Method for on-line multi-parameter collaborative control of degassing of aluminum alloy melt

Abstract

The present application relates to the technical field of industrial process automatic control, and especially relates to an aluminum alloy melt on-line degassing multi-parameter collaborative control method, which comprises collecting hydrogen atom diffusion flux, electrochemical potential and impedance real part signal, constructing a multi-field coupling state tensor through space-time alignment; setting up a dynamic cross-coupling Jacobian matrix, triggering matrix update by using impedance real part signal change rate, and establishing a prediction model to quantify the interference of control variables on hydrogen diffusion flux; inputting the state tensor into the prediction model to deduce evolution trajectory, and obtaining optimal control increment sequence by solving the target optimization function; and then collaboratively adjusting gas flow, bias voltage, stirring speed and vibration frequency. The present application realizes global optimization of melt electrochemistry and fluid dynamics state, eliminates system oscillation caused by single parameter adjustment, and improves the adaptability and quality stability of the degassing process.

Description

Technical Field

[0001] This invention relates to the field of industrial process automatic control technology, and in particular to a multi-parameter collaborative control method for online degassing of aluminum alloy melt. Background Technology

[0002] Online degassing of aluminum alloy melts refers to the process of removing dissolved hydrogen from molten aluminum in real time on a continuous casting production line. The multi-parameter collaborative control method incorporates key process variables such as protective gas flow rate, rotary degassing rotor speed, melt temperature, and ladle conveying speed into a unified control framework. Based on the outlet hydrogen content detection data, the linkage between various parameters is dynamically adjusted so that the system can automatically optimize the operation combination according to the changes in melt composition and maintain stable degassing quality.

[0003] Existing online degassing control systems for aluminum alloy melts employ an independent adjustment architecture for key process parameters such as gas flow rate, stirring speed, melt temperature, and flow rate. Each control loop lacks a collaborative optimization mechanism and real-time feedback closed loop centered on the final hydrogen content quality index. In scenarios with dynamic changes in melt composition, the system cannot adaptively adjust parameter combinations. For example, when converting 6061 aluminum alloy to 7075 alloy, abrupt changes in silicon and copper content significantly alter hydrogen diffusion behavior. Traditional systems mechanically execute preset parameters, resulting in abnormally high melt hydrogen reabsorption rates or intensified local oxidation within the degassing chamber. Hydrogen content detection values ​​repeatedly exceed the process allowable range, leading to significant fluctuations in ingot porosity defects. The system cannot dynamically correct multi-variable linkages based on online hydrogen measurement results at the outlet, severely limiting production process adaptability and causing continuous deterioration in product quality stability. This fully exposes the shortcomings of the process control system's adaptive capabilities under complex metallurgical conditions. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a multi-parameter collaborative control method for online degassing of aluminum alloy melts. This invention solves the technical problems of poor production process adaptability, unstable degassing effect, and product quality fluctuation caused by the independent control of multiple key process parameters and the lack of a collaborative optimization and feedback mechanism based on the final degassing quality.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: The present invention provides a multi-parameter coordinated control method for online degassing of molten aluminum alloy, comprising: Step 1: Collect the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the aluminum alloy melt. Spatiotemporally align the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal. Then, splice the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal to construct a multi-field coupled state tensor. Step 2: Establish a dynamic cross-coupling Jacobian matrix, calculate the rate of change of the real part of the impedance signal, and use the rate of change of the real part of the impedance signal to trigger the update of the dynamic cross-coupling Jacobian matrix. Combine the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model. Use the updated dynamic cross-coupling Jacobian matrix to quantify the degree of interference of control variables on the hydrogen atom diffusion flux signal. The control variables include gas flow rate, bias voltage, stirring speed, and vibration frequency. Step 3: Input the multi-field coupled state tensor into the prediction model to deduce the system state evolution trajectory for the next multiple steps, use the system state evolution trajectory to calculate the objective optimization function, and solve the objective optimization function to obtain the optimal control increment sequence; Step 4: Adjust the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency according to the optimal control increment sequence.

