Critical heat flux density prediction method based on micro-nano hierarchical structure and machine learning

Abstract

The invention belongs to the technical field of heat management of electronic equipment, and particularly relates to a critical heat flux density prediction method based on a micro-nano hierarchical structure and machine learning, which comprises the following steps: S1, providing a heat dissipation structure body with a micro-nano hierarchical structure surface for two-phase heat exchange with a cooling working medium; s2, temperature data and visual image data of the surface of the micro-nano hierarchical structure are collected in real time through a multi-mode monitoring module; and S3, the visual image data are processed through a data processing and predicting module, boiling critical heat flux is predicted through a machine learning algorithm, and the data processing and predicting module is in communication connection with the multi-mode monitoring module. According to the critical heat flux density prediction method based on the micro-nano hierarchical structure and machine learning, through the optimized micro-nano hierarchical structure, capillary force and liquid supplementing capacity are cooperatively enhanced, formation of dry spots is remarkably delayed from the physical level, and early and accurate prediction is improved and achieved.

Description

Technical Field

[0001] This invention relates to the field of thermal management technology for electronic devices, and in particular to a method for predicting critical heat flux density based on micro-nano hierarchical structures and machine learning. Background Technology

[0002] As integrated circuits and power devices evolve towards higher performance and smaller sizes, their heat flux density is increasing dramatically. Traditional single-phase heat dissipation methods, such as air cooling, are approaching their limits. Phase change heat transfer, especially boiling heat transfer, has become a key technology for addressing next-generation thermal challenges because it can utilize the latent heat of the working fluid and thus possesses extremely high heat dissipation capabilities. The performance upper limit of boiling heat transfer is determined by the critical heat flux. Decision, exceeding The heated surface will be covered by a vapor film, causing a rapid temperature rise and equipment failure. How can this be improved? And achieving accurate prediction is the current research focus.

[0003] In existing technologies, the heating surface is typically enhanced by constructing micron or nanostructures to increase nucleation sites, extend contact lines, and strengthen capillary forces, thereby improving… However, simple microstructures, in certain situations such as excessively small column spacing, have limited effect on improving capillary wicking capacity and may even inhibit liquid replenishment due to excessive flow resistance. Furthermore, the boiling process is highly random and nonlinear, with complex bubble dynamics, making traditional experience-based methods ineffective. The correlation-based approach has poor universality and cannot achieve real-time, accurate prediction. Therefore, there is an urgent need in this field for a method that can significantly improve... A comprehensive heat dissipation solution that can intelligently predict and avoid boiling over. Summary of the Invention

[0004] Based on the aforementioned technical problems, this invention proposes a critical heat flux density prediction method based on micro-nano hierarchical structures and machine learning.

[0005] This invention proposes a critical heat flux density prediction method based on micro / nano hierarchical structures and machine learning, comprising the following steps:

[0006] S1. Provide a heat dissipation structure body with a micro-nano hierarchical surface for two-phase heat exchange with the cooling working fluid.

[0007] S2. The temperature data and visual image data of the surface of the micro-nano hierarchical structure are collected in real time through the multimodal monitoring module.

[0008] S3. The visual image data is processed by the data processing and prediction module, and the boiling critical heat flux is predicted using a machine learning algorithm, wherein the physical descriptor of the boiling critical heat flux is denoted as... The data processing and prediction module is in communication connection with the multi-modal monitoring module.

[0009] S4, the signal receiving module is electrically connected with the prediction module, for receiving the early warning signal from the prediction module, and transmitting to the control circuit, and the control circuit executes the corresponding active control strategy according to the early warning signal.

[0010] Preferably, the micro-nano hierarchical structure surface comprises:

[0011] The micropillar array is arranged on a substrate, the diameter of the micropillar , the pitch , the height satisfies the preset diameter / pitch ratio , and between 0.1 and 0.5 to optimize the capillary wicking performance, which refers to the ability to absorb and conduct cooling working medium through capillary action, and the cooling working medium includes water and refrigerant.

