A steam boiler gas leakage detection method based on image processing

Abstract

The present application relates to the field of image data processing, and provides a steam boiler gas leakage detection method based on image processing, which comprises the following steps: setting an image acquisition unit to jointly cover a predetermined leakage risk area, acquiring corresponding reference images under different production conditions, and establishing an image reference library covering a target area; dividing a boiler operation time axis into multiple production states, and calling corresponding reference images for different states, performing difference, partition comparison and motion detection on real-time acquisition images, and obtaining suspected leakage cloud area under corresponding states; arranging a condensate water indicating patch, acquiring patch images, extracting color, brightness and texture features to form a patch humidity change curve, performing multi-source information fusion on time sequence features and production time states, and outputting corresponding steam leakage risk levels; generating alarm information according to the steam leakage risk levels, and recording and archiving in a monitoring and management system for energy consumption analysis and risk tracing.

Description

Technical Field

[0001] This invention relates to the field of image data processing, specifically to a method for detecting gas leaks in steam boilers based on image processing. Background Technology

[0002] Gas leak detection is a process that uses specialized equipment and technology to identify and locate gas leaks. Its purpose is to ensure safety, prevent energy waste, and protect the environment. It uses principles such as ultrasound, micro-flow, or tracer gas, combined with detection equipment, to issue early warnings and determine the source of the leak.

[0003] In food production workshops with high cleanliness requirements, steam boilers and their associated pipelines are typically located in equipment rooms close to the production line to facilitate the supply of high-temperature steam to processes such as sterilization, ingredient preparation, and filling. However, existing leak detection methods largely rely on a combination of fixed gas probes and manual inspections. While these methods can trigger alarms in cases of large-scale leaks and significant pressure anomalies, they still have significant shortcomings in the actual operating environment of cleanrooms. On the one hand, the walls and equipment surfaces of the workshop need to be kept smooth for easy cleaning and disinfection, making it unsuitable to install a large number of probes, brackets, and pipelines. This often limits the installation location of gas probes, making it difficult to cover a large number of high-risk areas. On the other hand, the presence of water mist generated during cleaning, white vapor clouds from normal exhaust, and condensation on equipment surfaces in cleanrooms makes it difficult for manual inspection to distinguish between normal water vapor and abnormal leaks in ordinary monitoring footage. This often requires judgment based on experience, which can easily lead to missed detections or misjudgments.

[0004] To address the aforementioned problems, this invention proposes a steam boiler gas leak detection method based on image processing. This method involves deploying cameras within the steam equipment's field of view to acquire baseline images of a predetermined area. Production status information is then segmented according to a timeline. Images are acquired of areas prone to leaks, and the condensate status is detected. Multi-source features are fused to determine the leak risk, and alarm information is generated and recorded based on the risk level. Summary of the Invention

[0005] In view of the existing problems mentioned above, a steam boiler gas leak detection method based on image processing is proposed.

[0006] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a steam boiler gas leakage detection method based on image processing, comprising:

[0007] Plan and deploy image acquisition units to cover the predetermined leakage risk area, acquire corresponding reference images under different production conditions, and establish an image reference library covering the target area;

[0008] By acquiring the production steam plan and cleaning plan, the operation timeline is divided into multiple production states, and corresponding reference images are called for different states. The real-time acquired images are differentially analyzed, compared in different regions, and motion detected to obtain the suspected leakage cloud area under the corresponding state.

[0009] Condensate indicator patches are placed in the predetermined leakage risk area. Patch images are continuously collected within a preset long time window. The patch moisture change curve is formed by extracting color, brightness and texture features. The moisture change trend, the shape and time sequence characteristics of suspected leakage clouds and production time status are fused by multi-source information to output the corresponding steam leakage risk level.

[0010] Alarm information is generated based on the risk level of steam leakage and recorded and archived in the monitoring and management system for energy consumption analysis and risk tracing.

[0011] As a preferred embodiment, the specific steps for covering the predetermined leakage risk area are as follows:

[0012] Based on the layout diagram and process pipeline diagram of the steam boiler equipment, and combined with the statistics of multiple leak points in the historical maintenance records, the predetermined leak risk areas are delineated; image acquisition units are fixedly installed at each predetermined leak risk area, and the height and pitch angle are adjusted by the mounting bracket to ensure that the effective field of view of the acquisition unit completely covers the corresponding connection parts, while the fields of view of adjacent acquisition units are appropriately overlapped.

