Critical heat flux density evaluation and early warning method based on dynamic recognition of dry patch image

Abstract

This invention discloses a critical heat flux density assessment and early warning method based on dynamic recognition of dry spot images, relating to the field of reactor safety analysis technology. The method includes the following steps: Real-time acquisition of images of the heated wall surface of a narrow rectangular flow channel using a high-speed camera system, followed by preprocessing; dry spot region segmentation and extraction of dry spot outline information; calculation of the geometric feature parameters of the dry spot based on the dry spot outline information and a preset spatial resolution calibration coefficient; construction of a liquid film stability index based on the geometric feature parameters and division of the liquid film stability region; construction of a coupled prediction model of dry spot dynamic expansion and wall superheat, and generation of a predicted remaining time value for the occurrence of the critical heat flux density; and construction of a graded early warning mechanism based on the liquid film stability index, dry spot expansion rate, and predicted remaining time value, triggering corresponding level early warning signals. This solution can transform visual information into accurate early warning decisions.

Description

Technical Field

[0001] This invention relates to the field of reactor safety analysis technology, specifically to a critical heat flux density assessment and early warning method based on dynamic identification of dry spot images. Background Technology

[0002] Narrow rectangular flow channel structures are widely used in many high heat flux density heat transfer devices due to their compact structure and large specific surface area, such as plate fuel research reactors, compact heat exchangers, and cooling channels for high-power electronic devices. In these applications, the equipment typically operates under extreme conditions, where the coolant undergoes intense boiling heat transfer as it flows over the heated wall. When the wall heat flux density exceeds a certain limit, namely the critical heat flux (CHF), a stable vapor film covers the heated wall, causing a sharp deterioration in heat transfer capacity and a rapid rise in wall temperature. This "boiling crisis" phenomenon must be avoided at all costs, as it can directly lead to the burnout of heat transfer elements and even cause the failure of the entire system.

[0003] Currently, the prediction and monitoring of heat-induced flow (CHF) in narrow rectangular channels mainly rely on empirical formulas fitted from a large amount of experimental data, or on indirect measurements of thermal-hydraulic parameters (such as pressure, temperature, and flow velocity) at the channel inlet and outlet. These methods have limitations. First, traditional methods infer CHF occurrence by monitoring macroscopic parameters, failing to directly observe the fundamental physical phenomena leading to CHF, namely the rupture of the liquid film and the formation and development of dry spots on the heated wall surface. Therefore, early warnings are often delayed and indirect, potentially triggering an alarm only after a boiling crisis has already occurred. Second, empirical formulas provide average or limiting parameters for the entire channel, failing to provide detailed spatial information such as the specific location, shape, and distribution of dry spots on the heated wall surface. Finally, empirical formulas are typically only effective within specific geometries and operating conditions; their predictive accuracy may decrease with changes in channel size, wall characteristics, or working fluid. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a critical heat flux density assessment and early warning method based on dynamic identification of dry spot images, thus solving the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a critical heat flux density assessment and early warning method based on dynamic identification of speckle images, comprising the following steps:

[0006] S1, a high-speed camera system is used to acquire a sequence of images of the heating wall surface of the narrow rectangular flow channel in real time, and the heating wall surface images in the sequence are preprocessed;

[0007] S2, perform dry spot region segmentation on the preprocessed heated wall image and extract the outer contour information of the dry spots;

[0008] S3. Calculate the geometric feature parameters of the dry spots based on the outer contour information of the dry spots and the preset spatial resolution calibration coefficients. The geometric feature parameters include at least area, roundness, distribution density, and spatial distribution entropy.

[0009] S4. Construct a liquid film stability index based on geometric feature parameters, and divide the liquid film stability region according to the liquid film stability index and the preset stability threshold.

[0010] S5, construct a coupled prediction model of dry spot dynamic expansion and wall superheat, and generate the remaining time prediction value of the critical heat flux density. The coupled prediction model of dry spot dynamic expansion and wall superheat adopts a layered architecture, including a dry spot dynamic feature extraction layer, a coupling mapping layer and a time extrapolation prediction layer.

[0011] S6. Based on the liquid film stability index, dry spot expansion rate, and remaining time prediction, a graded early warning mechanism is constructed and corresponding early warning signals are triggered.

[0012] Furthermore, in step S1, the preprocessing of the heated wall image includes: sequentially performing grayscale conversion, Gaussian filtering for noise reduction, and contrast-limited adaptive histogram equalization on the heated wall image.

[0013] Furthermore, step S2 includes the following steps:

[0014] S21, Perform preliminary speckle segmentation on the preprocessed heated wall image, and select an adaptive thresholding method or a deep learning network model for pixel-level separation based on speckle features to obtain a binarized heated wall image.