[0012] The micropillar surface and the substrate plane are both covered with a nanostructure layer, the nanostructure layer is any one of zinc oxide nanorod, carbon nanotube or metal oxide nanowire, and the surface roughness of the nanostructure layer is .

[0013] Through the above technical solution, the micro-nano hierarchical structure is composed of a micropillar array and a nanostructure grown thereon, and the optimized diameter / pitch ratio is preferably < 0.3, such as =10μm, =50μm, forming a sparse arrangement, allowing the nanostructure to fully exert its capillary advantage, forming a leading nano-wicking front end, which refers to the capillary front provided by the nanostructure, capable of quickly transporting liquid, synergizing with the capillary wicking performance of the micropillar, enhancing the overall liquid replenishment speed, greatly enhancing the liquid replenishment speed between the micropillar arrays, and the nanostructure further increases the specific surface area and capillary force, endowing the surface with superhydrophilicity, synergistically enhancing the capillary force and liquid replenishment capacity from the physical level, and significantly delaying the formation of dry spots.

[0014] Preferably, the multi-modal monitoring module comprises:

[0015] An infrared thermal imager is arranged on one side of the transparent heating substrate for monitoring the temperature distribution and dry spot formation of the micro-nano hierarchical structure surface.

[0016] A high-speed camera is arranged on the same side of the transparent heating substrate as the infrared thermal imager for capturing bubble dynamics image sequences during the boiling process.

[0017] By the technical scheme, the multi-modal perception system of the bottom vision and infrared combination is constructed, the infrared thermal imager can accurately identify the formation of local dry spots and overheating points, the high-speed camera records the full dynamic process of nucleation, growth, merging and separation of the bubbles at a high frame rate of more than 1000 fps, and provides a complete and real-time high-quality data source for subsequent intelligent prediction based on image sequences.

[0018] Preferably, the transparent heating substrate is one of sapphire, quartz or transparent conductive oxide coated glass, in particular one of a sapphire heater, a quartz heater or a transparent conductive oxide coated glass heater.

[0019] By the technical scheme, the transparent heating substrate used has good thermal conductivity, heat resistance and optical transparency, ensuring that the infrared thermal imager and the high-speed camera can pass through the substrate without interference, and realizing high-precision and in-situ monitoring of the surface state of the micro-nano hierarchical structure.

[0020] Preferably, the data processing and prediction module performs the following operations:

[0021] Step one, main feature analysis is performed on the collected image sequence, wherein the physical descriptor of the main feature analysis is denoted as , and the first main features are extracted to reduce the data dimension;

[0022] Step two, fast Fourier transform is performed on the time sequence of the first main feature , and the bubble departure frequency and the dry area fraction are calculated as key physical descriptors for representing the boiling state, wherein the physical descriptor of the Fourier transform is denoted as .

[0023] Step three, the bidirectional long short-term memory neural network is used to predict the future change of the main features in the next milliseconds by taking the main feature time sequence in the previous milliseconds as input, and then reconstruct the predicted future bubble shape, wherein the physical descriptor of the bidirectional long short-term memory neural network is denoted as model.

[0024] By the technical scheme, the module is the intelligent core of the system, which first uses unsupervised machine learning to reduce the dimension and extract features from the image sequence of the million-pixel level, and simplifies the complex visual data into the first main features, wherein the first main feature effectively captures the footprint distribution of the bubble image, and then Analysis, extract the key descriptors with clear physical meaning, namely bubble departure frequency and dryout area fraction , finally, using network learning long-term dependence of main feature time series, realize the prediction of future bubble dynamics evolution, this method overcomes the problem of inaccurate identification in high temperature difference area of traditional image segmentation method, realizes the deep understanding and advanced prediction of boiling state.

