Acoustic coalescence elimination method for hydrogen bubbles in molten aluminum

Abstract

The invention provides an acoustic coalescence elimination method for hydrogen bubbles in molten aluminum, and belongs to the technical field of molten aluminum bubble elimination. According to the method, sound field collaborative focusing is achieved by arranging a phase control array transducer set in a molten aluminum smelting furnace, primary gathered sound waves and efficient coalescence sound waves are applied in stages according to bubble size evolution in a double-frequency scanning mode, and the hydrogen bubbles in the molten aluminum are eliminated. A target frequency is dynamically adjusted in combination with a bubble resonance frequency optimization model to match the inherent frequency of a bubble, a sound-bubble coupling dynamic tracking system is used for monitoring bubble evolution triggering stage switching in real time, and a sound field focus point dynamically scans and covers a three-dimensional space area through phase delay control. A delay detection method is adopted to eliminate cavitation bubble interference and accurately measure the hydrogen content, sound intensity and frequency parameters are optimized according to a sound field parameter self-adaptive regulation and control model, and the technical problem that hydrogen bubbles in a molten aluminum deep area are difficult to be effectively driven by sound waves and cannot be fully coalesced and eliminated is solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of liquid aluminum bubble removal, and particularly relates to an acoustic coalescence removal method for hydrogen bubbles in liquid aluminum. BACKGROUND

[0002] In the process of removing hydrogen from liquid aluminum, ultrasonic acoustic radiation force is often used to drive hydrogen bubbles to coalesce and remove. The traditional method uses a single or a small number of transducers to apply a fixed frequency acoustic wave to the surface layer of the liquid aluminum, and uses acoustic radiation force to make hydrogen bubbles gather at the acoustic pressure node and float out of the liquid surface by colliding and merging into large bubbles. However, high-temperature liquid aluminum has a strong attenuation effect on acoustic waves, and the effective action distance of a single transducer is usually only about 150 millimeters, resulting in insufficient acoustic pressure intensity in the deep area of the liquid aluminum smelting furnace, and the micro hydrogen bubbles cannot obtain sufficient driving force for effective aggregation. At the same time, the fixed frequency acoustic wave cannot adapt to the resonance characteristics of bubbles of different sizes, and the response of small bubbles in the early stage and large bubbles in the later stage to the acoustic wave frequency is significantly different, and a single frequency cannot meet the coalescence needs in different stages. That is, the prior art has the technical problem that hydrogen bubbles in the deep area of the liquid aluminum cannot be effectively driven by acoustic waves and cannot be fully coalesced and removed. SUMMARY

[0003] Therefore, the application provides an acoustic coalescence removal method for hydrogen bubbles in liquid aluminum, which can solve the technical problem that hydrogen bubbles in the deep area of the liquid aluminum cannot be effectively driven by acoustic waves and cannot be fully coalesced and removed in the prior art.

[0004] The application is implemented in the following manner: The application provides an acoustic coalescence removal method for hydrogen bubbles in liquid aluminum, which arranges a phased array transducer group in the vertical direction inside a liquid aluminum smelting furnace, starts the phased array transducer group to enter a double-frequency scanning mode, applies a primary aggregation acoustic wave in the first stage, applies a high-efficiency coalescence acoustic wave according to a target frequency calculated by a bubble resonance frequency optimization model in the second stage, monitors the spatial distribution and size evolution of the bubble group in real time during the action of the acoustic wave by using an acoustic-bubble coupling dynamics tracking system, triggers a stage switching signal when the average diameter of the bubbles reaches a preset value, realizes dynamic scanning of the acoustic field focusing point by phase delay control of the phased array transducer group, measures the hydrogen content of the liquid aluminum during the intermittent period of the action of the acoustic wave by using a delay detection method, outputs the acoustic intensity and frequency parameters of the next period by an acoustic field parameter self-adaptive regulation model according to the hydrogen content detection result of the liquid aluminum, and ends the degassing process when the hydrogen content of the liquid aluminum is lower than a threshold value.

[0005] The phased array transducer group contains 6 to 12 transducer units, the spacing between adjacent transducer units is 150 to 250 mm, and the working frequency of each transducer unit is 20 to 40 kHz.

[0006] Among them, the phased array transducer group extends the range of sound waves through coordinated focusing. The sound waves emitted by multiple transducer units are superimposed in space to form an interference field. By controlling the phase difference of each transducer unit, the sound waves generate constructive interference at a predetermined position to form a high sound intensity region.

[0007] In the first stage, a primary focusing sound wave with a frequency of 30 to 50 kHz is applied for a duration of 120 to 180 s, and in the second stage, a high-efficiency coalescing sound wave is applied for a duration of 90 to 150 s.

[0008] In this process, the primary focusing sound wave drives tiny hydrogen bubbles with a diameter of less than 50 μm to generate sound radiation force, causing the tiny hydrogen bubbles to gather towards the sound pressure node or antinode. When the tiny hydrogen bubbles approach each other and collide, they merge into larger bubbles due to surface tension.

[0009] When the target frequency matches the natural frequency of the larger bubble, the larger bubble generates strong radial oscillations. The oscillating larger bubbles attract each other through the Bjerknes force, accelerating the coalescence process.

[0010] Among them, the bubble resonance frequency optimization model adopts a spiral progressive network structure. The output of the nth layer is not only transmitted to the n+1th layer, but also transmitted to the n+2th and n+3th layers through jump connections, forming a spiral information flow path.

[0011] The input layer receives a feature vector consisting of four parameters: average bubble diameter, molten aluminum temperature, hydrogen partial pressure, and sound pressure amplitude. The first hidden layer contains 32 neurons, and the number of neurons in the second to fifth hidden layers are 64, 128, 256, and 128, respectively. The output layer generates the optimized target frequency value.

[0012] Among them, the training of the bubble resonance frequency optimization model uses mean square error as the loss function to measure the deviation between the predicted target frequency and the experimental optimal resonance frequency. The Adam optimization algorithm is used to update the network weights, the initial learning rate is set to 0.001 and an exponential decay strategy is adopted.

[0013] Among them, the acoustic-bubble coupling dynamics tracking system discretizes the acoustic field space of molten aluminum into finite element meshes with side lengths of 10 to 15 mm. Each finite element mesh node stores the acoustic pressure amplitude and phase information generated by the phased array transducer group, and the bubble group is represented by the Lagrange particle method.

[0014] When the average diameter of the bubble reaches 100 to 150 μm, a stage switching signal is triggered. The acoustic-bubble coupling dynamics tracking system counts the instantaneous radius distribution of all bubble particles in the current aluminum liquid every 10 seconds and calculates the number-weighted average diameter of the bubble group.

[0015] Among them, the phase delay control sets different initial phase angles for each transducer unit by the control unit, and makes the sound field focus point move along the preset scanning path in three-dimensional space by continuously changing the value of the sound beam deflection angle.

[0016] The scanning path covers a three-dimensional spatial area 500 to 800 mm from the transducer surface inside the aluminum molten metal furnace, and the scanning path adopts a spiral upward trajectory to cover the entire effective volume of the aluminum molten metal furnace.

[0017] Among them, the delay time of the delayed detection method is 15 to 25 seconds, the cavitation bubbles collapse rapidly due to the internal and external pressure difference after the sound wave stops, the collapse time of the cavitation bubbles is 5 to 10 seconds, and the hydrogen bubbles can maintain a stable state for more than 60 seconds after the sound wave stops.

[0018] Among them, when the rate of decrease of hydrogen content in liquid aluminum is greater than While maintaining a constant sound intensity, when the rate of decrease in hydrogen content of the molten aluminum is less than The sound intensity will be increased by 10 to 15%.

[0019] Among them, the hydrogen content in the molten aluminum is below the threshold. The degassing process ends when the hydrogen content in the molten aluminum is below a certain level. And while it is still decreasing, the frequency will be finely adjusted to the target frequency value by ±2 to 3 kHz.

[0020] This invention employs a phased array transducer group to extend the effective range of sound waves through a cooperative focusing mechanism. Sound waves emitted by multiple transducer units superimpose in space to form an interference field. By controlling the phase difference of each transducer unit, constructive interference occurs at predetermined positions, creating a high-intensity acoustic region. This extends the effective range from 150 mm to 800 mm, overcoming the strong attenuation effect of sound waves in high-temperature molten aluminum. A dual-frequency scanning mode dynamically adjusts the sound wave frequency according to the bubble size evolution. In the first stage, primary focusing sound waves drive tiny hydrogen bubbles to aggregate into larger bubbles. In the second stage, the frequency is switched to the target frequency calculated by the bubble resonance frequency optimization model, causing the larger bubbles to generate strong radial oscillations and accelerating the aggregation process through Bjerknes force. In summary, this invention solves the technical problem mentioned in the background art where hydrogen bubbles in the deep regions of molten aluminum are difficult to be effectively driven by sound waves, thus hindering their aggregation and elimination. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a finite element simulation cloud map of the sound pressure distribution of the longitudinal section phased array acoustic field inside an aluminum molten metal furnace.

[0023] Figure 3 This is a graph showing the evolution of the average bubble diameter over time in dual-frequency scanning mode.

[0024] Figure 4 This diagram illustrates the dynamic process of larger bubbles coalescing under the influence of acoustic waves through the Bjerknes force, demonstrating efficient coalescence.