[0015] S22, Morphological correction is performed on the binarized heated wall image. Morphological closing operation is used to fill the voids inside the dry spots and connect the broken edges to obtain the corrected binarized mask image.

[0016] S23, the edge detection operator is used to scan the corrected binary mask image, extract the outer contour coordinate sequence of the dry spots, and the sub-pixel edge detection technology is used to interpolate and correct the contour coordinates of the dry spots.

[0017] Furthermore, in step S21, if the image signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, an adaptive thresholding method is used, and the neighborhood window size is dynamically set according to the channel width; if the image signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold, a deep learning model is used for segmentation.

[0018] Furthermore, in step S3, geometric feature parameters are calculated based on the binarized mask image using a connected component analysis algorithm.

[0019] Furthermore, step S4 includes the following steps:

[0020] S41, normalize the geometric feature parameters to generate coverage index, morphology index and order index;

[0021] S42, the liquid film stability index is calculated using a linear weighted model based on the coverage index, morphology index and order index;

[0022] S43, the liquid film stability region is divided according to the liquid film stability index and the preset stability threshold.

[0023] Furthermore, in step S43, the preset stability threshold includes a first stability threshold and a second stability threshold. When the liquid film stability index is greater than or equal to the first stability threshold, it is defined as a stable region; when the liquid film stability index is less than the first stability threshold but greater than the second stability threshold, it is defined as a fluctuating region; when the liquid film stability index is less than or equal to the second stability threshold, it is defined as a high-risk region.

[0024] Furthermore, in step S5, the speckle dynamics feature extraction layer is used to extract kinematic feature parameters of speckles from a continuous sequence of heated wall images. The kinematic feature parameters include speckle propagation rate and propagation acceleration.

[0025] The coupling mapping layer, based on the micro-liquid layer evaporation theory, establishes a nonlinear correlation to map the dry spot expansion rate into an estimated value of wall superheat.

[0026] The time extrapolation prediction layer, based on the current motion state of the dry spot, uses a second-order approximation equation with Taylor expansion to construct a prediction model for the remaining time of the critical heat flux density and generates a predicted value for the remaining time of the critical heat flux density.

[0027] Furthermore, in step S6, the tiered early warning mechanism includes:

[0028] If the liquid film stability index remains in the fluctuation range for a preset number of consecutive frames, a Level 1 warning will be triggered.

[0029] If the liquid film stability index is less than or equal to the second stability threshold or the instantaneous growth rate of the dry spot area exceeds the critical expansion rate, a level two warning is triggered.

[0030] If the predicted time remaining before the critical heat flux density occurs is less than the preset safety response time threshold, the highest priority alarm will be triggered.

[0031] This invention provides a method for assessing and warning critical heat flux density based on dynamic identification of speckle images, which has the following beneficial effects:

[0032] 1. This invention, based on high-speed imaging and image processing technology, directly captures the formation, expansion, and fusion process of dry spots on heated walls, transforming the traditional method of indirectly inferring CHF through macroscopic thermal parameters into in-situ observation of fundamental physical phenomena. It directly and accurately identifies and quantifies the physical origins leading to critical heat flux density—the formation, expansion, and evolution of dry spots—providing early warnings based on core physical phenomena. This fundamentally overcomes the lag and uncertainty inherent in traditional methods that rely on indirect measurement of macroscopic parameters, significantly improving the accuracy of critical heat flux density monitoring and early warning.

[0033] 2. By combining morphological optimization and sub-pixel contour correction with spatial calibration coefficients, the system can automatically calculate multi-dimensional geometric feature parameters such as the area, roundness, distribution density, and spatial distribution entropy of the dry spots. Furthermore, it can extract dynamic temporal feature parameters such as the expansion rate and expansion acceleration, thereby achieving quantitative characterization of the irregularity of the dry spot morphology, the disorder of its spatial distribution, and its evolution trend. This solves the limitation of traditional methods in providing fine spatial information.

[0034] 3. The liquid film stability index (LFSI) constructed in this invention is based on real-time visual features rather than empirical coefficients. By dynamically adjusting the weights through principal component analysis, it can adaptively reflect the influence of various characteristic parameters on CHF under different channel geometry, wall material and working fluid properties. The coupled prediction model maps visual kinematic parameters to wall superheat estimates, avoiding the dependence of traditional empirical formulas on specific structures and improving the robustness and generalization ability of the model when operating conditions change.

[0035] 4. From image acquisition, preprocessing, speckle segmentation, feature quantization, stability assessment, dynamic prediction to multi-level early warning decision-making, this invention provides a complete algorithm closed loop and system architecture. The segmentation model based on deep learning and the hierarchical switching strategy based on adaptive thresholding method balance accuracy and computational efficiency. The hierarchical early warning mechanism integrates steady-state index assessment and transient time extrapolation, achieving a unity of early warning sensitivity and reliability, and providing an engineering and automated integrated solution for monitoring boiling heat transfer in narrow rectangular flow channels. Attached Figure Description

[0036] Figure 1 This is a flowchart of the critical heat flux density assessment and early warning method based on dynamic identification of dry spot images according to the present invention.