[0025] Preferably, in the data processing and prediction module, the number of the first 10 to 100 main features extracted by the main feature analysis.

[0026] Through the above technical solutions, the data dimension is effectively reduced from the million pixel level to dozens to hundreds of main features, while retaining the key dynamic information of the boiling process, greatly reducing the calculation complexity, and ensuring the real-time performance of the prediction algorithm.

[0027] Preferably, in the data processing and prediction module, when the dryout area fraction sharp rise, judged as precursor, and trigger the warning signal, then use model prediction results, adjust the heating power or cooling flow in advance.

[0028] Through the above technical solutions, when the system approaches , will increase significantly, by monitoring the sharp change of the two key descriptors in real time, such as from 0.5 to 0.6 or more, this rule is verified by many experiments, which can realize the early and accurate identification of , and reserves valuable time window for active control.

[0029] When the number of dry spots around the bubble reaches the critical limit, thereby inhibiting the liquid replenishment, it will trigger , which will cause the surface temperature to rise, thereby further increasing the nucleation point, and then trigger dryout, the critical heat flux is calculated by the liquid flow driven by capillary action and the liquid evaporation amount at the dry spot:

[0030]

[0031] Critical heat flux; C0-constant function determined by experimental conditions, latent heat of vaporization, surface tension of fluid, liquid film thickness, dry spot expansion speed, dry spot area fraction.

[0032] Preferably, the method for two-phase heat exchange with the cooling working fluid described in S1 includes the following steps:

[0033] S11. Prepare a heat dissipation surface with a specific micro-nano hierarchical structure.

[0034] S12. Start the system to cause the cooling medium to boil on the heat dissipation surface.

[0035] S13. The surface temperature field and bubble image sequence are collected in real time through the multimodal monitoring module.

[0036] S14. The data processing and prediction module processes image data and extracts the bubble detachment frequency. and the area fraction of the arid region .

[0037] S15, based on The model predicts the bubble dynamics behavior at future time steps.

[0038] S16. Based on the changing trend of the physical descriptor and the prediction results, determine whether it is close to the boiling critical heat flux. If it is close to the boiling critical heat flux, execute one of the early warning or active control strategies. If it is not close to the boiling critical heat flux, return to step S13 and continue monitoring.

[0039] The above technical solution provides a complete, closed-loop intelligent heat dissipation workflow. From surface preparation, condition monitoring, data processing to prediction and control, it seamlessly integrates advanced surface engineering, multimodal sensing, and artificial intelligence algorithms to form a comprehensive solution that can proactively predict and avoid boiling over, significantly improving the reliability and safety of heat dissipation for high heat density equipment.

[0040] Preferably, in the surface of the micro / nano hierarchical structure, the diameter / spacing ratio of the micropillars is... <0.3, and the height of the micrometer column >20μm.

[0041] Through the above technical solution, this optimized parameter combination ensures that the nano wicking front end always leads the bulk wicking front end, achieving the best liquid transport enhancement effect, which is the key to the synergistic advantages of micro-nano hierarchical structures.

[0042] Preferably, the active control strategy in step S16 includes one or more of the following: reducing heating power, activating auxiliary cooling, or issuing an audible and visual alarm.

[0043] The above technical solution provides a tiered and flexible active control method, which can be used to predict... To mitigate risks, the system can automatically take one or more combined measures, such as reducing power to prevent a crisis, activating auxiliary cooling to enhance heat dissipation, or issuing alarms to remind users to intervene, which greatly enhances the system's proactive safety protection capabilities.

[0044] The beneficial effects of this invention are as follows:

[0045] 1. Through optimized micro / nano hierarchical structures, capillary force and liquid replenishment capabilities are synergistically enhanced, significantly delaying dry spot formation at a physical level and improving performance. Innovatively employing unsupervised machine learning Extract descriptors with clear physical meaning from the image: bubble detachment frequency. and the area fraction of the arid region This overcomes the problem of traditional image segmentation methods failing to accurately identify bubbles in areas of high temperature difference, and achieves [the ability to accurately identify bubbles]. Early and accurate predictions.