[0013] After the physical installation is completed, each image acquisition unit is calibrated once, and uniform resolution, exposure time and frame rate parameters are set. Geometric distortion correction and image coordinate system are performed by suspending a calibration plate on site.

[0014] As a preferred embodiment, the specific steps for establishing an image reference library covering the target area are as follows:

[0015] The production steam usage plan and cleaning plan in the production management system of the production workshop are connected to the data through the communication interface. The corresponding time interval and working condition identifier are configured for each process segment according to the plan. When the working condition is identified to be in a stable stage, the image acquisition unit is automatically triggered to continuously acquire multiple frames of images with the current parameters. The baseline image of the monitoring perspective under the working condition is generated by using inter-frame averaging and noise filtering. The images are indexed and stored according to the working condition type and camera number to form an image baseline library that covers all leakage risk areas and corresponds to various working conditions.

[0016] As a preferred embodiment, the specific steps for performing differential analysis, partition comparison, and motion detection on the real-time acquired images are as follows:

[0017] Based on the production steam usage plan and cleaning plan, each plan is converted into a time period with start time, end time, and status identifier, and the steam boiler operation time axis is divided according to multiple production states. When a real-time image frame is received at any time, the production state to which it belongs is first found in the time axis according to the timestamp of the image frame, and then the reference image corresponding to the state is called from the image reference library. The real-time image and the reference image are subdivided at the pixel level to obtain the difference image of the state.

[0018] Upon receiving the real-time image, the system first retrieves a reference image corresponding to the camera and production status based on the camera calibration parameters and resolution settings. The size, viewing angle, and brightness of both the real-time and reference images are then corrected to align them on the pixel grid. The aligned images are then converted to grayscale. For each pixel, the grayscale values ​​of both the real-time and reference images at that location are read, and the absolute value of the difference between the two is calculated using the following formula:

[0019] ,

[0020] in Represents pixel coordinates, Indicates real-time image at the pixel level grayscale value at that location Indicates the reference image in pixels The grayscale value at that location;

[0021] The results are written into the output image at the same location to generate a difference map that reflects the degree of difference between the current image and the reference image;

[0022] Based on the difference map, combined with the pre-set leakage risk area mask, the region is divided into regions for comparative analysis. Using threshold segmentation, connected component analysis and inter-frame motion vectors, the motion regions that persist in the current state and have diffusion and rising characteristics are identified and extracted as suspected leakage cloud regions. For the time period in the cleaning state, the baseline image established for high water mist conditions is used and the judgment threshold for short-term large-area changes is increased to filter out the planned cleaning water mist from the detection results and retain abnormal steam clouds that do not match the planned state.

[0023] As a preferred embodiment, the specific steps for extracting the patch moisture variation curve through color, brightness, and texture features are as follows:

[0024] Condensation indicator patches are affixed to the predetermined leakage risk area. The image acquisition unit performs fixed-point framing and continuous shooting of the local area of ​​each patch within a preset long time window. A region of interest is predefined for each patch. The hue, saturation, and brightness components of the region are extracted through color space conversion. At the same time, texture features such as surface roughness and the number of spots are extracted based on the gray-level co-occurrence matrix.

[0025] As a preferred embodiment, the specific steps for fusing multi-source information such as humidity change trends, the morphology and temporal characteristics of suspected leak clouds, and production time status to output the corresponding steam leak risk level are as follows:

[0026] The grayscale image of the region of interest (ROI) of the condensation indicator patch is quantized according to L levels, and a grayscale co-occurrence matrix is ​​constructed and normalized. Texture statistics of contrast, energy, uniformity, and entropy are calculated. Surface roughness is calculated based on these statistics and weighting coefficients obtained from patch material calibration, using the following formula:

[0027] ,

[0028] in Indicates contrast weight. Represents entropy weight. Indicates the uniformity weight. Represents contrast texture statistics. This represents entropy texture statistics. This represents a texture statistic indicating uniformity.

[0029] The local roughness is calculated block by block in the patch area using a sliding window method and compared with the reference roughness in the initial dry state to obtain the roughness increment of each sliding window. When the roughness increment is greater than or equal to the set roughness threshold, the center position of the corresponding window is marked as a spot pixel. The number of connected regions of the spots and the sum of their areas are counted through connected component analysis to form the spot quantity index and spot density index of the patch.