[0006] Furthermore, the multi-parameter coordinated control method for online degassing of aluminum alloy melt described in this invention, wherein the spatiotemporal alignment of the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal includes: The kinematic viscosity characteristic value of the aluminum alloy melt and the constant proportionality coefficient related to the physical structure of the degassing chamber are obtained. The spatial phase compensation delay time constant is generated by performing multiplication and division algebraic operations on the constant proportionality coefficient, the stirring speed and the kinematic viscosity characteristic value. The spatial phase compensation delay time constant is used to perform time shift compensation on the collected electrochemical potential signal so that the electrochemical potential signal and the hydrogen atom diffusion flux signal are aligned in the time dimension.

[0007] Furthermore, the multi-parameter collaborative control method for online degassing of aluminum alloy melt described in this invention, wherein the step of constructing a multi-field coupled state tensor by splicing the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal includes: The aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the current sampling period are concatenated in space to form a one-dimensional column vector. A preset number of one-dimensional column vectors of historical continuous sampling periods are obtained. The one-dimensional column vector of the current sampling period and the one-dimensional column vectors of the historical continuous sampling periods are arranged sequentially according to the time sequence to construct the multi-field coupled state tensor.

[0008] Furthermore, the multi-parameter collaborative control method for online degassing of aluminum alloy melt described in this invention, wherein calculating the rate of change of the real part of the impedance signal and triggering the update of the dynamic cross-coupled Jacobian matrix using the rate of change of the real part of the impedance signal includes: The difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period is calculated as the rate of change of the real part of the impedance signal. When the difference value is negative and the absolute value of the difference value continuously exceeds a preset difference threshold, the recursive least squares algorithm with a forgetting factor is triggered to update the dynamic cross-coupled Jacobian matrix.

[0009] Furthermore, in the online degassing multi-parameter collaborative control method for aluminum alloy melt described in this invention, the recursive least squares algorithm carrying a forgetting factor is used to update the dynamic cross-coupled Jacobian matrix, including: The system acquires the input increments of the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency for historical sampling periods; acquires the state measurement output increment within the multi-field coupled state tensor; iteratively calculates the internal partial derivative elements of the dynamic cross-coupled Jacobian matrix using the recursive least squares algorithm in combination with the input increments and the state measurement output increments; and obtains the cross-interference values ​​generated by the stirring speed and the bias voltage on the hydrogen atom diffusion flux signal.

[0010] Furthermore, the multi-parameter collaborative control method for online degassing of aluminum alloy melt described in this invention, wherein the step of combining the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model includes: The system state transition constant matrix, which is a pre-identified representation of the inertial hysteresis characteristics inside the melt, is combined with the updated dynamic cross-coupling Jacobian matrix to establish a predictive state space equation and construct the predictive model. The stirring speed limit value and the bias voltage limit value of the aluminum alloy melt are obtained, and the stirring speed limit value and the bias voltage limit value are set as nonlinear hard constraints.

[0011] Furthermore, the multi-parameter collaborative control method for online degassing of aluminum alloy melt described in this invention, wherein the calculation of the objective optimization function using the system state evolution trajectory includes: The predicted state-space equation is used to deduce the system state evolution trajectory for multiple future steps. The predicted hydrogen flux and predicted electrochemical potential values ​​within the system state evolution trajectory are extracted and dimensionless normalization mapping is performed. The sum of squared errors between the normalized predicted hydrogen flux and the normalized target degassing rate flux is calculated to obtain the hydrogen flux deviation penalty value. The sum of squared errors between the normalized predicted electrochemical potential and the normalized electrochemical potential safety benchmark is calculated to obtain the electrochemical potential deviation penalty value. The sum of squared increments of the normalized control variables is calculated to obtain the control energy consumption penalty value. The hydrogen flux deviation penalty value, the electrochemical potential deviation penalty value, and the control energy consumption penalty value are added together to generate the target optimization function.