[0037] Figure 2 This is a schematic diagram of the spatiotemporal evolution of the dry spot in the critical heat flux density assessment and early warning method based on dynamic recognition of dry spot images according to the present invention. Detailed Implementation

[0038] 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.

[0039] Example 1

[0040] Please see Figure 1 and Figure 2 This invention provides a method for assessing and warning critical heat flux density based on dynamic identification of speckle images, comprising the following steps:

[0041] S1, a high-speed camera system is used to acquire a sequence of images of the heating wall surface of the narrow rectangular flow channel in real time, and the heating wall surface images in the sequence are preprocessed;

[0042] S2, perform dry spot region segmentation on the preprocessed heated wall image and extract the outer contour information of the dry spots;

[0043] S3. Calculate the geometric feature parameters of the dry spots based on the outer contour information of the dry spots and the preset spatial resolution calibration coefficients. The geometric feature parameters include at least area, roundness, distribution density, and spatial distribution entropy.

[0044] S4. Construct a liquid film stability index based on geometric feature parameters, and divide the liquid film stability region according to the liquid film stability index and the preset stability threshold.

[0045] S5, construct a coupled prediction model of dry spot dynamic expansion and wall superheat, and generate the remaining time prediction value of the critical heat flux density. The coupled prediction model of dry spot dynamic expansion and wall superheat adopts a layered architecture, including a dry spot dynamic feature extraction layer, a coupling mapping layer and a time extrapolation prediction layer.

[0046] S6. Based on the liquid film stability index, dry spot expansion rate, and remaining time prediction, a graded early warning mechanism is constructed and corresponding early warning signals are triggered.

[0047] This method first uses a high-speed camera system deployed in the observation section to acquire a high-frame-rate image sequence of the heated wall surface within a narrow rectangular flow channel in real time. The high-frame-rate image sequence of the heated wall surface is an ordered set of multiple frames of digital images continuously acquired by the high-speed camera system at constant time intervals. All images in the high-frame-rate image sequence are taken from the same fixed observation area on the flow channel wall, ensuring that the images at different times are completely aligned in space, thereby enabling accurate tracking of the evolution of the same dry spot over time.

[0048] To highlight the analytical target and suppress interference, the acquired raw images need to undergo a series of preprocessing operations. Preprocessing includes sequentially performing grayscale conversion, Gaussian filtering for noise reduction, and contrast-limited adaptive histogram equalization on the heated wall image. The specific preprocessing workflow includes:

[0049] The heated wall image is converted into a single-channel grayscale image to simplify calculations. A Gaussian filtering algorithm with a preset kernel size is used to remove random noise, preferably a 3×3 to 7×7 pixel kernel size, with a standard deviation (σ) set to 0.5 to 1.5 to smooth noise while preserving high-frequency information at the edges of the dry spots. To maximize the contrast between the dry spots and the surrounding liquid film area, a contrast-limited adaptive histogram equalization (CLAHE) technique is employed. The heated wall image is divided into several non-overlapping blocks and equalized separately, effectively addressing potential uneven illumination within the narrow rectangular flow channel and maximizing the contrast between the dry spots and the surrounding liquid film area.

[0050] Step S2 involves performing speckle segmentation and precise contour localization on the preprocessed image. This step specifically includes the following process: S21, performing preliminary speckle segmentation on the preprocessed heated wall image. Based on the speckle features, an adaptive thresholding method or a deep learning network model is selected for pixel-level separation to obtain a binarized heated wall image. For images with significant grayscale differences, an adaptive thresholding method is used. That is, if the image signal-to-noise ratio is higher than a preset signal-to-noise ratio threshold, the adaptive thresholding method is used. The neighborhood window size is dynamically set according to the channel width to accurately separate speckles and liquid films with significant grayscale differences under local illumination changes. Specifically, the neighborhood window is set to a rectangle, and the size of the neighborhood window is set to 1 / 10 to 1 / 5 of the channel width, thereby accurately separating speckles and liquid films under local illumination changes. For complex morphological scenes (such as numerous speckles, overlapping boundaries, and irregular shapes), that is, if the image signal-to-noise ratio is lower than a preset signal-to-noise ratio threshold, a deep learning model is used for segmentation. A pre-trained U-Net is used. A deep learning network model is used for pixel-level precise separation. The deep learning network model adopts an encoder-decoder architecture: the encoder contains a 4-layer downsampling module, each layer consisting of two 3×3 convolutional layers and one 2×2 max pooling layer with a stride of 2. The activation function is ReLU, which is used to extract high-dimensional features of the speckle. The upsampling module fuses the feature maps of the encoder layer and the decoder layer through skip connections to restore the spatial details of the speckle edges. The model training dataset contains 100,000 speckle images under different working conditions in a narrow rectangular flow channel. The image resolution is uniformly 256×256 pixels. The speckle labeling standard is to use the gray-level abrupt change between the speckle and the liquid film as the boundary, including the complete speckle region and excluding the areas obscured by steam bubbles (bubbles are labeled as background). The Dice loss function (to deal with the class imbalance problem of low speckle pixel ratio) and the Adam optimizer are used. Training is terminated when the pixel accuracy on the validation set is ≥98% and the Dice coefficient is ≥0.95. After training, the generalization ability of the model is verified by 5000 test set images that were not used in the training.