[0030] The features are normalized and compared with the baseline features of the patch in its initial dry state to calculate the moisture index as a function of time. A moisture change curve is generated by time smoothing. The moisture change trend is correlated with the temporal characteristics of the suspected leak cloud above the patch obtained by differential detection. Simultaneously, considering the production time status, when the operating conditions are such that the steam supply should be relatively dry, but the moisture index remains above the threshold for a long period, and there is a spatial correspondence between the cloud features and the patch location, a high-risk situation of continuous steam leakage is identified. A comprehensive score is calculated by weighting the absolute value of the patch moisture, its growth rate, and the cloud feature score. The formula is as follows:

[0031] ,

[0032] in Indicates the weight of moisture level. Indicates the weight of the rate of humidification. Indicates cloud cluster weight. The score represents the absolute degree of humidity. The score indicates the rate at which humidity increases. Indicates the feature score of the cloud cluster;

[0033] The comprehensive score is compared with preset multi-level risk thresholds, which are low, medium, and high risk.

[0034] As a preferred embodiment, the specific steps for generating alarm information based on the steam leakage risk level are as follows:

[0035] According to the preset risk classification rules, risks are mapped to different levels of alarm events. Low risk is marked as prompt level, medium risk as warning level, and high risk as alarm level. A unique leakage event number is generated for events triggered by medium and high risks. Based on the calibration information of the acquisition unit and the position of the leakage cloud in the image, the corresponding device name and spatial location identifier are calculated. At the same time, the operation information is read and the various information is encapsulated into a structured alarm record.

[0036] Beneficial effects

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] 1. By acquiring the production steam consumption plan and cleaning plan, the operating time axis of the steam boiler is divided into multiple production states. During real-time detection, the current production state is first determined, and then the matching reference image is called for differential analysis. In this way, false alarms caused by cleaning conditions can be suppressed in the complex environment of cleanrooms where water mist and exhaust steam exist for a long time.

[0039] 2. By attaching condensate indicator patches to the predetermined leakage risk area and utilizing the patch's color and texture changes under long-term humidity, images of the patch are continuously acquired over a long period of time to construct humidity indexes and humidity change curves. This can detect small, continuous, and hidden vapor leaks in advance, achieving an early warning effect that is difficult to achieve with traditional concentration thresholds and instantaneous image detection. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation on the scope of this application.

[0041] Figure 1 This is a flowchart illustrating the present invention;

[0042] Figure 2 This is a comparison diagram of the effects of the present invention and the prior art, where gray bars represent the prior art and black bars represent the present invention. Detailed Implementation

[0043] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0044] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a method for detecting gas leaks in steam boilers based on image processing. The method includes the following steps:

[0045] Step S1: Set up the camera's field of view and acquire a reference image;

[0046] Step S2: Divide production status information and detect the image area;

[0047] Step S3: Detect the condensate status and assess the risk of leakage;

[0048] Step S4: Generate alarm information and record it for archiving.

[0049] This method is implemented in the order of S1–S4, and its overall process is as follows:

[0050] Plan and deploy image acquisition units to cover the predetermined leakage risk area, acquire corresponding reference images under different production conditions, and establish an image reference library covering the target area;

[0051] By acquiring the production steam plan and cleaning plan, the operation timeline is divided into multiple production states, and corresponding reference images are called for different states. The real-time acquired images are differentially analyzed, compared in different regions, and motion detected to obtain the suspected leakage cloud area under the corresponding state.

[0052] Condensate indicator patches are placed in the predetermined leakage risk area. Patch images are continuously collected within a preset long time window. The patch moisture change curve is formed by extracting color, brightness and texture features. The moisture change trend, the shape and time sequence characteristics of suspected leakage clouds and production time status are fused by multi-source information to output the corresponding steam leakage risk level.

[0053] Alarm information is generated based on the risk level of steam leakage and recorded and archived in the monitoring and management system for energy consumption analysis and risk tracing.