[0012] Furthermore, in the online degassing multi-parameter coordinated control method for aluminum alloy melt described in this invention, solving the objective optimization function to obtain the optimal control increment sequence includes: The objective optimization function and the nonlinear hard constraint are combined to establish a quadratic programming subproblem. The second-order partial derivative of the objective optimization function is calculated using the finite difference method to construct the Hessian matrix. The gradient search direction of the control variable is calculated using the Hessian matrix. The optimal control increment sequence that minimizes the value of the objective optimization function is obtained by iterative optimization along the gradient search direction using the interior point method. The current sampling period increment command is extracted from the optimal control increment sequence.

[0013] Furthermore, the multi-parameter coordinated control method for online degassing of aluminum alloy melt according to the present invention, wherein adjusting the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency according to the optimal control increment sequence includes: The incremental command of the current sampling period is parsed to obtain the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. The incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency are added to the actual control values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency of the previous sampling period to generate an absolute drive command sequence including the target values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency.

[0014] Furthermore, the multi-parameter coordinated control method for online degassing of aluminum alloy melt according to the present invention, wherein adjusting the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency includes: The gas flow rate is adjusted according to the target gas flow rate value in the absolute drive command sequence, the bias voltage target value in the absolute drive command sequence is converted into a duty cycle signal to adjust the bias voltage, the stirring speed target value in the absolute drive command sequence is converted into an analog voltage signal to adjust the stirring speed, and the vibration frequency target value in the absolute drive command sequence is converted into a frequency-hopping sinusoidal excitation signal to adjust the vibration frequency.

[0015] Beneficial effects of this invention: This invention presents a multi-parameter collaborative control method for online degassing of aluminum alloy melts. Through the synchronous acquisition of multi-source spatiotemporal characteristic data and the construction of a state tensor, it utilizes a spatial phase compensation delay time constant to eliminate signal transmission lag caused by physical position deviations of different sensors. This achieves precise alignment of the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal in the time dimension, providing a real-time and consistent data foundation for subsequent multivariate coupled calculations. By establishing a dynamic cross-coupling Jacobian matrix and introducing a recursive least squares algorithm carrying a forgetting factor, the system can identify online the drastic fluctuations in melt kinematic viscosity caused by alloy grade switching or eutectic phase transformation. It dynamically quantifies the cross-interference values ​​of stirring speed and bias voltage on hydrogen atom diffusion flux, overcoming the physical interference between loops caused by the single-variable independent control of traditional systems, and improving the flexibility and accuracy of process parameter adjustment. Based on a multi-objective collaborative prediction and optimization mechanism, this invention utilizes a predictive model to deduce the system state evolution trajectory over multiple future steps. Under nonlinear hard constraints such as the limit values ​​of stirring speed and bias voltage, the globally optimal control increment sequence is obtained by solving an objective optimization function that includes penalties for hydrogen flux deviation, electrochemical potential deviation, and control energy consumption. This achieves comprehensive optimization among degassing efficiency, anti-oxidation stability, and system energy consumption. The dynamic reconfiguration and collaborative drive mechanism of the underlying actuator transforms the optimal control increment sequence into a specific absolute drive command sequence, synchronously adjusting gas flow rate, bias voltage, stirring speed, and vibration frequency. This eliminates system oscillations and quality fluctuations caused by single-parameter adjustments, significantly enhancing the adaptability of the online degassing process under complex metallurgical conditions and maintaining the absolute stability of aluminum alloy degassing quality. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the multi-parameter coordinated control method for online degassing of aluminum alloy melt according to the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] Please see Figure 1 The present invention provides a multi-parameter coordinated control method for online degassing of aluminum alloy melt, comprising: Step 1: Collect the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the aluminum alloy melt. Spatiotemporally align the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal. Then, splice the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal to construct a multi-field coupled state tensor. Step 2: Establish a dynamic cross-coupling Jacobian matrix, calculate the rate of change of the real part of the impedance signal, and use the rate of change of the real part of the impedance signal to trigger the update of the dynamic cross-coupling Jacobian matrix. Combine the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model. Use the updated dynamic cross-coupling Jacobian matrix to quantify the degree of interference of control variables on the hydrogen atom diffusion flux signal. The control variables include gas flow rate, bias voltage, stirring speed, and vibration frequency. Step 3: Input the multi-field coupled state tensor into the prediction model to deduce the system state evolution trajectory for the next multiple steps, use the system state evolution trajectory to calculate the objective optimization function, and solve the objective optimization function to obtain the optimal control increment sequence; Step 4: Adjust the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency according to the optimal control increment sequence.