[0046] 2. By setting The model can not only analyze the current state but also predict future bubble dynamics evolution, providing a valuable time window for active control. The seamless integration of surface engineering, advanced sensing, and artificial intelligence algorithms forms a complete, intelligent, and highly reliable heat dissipation solution. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of a critical heat flux density prediction method based on micro-nano hierarchical structures and machine learning proposed in this invention.

[0048] Figure 2 This is a scanning electron microscope image of the surface of a micro-nano hierarchical structure, which is proposed in this invention as a critical heat flux density prediction method based on micro-nano hierarchical structure and machine learning.

[0049] Figure 3 This is a schematic diagram of boiling bubble image processing for a critical heat flux density prediction method based on micro-nano hierarchical structure and machine learning proposed in this invention.

[0050] Figure 4 This invention proposes a critical heat flux density prediction method based on micro / nano hierarchical structures and machine learning. Model prediction A flowchart.

[0051] In the diagram: 1. Sapphire heater; 2. Micro-nano hierarchical surface structure; 3. Infrared thermal imager; 4. High-speed camera; 5. Data processing and prediction module; 6. Control circuit; 7. Signal receiving module; 8. Model. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0053] Reference Figures 1-4 A critical heat flux density prediction method based on micro / nano hierarchical structures and machine learning includes the following steps:

[0054] S1. Provide a heat dissipation structure body with a micro-nano hierarchical structure surface 2 for two-phase heat exchange with the cooling working fluid.

[0055] S2. The temperature data and visual image data of the surface 2 of the micro-nano hierarchical structure are collected in real time through the multimodal monitoring module.

[0056] S3. The visual image data is processed by the data processing and prediction module 5, and the boiling critical heat flux is predicted using a machine learning algorithm. The physical descriptor of the boiling critical heat flux is denoted as... The data processing and prediction module 5 is connected to the multimodal monitoring module.

[0057] S4. The signal receiving module 7 is electrically connected to the prediction module 5 to receive the early warning signal from the prediction module 5 and transmit it to the control circuit 6. The control circuit 6 executes the corresponding active control strategy according to the early warning signal.

[0058] Micro-nano hierarchical structure surface 2 includes:

[0059] The array consists of micropillars arranged on a substrate, with the diameter of the micropillars being... ,spacing ,high Meets the preset diameter / spacing ratio ,and The value should be between 0.1 and 0.5 to optimize capillary wicking performance, which refers to the ability to absorb and conduct cooling fluids through capillary action. Cooling fluids include water and refrigerants.

[0060] Both the surface of the micron-sized pillars and the substrate plane are covered with a nanostructure layer, which can be any one of zinc oxide nanorods, carbon nanotubes, or metal oxide nanowires. The surface roughness of the nanostructure layer is... This micro / nano hierarchical structure consists of a micrometer-scale column array and nanostructures grown on it, with an optimized diameter / spacing ratio. < 0.3, such as =10μm, =50μm, forming a sparse arrangement, allowing the nanostructure to fully exert its capillary advantages and form a leading nano-wicking front. The nano-wicking front refers to the capillary action front provided by the nanostructure, which can rapidly transport liquid. In synergy with the capillary wicking performance of the microstructure, it enhances the overall liquid replenishment rate and greatly enhances the liquid replenishment rate between the micropillars. The nanostructure further increases the specific surface area and capillary force, endows the surface with superhydrophilicity, and synergistically enhances the capillary force and liquid replenishment ability from a physical level, significantly delaying the formation of dry spots.

[0061] The multimodal monitoring module includes:

[0062] Infrared thermal imager 3, configured on one side of the transparent heating substrate, is used to monitor the temperature distribution and dry spot formation on the surface 2 of the micro-nano hierarchical structure.