[0054] The specific steps for setting up the camera's field of view and acquiring a reference image are as follows:

[0055] Based on the layout diagram and process pipeline diagram of the steam boiler equipment, and combined with the statistics of frequently leaking points in historical maintenance records, the locations of the steam boiler body shell, upper steam drum connection, steam transmission pipeline valves, flange interfaces and welded joints are designated as predetermined leakage risk areas. Image acquisition units are fixedly installed at each predetermined leakage risk area, such as industrial cameras with dustproof and waterproof shells. The height and pitch angle are adjusted by the mounting bracket so that the effective field of view of the camera completely covers the corresponding connection parts. At the same time, the fields of view of adjacent cameras are appropriately overlapped so that if a single camera fails, the neighboring camera can still provide compensation monitoring.

[0056] After physical installation is completed, each image acquisition unit is calibrated once, and uniform resolution, exposure time and frame rate parameters are set. Geometric distortion correction and image coordinate system one are performed by suspending a calibration board on site to ensure that images acquired at different times are comparable. Specifically, geometric distortion correction and image coordinate system one are performed by temporarily placing a calibration board with known size and regular marked points in the camera's monitoring field of view, using calibration algorithms to obtain the intrinsic and extrinsic parameters of each image acquisition unit, and performing distortion correction and perspective transformation on the acquired images accordingly. At the same time, the coordinates of the calibration board in the image are mapped to a unified reference coordinate system, thereby completing geometric distortion correction and image coordinate system one.

[0057] The production steam consumption plan and cleaning plan in the production management system of the production workshop are connected to the image processing system through the communication interface. The corresponding time interval and working condition identifier are configured for each process segment according to the plan. When a certain working condition is identified to be in a stable stage, such as when the pressure and flow rate remain within the preset fluctuation range for a period of time, the image acquisition units are automatically triggered to continuously acquire multiple frames of images with the current parameters. The baseline image of the monitoring perspective under the working condition is generated by using inter-frame averaging and noise filtering. The images are indexed and stored according to the working condition type and camera number to form an image baseline library that covers all leakage risk areas and corresponds to the production steam supply working condition, high water mist cleaning working condition, shutdown and standby working condition, etc.

[0058] By precisely defining the monitoring range around structures prone to leakage during the installation phase and building sub-condition benchmarks during the benchmark acquisition phase, the subsequent detection process no longer relies on a single static background. Instead, it can call up matching benchmark images for comparison under different production conditions, which is different from the existing image monitoring solutions that only make rough background models based on fixed images.

[0059] The specific steps for dividing production status information and detecting the image area are as follows:

[0060] Based on the production steam supply plan and cleaning plan, each plan is converted into a time period with start time, end time, and status identifier. The steam boiler operation time axis is divided into various production states such as production steam supply, high water mist cleaning, and shutdown standby. When a real-time image frame is received at any time, the production state corresponding to the image frame is first found in the time axis according to the timestamp of the image frame. Then, the reference image corresponding to the state is called from the image reference library. The real-time image and the reference image are subtracted at the pixel level to obtain the difference image of the state.

[0061] Specifically, the process involves pixel-level difference analysis between the real-time image and the reference image to obtain a difference map for that state. Upon receiving the real-time image at the current moment, the process first calls up the reference image corresponding to the camera and production state based on the camera calibration parameters and resolution settings. Then, the size, viewing angle, and brightness of both the real-time and reference images are corrected to align them on the pixel grid. The aligned images are then converted to grayscale. For each pixel, the grayscale values ​​of the real-time and reference images at that location are read, and the absolute value of the difference between the two is calculated using the following formula:

[0062] ,

[0063] in Represents pixel coordinates, Indicates real-time image at the pixel level grayscale value at that location Indicates the reference image in pixels The grayscale value at that location;

[0064] The results are written into the output image at the same location to generate a difference map that reflects the degree of difference between the current image and the reference image;

[0065] Based on the differential map, and combined with the pre-set leakage risk area mask, the key areas such as the boiler body, valves, and flanges are divided into regions for comparative analysis. Using methods such as threshold segmentation, connected component analysis, and inter-frame motion vectors, motion areas that persist in the current state and have diffusion and rising characteristics are identified and extracted as suspected leakage cloud areas. For time periods in the cleaning state, a baseline image established for high water mist conditions is used, and the judgment threshold for short-term large-area changes is increased to filter out planned cleaning water mist from the detection results and retain abnormal steam clouds that do not match the planned state, thus realizing a differential detection mechanism that selects the baseline by state and checks for anomalies by region.