[0051] Specifically, when the grayscale difference between the speckle and the liquid film is significant, the ratio of image signal (grayscale value of the speckle region) to noise (background fluctuations, reflection interference) is high. A high signal-to-noise ratio (SNR) indicates that the effective signal in the image is prominent. In this case, the adaptive thresholding method (local statistical segmentation) can efficiently separate the target from the background. For complex scenes with numerous speckles, overlapping boundaries, and irregular shapes, the macroscopic statistical characteristics are high speckle distribution density (number of speckles per unit area or area ratio). Low SNR corresponds to complex scenes with blurred grayscale contrast and strong interference. When the density is high, traditional thresholding methods are prone to undersegmentation or oversegmentation, and pixel-level accurate separation must be achieved by relying on the feature learning capabilities of deep learning. The formula for calculating the SNR is as follows: ,in The difference in grayscale mean between the dry spot and liquid film regions. The standard deviation of the background noise is used as the preset signal-to-noise ratio threshold in this embodiment, which is 25dB.

[0052] S22, perform morphological correction on the binarized heated wall image, using morphological closing operation to fill the voids inside the dry spots and connect the broken edges to obtain the corrected binarized mask image; specifically, perform morphological closing operation (dilation followed by erosion) on the binarized heated wall image, using circular structuring elements with a radius of 2 to 3 pixels to fill the voids inside the dry spots and connect the broken edges to improve the integrity of the dry spot outline, thus obtaining the corrected binarized mask image.

[0053] S23, precise contour localization. S23 uses edge detection operators (such as Canny, Sober) to scan the corrected binarized mask image, extracts the outer contour coordinate sequence of the speckle, and uses sub-pixel edge detection technology to interpolate and correct the speckle contour coordinates.

[0054] Before performing quantitative analysis on the identified dry spots, this method first performs a spatial calibration step: a reference object with known geometric dimensions, such as a standard scale grid, is placed in the visible area of ​​the narrow rectangular flow channel. By comparing the pixel distance in the image with the actual physical distance, the spatial resolution calibration coefficient k (unit: mm / pixel) is calculated as the preset spatial resolution calibration coefficient. This coefficient serves as a global parameter throughout all subsequent physical quantity conversion processes. After calibration, the reference object is removed, and a high-speed camera system is activated to continuously acquire images of the heated wall surface at a frame rate of no less than 200 fps. The heating area of ​​interest is automatically extracted using a preset flow channel geometry template.

[0055] Based on the spatial resolution calibration coefficient, the system converts the number of speckle pixels in the image into data with actual physical meaning. After successfully segmenting the speckles, the system automatically calculates geometric feature parameters, including area, roundness, distribution density, and spatial distribution entropy, based on the binarized mask image and using a connected component analysis algorithm. In this embodiment, the formula for calculating the area is: Where N is the total number of pixels in the connected region, and k (mm / pixel) is the spatial resolution calibration coefficient of the camera; the formula for calculating roundness is: Where P is the perimeter of the dry spot, which is the outline length of the dry spot obtained by the edge detection algorithm and is used to quantify the irregularity of the dry spot shape; the distribution density (D) is the ratio of the total area of ​​all dry spots in the field of view to the heating area of ​​the flow channel, as well as the number of dry spots per unit area, used to characterize the density of boiling; the spatial distribution entropy (S) is calculated using the Shannon entropy formula to characterize the probability distribution of dry spots in each grid, in order to quantify the disorder of the dry spot distribution.

[0056] Based on step S3, this method further performs visualization and storage operations. The system generates a real-time dynamic distribution map of dry spots according to the segmentation results, and displays the identified dry spot contours and location information on the original video stream in pseudo-color or highlight overlay to form a visual monitoring interface. At the same time, the system synchronously records and stores image sequences with dry spot feature annotations and corresponding feature parameter datasets to establish a full-cycle boiling process database.