[0063] A high-speed camera 4, configured on the transparent heating substrate and located on the same side as the infrared thermal imager 3, is used to acquire image sequences of bubble dynamics during the boiling process, thus constructing a multimodal perception system that combines bottom vision and infrared. The infrared thermal imager 3 can accurately identify the formation of local dry spots and hot spots. The high-speed camera 4 records the entire dynamic process of bubble nucleation, growth, merging, and detachment at a high frame rate of more than 1000fps, providing a complete, real-time, high-quality data source for subsequent intelligent prediction based on image sequences.

[0064] The transparent heating substrate is one of sapphire, quartz, or transparent conductive oxide coated glass, specifically one of sapphire heater 1, quartz heater, or transparent conductive oxide coated glass heater. The selected transparent heating substrate has good thermal conductivity, heat resistance, and optical transparency, ensuring that the infrared thermal imager 3 and high-speed camera 4 can pass through the substrate without interference, and realize high-precision, in-situ monitoring of the state of the micro-nano hierarchical structure surface 2.

[0065] Data processing and prediction module 5 performs the following operations:

[0066] Step 1: Perform principal feature analysis on the acquired image sequence, where the physical descriptor for principal feature analysis is denoted as... Before extraction Individual main features To reduce data dimensionality;

[0067] Step 2: For the first primary feature Perform a Fast Fourier Transform on the time series to calculate its bubble detachment frequency. and the area fraction of the arid region As a key physical descriptor characterizing the boiling state, the physical descriptor of the Fourier transform is denoted as... .

[0068] Step 3: Utilize a bidirectional long short-term memory neural network, previously... Using millisecond-level main feature time series as input, predict the future. The main feature changes in milliseconds are used to reconstruct the predicted future bubble shape, where the physical descriptor of the bidirectional long short-term memory neural network is denoted as... Model 8, this module is the intelligent core of the system, and it first employs unsupervised machine learning. Dimensionality reduction and feature extraction are performed on megapixel-level image sequences to simplify complex visual data. There are 1 main features, of which the first main feature is The footprint distribution of the bubble image was effectively captured, and then analyzed through the PC1 time series... Analysis revealed key descriptors with clear physical meaning, namely the bubble detachment frequency. and the area fraction of the arid region Finally, using Model 8 network learns the long-term dependencies of the main feature time series to predict the future evolution of bubble dynamics. This method overcomes the problem of inaccurate identification of high temperature difference regions by traditional image segmentation methods and achieves a deep understanding and advanced prediction of boiling state.

[0069] In the data processing and prediction module 5, the main feature analysis extracts the first... The number of principal features ranges from 10 to 100, effectively reducing the data dimension from millions of pixels to tens to hundreds of principal features. While retaining key dynamic information of the boiling process, this significantly reduces computational complexity and ensures the real-time performance of the prediction algorithm.

[0070] In the data processing and prediction module 5, when the area fraction of the arid region is monitored... A sharp increase, judged as Precursors, triggering warning signals, and then utilizing Based on the predictions of Model 8, heating power or cooling flow rate can be adjusted in advance when the system approaches... hour, There will be a significant surge, which can be observed by monitoring the sharp changes in these two key descriptors in real time, such as... A rapid increase from 0.5 to over 0.6 can achieve [the following]: Early and accurate identification allows for a valuable time window for proactive control.

[0071] When the number of dry spots around the bubble reaches a critical limit, thus inhibiting liquid replenishment, it triggers... This will cause the surface temperature to rise, thereby increasing the number of nucleation sites and triggering drying. The critical heat flux density is calculated using the liquid flow rate driven by capillary action and the amount of liquid evaporation at the dry spot:

[0072]

[0073] - Critical heat flux density; C0 - A constant function determined by experimental conditions. -Latent heat of vaporization - Surface tension of the fluid, -Liquid film thickness, - Dry spot expansion rate, - Dry spot area fraction.