[0066] The specific steps for detecting the condensate condition and assessing leakage risk are as follows:

[0067] In areas with predetermined leakage risks, such as below valves, flange interfaces, and welded joints, condensate indicator patches are affixed. These condensate indicator patches are preferably made of materials that are sensitive to long-term moisture and whose color and surface texture change with moisture content. The image acquisition unit performs fixed-point framing and continuous shooting of a local area of ​​each patch within a preset long-term window, and predefines a region of interest for each patch. The hue, saturation, and brightness components of the region are extracted through color space conversion, and texture features such as surface roughness and the number of spots are extracted based on the gray-level co-occurrence matrix.

[0068] Specifically, the method involves extracting surface roughness and spot quantity texture features based on the gray-level co-occurrence matrix (GLCM). The gray-level image of the region of interest (ROI) of the condensation indicator patch is quantized according to L levels, and a GLCM is constructed and normalized to obtain the GLCM. Texture statistics of contrast, energy, uniformity, and entropy are calculated. Surface roughness is then calculated based on these statistics and weighting coefficients obtained from patch material calibration, using the following formula:

[0069] ,

[0070] in Indicates contrast weight. Indicates the uniformity weight. Represents entropy weight. Represents contrast texture statistics. This represents a statistical measure of texture uniformity. Represents entropy texture statistics;

[0071] The local roughness is calculated block by block in the patch area using a sliding window method and compared with the reference roughness in the initial dry state to obtain the roughness increment of each sliding window. When the roughness increment is greater than or equal to the set roughness threshold, the center position of the corresponding window is marked as a spot pixel. The number of connected regions of the spots and the sum of their areas are counted through connected component analysis to form the spot quantity index and spot density index of the patch.

[0072] The above characteristics are normalized and compared with the baseline characteristics of the patch in its initial dry state to calculate the moisture index changing over time. A moisture change curve is generated by time smoothing. The moisture change trend is correlated with the temporal characteristics of the suspected leak cloud above the patch, such as the frequency, duration, and area change. Simultaneously, considering the production time status, when the operating condition is that the production steam supply should be relatively dry, but the moisture index remains above the threshold for a long period, and the cloud characteristics have a spatial correspondence with the patch location, a high-risk situation of continuous steam leakage is identified. A comprehensive score is calculated by weighting the absolute value of the patch moisture, its growth rate, and the cloud characteristic score. The formula is:

[0073] ,

[0074] in Indicates the weight of moisture level. Indicates the weight of the rate of humidification. Indicates cloud cluster weight. The score represents the absolute degree of humidity. The score indicates the rate at which humidity increases. Indicates the feature score of the cloud cluster;

[0075] The comprehensive score is compared with preset multi-level risk thresholds, which are low, medium and high risk. For example, the thresholds can be set to 0.3 for low risk, 0.6 for medium risk and greater than 0.6 for high risk and lasting more than 10 minutes.

[0076] Compared with existing technologies that rely solely on instantaneous steam cloud morphology and flow anomalies for short-term judgment, this method introduces condensate indicator patches to amplify minute leaks that are difficult to observe directly into image humidity curves that can be tracked for a long time. By integrating cloud characteristics and production status from multiple sources, this method enables early identification of hidden and continuous leaks, provided that the pressure and flow are still within the normal range and the workshop needs to maintain clean conditions.

[0077] The specific steps for generating alarm information and recording and archiving it are as follows:

[0078] According to the preset risk classification rules, risks are mapped to different levels of alarm events. Low risk is marked as alert level, medium risk as warning level, and high risk as alarm level. A unique leakage event number is generated for events triggered by medium and high risks. Based on camera calibration information and the position of the leakage cloud in the image, the corresponding equipment name and spatial location identifier are calculated. At the same time, the current production time status, boiler pressure and steam flow and other operating information are read. The key information such as leakage risk level, event number, occurrence time, duration, leakage location, moisture index and change curve of associated patch, and comprehensive score are encapsulated into a structured alarm record.

[0079] Compared to existing technologies that simply trigger a one-time alarm when pressure exceeds limits and lack deep integration with energy consumption and hygiene management, this step integrates image leak detection results, energy consumption data, and production management information into a unified monitoring and management system, forming a recording and analysis mechanism from alarm to energy consumption to hygiene. This expands leak detection from a simple safety alarm into a comprehensive management tool for energy consumption management and risk traceability.