[0057] S4. Construct a liquid film stability index based on geometric feature parameters, and divide the liquid film stability region according to the liquid film stability index and a preset stability threshold. This invention constructs a Liquid Film Stability Index (LFSI) based on the extracted geometric feature parameters. The liquid film stability index is a dimensionless value from 0 to 1 (1 represents stability, 0 represents the occurrence of critical heat flux density), used to comprehensively characterize the health state of the liquid film. The construction and calculation of the liquid film stability index specifically includes the following steps:

[0058] S41, to eliminate the influence of different dimensions, the geometric feature parameters are normalized to generate coverage, morphology, and orderliness indices. Among these, the total area of ​​the dry spots (…) ) Calculate coverage index ( The formula for calculating the coverage index is: = 1—min(1, / In the formula, This represents the total area of ​​the dry spots. The critical value of the total dry spot area at which the critical heat flux density occurs; using the average roundness of the dry spot ( ) as a morphological indicator ( ), used to reflect the smoothness of the dry spot edges; orderliness index ( The spatial distribution entropy is used for calculation, and the formula is as follows: In the formula, The maximum theoretical entropy value is the index that reflects the degree of disorder in the distribution of dry spots.

[0059] S42, Based on the coverage index, morphology index, and order index, the liquid film stability index (LFSI) is calculated using a linear weighted model. The mathematical expression is as follows: In this context, α, β, and γ are the weighting coefficients of the coverage index, morphology index, and order index, respectively, and α+β+γ=1. The specific values ​​of the weighting coefficients are determined by principal component analysis (PCA) of the boiling experimental data. The results show that the dry spot area has the greatest impact on the critical heat flux density, followed by the morphology.

[0060] S43. Based on the liquid film stability index and the preset stability threshold, divide the liquid film stability region and establish the correspondence between liquid film stability and stability. The preset stability threshold includes a first stability threshold and a second stability threshold. When the liquid film stability index is greater than or equal to the first stability threshold, it is defined as a stable region; when the liquid film stability index is less than the first stability threshold but greater than the second stability threshold, it is defined as a fluctuating region; when the liquid film stability index is less than or equal to the second stability threshold, it is defined as a high-risk region.

[0061] In this embodiment, the first stability threshold is 0.8 and the second stability threshold is 0.3. That is, when LFSI ≥ 0.8, it is defined as the stable region, which is in the safe nucleo-boiling stage; when 0.3 < LFSI < 0.8, it is defined as the fluctuating region, which triggers a first-level warning to indicate the occurrence of local heat transfer deterioration; when LFSI ≤ 0.3, it is defined as the high-risk region, which triggers a second-level warning to prevent the occurrence of critical heat flux density (CHF).

[0062] S5, a coupled prediction model for dry spot dynamic expansion and wall superheat is constructed. This coupled prediction model adopts a layered architecture, including a dry spot dynamic feature extraction layer, a coupling mapping layer, and a time extrapolation prediction layer. To address the challenge of traditional methods failing to quantify the critical heat flux (CHF) occurrence time in advance, this invention further constructs a coupled prediction model for dry spot dynamic expansion and wall superheat. This coupled prediction model for dry spot dynamic expansion and wall superheat adopts a layered architecture design, specifically including the following three logical processing layers.

[0063] The first layer is the speckle dynamics feature extraction layer. This layer is used to extract the kinematic feature parameters of speckles from a continuous sequence of binarized heated wall images. The kinematic feature parameters include speckle propagation rate and propagation acceleration. A sliding time window (window length Δt) is used to perform first-order and second-order difference operations on the speckle area A(t) to generate the speckle propagation rate. and extended acceleration The formula for calculating the dry spot spread rate is: ,in, The dry spot expansion rate at the current time t (physical unit: area / time, (mm² / s)). The first derivative of the dry spot area with respect to time represents the theoretical instantaneous rate of change of the dry spot area. The total area of ​​the dry spots at the current time t (physical unit: e.g., (mm²)); The end time of the previous sliding time window The total area of ​​dry spots (and) (Units consistent) The length of the sliding time window (time unit: e.g., s or ms) is the time interval between two adjacent dry spot area acquisitions and calculations. The formula for calculating the spreading acceleration is: ,in, The acceleration of the dry spot expansion at the current time t (physical unit: area / time², mm² / s²). The second derivative of the dry spot area with respect to time represents the theoretical instantaneous rate of change of the dry spot expansion rate. Let be the current time t, the rate of dry spot expansion. Last sliding time window end time The rate of dry spot expansion.