[0025] Figure 5 This is a historical curve showing the decrease in hydrogen content in molten aluminum over cumulative time during the complete degassing process. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0027] like Figure 1 The diagram shows a flowchart of an acoustic coalescence and removal method for hydrogen bubbles in molten aluminum provided by this invention. The method includes the following steps:

[0028] S01. A phased array transducer group is arranged vertically inside the aluminum melt smelting furnace. The phased array transducer group contains 6 to 12 transducer units. The spacing between adjacent transducer units is 150 to 250 mm. The operating frequency of each transducer unit is 20 to 40 kHz.

[0029] S02. Start the phased array transducer group to enter the dual-frequency scanning mode. In the first stage, apply a primary focusing acoustic wave with a frequency of 30 to 50 kHz for a duration of 120 to 180 s. In the second stage, apply a high-efficiency coalescing acoustic wave with a duration of 90 to 150 s based on the target frequency calculated by the bubble resonance frequency optimization model.

[0030] S03. During the acoustic wave action, an acoustic-bubble coupling dynamic tracking system is used to monitor the spatial distribution and size evolution of the bubble group in real time. When the average diameter of the bubble reaches 100 to 150 μm, a stage switching signal is triggered.

[0031] S04. Dynamic scanning of the sound field focal point is achieved by controlling the phase delay of the phased array transducer group. The scanning path covers a three-dimensional spatial area 500 to 800 mm away from the transducer surface inside the aluminum molten metal furnace.

[0032] S05. During the interval of the sound wave action, the hydrogen content of the aluminum liquid is measured by a delayed detection method. The delay time of the delayed detection method is 15 to 25 seconds. The accurate measurement is achieved by utilizing the time difference between the collapse of cavitation bubbles and hydrogen bubbles.

[0033] S06. Based on the detection results of hydrogen content in molten aluminum, the sound field parameter adaptive control model outputs the sound intensity and frequency parameters for the next cycle. The degassing process ends when the hydrogen content in molten aluminum is lower than the threshold of 0.15 mL / 100 g.

[0034] The principle of extending the effective range of sound waves through the phased array transducer group is as follows: the sound waves emitted by multiple transducer units are superimposed in space to form an interference field. By controlling the phase difference of each transducer unit, the sound waves generate constructive interference at a predetermined position to form a high sound intensity region. The focusing mechanism extends the effective range of a single transducer unit from 150mm to 800mm, overcoming the strong attenuation effect of sound waves in high-temperature aluminum liquid.

[0035] The technical principle of the dual-frequency scanning mode is as follows: In the first stage, the primary focusing acoustic wave with a wavelength of 30 to 50 kHz has a long wavelength and strong penetration ability, which can drive tiny hydrogen bubbles with a diameter of less than 50 μm to generate acoustic radiation force, causing the tiny hydrogen bubbles to gather towards the sound pressure node or antinode. When the tiny hydrogen bubbles approach each other and collide, they merge into larger bubbles with a diameter of 100 to 150 μm due to surface tension. In the second stage, the system switches to the target frequency calculated by the bubble resonance frequency optimization model. When the target frequency matches the natural frequency of the larger bubbles, the larger bubbles generate strong radial oscillations. The oscillating larger bubbles attract each other through the Bjerknes force, accelerating the merging process. The merged millimeter-sized bubbles rise rapidly and are discharged from the molten aluminum because the buoyancy is much greater than the acoustic radiation force.

[0036] The bubble resonance frequency optimization model adopts a spiral progressive network structure, expanding the receptive field layer by layer through a spiral path, achieving a balance between global feature capture and local detail preservation in the fifth layer. The specific structure of the bubble resonance frequency optimization model is as follows: the input layer receives a feature vector composed of four parameters: average bubble diameter, aluminum liquid temperature, hydrogen partial pressure, and sound pressure amplitude. The first hidden layer contains 32 neurons to extract primary features. The second to fifth hidden layers use a spiral connection method, meaning the output of the nth layer is not only transmitted to the n+1th layer but also to the n+2th and n+3th layers through skip connections, forming a spiral information flow path. The number of neurons in each layer is 64, 128, 256, and 128 respectively. The sixth hidden layer converges and fuses multi-scale features, and the output layer generates the optimized target frequency value. The steps for establishing the training dataset for the bubble resonance frequency optimization model specifically include: applying a series of sound waves with frequencies from 20kHz to 100kHz at 5kHz intervals to bubbles of different diameters under laboratory conditions; recording the oscillation amplitude and coalescence rate of the bubbles using a high-speed camera system; marking the frequency that maximizes the coalescence rate as the optimal resonance frequency for the bubble size; and collecting data covering average bubble diameters from 50μm to 500μm, aluminum melt temperatures from 680℃ to 750℃, hydrogen partial pressures from 50Pa to 500Pa, and sound pressure amplitudes from 1... Up to 10 The training set consists of 12,000 sets of experimental data. The specific steps for training the bubble resonance frequency optimization model include: using mean squared error as the loss function to measure the deviation between the predicted target frequency and the experimental optimal resonance frequency; using the Adam optimization algorithm to update the network weights; setting the initial learning rate to 0.001 and adopting an exponential decay strategy; using a batch size of 64; performing 5,000 training iterations; and stopping training when the validation set loss no longer decreases for 100 consecutive iterations.

[0037] The core advantage of the spiral progressive network structure lies in its unique information propagation mechanism. Traditional feedforward networks, where information is only transmitted between adjacent layers, are prone to gradient vanishing and feature degradation problems in deep networks. Spiral connections, however, allow local features extracted from shallow layers to be directly transmitted across intermediate layers to deeper layers, where they are fused with global features. This mechanism enables the bubble resonance frequency optimization model to capture both the macroscopic influence of aluminum melt temperature and sound pressure amplitude on the overall sound field environment, and the microscopic effects of average bubble diameter and hydrogen partial pressure on local bubble dynamics, while learning the nonlinear mapping relationship between bubble resonance frequency and multiple parameters. This avoids losing crucial scale details during deep abstraction. The equilibrium point achieved by the spiral path at the fifth layer gives the bubble resonance frequency optimization model strong generalization ability for bubble size distribution under different working conditions. When multiple scales of bubbles exist simultaneously in the molten aluminum, the model can comprehensively evaluate the response characteristics of bubbles at each scale and output a compromise frequency that balances the initial aggregation of small bubbles and the efficient removal of large bubbles, significantly improving the overall efficiency and robustness of the degassing process.

[0038] The acoustic-bubble coupling dynamics tracking system is implemented based on a multi-scale bubble evolution tracking algorithm of acoustic-bubble coupling dynamics. This algorithm discretizes the acoustic field of molten aluminum into a finite element mesh with side lengths of 10 to 15 mm. Each finite element mesh node stores the acoustic pressure amplitude and phase information generated by the phased array transducer group. The bubble swarm is represented using the Lagrange particle method, with each bubble particle carrying four dynamic attributes: three-dimensional spatial coordinates, instantaneous radius, radial oscillation velocity, and hydrogen content. The multi-scale bubble evolution tracking algorithm of acoustic-bubble coupling dynamics calculates the change in bubble radius over time based on the dynamic equation describing the radial oscillation behavior of bubbles under acoustic pressure. This dynamic equation comprehensively considers the coupling effects of liquid inertia, surface tension, viscous damping, and acoustic pressure excitation. The algorithm also introduces the Bjerknes force model to describe the interaction between bubbles, calculating the attractive or repulsive forces between any two bubbles caused by their respective secondary acoustic fields generated by oscillation. When the bubble oscillation phases are the same, an attractive force is generated, promoting coalescence. When the distance between the centers of two bubbles is less than 1.2 times the sum of their radii, the merging criterion is triggered, and the two bubbles merge into a new bubble. The radius of the new bubble is determined by the principle of volume conservation, and the three-dimensional spatial coordinates of the new bubble are taken from the centroids of the original two bubbles. The multi-scale bubble evolution tracking algorithm of the acoustic-bubble coupled dynamics uses the fourth-order Runge-Kutta method combined with an adaptive time step strategy to solve the differential equations of bubble motion. The time step is dynamically adjusted from 0.1 to 1.0 μs according to the bubble oscillation frequency, and tracks the complete evolution process from a hydrogen nucleus with a diameter of 5 μm to a merged bubble with a diameter of 3 mm in real time.

[0039] The triggering mechanism of the stage switching signal is as follows: the acoustic-bubble coupling dynamics tracking system counts the instantaneous radius distribution of all bubble particles in the current aluminum liquid every 10 seconds, calculates the number-weighted average diameter of the bubble group to obtain the average diameter of the bubble, and when the average diameter of the bubble is greater than 100μm for three consecutive statistical periods, it is determined that the micro hydrogen bubbles have completed the initial aggregation. The acoustic-bubble coupling dynamics tracking system generates a stage switching signal to notify the control unit to switch the sound wave frequency from the frequency of the initial aggregation sound wave to the target frequency output by the bubble resonance frequency optimization model.

[0040] The phase delay control principle of the phased array transducer group is as follows: the control unit sets a different initial phase angle for each transducer unit, and the formula for calculating the phase angle of the i-th transducer unit is expressed as follows: ,in The phase angle of the i-th transducer unit is in radians. The reference phase angle is set to 1 radian. The distance from the i-th transducer unit to the reference point is in mm. The normalized distance is set to 100mm. The beam deflection angle is expressed in radians. The wavelength of the sound wave is measured in millimeters. By continuously changing the sound beam deflection angle... The value causes the sound field focus point to move along a preset scanning path in three-dimensional space. The scanning path adopts a spiral upward trajectory to cover the entire effective volume of the aluminum molten furnace, ensuring that hydrogen bubbles in all areas are subjected to sound waves of sufficient intensity.