[0080] like Figure 2The image shows a comparison of the performance of a steam boiler gas leak detection method based on image processing. The horizontal axis lists the key performance indicators, and the vertical axis represents the exemplified performance scores, ranging from 0 to 100. The higher the value, the better the performance. The aim is to intuitively demonstrate the expected improvement of the present invention in key capabilities compared to typical prior art.

[0081] Example 2:

[0082] Based on the above embodiment 1, a steam boiler gas leak detection method based on image processing is provided in a clean production workshop setting for food production, specifically as follows:

[0083] Step 1: Install electric sliding tables running along tracks above and on both sides of the steam boiler. A single high-resolution image acquisition unit is fixed on each sliding table. QR code labels with unique codes are affixed to the boiler body shell, upper steam drum connection pipes, steam transmission pipeline valves, flange interfaces, and welded joints. The QR code locations correspond to leakage risk points in the equipment layout diagram. The sliding tables are driven according to a preset inspection path and stopping rhythm. They automatically stop when each QR code label enters the field of view and the image is clear. The QR code decoding result is used to determine the leakage risk area number corresponding to the current shooting location. By controlling the camera's tilt angle and zoom parameters, a single frame image completely covers the connection part corresponding to that number. Based on the boiler operating parameters, different operating conditions are determined. During multiple rounds of inspections under each operating condition, multiple frames of images are continuously acquired at the same QR code number location. Automatic registration and perspective unification are performed using the QR code identifiers. Time averaging and noise filtering are applied to the multiple frames to generate a baseline image of the leakage risk area under the corresponding operating condition. The image is stored in the image baseline library using the QR code number, operating condition type, and timestamp as indexes.

[0084] Step 2: Based on the steam consumption plan and cleaning plan, the steam boiler operating timeline is divided into multiple production states, and an image cache queue updated by time is maintained for each state. When a real-time image frame is received at any moment, its production state is first determined based on the timestamp. The most recent historical images are extracted from the cache queue corresponding to that state, and the mean value in the time direction is calculated in pixels to form a dynamic reference image for the current state. The real-time image and the dynamic reference image are compared in terms of regional features within the leakage risk area mask. It is preferred to divide the detection sub-areas according to the boiler body, valves, flanges, etc. The color, brightness, and texture feature vectors of each sub-area are statistically analyzed, and the similarity between the real-time features and the reference features is calculated. When the similarity is lower than a preset threshold and appears continuously in multiple frames, the sub-area is marked as an abnormal state area. Then, combined with motion vector information, short-term noise is filtered out, and only motion abnormal areas with continuously rising and spreading characteristics are retained as suspected leakage cloud areas.

[0085] Step 3: Affix condensate indicator patches below valves, flange interfaces, and welded joints. Take photos of local areas of the patches at fixed points within a preset long-term window. For each patch, define a predefined region of interest, convert the region to a grayscale image, calculate the average grayscale value at the current moment and the proportion of dark pixels below the set wetness threshold, and compare it with the baseline grayscale and baseline dark pixel proportion of the initial dry state to construct a humidity index. Then, perform a time-moving average on the humidity index to obtain a smooth humidity change curve. Correlate the absolute value of humidity and its growth rate within the window with the temporal characteristics of the suspected leak cloud above the patch, such as the frequency of occurrence, duration, and area change, obtained by differential detection. When the production time state belongs to a relatively dry production steam supply state, and the humidity index is continuously higher than the preset threshold within the long-term window, and the center of gravity of the cloud coincides with the patch position in space, a high-risk situation of continuous steam leakage is determined. Calculate the leakage risk score by combining the humidity magnitude score, growth rate score, and cloud feature score according to preset weights, and compare it with multi-level risk thresholds to obtain low, medium, and high leakage risk levels.

[0086] Step 4: Continuously receive leakage risk scores from each monitoring point. When the same device reaches low, medium, and high risk levels multiple times within a preset time window, it is first categorized into three alarm states: observation, early warning, and alarm, using a hysteresis-based grading rule. Upon first entering the early warning or alarm state, a leakage event index is automatically generated. During the event lifecycle, the maximum risk level and cumulative duration under this index are updated in real time. Spatial coordinates and device names are directly obtained from the 3D device model through a lookup table. Simultaneously, the production status, boiler pressure, steam flow, and patch moisture time series are recorded. The above information is written into the historical event database in the form of an event index with time series attachments. Only the current status indication is retained in the real-time alarm table, achieving the goal of aggregated management of the same leakage process by device and time.