[0064] The second layer is the coupling mapping layer. This layer inverts the wall thermal state, which cannot be directly measured, through kinematic characteristic parameters. Based on the micro-liquid layer evaporation theory, the expansion of dry spots is driven by wall superheat. This invention establishes a nonlinear semi-empirical correlation to map the visual dry spot expansion rate to an estimated value (ΔTest) of wall superheat at the thermal level. The formula for calculating the estimated value of wall superheat is as follows: Where, Attotal is the total area of ​​the heated wall surface, η is the expansion coefficient, which is related to the physical properties of the working fluid (such as latent heat of vaporization and surface tension), and in this embodiment, the value is 0.6 to 0.8 after calibration; λ is the thermal coupling coefficient, which reflects the thermal conductivity of the wall material, and after experimental calibration, λ=2.5; μ is the area correction coefficient, which is used to compensate for the feedback effect of local heat transfer deterioration caused by large-area dry spots on the overall superheat, and in this embodiment, μ=20; The initial boiling superheat constant is set to 15℃ based on the experimental conditions. Through this model, the system essentially constructs a virtual thermocouple, enabling real-time sensing of sudden temperature rises on the wall surface even in the absence of a temperature sensor.

[0065] The third layer is the time extrapolation prediction layer, which is used to generate the final warning time signal. The system defines the critical criterion for the occurrence of critical heat flux density (CHF) as the dry spot area reaching the critical coverage rate. Based on the current motion state of the dry spot, a second-order approximation equation with Taylor expansion is used to construct the remaining time for the occurrence of the critical heat flux density (CHF). The prediction model for the remaining time of the critical heat flux density is generated, and the mathematical expression for the prediction model for the remaining time of the critical heat flux density is: The equation is analyzed to obtain multiple real solutions with respect to the time variable. The smallest positive real solution is taken as the predicted time remaining for the occurrence of the critical heat flux density. In the formula, The critical dry spot coverage rate is the ratio of the total dry spot area to the heated area of ​​the flow channel when the critical heat flux (CHF) occurs. It is the core critical criterion for determining the occurrence of CHF. Let be the total area of ​​the dry spots at the current time t. Let be the current time t, the rate of dry spot expansion. The remaining time from the current time t to the occurrence of CHF is to be predicted; The fixed coefficients for the second-order approximation of the Taylor expansion are the standard coefficients of the second derivative term; Let be the acceleration of the dry spot expansion at the current time t.

[0066] Through the above three-layer architecture, this invention transforms the originally discrete image features into continuous physical field deduction, which not only solves the lag problem of relying solely on area threshold alarms, but also achieves accurate prediction of the timing of CHF occurrence.

[0067] S6. Based on the liquid film stability index, dry spot expansion rate, and remaining time prediction, a graded early warning mechanism is constructed and corresponding early warning signals are triggered. This method establishes a graded early warning mechanism based on the above analysis. If the liquid film stability index is in the fluctuation range for a preset number of consecutive frames, a first-level early warning is triggered. In this embodiment, the preset number of consecutive frames is when the liquid film stability index is detected to be in the fluctuation range for 10 consecutive frames, triggering a first-level early warning, indicating that significant dry spots have appeared on the heating surface, but the liquid film still has rewetting capability, and the system needs to prompt the operator to pay attention.

[0068] A Level 2 warning is triggered when any of the following conditions are met: a) The liquid film stability index (LSFI) is less than or equal to the second stability threshold, at which point the LSFI value falls below the safety threshold of 0.3; b) The instantaneous growth rate of the dry spot area. Exceeding the critical expansion rate ( ), that is > In this embodiment, Defined as 10 mm² / s, this value was obtained by analyzing the balance point between the evaporation rate of the liquid film and the retraction rate of the contact line under different superheat conditions.

[0069] When the predicted time remaining before the critical heat flux density occurs is less than the system's preset safety response time threshold (in this embodiment, the safety response time threshold is 500ms), the system will exceed the threshold judgment of the liquid film stability index (LFSI), cover all other warning states, and forcibly trigger the highest priority blocking alarm.

[0070] To ensure the engineering practicality of this solution, the entire algorithm has been highly optimized, employing a parallel acquisition and computation strategy. Specifically, the image acquisition unit continuously operates at 200 fps (one frame every 5 milliseconds) to ensure complete recording of all details of the boiling process without losing any data for subsequent verification. In the real-time early warning stage, the system adopts the following mode: after each computation task is completed, the system immediately reads the latest acquired frame for the next round of analysis, ignoring intermediate frames generated during computation. Although the processing time for a single frame is approximately 50 milliseconds, meaning the system does not process every frame, considering the thermal inertia of the heating wall temperature rise and heat transfer deterioration (typically requiring a timescale of several hundred milliseconds), a monitoring frequency of 50 milliseconds is far faster than the speed at which physical failures occur. Therefore, the system is fully capable of issuing commands before the equipment burns out, meeting the real-time protection requirements of industrial sites.