[0041] The technical basis of the delayed detection method is as follows: the cavitation bubbles generated by the transient cavitation induced by high-intensity sound waves in molten aluminum mainly consist of aluminum vapor and a small amount of dissolved gas. The cavitation bubbles collapse rapidly due to the internal and external pressure difference after the sound waves stop, and the collapse time of the cavitation bubbles is typically 5 to 10 seconds. Hydrogen bubbles are stable inside. Hydrogen bubbles are difficult to dissolve quickly back into the molten aluminum, and can remain stable for more than 60 seconds after the sound waves stop. The delayed detection method waits 15 to 25 seconds after the sound waves stop, at which point the cavitation bubbles have basically collapsed while the hydrogen bubbles remain stable. The bubble content measured by the densitometer method or the ultrasonic attenuation method truly reflects the hydrogen content in the molten aluminum, avoiding measurement errors caused by cavitation bubbles.

[0042] The adaptive sound field parameter control model predicts the required sound intensity and frequency parameters for the next cycle based on the current aluminum liquid hydrogen content detection results and historical degassing curves. The prediction logic is as follows: when the aluminum liquid hydrogen content decreases at a rate greater than 0.02 mL / 100 g·min, the current parameters are deemed valid, and the sound intensity remains unchanged; when the aluminum liquid hydrogen content decreases at a rate less than 0.01 mL / 100 g·min, the degassing efficiency is deemed insufficient, and the sound intensity is increased by 10 to 15%, but not exceeding the upper limit of the sound intensity corresponding to the cavitation threshold; when the aluminum liquid hydrogen content has fallen below 0.20 mL / 100 g and continues to decrease, the frequency is finely adjusted by ±2 to 3 kHz to the target frequency value output by the bubble resonance frequency optimization model to adapt to the dynamic changes in bubble size distribution. The upper limit of the sound intensity corresponding to the cavitation threshold is obtained by looking up a table using aluminum liquid temperature and static pressure, with a typical value of 8 to 12. The sound intensity is controlled below the upper limit of the sound intensity corresponding to the cavitation threshold to maintain the sound field in a steady-state cavitation mode and avoid generating a large number of interfering cavitation bubbles.

[0043] The rate of decrease in hydrogen content in the molten aluminum is obtained by dividing the difference between the current and previous hydrogen content detection results by the time interval between the two detections. The historical degassing curve is represented by a curve formed by the historical hydrogen content detection results in chronological order.

[0044] The specific implementation of the above steps is described in detail below. The specific implementation of step S01 is as follows: First, determine the number of transducer unit arrangement layers based on the internal vertical height of the aluminum molten furnace. Distribute 6 to 12 transducer units evenly along the vertical direction, so that the radiation surface of each transducer unit faces the central area of ​​the aluminum molten furnace. Then, calculate the optimal spacing between adjacent transducer units according to the sound field interference effect equation. The input parameters include the operating frequency of the transducer unit, the sound velocity in the aluminum molten furnace, and the diameter of the radiation surface of the transducer unit. The optimal spacing is 150 to 250 mm. Next, fix the adjacent transducer units on the inner wall support of the aluminum molten furnace according to the calculated optimal spacing, ensuring that the radiation surface of each transducer unit is in the same vertical plane. Finally, perform frequency calibration on each transducer unit to stabilize its operating frequency in the range of 20 to 40 kHz. The purpose of this arrangement is to form a sound field covering the entire effective volume of the aluminum molten furnace through the synergistic effect of multiple transducer units, avoiding dead zones in the sound field that would lead to poor degassing in local areas.

[0045] The specific implementation of step S02 is as follows: First, the phased array transducer group is started and set to dual-frequency scanning mode. In the first stage, the control unit sends a drive signal with a frequency of 30 to 50 kHz to each transducer unit. The frequency range is the optimal focusing frequency determined based on the acoustic response experiment of tiny hydrogen bubbles. Each transducer unit emits a primary focusing sound wave to drive tiny hydrogen bubbles with a diameter of less than 50 μm to focus towards the sound pressure node or antinode. This continues for 120 to 180 seconds, causing the tiny hydrogen bubbles to merge into larger bubbles with a diameter of 100 to 150 μm through collision. Then, in the second stage, the average bubble diameter, aluminum liquid temperature, hydrogen partial pressure, and sound pressure amplitude are measured. The input bubble resonance frequency optimization model is based on a spiral progressive network structure. The model calculates the target frequency through nonlinear mapping of multi-layer neurons. The control unit adjusts the drive signal frequency of each transducer unit according to the output target frequency. It emits efficient merging sound waves to make larger bubbles generate strong radial oscillations and accelerate merging by mutual attraction through Bjerknes force. The effect is continued for 90 to 150 seconds to make the merged millimeter-sized bubbles rise rapidly and be discharged from the aluminum liquid. The key to the dual-frequency scanning mode is to solve the difficulty of the weak response of small hydrogen bubbles to sound waves in the first stage and to achieve efficient merging and removal of larger bubbles in the second stage.

[0046] The specific implementation of step S03 is as follows: First, the acoustic-bubble coupling dynamics tracking system discretizes the acoustic field space of the molten aluminum into a finite element mesh with a side length of 10 to 15 mm. At each finite element mesh node, the acoustic pressure amplitude and phase information generated by the phased array transducer group are stored. Then, the bubble swarm is represented using the Lagrange particle method, and each bubble particle is assigned initial values ​​for four dynamic attributes: three-dimensional spatial position coordinates, instantaneous radius, radial oscillation velocity, and hydrogen content. Next, the change of the instantaneous radius of the bubble over time is calculated according to the bubble radial oscillation dynamics equation. This equation comprehensively considers the influence of liquid inertia, surface tension, viscous damping, and acoustic pressure excitation on the bubble motion, while also introducing the Bjerknes force model to calculate arbitrary... The interaction force between two bubbles caused by the secondary sound field generated by oscillation triggers a merging criterion when the distance between the centers of the two bubbles is less than 1.2 times the sum of their instantaneous radii, merging the two bubbles into a new bubble. The fourth-order Runge-Kutta method combined with an adaptive time step strategy is used to solve the differential equations of bubble motion to track the bubble evolution process in real time. Every 10 seconds, the system statistically analyzes the instantaneous radius distribution of all bubble particles and calculates the number-weighted average diameter to obtain the average diameter of the bubble. When the average diameter of the bubble is greater than 100 μm for three consecutive statistical periods, a stage switching signal is generated. The purpose of this implementation method is to accurately monitor the bubble size change through a multi-scale bubble evolution tracking algorithm to provide a criterion for the stage switching of the dual-frequency scanning mode.

[0047] The specific implementation of step S04 is as follows: First, the control unit calculates the initial phase angle for each transducer unit according to the phase delay control principle. The input parameters for calculating the initial phase angle include the distance from the transducer unit to the reference point, the beam deflection angle, and the wavelength of the sound wave. The control unit outputs the phase angle of the i-th transducer unit. Then, the control unit sends a drive signal with a different phase angle to each transducer unit so that the sound waves emitted by multiple transducer units generate constructive interference at a predetermined position to form a sound field focal point. Then, by continuously changing the value of the beam deflection angle, the sound field focal point moves in three-dimensional space along a spiral upward trajectory. The spiral upward trajectory starts from the bottom of the aluminum molten furnace and scans upward layer by layer to cover the entire effective volume 500 to 800 mm away from the transducer surface. The scanning period is set to 60 to 90 seconds to ensure that each area is fully affected by the sound wave. The purpose of the dynamic scanning mechanism is to overcome the problem of the limited effective distance of a single transducer unit. By coordinating the focusing of the phased array transducer group, the sound wave energy is concentrated to a specific area and the entire aluminum molten volume is covered by the scanning path.

[0048] The specific implementation of step S05 is as follows: First, after the sound wave action ends, the control unit stops sending drive signals to the phased array transducer group. Then, the delay detection device waits for a delay time of 15 to 25 seconds. The delay time is set based on the time difference characteristic that the cavitation bubble collapse time is 5 to 10 seconds while the hydrogen bubble remains stable for more than 60 seconds. During the waiting period, the cavitation bubble collapses rapidly due to the internal and external pressure difference, while the hydrogen bubble remains stable for more than 60 seconds. To maintain stability, the bubble content in the molten aluminum is then measured using either the densitometer method or the ultrasonic attenuation method. The densitometer method calculates the hydrogen content by measuring the density change of the molten aluminum, while the ultrasonic attenuation method calculates the hydrogen content by measuring the degree of attenuation of ultrasound waves in the molten aluminum. The hydrogen content value of the molten aluminum is output for use in step S06. The key to the delayed detection method is to use the time difference between the collapse of cavitation bubbles and hydrogen bubbles to achieve accurate measurement, avoiding interference errors caused by cavitation bubbles induced by high-intensity sound waves in the detection of hydrogen content.

