorption bandwidth, spectral reflectivity and normalized absorption index data; For N bands of each pixel point, a dynamic adaptive method is used to obtain D bands; For D bands, a weighted average method is used to obtain the band comprehensive data of each pixel point, and the band comprehensive data of each pixel point is weighted and fused to obtain the band comprehensive data of the abnormal area.

6. The method of claim 5, wherein, The specific way that the dynamic adaptive method is used to obtain D bands for N bands of each pixel point includes: When the multispectral camera is triggered, the multispectral camera obtains N band data of each pixel point in the abnormal area, and a spectral angle matching analysis correction method is used to obtain the response intensity data of N bands of each pixel point to the abnormal area; The response intensity of N bands to the abnormal area is sorted in descending order, and an expert experience method is used to set a response threshold, and bands greater than or equal to the response threshold are retained, and bands less than the response threshold are removed, so that D bands are retained for each pixel point.

7. The method of claim 6, wherein, The specific method for obtaining the response intensity data of each pixel point on the abnormal area by using the spectral angle matching analysis correction method comprises: constructing the N-band radiation intensity, absorption bandwidth, spectral reflectivity and normalized absorption index data of the pixel point into N spectral vectors; calculating the spectral angle between the spectral vector of each band and the spectral feature vector of the abnormal area by using a spectral angle formula; for each band, using a Pearson correlation method to calculate the linear relationship between the spectral vector of each band and the spectral feature vector of the abnormal area to obtain a correlation value; using the correlation value to correct the spectral angle between the spectral vector of each band and the spectral data vector of the abnormal area to obtain the correlation intensity data of each band; and calculating the response intensity data from the correlation intensity data.

8. The method of claim 7, wherein, The method for obtaining the spectral feature vector of the abnormal area comprises: for each pixel point, extracting the spectral data of the N bands, constructing the spectral data into a spectral vector, and using a weighted average method to perform weighted calculation on the spectral vectors of the H pixel points to obtain the spectral feature vector of the abnormal area.

9. The method of claim 8, wherein, When the area abnormality is caused by the equipment itself, the specific method for triggering the infrared thermal imaging camera and the multispectral camera to perform high-frequency inspection on the abnormal area comprises: for the abnormal area caused by the equipment itself, triggering high-frequency inspection, wherein the inspection frequency of the infrared thermal imaging camera is set to 2u, and the inspection frequency of the multispectral camera is set to 2f.

Images