[0071] The invention ensures that the total processing time for a single frame image is less than 50 milliseconds, thereby enabling real-time matching with high-speed cameras at frame rates of 200 fps or even higher, achieving zero-delay capture and early warning of boiling crisis precursors. The invention also includes a system for implementing the above method, comprising an image acquisition unit, a data processing unit executing the above algorithm, a visualization and storage unit for real-time presentation of the spatiotemporal evolution of dry spots and saving historical data, and an early warning information dissemination unit. Specifically, the image acquisition unit consists of a high-speed camera (model Basler acA2500-14gm), a heat-resistant macro lens, and a quartz glass observation window (with an inner SiO2-based anti-fog coating). The high-speed camera's frame rate and resolution meet the requirements for dynamic capture of dry spots. It is fixed by a damping bracket, and the lens optical axis is perpendicular to the heated wall surface to ensure image clarity. The inner side of the quartz glass observation window is coated with a high-temperature resistant anti-fog coating, adapting to the high-heat environment of the flow channel.

[0072] The industrial control host with a GPU acceleration module in the data processing unit has built-in image processing, speckle recognition, binarized mask image generation, LFSI calculation and coupled prediction model algorithms to ensure that the processing time of a single frame of image meets the real-time requirements.

[0073] The visualization and storage unit includes an industrial touch screen and a data storage module, which is used to display the dynamic distribution map of dry spots, feature parameters and early warning information in real time, and synchronously store image sequences and datasets with dry spot feature annotations, and manage data hierarchically according to preset rules.

[0074] The early warning information release unit consists of an audible and visual alarm and an RS485 communication module. It outputs corresponding alarm signals according to the early warning level and can be linked with the equipment control system to output control commands. The first-level early warning outputs low-frequency audible and visual alarms, the second-level early warning outputs medium-frequency audible and visual alarms and sends a power reduction suggestion signal, and the highest priority alarm outputs high-frequency audible and visual alarms and sends an emergency stop signal.

[0075] During system operation, the image acquisition unit continuously acquires images at 200fps, the data processing unit performs preprocessing, segmentation, quantization, LFSI calculation and coupling prediction in real time, the visualization unit displays the dynamic distribution map of dry spots and early warning information simultaneously, and the early warning unit outputs corresponding signals according to the judgment results. The whole system realizes real-time and accurate early warning of CHF in narrow rectangular flow channels.

[0076] To further illustrate the technical solution of the present invention and its practical application effect, this embodiment constructs a simulated narrow rectangular flow channel boiling heat transfer monitoring scenario and uses synthetic data to verify the effectiveness of the method.

[0077] This embodiment simulates a vertically placed narrow rectangular flow channel with the following geometric dimensions: length 500mm, width 40mm, and slit gap 2mm. The simulation conditions are set as follows: system pressure 0.1 MPa, coolant (water) mass flow rate G = 600 kg / (m²⋅s), and inlet subcooling 50°C. The simulated high-speed camera has a resolution of 512×256 pixels, and the frame rate is set to 200 fps (frame interval 5ms). Based on the correspondence between the flow channel width (40mm) and the image width (256 pixels), the spatial resolution calibration coefficient K≈0.156 mm / pixel is calculated. That is, one pixel represents an actual area of ​​approximately 0.024 mm².

[0078] In this embodiment, the key parameters of the data processing unit algorithm are configured as follows: the Gaussian filter kernel size is 5×5, the standard deviation σ = 1.0; the adaptive threshold window size is 31×31 pixels, and the constant term C = 2; according to the characteristics of this flow channel, the weight coefficients are set as α = 0.5, β = 0.3, and γ = 0.2.

[0079] The image processing unit executes the following closed-loop process: it reads the original RAW format image in real time, performs CLAHE histogram equalization, and eliminates bubble shadow interference. Then, the preprocessed image is input into the U-Net segmentation network, which outputs a binarized mask. Subsequently, feature calculations are performed: based on the spatial calibration coefficient (K=0.04 mm / pixel), the physical area A of the speckle is calculated; the shape factor (circularity C) and spatial distribution entropy S are also calculated.

[0080] The simulation process was set to increase the heat flux density linearly from 0.5 MW / m2 to the point where CHF occurs (approximately 1.2 MW / m2). Figure 2The schematic diagram of the spatiotemporal evolution of dry spots shows the monitoring snapshots of the system at four key moments: T1 = 2.0s: Only discrete small bubbles are present on the heating surface, the total area of ​​the detected dry spots is A≈5 mm2, LFSI = 0.92, and the system is judged to be "stable". T2 = 5.5s: Local dry spots begin to grow, the edges become rough, the LFSI drops to 0.65, and the system successfully triggers a level one warning. T3 = 8.2s: Dry spots appear in the center of the flow channel and merge, forming an irregular large dry spot with an area of ​​approximately 120 mm2. The roundness drops below 0.4, the LFSI drops sharply to 0.28, and the system immediately triggers a level two high-risk alarm. T = 8.5s (CHF occurrence moment): Simulation data shows that without intervention, the wall temperature will skyrocket at this moment.