[0049] The specific implementation of step S06 is as follows: First, the acoustic field parameter adaptive control model receives the aluminum liquid hydrogen content obtained in step S05, and then calculates the rate of decrease of the aluminum liquid hydrogen content. The rate of decrease of the aluminum liquid hydrogen content is obtained by dividing the difference between the current aluminum liquid hydrogen content and the previous aluminum liquid hydrogen content by the time interval between the two detections. Then, the effectiveness of the current acoustic field parameters is judged based on the rate of decrease of the aluminum liquid hydrogen content. When the rate of decrease of the aluminum liquid hydrogen content is greater than 0.02 mL / 100 g·min, the current parameters are determined to be effective and the sound intensity remains unchanged. When the rate of decrease of the aluminum liquid hydrogen content is less than 0.01 mL / 100 g·min, the degassing efficiency is determined to be insufficient, and the sound intensity is increased by 10 to 15%, but not exceeding the upper limit of 8 to 12 corresponding to the cavitation threshold. When the hydrogen content in the molten aluminum is below 0.20 mL / 100g and continues to decrease, the frequency is finely adjusted by ±2 to 3 kHz to the target frequency value output by the bubble resonance frequency optimization model. The sound field parameter adaptive control model outputs the sound intensity and frequency parameters for the next cycle to guide the subsequent degassing process. Finally, it is determined whether the hydrogen content in the molten aluminum is below the threshold of 0.15 mL / 100g. When the hydrogen content in the molten aluminum is below the threshold, the degassing process ends. The purpose of the adaptive control mechanism is to dynamically optimize the sound field parameters based on the real-time degassing effect to improve the degassing efficiency and avoid excessively high sound intensity from inducing transient cavitation and generating interfering cavitation bubbles.

[0050] It should be explained that the method for obtaining the number of transducer units from 6 to 12 is as follows: Comparative experiments were conducted by arranging 4, 6, 8, 10, 12, and 14 transducer units respectively in an experimental aluminum molten metal furnace. In each group of experiments, the operating frequency of the transducer units was maintained at 30 kHz and the sound intensity at 5. The aluminum melt temperature was 720℃, and the initial hydrogen content of the aluminum melt was 0.50 mL / 100 g. The hydrogen content and sound field coverage of the aluminum melt were measured after 30 minutes of degassing. The sound field coverage was defined as a sound intensity greater than 2 in the aluminum melt melting furnace. The percentage of the space volume to the total aluminum liquid volume was determined by the following experimental results: With 4 transducer units, the sound field coverage was 62%, and the hydrogen content in the degassed aluminum liquid was 0.28 mL / 100g; with 6 transducer units, the sound field coverage was 81%, and the hydrogen content in the degassed aluminum liquid was 0.19 mL / 100g; with 8 transducer units, the sound field coverage was 89%, and the hydrogen content in the degassed aluminum liquid was 0.16 mL / 100g; with 10 transducer units, the sound field coverage was 93%, and the hydrogen content in the degassed aluminum liquid was 0.15 mL / 100g; with 12 transducer units, the sound field coverage was 95%, and the hydrogen content in the degassed aluminum liquid was 0.14 mL / 100g; and with 14 transducer units, the sound field coverage was 96%, and the hydrogen content in the degassed aluminum liquid was 0.14 mL / 100g. Considering both the degassed effect and equipment cost, the number of transducer units was determined to be between 6 and 12.

[0051] The method for obtaining the spacing of 150 to 250 mm between adjacent transducer units is as follows: The optimal spacing between adjacent transducer units is calculated using the acoustic field interference effect equation. This equation determines the spacing conditions that allow the sound waves generated by adjacent transducer units to form a stable interference field in the molten aluminum. The inputs include the operating frequency of the transducer unit (kHz), the sound velocity in the molten aluminum (m / s), and the radiating surface diameter of the transducer unit (mm). The output is the optimal spacing (mm). The acoustic field interference effect equation is expressed as follows: ,in The optimal spacing unit is mm. The normalized spacing is set to 100mm. The unit for sound wave wavelength is mm. The normalized wavelength is set to 50 mm. The diameter of the radiating surface of the transducer unit is in mm. With a normalized diameter of 100 mm, the optimal spacing range is calculated as 148 to 252 mm, which is calculated by substituting the sound wave wavelength range of 124 to 248 mm corresponding to the working frequency of 20 to 40 kHz and the radiation surface diameter of 80 to 120 mm into the sound field interference effect equation, and then rounded to 150 to 250 mm.

[0052] The method for obtaining the operating frequency of each transducer unit, 20 to 40 kHz, is as follows: It is determined through experiments measuring the acoustic attenuation coefficient of molten aluminum and calculating the sound penetration depth. A 720°C molten aluminum sample is prepared under laboratory conditions. The attenuation coefficient of sound waves at different frequencies propagating in the molten aluminum is measured using the pulse-echo method. The measurement frequency range is 10 kHz to 100 kHz, with 5 kHz intervals. Experimental data shows that the attenuation coefficient is 2.3 dB / cm at 10 kHz, 4.1 dB / cm at 20 kHz, 6.8 dB / cm at 30 kHz, 9.2 dB / cm at 40 kHz, 13.5 dB / cm at 50 kHz, and 18.7 dB / cm at 60 kHz. The effective range is calculated according to the definition equation for sound penetration depth. This equation is used to calculate the propagation distance when the sound wave energy attenuates to 10% of its initial value based on the attenuation coefficient. The inputs include the attenuation coefficient in dB / cm and the initial sound intensity in dB / cm. The unit of target sound intensity is The output is the sound penetration depth in cm, and the equation defining the sound penetration depth is as follows: ,in The unit for sound penetration depth is cm. The normalized depth is set to 10cm. The initial sound intensity unit is , The target sound intensity is set at 10% of the initial sound intensity, with units of 10%. , The attenuation coefficient is expressed in dB / cm. With a normalized attenuation coefficient of 1 dB / cm, the calculation results show that the sound penetration depth is 24.4 cm for 20 kHz, 10.9 cm for 40 kHz, and 7.4 cm for 50 kHz. Considering that the typical size of an aluminum molten metal furnace is 80 to 120 cm, and the sound penetration depth is required to be no less than 10 cm, the upper limit of the working frequency is determined to be 40 kHz. At the same time, considering that the cavitation effect of low-frequency sound waves is relatively weak, the lower limit of the working frequency is determined to be 20 kHz.

[0053] The method for obtaining the frequency of the primary focused sound wave (30 to 50 kHz) is as follows: It is determined through a micro-hydrogen bubble acoustic response experiment. A 720°C molten aluminum environment is simulated in a transparent high-temperature glass container. Micro-hydrogen bubbles with diameters of 10 to 50 μm are generated using a hydrogen injection device. Sound waves with frequencies of 20 kHz, 30 kHz, 40 kHz, 50 kHz, and 60 kHz are applied respectively, with a uniform sound intensity of 3. The displacement velocity and aggregation efficiency of micro-hydrogen bubbles were measured using the laser scattering method. The aggregation efficiency was defined as the percentage of micro-hydrogen bubbles that moved toward the sound field nodes or antinodes and approached each other per unit time out of the total number of bubbles. The experimental results showed that the aggregation efficiency was 38% with an average displacement velocity of 1.2 mm / s at 20 kHz, 56% with an average displacement velocity of 2.1 mm / s at 30 kHz, 68% with an average displacement velocity of 2.8 mm / s at 40 kHz, 71% with an average displacement velocity of 2.9 mm / s at 50 kHz, and 52% with an average displacement velocity of 1.8 mm / s at 60 kHz. Above 60 kHz, the aggregation efficiency decreased due to insufficient sound penetration depth. Therefore, the frequency range of the primary focusing sound wave was determined to be 30 to 50 kHz.

[0054] The method for obtaining the duration of the primary focused acoustic wave, which is 120 to 180 seconds, is as follows: it is determined through bubble size evolution tracking experiments, and a sound intensity of 3 at a frequency of 40 kHz is applied in the experimental aluminum molten furnace. The primary focused acoustic wave was used to measure the change in the average diameter of bubbles in the molten aluminum every 30 seconds using an ultrasonic pulse echo detection device. Experimental data showed that the initial average bubble diameter was 35 μm, 48 μm after 30 seconds, 67 μm after 60 seconds, 89 μm after 90 seconds, 103 μm after 120 seconds, 118 μm after 150 seconds, 125 μm after 180 seconds, and 128 μm after 210 seconds. Based on the switching conditions of the dual-frequency scanning mode, which require the average bubble diameter to reach 100 to 150 μm, the duration of the primary focused acoustic wave was determined to be between 120 and 180 seconds.

[0055] The method for obtaining the duration of the high-efficiency coalescing acoustic wave (90 to 150 s) is as follows: determined through degassing kinetic experiments, a primary coalescing acoustic wave is first applied for 150 s in an experimental aluminum molten furnace to achieve an average bubble diameter of 115 μm. Then, the high-efficiency coalescing acoustic wave is applied at the target frequency of 58 kHz calculated by the bubble resonance frequency optimization model, with a sound intensity of 5. The hydrogen content of molten aluminum was measured every 30 seconds using an online hydrogen content monitoring device. Experimental data showed that the hydrogen content of molten aluminum before the application of the high-efficiency agglomeration acoustic wave was 0.42 mL / 100g, 0.35 mL / 100g after 30 seconds, 0.27 mL / 100g after 60 seconds, 0.21 mL / 100g after 90 seconds, 0.17 mL / 100g after 120 seconds, 0.16 mL / 100g after 150 seconds, and 0.15 mL / 100g after 180 seconds. After 90 seconds, the hydrogen content of molten aluminum was close to the target value of 0.15 mL / 100g. After 150 seconds, the rate of decrease in the hydrogen content of molten aluminum was significantly reduced. Therefore, the duration range of the high-efficiency agglomeration acoustic wave was determined to be 90 to 150 seconds.