[0087] The embodiments of the present invention described above are subject to modification and change of method by those skilled in the art without departing from the embodiments and broader aspects of the present invention. The appended claims are intended to include all such modifications and changes of method that do not depart from the present invention.

Claims

1. A method for detecting gas leaks in a steam boiler based on image processing, characterized in that, include: Plan and deploy image acquisition units to cover the predetermined leakage risk area, acquire corresponding reference images under different production conditions, and establish an image reference library covering the target area; By acquiring the production steam plan and cleaning plan, the operation timeline is divided into multiple production states, and corresponding reference images are called for different states. The real-time acquired images are differentially analyzed, compared in different regions, and motion detected to obtain the suspected leakage cloud area under the corresponding state. Condensate indicator patches are placed in the predetermined leakage risk area. Patch images are continuously collected within a preset long time window. The patch moisture change curve is formed by extracting color, brightness and texture features. The moisture change trend, the shape and time sequence characteristics of suspected leakage clouds and production time status are fused by multi-source information to output the corresponding steam leakage risk level. Alarm information is generated based on the risk level of steam leakage and recorded and archived in the monitoring and management system for energy consumption analysis and risk tracing.

2. The method for detecting gas leaks in a steam boiler based on image processing according to claim 1, characterized in that, The specific steps for covering the predetermined leakage risk area are as follows: Based on the layout diagram and process pipeline diagram of the steam boiler equipment, and combined with the statistics of multiple leak points in the historical maintenance records, the predetermined leak risk areas are delineated; image acquisition units are fixedly installed at each predetermined leak risk area, and the height and pitch angle are adjusted by the mounting bracket to ensure that the effective field of view of the acquisition unit completely covers the corresponding connection parts, while the fields of view of adjacent acquisition units are appropriately overlapped. After the physical installation is completed, each image acquisition unit is calibrated once, and uniform resolution, exposure time and frame rate parameters are set. Geometric distortion correction and image coordinate system unification are performed by suspending a calibration plate on site.

3. The method for detecting gas leaks in a steam boiler based on image processing according to claim 1, characterized in that, The specific steps for establishing an image reference library covering the target area are as follows: The production steam usage plan and cleaning plan in the production management system of the production workshop are connected to the data through the communication interface. According to the plan, the corresponding time interval and working condition identifier are configured for each process segment. When the working condition is identified to be in a stable stage, the image acquisition unit is automatically triggered to continuously acquire multiple frames of images with the current parameters. The baseline image of the monitoring perspective under the working condition is generated by using inter-frame averaging and noise filtering. The images are indexed and stored according to the working condition type and camera number to form an image baseline library covering all leakage risk areas and corresponding to various working conditions.

4. The method for detecting gas leaks in a steam boiler based on image processing according to claim 1, characterized in that, The specific steps for performing differential analysis, partition comparison, and motion detection on real-time acquired images are as follows: Based on the production steam usage plan and cleaning plan, each plan is converted into a time period with start time, end time, and status identifier, and the steam boiler operation time axis is divided according to multiple production states. When a real-time image frame is received at any time, the production state to which it belongs is first found in the time axis according to the timestamp of the image frame, and then the reference image corresponding to the state is called from the image reference library. The real-time image and the reference image are differentially analyzed at the pixel level to obtain the difference image of that state.

5. The image processing-based method for detecting gas leaks in a steam boiler according to claim 4, characterized in that, The specific steps for performing differential analysis, partition comparison, and motion detection on real-time acquired images also include: Upon receiving the real-time image, the system first retrieves a reference image corresponding to the camera and production status based on the camera calibration parameters and resolution settings. The size, viewing angle, and brightness of both the real-time and reference images are then corrected to align them on the pixel grid. The aligned images are then converted to grayscale. For each pixel, the grayscale values ​​of both the real-time and reference images at that location are read, and the absolute value of the difference between the two is calculated using the following formula: ， in Represents pixel coordinates, Indicates real-time image at the pixel level grayscale value at that location Indicates the reference image in pixels The grayscale value at that location; The results are written into the output image at the same location to generate a difference map that reflects the degree of difference between the current image and the reference image.