[0020] In the application scenario of online degassing process control technology for aluminum alloy melt, the underlying sensor network continuously outputs a large amount of analog data characterizing the metallurgical physical state. The control system needs to perform digital conversion processing on the acquired heterogeneous physical signals. The control system acquires the hydrogen atom diffusion flux signal, electrochemical potential signal, and real part impedance signal of the aluminum alloy melt. The hydrogen atom diffusion flux signal is represented by a microampere-level current data sequence reflecting the microscopic physical rate of hydrogen evolution inside the melt. The electrochemical potential signal is represented by a millivolt-level voltage data sequence characterizing the intensity of the interface redox reaction. The real part impedance signal is represented by a high-frequency AC resistance data sequence indicating the evolution of the internal liquid-solid structure of the crystal phase. Because there is a physical spatial installation distance between the probes measuring the hydrogen atom diffusion flux signal and the electrochemical potential signal inside the degassing chamber, there is a time difference due to fluid physics transmission when the data from different physical fields are transmitted to the control system. A time delay compensation calculation logic is established within the control system to obtain the kinematic viscosity characteristic value of the aluminum alloy melt and the stirring speed of the current sampling period. The control system generates a spatial phase compensation delay time constant by performing algebraic operations of multiplication and division using a constant proportionality coefficient, the stirring speed, and the kinematic viscosity characteristic value. The control system then uses this spatial phase compensation delay time constant to perform time-shift compensation on the acquired electrochemical potential signal, ensuring that the time-shifted electrochemical potential signal and the hydrogen atom diffusion flux signal are perfectly aligned on the same physical time slice. The control system concatenates the aligned hydrogen atom diffusion flux signal, the electrochemical potential signal, and the real part of the impedance signal for the current sampling period into a one-dimensional column vector within a data matrix. The control system retrieves a set number of historical continuous sampling periods' one-dimensional column vectors from memory. The control system then sequentially arranges and concatenates the one-dimensional column vectors of the current sampling period and the historical continuous sampling periods according to the time sequence to construct a multi-field coupled state tensor.

[0021] In complex melt degassing calculation environments with strong multivariate coupling, conventional independent proportional-integral-derivative (PI-DE) control rules are prone to mutual interference between control variables, leading to system calculation divergence and numerical oscillations. The control system establishes a dynamic cross-coupling Jacobian matrix in the memory address space. The internal elements of this dynamic cross-coupling Jacobian matrix are composed of the partial derivatives of the state variables with respect to the input variables. In real-world business scenarios involving continuous casting alloy grade switching, such as when a factory switches its production line from producing 6-series aluminum alloys to 7-series aluminum alloys, abrupt changes in the silicon and copper content within the aluminum alloy melt cause drastic physical fluctuations in melt viscosity. The control system calculates in real-time the difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period as the rate of change of the real part of the impedance signal. When the difference value shows a negative value drop and continuously exceeds a preset difference threshold, it indicates that the aluminum alloy melt is about to undergo a eutectic phase transition. This eutectic phase transition precursor causes the rate of change of the real part of the impedance signal to meet the algorithm trigger condition. The control system uses the rate of change of the real part of the impedance signal to trigger a recursive least squares algorithm carrying a forgetting factor to update the dynamic cross-coupling Jacobian matrix. The control system acquires input increment data of gas flow rate, bias voltage, stirring speed, and vibration frequency from historical sampling periods. The control system acquires state measurement output increment data within the multi-field coupled state tensor. The control system uses the recursive least squares algorithm to iteratively calculate the internal partial derivative elements of the dynamic cross-coupling Jacobian matrix by combining the input increment data and the state measurement output increment data. A prediction model is established by combining the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix. The control system uses the updated dynamic cross-coupling Jacobian matrix to quantify the interference degree of control variables on the hydrogen atom diffusion flux signal; these control variables include the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency.