[0081] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A critical heat flux density assessment and early warning method based on dynamic recognition of speckle images, characterized in that: Includes the following steps: S1, a high-speed camera system is used to acquire a sequence of images of the heating wall surface of the narrow rectangular flow channel in real time, and the heating wall surface images in the sequence are preprocessed; S2, perform dry spot region segmentation on the preprocessed heated wall image and extract the outer contour information of the dry spots; S3. Calculate the geometric feature parameters of the dry spots based on the outer contour information of the dry spots and the preset spatial resolution calibration coefficients. The geometric feature parameters include at least area, roundness, distribution density, and spatial distribution entropy. S4. Construct a liquid film stability index based on geometric feature parameters, and divide the liquid film stability region according to the liquid film stability index and the preset stability threshold. S5, construct a coupled prediction model of dry spot dynamic expansion and wall superheat, and generate the remaining time prediction value of the critical heat flux density. The coupled prediction model of dry spot dynamic expansion and wall superheat adopts a layered architecture, including a dry spot dynamic feature extraction layer, a coupling mapping layer and a time extrapolation prediction layer. S6. Based on the liquid film stability index, dry spot expansion rate, and remaining time prediction, a graded early warning mechanism is constructed and corresponding early warning signals are triggered.

2. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 1, characterized in that: In step S1, the preprocessing of the heated wall image includes: performing grayscale conversion, Gaussian filtering for noise reduction, and contrast-limited adaptive histogram equalization on the heated wall image in sequence.

3. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 2, characterized in that: Step S2 includes the following steps: S21, Perform preliminary speckle segmentation on the preprocessed heated wall image, and select an adaptive thresholding method or a deep learning network model for pixel-level separation based on speckle features to obtain a binarized heated wall image. S22, Morphological correction is performed on the binarized heated wall image. Morphological closing operation is used to fill the voids inside the dry spots and connect the broken edges to obtain the corrected binarized mask image. S23, the edge detection operator is used to scan the corrected binary mask image, extract the outer contour coordinate sequence of the dry spots, and the sub-pixel edge detection technology is used to interpolate and correct the contour coordinates of the dry spots.

4. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 3, characterized in that: In step S21, if the image signal-to-noise ratio is higher than the preset signal-to-noise ratio threshold, an adaptive thresholding method is used, and the neighborhood window size is dynamically set according to the channel width; if the image signal-to-noise ratio is lower than the preset signal-to-noise ratio threshold, a deep learning model is used for segmentation.

5. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 3, characterized in that: In step S3, geometric feature parameters are calculated using a connected component analysis algorithm based on the binarized mask image.

6. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 1, characterized in that: Step S4 includes the following steps: S41, normalize the geometric feature parameters to generate coverage index, morphology index and order index; S42, the liquid film stability index is calculated using a linear weighted model based on the coverage index, morphology index and order index; S43, the liquid film stability region is divided according to the liquid film stability index and the preset stability threshold.

7. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 6, characterized in that: In step S43, the preset stability thresholds include a first stability threshold and a second stability threshold. When the liquid film stability index is greater than or equal to the first stability threshold, it is defined as a stable region; when the liquid film stability index is less than the first stability threshold but greater than the second stability threshold, it is defined as a fluctuating region; when the liquid film stability index is less than or equal to the second stability threshold, it is defined as a high-risk region.

8. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 7, characterized in that: In step S5, the speckle dynamics feature extraction layer is used to extract kinematic feature parameters of speckles from a continuous sequence of heated wall images. The kinematic feature parameters include speckle propagation rate and propagation acceleration. The coupling mapping layer, based on the micro-liquid layer evaporation theory, establishes a nonlinear correlation to map the dry spot expansion rate into an estimated value of wall superheat. The time extrapolation prediction layer, based on the current motion state of the dry spot, uses a second-order approximation equation with Taylor expansion to construct a prediction model for the remaining time of the critical heat flux density and generates a predicted value for the remaining time of the critical heat flux density.

9. The critical heat flux density assessment and early warning method based on dynamic recognition of speckle images according to claim 8, characterized in that: In step S6, the tiered early warning mechanism includes: If the liquid film stability index remains in the fluctuation range for a preset number of consecutive frames, a Level 1 warning will be triggered. If the liquid film stability index is less than or equal to the second stability threshold or the instantaneous growth rate of the dry spot area exceeds the critical expansion rate, a level two warning is triggered. If the predicted time remaining before the critical heat flux density occurs is less than the preset safety response time threshold, the highest priority alarm will be triggered.

Images