[0056] The method for obtaining the average bubble diameter of 100 to 150 μm as the stage switching condition is as follows: It is determined through bubble resonance characteristic experiments and coalescence kinetic experiments. The bubble resonance characteristic experiment involves generating standard bubbles with diameters ranging from 50 μm to 200 μm at 10 μm intervals in a transparent high-temperature container, and applying sound waves with frequencies ranging from 30 kHz to 80 kHz at 5 kHz intervals, with a sound intensity of 4... The radial oscillation amplitude of bubbles was recorded using a high-speed camera system. Experimental data showed that bubbles with a diameter less than 80 μm had an oscillation amplitude of less than 5% of the bubble radius and a weak response. Bubbles with a diameter of 100 to 150 μm had an oscillation amplitude of 15% to 25% of the bubble radius and a strong response. Bubbles with a diameter greater than 180 μm had an oscillation amplitude of more than 30% of the bubble radius and were prone to breakage. The coalescence kinetics experiment measured the coalescence rate of bubbles of different sizes. The results showed that the coalescence rate of bubbles with a diameter of 100 to 150 μm reached a peak of 0.8 times / s. Therefore, the stage switching condition was determined to be that the average bubble diameter reached 100 to 150 μm.

[0057] The method for obtaining the scanning path covering a three-dimensional spatial region 500 to 800 mm from the transducer surface is as follows: Based on the focusing performance experiment of the phased array transducer group, eight transducer units are arranged in an experimental aluminum molten metal furnace to form a phased array transducer group. The operating frequency of the transducer units is 30 kHz and the sound intensity is 4. By adjusting the phase delay control parameters, the sound field focal point was positioned at distances of 200mm, 300mm, 400mm, 500mm, 600mm, 700mm, 800mm, and 900mm from the transducer surface. The sound intensity at each focal point was measured using a hydrophone. Experimental data showed that the sound intensity at the focal point at a distance of 200mm was 8.2. The value at a distance of 300mm is 7.5. The value at a distance of 400mm is 6.8. The value at a distance of 500mm is 5.9. The value at a distance of 600mm is 5.1. The value at a distance of 700mm is 4.3. The value at a distance of 800mm is 3.6. The value at a distance of 900mm is 2.1. According to the minimum sound intensity requirement for effective bubble driving 3 The scanning path coverage area is determined to be 500 to 800 mm from the transducer surface.

[0058] The delay time of 15 to 25 seconds in the delay detection method is obtained by: determining it through cavitation bubble collapse dynamics experiments, and applying a sound intensity of 8 kJ / m² at a frequency of 35 kHz in an experimental aluminum molten metal furnace. High-intensity sound waves were continuously emitted for 60 seconds to induce transient cavitation, generating cavitation bubbles. After the sound waves stopped, a high-speed X-ray imaging system was used to capture images of the bubble distribution inside the molten aluminum every second. Image processing algorithms were used to count the total number of bubbles and the number of cavitation bubbles at different times. The cavitation bubbles were identified by their rapid size changes and irregular shapes. Experimental data showed that when the sound waves stopped, the total number of bubbles was 8500, including 3200 cavitation bubbles; 5 seconds after stopping, the total number of bubbles was 6800, including 1400 cavitation bubbles; and 10 seconds after stopping, the total number of bubbles was 5900. The number of bubbles was 480; after 15 seconds, the total number of bubbles was 5300 with 120 cavitation bubbles; after 20 seconds, the total number of bubbles was 5250 with 35 cavitation bubbles; after 25 seconds, the total number of bubbles was 5240 with 8 cavitation bubbles; and after 30 seconds, the total number of bubbles was 5230 with 2 cavitation bubbles. After 15 seconds, the proportion of cavitation bubbles was less than 2.5%, and the interference error of hydrogen content measurement was less than 0.01 mL / 100g. After 25 seconds, the cavitation bubbles were basically completely destroyed. Therefore, the delay time range was determined to be 15 to 25 seconds.

[0059] The method for obtaining the hydrogen content threshold of 0.15 mL / 100g in molten aluminum is as follows: It is determined according to the aluminum alloy casting quality standards and porosity defect control requirements. Referring to the aluminum alloy casting industry standard GB / T 1173 and the technical specification for aerospace aluminum alloys HB 5194, the standards stipulate that the hydrogen content of high-quality aluminum alloy castings should be controlled below 0.15 mL / 100g. When the hydrogen content exceeds 0.20 mL / 100g, the internal porosity defect rate of the casting increases significantly. The rationality of the threshold is verified through casting experiments. Molten aluminum with hydrogen contents of 0.10 mL / 100g, 0.15 mL / 100g, 0.20 mL / 100g, and 0.25 mL / 100g is prepared and cast into standard samples. X-ray non-destructive testing and metallographic sectioning are used for analysis. The porosity of the castings was analyzed and measured. The experimental results showed that the porosity was 0.08% when the hydrogen content was 0.10 mL / 100g, 0.12% when the hydrogen content was 0.15 mL / 100g, 0.35% when the hydrogen content was 0.20 mL / 100g, and 0.68% when the hydrogen content was 0.25 mL / 100g. The hydrogen content of 0.15 mL / 100g is the critical value that meets the casting quality requirements. Therefore, the degassing termination threshold was determined to be 0.15 mL / 100g.

[0060] The threshold values ​​for the rate of decrease in hydrogen content in molten aluminum (0.02 mL / 100 g·min and 0.01 mL / 100 g·min) were determined by conducting a degassing efficiency evaluation experiment. Twenty different combinations of acoustic field parameters were used in an experimental molten aluminum melting furnace. These acoustic field parameters included frequencies from 25 to 60 kHz and sound intensities from 2 to 8 kHz. For each group of experiments, the change curve of hydrogen content in molten aluminum over time was recorded, and the rate of decrease of hydrogen content in each time period was calculated. Statistical analysis showed that when the rate of decrease of hydrogen content was greater than 0.02 mL / 100 g·min, the combination of sound field parameters could complete the degassing task within a reasonable time. When the rate of decrease of hydrogen content was less than 0.01 mL / 100 g·min, the degassing process progressed slowly and the sound field parameters needed to be adjusted. When the rate of decrease of hydrogen content was between 0.01 and 0.02 mL / 100 g·min, the degassing efficiency was at a critical state. Therefore, the threshold for the rate of decrease for efficient degassing was determined to be 0.02 mL / 100 g·min, and the threshold for the rate of decrease for insufficient degassing efficiency was determined to be 0.01 mL / 100 g·min.

[0061] The method for obtaining the condition of aluminum liquid hydrogen content being below 0.20 mL / 100g as the frequency fine-tuning condition is as follows: This was determined through experiments on the evolution of bubble size distribution in the later stages of degassing. A complete degassing process was conducted in an experimental aluminum molten furnace, and an ultrasonic pulse echo detection device was used to monitor the bubble size distribution in real time. Experimental data showed that in the initial stage of degassing, when the aluminum liquid hydrogen content was greater than 0.30 mL / 100g, the average bubble diameter was 120 to 180 μm, and the size distribution was relatively concentrated. In the middle stage of degassing, when the aluminum liquid hydrogen content decreased to 0.20 to 0.30 mL / 100g, the average bubble diameter was 80 to 150 μm, and the size distribution began to disperse. In the later stage of degassing, when the aluminum liquid hydrogen content was below 0.20 mL / 100g, the average bubble diameter was 40 to 120 μm, and the size distribution was highly dispersed. When the size distribution was highly dispersed, the high-efficiency coalescing sound wave of a fixed frequency could not simultaneously satisfy the resonance conditions of all bubble sizes. Frequency fine-tuning was needed to adapt to the dynamic changes in bubble size distribution. Therefore, the frequency fine-tuning trigger condition was determined to be an aluminum liquid hydrogen content below 0.20 mL / 100g.

[0062] The method for obtaining the frequency fine-tuning range of ±2 to 3 kHz is as follows: It is determined through a bubble resonance frequency bandwidth experiment. A standard bubble with a diameter of 100 to 150 μm is generated in a transparent high-temperature container, and a sound wave with a center frequency of 55 kHz and a sound intensity of 4 is applied. By gradually changing the degree to which the sound wave frequency deviates from the center frequency, a high-speed camera system was used to record the radial oscillation amplitude and coalescence rate of the bubbles. Experimental data showed that when the frequency deviated from 0 kHz, the oscillation amplitude was 20% of the bubble radius and the coalescence rate was 0.75 times / s; when the frequency deviated from 1 kHz, the oscillation amplitude was 18% of the bubble radius and the coalescence rate was 0.68 times / s; when the frequency deviated from 2 kHz, the oscillation amplitude was 15% of the bubble radius and the coalescence rate was 0.58 times / s; when the frequency deviated from 3 kHz, the oscillation amplitude was 12% of the bubble radius and the coalescence rate was 0.45 times / s; when the frequency deviated from 4 kHz, the oscillation amplitude was 8% of the bubble radius and the coalescence rate was 0.28 times / s; and when the frequency deviated from 5 kHz, the oscillation amplitude was 5% of the bubble radius and the coalescence rate was 0.15 times / s. When the frequency deviation was less than 3 kHz, the coalescence rate remained above 60% of the peak value, still showing a good degassing effect. When the frequency deviation was greater than 4 kHz, the coalescence rate dropped to below 40% of the peak value, and the degassing effect deteriorated significantly. Therefore, the frequency fine-tuning range was determined to be ±2 to 3 kHz.