[0074] The method for two-phase heat exchange with the cooling working fluid in S1 includes the following steps:

[0075] S11. Prepare a heat dissipation surface with a specific micro-nano hierarchical structure.

[0076] S12. Start the system to cause the cooling medium to boil on the heat dissipation surface.

[0077] S13. The surface temperature field and bubble image sequence are collected in real time through the multimodal monitoring module.

[0078] S14, Data Processing and Prediction Module 5 processes image data and extracts bubble detachment frequency. and the area fraction of the arid region .

[0079] S15, based on Model 8 predicts the bubble dynamics behavior at future time steps.

[0080] S16. Based on the changing trend of the physical descriptor and the prediction results, determine whether it is close to the boiling critical heat flux. If it is close to the boiling critical heat flux, execute one of the early warning or active control strategies. If it is not close to the boiling critical heat flux, return to step S13 and continue monitoring. It provides a complete and closed-loop intelligent heat dissipation workflow. From surface preparation, state monitoring, data processing to prediction and control, it seamlessly integrates advanced surface engineering, multimodal sensing and artificial intelligence algorithms to form a comprehensive solution that can actively predict and avoid boiling crises, significantly improving the reliability and safety of heat dissipation for high heat density equipment.

[0081] In the micro / nano hierarchical surface 2, the diameter / spacing ratio of the micron pillars <0.3, and the height of the micrometer column With a diameter of >20μm, this optimized parameter combination ensures that the nano wicking front always leads the bulk wicking front, achieving the best liquid transport enhancement effect, which is the key to the synergistic advantages of micro-nano hierarchical structures.

[0082] The active control strategy in step S16 includes one or more of the following: reducing heating power, activating auxiliary cooling, or issuing audible and visual alarms. This provides a tiered and flexible active control mechanism, which can be activated once a warning is issued. To mitigate risks, the system can automatically take one or more combined measures, such as reducing power to prevent a crisis, activating auxiliary cooling to enhance heat dissipation, or issuing alarms to remind users to intervene, which greatly enhances the system's proactive safety protection capabilities.

[0083] Through optimized micro / nano hierarchical structures, capillary forces and liquid replenishment capabilities are synergistically enhanced, significantly delaying dry spot formation at a physical level and improving performance. Innovatively employing unsupervised machine learning Extract descriptors with clear physical meaning from the image: bubble detachment frequency. and the area fraction of the arid region This overcomes the problem of traditional image segmentation methods failing to accurately identify bubbles in areas of high temperature difference, and achieves [the ability to accurately identify bubbles]. Early and accurate predictions.

[0084] Working principle: The system operation process of this invention is as follows: Cooling water wets the heat dissipation surface, heating begins, the water begins to boil, high-speed camera 4 acquires images at 2000 fps, and transmits them to data processing and prediction module 5. Data processing and prediction module 5 processes the image sequence. Processing and calculating the real-time area fraction of the dried-up region. and bubble detachment frequency Meanwhile, pre-trained with historical data Model 8 is based on the first 200ms Sequence, predicting the next 60ms When the heat flux rises to near At that time, the data processing and prediction module 5 detected... The voltage rapidly increased from 0.5 to above 0.6, prompting the data processing and prediction module 5 to immediately issue a warning signal and reduce the heating power by 20% via the control circuit 6, successfully preventing further damage. After the occurrence of this, the surface temperature returned to normal.

[0085] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting critical heat flux density based on micro / nano hierarchical structures and machine learning, characterized in that: The method comprises the following steps: S1, providing a heat dissipation structure body with a micro-nano hierarchical structure surface (2) for two-phase heat exchange with a cooling working medium; S2, collecting temperature data and visual image data of the micro-nano hierarchical structure surface (2) in real time through a multi-modal monitoring module; S3, processing the visual image data by a data processing and prediction module (5) and predicting the boiling critical heat flux using a machine learning algorithm, wherein the physical descriptor of the boiling critical heat flux is denoted by , the data processing and prediction module (5) being communicatively connected with the multi-modal monitoring module; S4, the signal receiving module (7) is electrically connected with the prediction module (5), used for receiving a warning signal from the prediction module (5) and transmitting to the control circuit (6), and the control circuit (6) executes a corresponding active control strategy according to the warning signal.

2. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 1, characterized in that: The micro-nano hierarchical structure surface (2) comprises; microposts on a substrate, the microposts having diameters , pitches , heights satisfying a predetermined diameter / pitch ratio , and between 0.1 and 0.5 to optimize capillary wicking performance.

3. The micropost surface and the substrate plane are both covered with a nanostructured layer, the nanostructured layer being any one of zinc oxide nanorods, carbon nanotubes or metal oxide nanowires, the nanostructured layer having a surface roughness .

4. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 2, characterized in that: The multi-modal monitoring module comprises; The infrared thermal imager (3) is arranged on one side of the transparent heating substrate and is used for monitoring the temperature distribution and dry spot formation of the micro-nano hierarchical structure surface (2); The high-speed camera (4) is arranged on the same side of the transparent heating substrate as the infrared thermal imager (3) and is used for collecting bubble dynamics image sequences in the boiling process.

5. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 3, characterized in that: The transparent heating substrate is one of sapphire, quartz or transparent conductive oxide coated glass.

6. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 4, characterized in that: The data processing and prediction module (5) performs the following operations; Step one, using U-Net segmentation network to preprocess the bubble image, accurately identify the bubble boundary and dry area, and perform main feature analysis on the collected image sequence, wherein the physical descriptor of the main feature analysis is denoted as , and the first main features are extracted to reduce the data dimension; Step two, Fast Fourier Transform of the time series of the first principal feature to calculate its bubble departure frequency and dryout area fraction as key physical descriptors of the boiling regime, where the Fourier transformed physical descriptors are denoted by ; Step 3: Utilize a bidirectional long short-term memory neural network, previously... Using millisecond-level main feature time series as input, predict the future. The main feature changes in milliseconds are used to reconstruct the predicted future bubble shape, where the physical descriptor of the bidirectional long short-term memory neural network is denoted as... Model (8).

7. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 5, characterized in that: In the data processing and prediction module (5), the main feature analysis extracts the first... The number of principal features ranges from 10 to 100.

8. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 6, characterized in that: The data processing and prediction module (5) monitors the area fraction of the dryout region Sharp rise, determined as The precursor, and trigger an early warning signal, and then use Model (8) to predict the results, adjust the heating power or cooling flow in advance.

9. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 7, characterized in that: The method for two-phase heat exchange with a cooling working medium in S1 comprises the following steps: S11, preparing a heat dissipation surface with a specific micro-nano hierarchical structure; S12, starting the system to make the cooling working medium boil on the heat dissipation surface; S13, collecting the temperature field and bubble image sequence of the surface in real time through a multi-modal monitoring module; S14, the data processing and prediction module (5) processes the image data and extracts the bubble detachment frequency and the area fraction of dry regions ; S15, based on The model (8) predicts the bubble dynamics behavior at future time steps; S16, judging whether the boiling critical heat flux is approached according to the change trend of the physical descriptor and the prediction result, and when the boiling critical heat flux is approached, one of the warning or active control strategy is executed, and if the boiling critical heat flux is not approached, the step S13 is returned to continue monitoring.

10. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 8, characterized in that: In the micro-nano hierarchical structure surface (2), the diameter / pitch ratio of the micropillars is <0.3, and the height of the micropillars > 20 pm.

11. The micro-nano hierarchical structure and machine learning based critical heat flow density prediction method according to claim 9, characterized in that: The active control strategy in step S16 comprises one or more than two of reducing the heating power, starting auxiliary cooling or issuing an audible and visual alarm.

Images