[0022] To avoid microporosity defects within the cast grains caused by relying on current data feedback for delayed control calculations, the computation module uses a discrete state-space transition equation matrix deployed in the microprocessor as the prediction model. The control system inputs multi-field coupled state tensors into the prediction model and performs forward difference operations based on the assumed future multi-step control sequence to deduce the system state evolution trajectory for the next few steps. Simultaneously, the logic unit acquires the stirring speed limit and bias voltage limit values ​​of the aluminum alloy melt and sets these limits as nonlinear hard constraint physical boundaries. After acquiring the system state evolution trajectory, the processor extracts the predicted hydrogen flux and predicted electrochemical potential values, calculates the difference between the predicted hydrogen flux and the target degassing rate flux to obtain the hydrogen flux deviation penalty value, and simultaneously calculates the difference between the system state evolution trajectory and the electrochemical potential safety benchmark to obtain the electrochemical potential deviation penalty value. Finally, it calculates the control energy consumption penalty value by calculating the sum of squared increments of the control variables. Subsequently, the algorithm module adds the hydrogen flux deviation penalty, electrochemical potential deviation penalty, and control energy consumption penalty to generate the objective optimization function, and then merges the objective optimization function with the nonlinear hard constraints to establish a quadratic programming subproblem. The underlying solver uses the Hessian matrix to approximate the gradient search direction of the control variables, and uses the interior point method to iteratively search for optimization along the gradient search direction in the multidimensional solution space, thereby obtaining the optimal control increment sequence that minimizes the objective optimization function.

[0023] After obtaining the optimal control increment sequence in the global dimension, the control system extracts the current sampling period increment command from the optimal control increment sequence, performs data parsing processing on the current sampling period increment command, and independently extracts the discrete gas flow rate increment value, bias voltage increment value, stirring speed increment value, and vibration frequency increment value. The arithmetic logic unit algebraically adds the gas flow rate increment value, bias voltage increment value, stirring speed increment value, and vibration frequency increment value to the actual control values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency in the previous sampling period, respectively. After the algebraic addition operation, the command generation module constructs an absolute drive command sequence including the target values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. To achieve synchronous adjustment of physical equipment, the communication drive component outputs a standard analog current control data packet according to the gas flow rate target value in the absolute drive command sequence, adjusts the proportional valve, and thus adjusts the gas flow rate; at the same time, it converts the bias voltage target value in the absolute drive command sequence into a duty cycle communication signal sent by the control board, adjusts the high-power power supply output, and thus changes the bias voltage. In addition, the digital-to-analog converter circuit converts the target value of stirring speed in the absolute drive command sequence into an analog voltage control data packet, adjusts the operating frequency of the servo frequency converter, and thus changes the stirring speed; finally, the programmable logic device converts the target value of vibration frequency in the absolute drive command sequence into a frequency hopping control command generated by the direct digital frequency synthesizer, adjusts the piezoelectric sound generator, and thus changes the vibration frequency.

[0024] The control system needs to process asynchronous data streams collected by sensors at different locations. The control system acquires the kinematic viscosity characteristic values ​​of the aluminum alloy melt, which are specifically a digital representation of the physical quantity reflecting the internal frictional resistance of the molten metal. Considering the spatial distribution differences of the flow field inside the degassing chamber, algebraic operations of multiplication and division are performed using a constant proportionality coefficient, stirring speed, and kinematic viscosity characteristic values ​​to generate a spatial phase compensation delay time constant with time dimensions. The control system uses this spatial phase compensation delay time constant to perform time-shift compensation on the acquired electrochemical potential signal. In the actual production scenario of casting high-viscosity VII series aluminum alloys, this translation compensation operation can eliminate the signal transmission time difference caused by different physical installation positions of the sensors, ensuring that the electrochemical potential signal after translation compensation is completely aligned with the captured hydrogen atom diffusion flux signal in the same physical time dimension.