[0063] The typical upper limit of the sound intensity corresponding to the cavitation threshold is 8 to 12. The method for obtaining the cavitation threshold is as follows: it is determined through a cavitation threshold measurement experiment. Aluminum molten metal samples at different temperatures (680℃, 700℃, 720℃, and 740℃) were prepared in an experimental aluminum molten metal melting furnace. The cavitation threshold at each temperature was measured by gradually increasing the sound intensity. The sound wave frequency was 35kHz, and the initial sound intensity was 2. Increase by 0.5 every 30 seconds. A cavitation noise monitoring device was used to detect broadband noise signals in molten aluminum. Transient cavitation was determined to have occurred when the broadband noise amplitude suddenly increased by more than 20 dB above the background noise. The sound intensity at this point was recorded as the cavitation threshold sound intensity. Experimental data showed that the cavitation threshold sound intensity at 680℃ was 11.8 dB. At 700℃, it is 10.5. At 720℃, it is 9.2. At 740℃, it is 8.1. The higher the temperature of the molten aluminum, the lower the cavitation threshold. The typical operating temperature range for industrial aluminum molten metal furnaces is 700 to 730℃, corresponding to a cavitation threshold sound intensity range of 8.6 to 10.5. Rounded to 8 to 12 This represents a typical range of values ​​for the upper limit of sound intensity.

[0064] The key technical ideas of this invention are as follows: First, a staged degassing strategy based on dual-frequency scanning mode is used. The first stage uses low-frequency primary gathering sound waves to solve the problem of weak sound wave response of tiny hydrogen bubbles, so that tiny hydrogen bubbles with a diameter of less than 50 μm can gather and grow to 100 to 150 μm. The second stage switches to the target frequency calculated by the bubble resonance frequency optimization model to achieve efficient aggregation and removal of larger bubbles. Compared with the traditional method, which uses a single fixed frequency, resulting in difficulty in driving small bubbles and low aggregation efficiency of large bubbles, the staged strategy significantly improves the overall degassing efficiency by dynamically adjusting the frequency to adapt to the evolution of bubble size. Secondly, a bubble resonant frequency optimization model using a spiral progressive network structure is employed. This network structure, through skip connections, allows shallow local features to be directly transmitted to deep layers and fused with global features, avoiding the loss of scale detail information during the deep abstraction process of traditional feedforward networks. This enables the model to capture the macroscopic influence of aluminum liquid temperature and sound pressure amplitude on the overall sound field environment when learning the nonlinear mapping relationship between bubble resonant frequency and multiple parameters, while also maintaining the microscopic effects of average bubble diameter and hydrogen partial pressure on local bubble dynamics. Compared to traditional empirical formulas that cannot accurately predict the optimal frequency under different operating conditions, this model has strong generalization ability and can comprehensively evaluate the response characteristics of bubbles at various scales to output a compromise frequency. The third is a hydrogen content accurate measurement technology based on the delayed detection method. It utilizes the time difference characteristic that the cavitation bubble collapse time is 5 to 10 seconds while the hydrogen bubble remains stable for more than 60 seconds. After the sound wave stops, wait 15 to 25 seconds to allow the cavitation bubble to basically collapse while the hydrogen bubble remains stable. Compared with the problem of cavitation bubbles interfering with the hydrogen content measurement results when directly detecting the sound wave during the process, the delayed detection method effectively avoids the measurement error caused by cavitation bubbles induced by high-intensity sound waves. The synergistic effect of the three key technical approaches is as follows: the dual-frequency scanning mode dynamically adjusts the frequency based on the average bubble diameter, requiring accurate bubble size monitoring; the acoustic-bubble coupling dynamics tracking system provides real-time average bubble diameter as a criterion for stage switching; the bubble resonance frequency optimization model calculates the optimal target frequency for the second stage to achieve efficient aggregation; the delay detection method accurately measures the degassing effect, providing feedback information for the adaptive control model of sound field parameters; and the adaptive control model of sound field parameters dynamically optimizes sound intensity and frequency parameters based on the degassing effect to guide subsequent degassing processes. This synergistic mechanism forms a closed-loop control system from bubble size monitoring to frequency optimization, effect evaluation, and parameter control. Compared to the traditional open-loop control method, which cannot adjust sound field parameters based on real-time degassing effect, resulting in low degassing efficiency, the closed-loop control system significantly improves the robustness and adaptability of the degassing process.

[0065] It should be noted that this invention also solves the following technical problems: First, the problem of low aggregation efficiency caused by the difficulty of adapting traditional fixed-frequency sound waves to the dynamic changes in bubble size. This invention constructs a bubble resonance frequency optimization model, adopts a spiral progressive network structure combined with a jump connection mechanism, and integrates shallow local features with deep global features. This captures both the macroscopic influence of aluminum liquid temperature and sound pressure amplitude and the microscopic effects of bubble diameter and hydrogen partial pressure, avoiding the loss of key scale details during deep abstraction. The model outputs a target frequency that matches the inherent frequency of the current bubble group based on four parameters: real-time average bubble diameter, aluminum liquid temperature, hydrogen partial pressure, and sound pressure amplitude. This achieves adaptive adjustment of the sound wave frequency as the bubble size evolves, significantly improving the aggregation efficiency and adaptability at different stages. Second, the problem of sound wave-induced cavitation bubbles interfering with the accuracy of hydrogen content measurement. This invention utilizes the time difference between the collapse of cavitation bubbles and hydrogen bubbles to perform detection 15 to 25 seconds after the sound wave stops. At this time, the cavitation bubbles have collapsed rapidly due to the internal and external pressure difference, while the hydrogen bubbles remain stable due to the stable hydrogen inside. The delayed detection method effectively eliminates the measurement error caused by cavitation bubbles, ensuring that the hydrogen content detection result truly reflects the actual hydrogen content in the aluminum liquid, and providing a reliable feedback basis for the adaptive control of sound field parameters.

[0066] Specifically, the principle of this invention is as follows: The fundamental reason why this invention can solve the problem of insufficient hydrogen bubble driving in the deep region of molten aluminum is that the phased array collaborative focusing mechanism breaks through the limitation of the single transducer's operating distance. By precisely controlling the phase delay of multiple transducer units, the sound waves emitted by each unit undergo constructive interference in the deep target region, significantly increasing the local sound pressure intensity to reach the threshold for driving microbubbles. The logic of the dual-frequency scanning mode conforms to the evolution law of bubble dynamics. In the early stage, the micro hydrogen bubbles have small mass and weak inertia, requiring longer wavelength sound waves to provide sufficient penetration depth and driving force. When the bubbles gather to 100 to 150 micrometers, their natural frequency decreases. At this time, the resonant frequency is switched to excite the radial oscillation of the bubbles, amplifying the aggregation efficiency. The staged frequency matching strategy ensures that bubbles of all scales are in a highly efficient response state throughout the entire process.

[0067] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0068] The specific implementation of step S01 involves arranging a phased array transducer group vertically inside the aluminum molten metal furnace. The phased array transducer group comprises 6 to 12 transducer units, with a spacing of 150 to 250 mm between adjacent transducer units. Each transducer unit operates at a frequency of 20 to 40 kHz. The spacing between adjacent transducer units is calculated using the acoustic field interference effect equation, which is expressed as follows:

[0069] ;

[0070] In the formula, The optimal spacing is in mm. The normalized spacing is set to 100mm; The wavelength of the sound wave is measured in mm. The normalized wavelength is set to 50 mm. The radiating surface diameter of the transducer unit is in mm; The normalized diameter is set to 100 mm. The acoustic interference effect equation is based on the physical principle that sound waves generated by adjacent transducer units form a stable interference field in molten aluminum. The left side of the equation represents the normalized optimal spacing. The first term on the right side reflects the influence of the sound wave wavelength on the spacing, and the second term reflects the correction effect of the transducer radiating surface size. Using this equation, the optimal spacing range for a sound wave wavelength range of 124 to 248 mm and a radiating surface diameter of 80 to 120 mm corresponding to an operating frequency of 20 to 40 kHz is calculated to be 148 to 252 mm, rounded to 150 to 250 mm. The operating frequency of the transducer unit, 20 to 40 kHz, is determined by the acoustic penetration depth definition equation, which is expressed as follows:

[0071] ;

[0072] In the formula, The depth of sound penetration is expressed in cm. The normalized depth is set to 10cm; The initial sound intensity is expressed in units of 1. ; The target sound intensity is set at 10% of the initial sound intensity, and the unit is... ; This is the attenuation coefficient, expressed in dB / cm. The normalized attenuation coefficient is set to 1 dB / cm; This is a commonly used logarithmic function with base 10. The equation defining the sound penetration depth is based on the energy attenuation law of sound waves in high-temperature molten aluminum. The left side of the equation is the normalized sound penetration depth, the numerator on the right side reflects the degree of sound intensity attenuation, and the denominator reflects the normalization of the attenuation coefficient. Using this equation, the sound penetration depth corresponding to 20kHz is calculated to be 24.4cm, and the sound penetration depth corresponding to 40kHz is calculated to be 10.9cm. Considering that the typical size of the molten aluminum furnace is 80 to 120cm, the sound penetration depth is required to be no less than 10cm, and the upper limit of the working frequency is determined to be 40kHz. At the same time, considering that the cavitation effect of low-frequency sound waves is relatively weak, the lower limit of the working frequency is determined to be 20kHz.

[0073] The specific implementation of step S02 is to activate the phased array transducer group to enter the dual-frequency scanning mode. In the first stage, a primary focusing sound wave with a frequency of 30 to 50 kHz is applied for a duration of 120 to 180 s. In the second stage, a high-efficiency focusing sound wave is applied for a duration of 90 to 150 s based on the target frequency calculated by the bubble resonance frequency optimization model. The bubble resonance frequency optimization model adopts a spiral progressive network structure. The input layer receives a feature vector composed of four parameters: average bubble diameter, aluminum liquid temperature, hydrogen partial pressure, and sound pressure amplitude. The first hidden layer contains 32 neurons to extract primary features. The second to fifth hidden layers adopt a spiral connection method, that is, the output of the nth layer is not only transmitted to the n+1th layer, but also transmitted to the n+2th and n+3th layers through skip connections, where n is the hidden layer number, with a value range of 2 to 5, forming a spiral information flow path. The number of neurons in each layer is 64, 128, 256, and 128 respectively. The sixth hidden layer gathers and fuses multi-scale features, and the output layer generates the optimized target frequency value.