6. The image processing-based method for detecting gas leaks in a steam boiler according to claim 4, characterized in that, The specific steps for performing differential and partition comparison and motion detection on real-time acquired images also include: Based on the difference map, combined with the pre-set leakage risk area mask, the region is divided into regions for comparative analysis. Using threshold segmentation, connected component analysis, and inter-frame motion vectors, the motion regions that persist in the current state and have diffusion and rising characteristics are identified and extracted as suspected leakage cloud regions. For the time period in the cleaning state, the baseline image established for high water mist conditions is used, and the judgment threshold for short-term large-area changes is increased to filter out the planned cleaning water mist from the detection results and retain abnormal steam clouds that do not match the planned state.

7. The method for detecting gas leaks in a steam boiler based on image processing according to claim 1, characterized in that, The specific steps for extracting the patch moisture variation curve through color, brightness, and texture features are as follows: Condensation indicator patches are affixed to the predetermined leakage risk area. The image acquisition unit performs fixed-point framing and continuous shooting of the local area of ​​each patch within a preset long time window. A region of interest is predefined for each patch, and the hue, saturation, and brightness components of the region are extracted through color space conversion. At the same time, texture features such as surface roughness and the number of spots are extracted based on the gray-level co-occurrence matrix.

8. The image processing-based method for detecting gas leaks in a steam boiler according to claim 1, characterized in that, The specific steps for fusing multi-source information, including humidity variation trends, the morphology and temporal characteristics of suspected leak clouds, and production time status, to output the corresponding steam leak risk level are as follows: The grayscale image of the region of interest (ROI) of the condensation indicator patch is quantized according to L levels, and a grayscale co-occurrence matrix is ​​constructed and normalized. Texture statistics of contrast, energy, uniformity, and entropy are calculated. Surface roughness is calculated based on these statistics and weighting coefficients obtained from patch material calibration, using the following formula: ， in Indicates contrast weight. Represents entropy weight. Indicates the uniformity weight. Represents contrast texture statistics. Represents entropy texture statistics. This represents the texture statistics for uniformity.

9. The method for detecting gas leaks in a steam boiler based on image processing according to claim 8, characterized in that, The specific steps for fusing multi-source information, including humidity variation trends, the morphology and temporal characteristics of suspected leak clouds, and production time status, to output the corresponding steam leak risk level, also include: The local roughness is calculated block by block in the patch area using a sliding window method and compared with the reference roughness in the initial dry state to obtain the roughness increment of each sliding window. When the roughness increment is greater than or equal to the set roughness threshold, the center position of the corresponding window is marked as a spot pixel. The number of connected regions of the spots and the sum of their areas are counted through connected component analysis to form the spot quantity index and spot density index of the patch. The features are normalized and compared with the baseline features of the patch in its initial dry state to calculate the moisture index as a function of time. A moisture change curve is generated by time smoothing. The moisture change trend is correlated with the temporal characteristics of the suspected leak cloud above the patch obtained by differential detection. Simultaneously, considering the production time status, when the operating conditions are such that the steam supply should be relatively dry, but the moisture index remains above the threshold for a long period, and there is a spatial correspondence between the cloud features and the patch location, a high-risk situation of continuous steam leakage is identified. A comprehensive score is calculated by weighting the absolute value of the patch moisture, its growth rate, and the cloud feature score. The formula is as follows: ， in Indicates the weight of moisture level. Indicates the weight of the rate of humidification. Indicates cloud cluster weight. The score represents the absolute degree of humidity. The score indicates the rate at which humidity increases. Indicates the feature score of the cloud cluster; The comprehensive score is compared with the preset multi-level risk thresholds, which are low, medium and high risk.

10. The method for detecting gas leaks in a steam boiler based on image processing according to claim 1, characterized in that, The specific steps for generating alarm information based on the steam leak risk level are as follows: According to the preset risk classification rules, risks are mapped to different levels of alarm events. Low risk is marked as prompt level, medium risk as warning level, and high risk as alarm level. A unique leakage event number is generated for events triggered by medium and high risks. Based on the calibration information of the acquisition unit and the position of the leakage cloud in the image, the corresponding device name and spatial location identifier are calculated. At the same time, the operation information is read and the various information is encapsulated into a structured alarm record.

Images