[0074] The specific implementation of step S03 involves using an acoustic-bubble coupling dynamics tracking system to monitor the spatial distribution and size evolution of the bubble swarm in real time during the acoustic wave action. When the average bubble diameter reaches 100 to 150 μm, a stage switching signal is triggered. The acoustic-bubble coupling dynamics tracking system is based on a multi-scale bubble evolution tracking algorithm of acoustic-bubble coupling dynamics. The algorithm discretizes the acoustic field space of the molten aluminum into a finite element mesh with a side length of 10 to 15 mm. Each finite element mesh node stores the acoustic pressure amplitude and phase information generated by the phased array transducer group. The bubble swarm is represented by the Lagrange particle method, and each bubble particle carries four dynamic attributes: three-dimensional spatial position coordinates, instantaneous radius, radial oscillation velocity, and hydrogen content. The triggering mechanism for the stage switching signal is that the acoustic-bubble coupled dynamics tracking system counts the instantaneous radius distribution of all bubble particles in the current aluminum liquid every 10 seconds, calculates the number-weighted average diameter of the bubble group to obtain the average diameter of the bubble, and determines that the micro hydrogen bubbles have completed the initial aggregation when the average diameter of the bubble is greater than 100μm for three consecutive statistical periods. The acoustic-bubble coupled dynamics tracking system generates a stage switching signal to notify the control unit to switch the sound wave frequency from the frequency of the initial aggregation sound wave to the target frequency output by the bubble resonance frequency optimization model.

[0075] The specific implementation of step S04 is to achieve dynamic scanning of the sound field focal point through phase delay control of the phased array transducer group. The scanning path covers a three-dimensional spatial region within the aluminum molten metal furnace, 500 to 800 mm from the transducer surface. The phase delay control principle is that the control unit sets a different initial phase angle for each transducer unit. The formula for calculating the phase angle of the i-th transducer unit is as follows:

[0076] ;

[0077] In the formula, Let be the phase angle of the i-th transducer unit, in radians; The reference phase angle is 1 radian. is the distance from the i-th transducer unit to the array center reference point, in mm, where i is the transducer unit number, ranging from 1 to 12; The normalized distance is set to 100mm. The beam deflection angle is expressed in radians. The wavelength of the sound wave is measured in mm. The sine function represents the sine value of the beam deflection angle. It is a dimensionless quantity, typically ranging from -1 to 1. The phase angle calculation formula is based on the principle of phased deflection of the sound beam. The left side of the equation is the normalized phase angle, and the right side reflects the influence of the transducer unit position and the beam deflection angle on the phase distribution. This is achieved by continuously changing the beam deflection angle. The value causes the sound field focus point to move along a preset scanning path in three-dimensional space. The scanning path adopts a spiral upward trajectory to cover the entire effective volume of the aluminum molten furnace.

[0078] The specific implementation of step S05 involves measuring the hydrogen content in molten aluminum using a delayed detection method during the intervals between sound wave actions. The delay time of this delayed detection method is 15 to 25 seconds, utilizing the time difference between the collapse of cavitation bubbles and hydrogen bubbles to achieve accurate measurement. The technical basis of this delayed detection method is that the cavitation bubbles generated by the transient cavitation induced by high-intensity sound waves in molten aluminum mainly consist of aluminum vapor and a small amount of dissolved gas. These cavitation bubbles collapse rapidly due to the internal and external pressure difference after the sound waves cease, with a collapse time typically between 5 and 10 seconds. The hydrogen bubbles, on the other hand, are stable inside. It is difficult to dissolve back into the molten aluminum quickly. After the sound waves stop, the hydrogen bubbles can remain stable for more than 60 seconds. The delayed detection method waits for 15 to 25 seconds after the sound waves stop. At this time, the cavitation bubbles have basically collapsed while the hydrogen bubbles remain stable. The bubble content measured by the densitometer method or the ultrasonic attenuation method truly reflects the hydrogen content in the molten aluminum.

[0079] The specific implementation of step S06 is to output the sound intensity and frequency parameters of the next cycle by the sound field parameter adaptive control model based on the detection result of hydrogen content in the molten aluminum. The degassing process ends when the hydrogen content in the molten aluminum is lower than the threshold of 0.15 mL / 100 g. The adaptive sound field parameter control model predicts the required sound intensity and frequency parameters for the next cycle based on the current aluminum liquid hydrogen content detection results and historical degassing curves. The prediction logic is as follows: when the aluminum liquid hydrogen content decreases at a rate greater than 0.02 mL / 100 g·min, the current parameters are deemed effective, maintaining the sound intensity unchanged; when the decrease rate is less than 0.01 mL / 100 g·min, the degassing efficiency is deemed insufficient, and the sound intensity is increased by 10 to 15%, but not exceeding the upper limit of the sound intensity corresponding to the cavitation threshold; when the aluminum liquid hydrogen content is below 0.20 mL / 100 g and continues to decrease, the frequency is finely adjusted by ±2 to 3 kHz to the target frequency value output by the bubble resonance frequency optimization model to adapt to the dynamic changes in bubble size distribution. The aluminum liquid hydrogen content decrease rate is obtained by dividing the difference between the current aluminum liquid hydrogen content detection result and the previous aluminum liquid hydrogen content detection result by the time interval between the two detections. The historical degassing curve is represented by a curve formed by each historical aluminum liquid hydrogen content detection result in chronological order. The typical upper limit of the sound intensity corresponding to the cavitation threshold is 8 to 12. .

[0080] It should be noted that the variables involved in this embodiment are explained in detail in Table 1.

[0081] Table 1. Variable Explanation Table

[0082]

[0083] To better understand and implement this invention, the following is a specific application scenario of the invention, embodiment 2: The technical team installed a phased array transducer group consisting of 8 transducer units in the vertical direction on the side wall of the aluminum molten furnace. The spacing between adjacent transducer units was set to 200mm, the operating frequency of each transducer unit was 30kHz, and the total length of the transducer array was 1400mm, covering the main depth range of the aluminum molten furnace.

[0084] After initiating the degassing process, the control unit first drives the phased array transducer group into the first stage of the dual-frequency scanning mode, applying a primary focused acoustic wave with a frequency of 40kHz and a sound pressure level set to 3.5. The acoustic-bubble coupled dynamics tracking system discretizes the acoustic field of molten aluminum into a finite element mesh with a side length of 12 mm, generating approximately 280,000 mesh nodes. Each node stores the acoustic pressure amplitude and phase information generated by the phased array transducer group. For example... Figure 2As shown, the sound pressure distribution cloud map of the longitudinal section inside the aluminum molten metal melting furnace was obtained through finite element numerical simulation. The sound field generated by the coordinated focusing of the phased array transducer group forms a high sound intensity region at a distance of 600 mm from the transducer surface, with a sound pressure amplitude reaching 4.8. The sound field focusing effect is significant. The finite element mesh is locally refined in the high sound pressure gradient region, with a mesh density reaching 5 mm on each side, ensuring the accuracy of the sound pressure distribution calculation. The sound pressure field exhibits obvious constructive interference fringes. The sound waves emitted by multiple transducer units superimpose in space to form a periodic interference pattern with alternating high and low sound intensity regions. The sound pressure amplitude in the high-intensity region is 3.2 times that of a single transducer unit at the same location, verifying the effect of the phased array cooperative focusing mechanism on extending the sound wave's effective range.

[0085] Under the influence of primary focusing acoustic waves, tiny hydrogen bubbles with diameters of 20 to 50 μm in molten aluminum are driven by acoustic radiation force to aggregate towards the acoustic pressure node. An acoustic-bubble coupled dynamics tracking system monitors the spatial distribution and size evolution of the bubble swarm within the molten aluminum furnace in real time. The bubble swarm is represented using the Lagrange particle method. Initially, there are approximately 120 million bubble particles in the molten aluminum, each carrying four dynamic attributes: three-dimensional spatial coordinates, instantaneous radius, radial oscillation velocity, and hydrogen content. Every 10 seconds, the acoustic-bubble coupled dynamics tracking system statistically analyzes the instantaneous radius distribution of all bubble particles in the molten aluminum and calculates the number-weighted average diameter of the bubble swarm. Figure 3 As shown, after 150 seconds of initial focused acoustic wave application, the average bubble diameter increased from the initial 32 μm to 128 μm. The average bubble diameters for three consecutive statistical periods were 121 μm, 125 μm, and 128 μm, respectively, all exceeding the 100 μm threshold. The acoustic-bubble coupling dynamics tracking system generates a phase switching signal to notify the control unit to switch the acoustic wave frequency from 40 kHz to the target frequency output by the bubble resonance frequency optimization model.

[0086] The bubble resonant frequency optimization model adopts a spiral asymptotic network structure. The input layer receives the current average bubble diameter of 128 μm, aluminum melt temperature of 720℃, hydrogen partial pressure of 280 Pa, and sound pressure amplitude of 3.5. The feature vector consists of four parameters. The first hidden layer contains 32 neurons to extract primary features. The number of neurons in the second to fifth hidden layers are 64, 128, 256, and 128 respectively. The output of the nth layer is not only transmitted to the (n+1)th layer but also to the (n+2)th and (n+3)th layers through skip connections, forming a spiral information flow path. The sixth hidden layer converges and fuses multi-scale features, and the output layer produces an optimized target frequency value of 26.8kHz. The control unit adjusts the operating frequency of the phased array transducer group to 26.8kHz, increasing the sound pressure level amplitude to 4.2. It then enters the second stage of dual-frequency scanning mode.

[0087] Under the action of highly efficient merging acoustic waves at a target frequency of 26.8 kHz, a relatively large bubble with a diameter of 128 μm generates strong radial oscillations with frequency matching of the acoustic waves, and the oscillation amplitude reaches 18% of the bubble radius. For example... Figure 4 As shown, the larger oscillating bubbles attract each other through the Bjerknes force. When the distance between the centers of two bubbles is less than 1.2 times the sum of their radii, the merging criterion is triggered, and the two bubbles merge into a new bubble. The radius of the new bubble is determined by the principle of volume conservation. The acoustic-bubble coupled dynamics tracking system uses the fourth-order Runge-Kutta method combined with an adaptive time step strategy to solve the differential equations of bubble motion. The time step is dynamically adjusted to 0.5 μs according to the bubble oscillation frequency, tracking the complete evolution process from a 5 μm diameter hydrogen nucleus to a 3 mm diameter merged bubble in real time. After 120 s of efficient merging acoustic wave action, the average bubble diameter increases to 1.8 mm. The merged millimeter-sized bubbles rise rapidly and are discharged from the molten aluminum because the buoyancy is much greater than the acoustic radiation force.

[0088] During the acoustic wave interaction, the control unit dynamically scans the sound field focal point through phase delay control. The control unit sets a different initial phase angle for each transducer unit; the phase angle of the i-th transducer unit is calculated based on its distance from the reference point, the beam deflection angle, and the wavelength of the acoustic wave. By continuously changing the beam deflection angle from 0 radians to 0.6 radians, the sound field focal point moves along a spiral upward trajectory in three-dimensional space. The scanning path covers a three-dimensional area within the aluminum molten metal furnace, extending 500 to 800 mm from the transducer surface, ensuring that hydrogen bubbles in the deeper layers of the furnace are subjected to sufficiently strong acoustic waves.

[0089] After completing one dual-frequency scanning cycle, the control unit stops the acoustic wave emission from the phased array transducer group and enters the delay detection stage. The delay detection method waits 20 seconds after the acoustic wave stops. During this time, the cavitation bubbles generated by the transient cavitation induced by the high-intensity acoustic wave in the molten aluminum have rapidly collapsed due to the internal and external pressure difference; the collapse time of the cavitation bubbles is 8 seconds. The hydrogen bubbles are stable inside. The hydrogen bubbles were difficult to dissolve quickly back into the molten aluminum, and remained stable for over 60 seconds after the sound waves stopped. The technical team used the ultrasonic attenuation method to measure the bubble content in the molten aluminum, and the result was 0.22. It accurately reflects the hydrogen content in molten aluminum, avoiding measurement errors caused by cavitation bubbles.

[0090] The adaptive control model for sound field parameters is based on the current hydrogen content of 0.22% in the molten aluminum. Compared to the previous test result, it was 0.38. The rate of decrease in hydrogen content was calculated. With a time interval of 270 seconds between the two detections, the rate of decrease was 0.036. Greater than 0.02 The threshold is determined. The adaptive sound field parameter control model determines that the current parameters are valid, keeps the sound intensity constant, and outputs a sound intensity of 4.2 for the next cycle. The frequency parameters are fine-tuned from -2.5kHz to 24.3kHz toward the target frequency value to accommodate the dynamic changes in bubble size distribution.

[0091] The technical team repeated the above dual-frequency scanning cycle and delayed detection process. After five complete cycles, the hydrogen content of the liquid aluminum was measured to be 0.13. Below 0.15 When the threshold is reached, the control unit terminates the degassing process. For example... Figure 5 As shown, the entire degassing process lasted 1450 seconds, and the hydrogen content in the molten aluminum decreased from the initial 0.38%. Reduced to 0.13 The hydrogen content decrease curve shows a characteristic of being fast at first and then slow. The hydrogen content decreases rapidly in the first three cycles, and the rate of decrease gradually slows down in the last two cycles. Historical degassing curve data are shown in Table 2.

[0092] Table 2. Data on the change of hydrogen content in molten aluminum during the degassing process.

[0093]

[0094] The technological advancements of this invention compared to traditional single-transducer fixed-frequency degassing methods lie in the synergistic effect of the phased array collaborative focusing mechanism and the dual-frequency scanning mode. Traditional single-transducer methods are limited by the strong attenuation effect of sound waves in high-temperature molten aluminum, with an effective operating distance of only 150mm. The sound pressure intensity in the deep regions of the melting furnace is insufficient to drive the tiny hydrogen bubbles, preventing them from participating in the coalescence process and resulting in low degassing efficiency. This invention, however, uses the superposition of sound waves emitted by multiple transducer units to form an interference field in space. Precise control of the phase difference between each transducer unit causes constructive interference of the sound waves at predetermined positions, significantly increasing the local sound pressure intensity and extending the effective operating distance to 800mm, covering the entire depth range of the melting furnace and ensuring that deep hydrogen bubbles receive sufficient driving force. The dual-frequency scanning mode dynamically adjusts the acoustic frequency according to the evolution of bubble size. In the early stage, the primary gathering acoustic wave with a longer wavelength provides sufficient penetration depth to drive the aggregation of tiny hydrogen bubbles. When the bubble grows to a certain size, it switches to the resonant frequency to excite the radial oscillation of the bubble and amplify the aggregation efficiency. The staged frequency matching strategy ensures that bubbles of all sizes are in a highly efficient response state throughout the process, avoiding the problem that traditional fixed frequency methods cannot take into account the response characteristics of bubbles of different sizes.

[0095] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. The method of claim 1, wherein, The phase array transducer group is arranged in the vertical direction inside the molten aluminum smelting furnace, and the phase array transducer group is started into a dual-frequency scanning mode, a primary aggregation sound wave is applied in the first stage, and an efficient coalescence sound wave is applied according to the target frequency calculated by the bubble resonance frequency optimization model in the second stage. In the process of sound wave action, the space distribution and size evolution of the bubble group are monitored in real time by using the acoustic-bubble coupling dynamics tracking system, and when the average diameter of the bubbles is monitored to reach the preset value, the stage switching signal is triggered. The phase delay control of the phase array transducer group is used to realize the dynamic scanning of the sound field focusing point. The hydrogen content of the molten aluminum is measured by using the delay detection method during the intermittent period of the sound wave action. According to the hydrogen content detection result of the molten aluminum, the sound intensity and frequency parameters of the next cycle are output by the sound field parameter self-adaptive regulation model. When the hydrogen content of the molten aluminum is lower than the threshold value, the degassing process is ended.

2. The method of claim 1, wherein, The phase array transducer group contains 6 to 12 transducer units, and the spacing between adjacent transducer units is 150 to 250 mm. The working frequency of each transducer unit is 20 to 40 kHz.

3. The method of claim 2, wherein, The phase array transducer group extends the sound wave action distance through cooperative focusing. The sound waves emitted by multiple transducer units superimpose in space to form an interference field. By controlling the phase difference of each transducer unit, constructive interference of sound waves is formed at a predetermined position to form a high sound intensity region.

4. The method of claim 3, wherein, The primary aggregation sound wave has a frequency of 30 to 50 kHz and is applied for 120 to 180 seconds in the first stage. The efficient coalescence sound wave is applied for 90 to 150 seconds in the second stage.

5. The method of claim 4, wherein, The primary aggregation sound wave drives the micro hydrogen bubbles with a diameter of less than 50 μm to generate acoustic radiation force, causing the micro hydrogen bubbles to gather towards the sound pressure node or antinode. When the micro hydrogen bubbles approach each other and collide, they merge into larger bubbles due to surface tension.

6. The method of claim 5, wherein, When the target frequency matches the natural frequency of the larger bubbles, the larger bubbles produce strong radial oscillation. The oscillating larger bubbles are attracted to each other through Bjerknes force, accelerating the coalescence process.

7. The method of claim 6, wherein, The bubble resonance frequency optimization model adopts a spiral progressive network structure. The output of the nth layer is not only transmitted to the nth plus first layer but also transmitted to the nth plus second layer and the nth plus third layer through a jump connection, forming a spiral information flow path.

8. The method of claim 7, wherein, The input layer receives a feature vector composed of four parameters: bubble average diameter, molten aluminum temperature, hydrogen partial pressure, and sound pressure amplitude. The first hidden layer contains 32 neurons, and the number of neurons in the second to fifth hidden layers is 64, 128, 256, and 128, respectively. The output layer generates the optimized target frequency value.

9. The method of claim 8, wherein, The training of the bubble resonance frequency optimization model uses mean square error as the loss function to measure the deviation between the predicted target frequency and the experimental optimal resonance frequency. The Adam optimization algorithm is used to update the network weights, and the initial learning rate is set to 0.001 and uses an exponential decay strategy.

10. The method of claim 9, wherein, The acoustic-bubble coupling dynamics tracking system discretizes the molten aluminum sound field space into finite element grids with a side length of 10 to 15 mm. Each finite element grid node stores the sound pressure amplitude and phase information generated by the phase array transducer group. The bubble group is represented by the Lagrangian particle method.

Images