[0025] The aligned multi-source physical signals need to be structurally recombined to meet the input format requirements of high-dimensional matrix operations. The control system concatenates the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the current sampling period in the spatial dimension, splicing them into a one-dimensional column vector with three-dimensional features. In order to capture the dynamic temporal evolution of the degassing state inside the melt, the control system extracts a preset number of one-dimensional column vectors of historical continuous sampling periods from the data buffer, and the system scheduler arranges the one-dimensional column vector of the current sampling period and the one-dimensional column vectors of the historical continuous sampling periods sequentially according to the time sequence. The control system pre-configures the specific value of the preset number as 19, so that the spliced ​​multi-field coupled state tensor strictly contains the feature data of 20 consecutive physical sampling periods in the time dimension, thereby providing an initial state data input matrix with a 20-row, 3-column deterministic dimensional boundary for predicting the state space equation. In the actual physical scenario of tracking the trend of aluminum melt degassing quality fluctuations, the sequential arrangement operation expands the discrete single-point time vector into a multi-field coupled state tensor containing the time series span feature.

[0026] Sudden changes in the internal physical state of molten aluminum alloy are often accompanied by abnormal fluctuations in high-frequency electrical signals. The control system calculates the difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period by executing subtraction instructions. The data comparator inside the system uses this difference value as the rate of change of the real part of the impedance signal. In scenarios where a local temperature drop in the melt leads to a precursor to a eutectic phase transition from liquid to solid state, the microscopic changes in the crystal structure will cause an increase in the negative drop value of the impedance characteristic. When the difference value exceeds a preset difference threshold for multiple consecutive instruction cycles, the logic judgment of continuous threshold exceedance can filter out occasional electromagnetic interference noise signals and trigger the recursive least squares algorithm with a forgetting factor in the underlying controller to update the matrix parameters of the dynamically cross-coupled Jacobian matrix.

[0027] The matrix parameter update process requires extracting input and output behavior feature data of the system over a past period. The control system obtains the input increments of gas flow rate, bias voltage, stirring speed, and vibration frequency for historical sampling periods from the historical data stack. These input increments specifically represent the magnitude of action changes of the actuator within adjacent periods. The control system simultaneously acquires the state measurement output increments within the multi-field coupled state tensor. The recursive least squares algorithm combines the input increments and state measurement output increments to perform multiple iterative calculations in the matrix space, refreshing the internal partial derivative elements of the dynamic cross-coupling Jacobian matrix. In degassing control scenarios where different process parameters physically interfere with each other, the iterative calculation operation can quantify and analyze the cross-interference values ​​of the hydrogen atom diffusion flux signal caused by changes in the stirring speed of the variable frequency motor and the bias voltage of the graphite electrode.

[0028] After obtaining the quantified cross-interference characteristics, the system needs to establish a mathematical computational environment for future trend prediction. The control system uses the updated dynamic cross-coupling Jacobian matrix to establish a predictive state-space equation reflecting the inertial hysteresis characteristics inside the melt, and uses the predictive state-space equation to construct a predictive model for forward-looking logic calculations. Mechanical equipment and electrical components in industrial settings have objective physical limits. The control system obtains the stirring speed limit and bias voltage limit values ​​for the aluminum alloy melt. To prevent excessively high stirring speeds from causing vortices on the melt surface that draw in external air, and to prevent excessively high bias voltages from causing localized electrochemical oxidation and breakdown, the control system forcibly sets the stirring speed limit and bias voltage limit values ​​as nonlinear hard constraints within the solution space of the control algorithm.

[0029] The core operational logic of the control system lies in finding the optimal coordinated control command with the lowest cost through look-ahead prediction. The control system uses predictive state-space equations combined with a pre-set hypothetical input control sequence to deduce the system state evolution trajectory over multiple future steps. Specifically, the system state evolution trajectory is represented as a multi-dimensional data matrix containing predicted hydrogen flux and potential values ​​for multiple future time points. Subsequently, the computational unit calculates the numerical difference between the system state evolution trajectory and the target degassing rate flux to obtain the hydrogen flux deviation penalty value, and simultaneously calculates the numerical difference between the system state evolution trajectory and the electrochemical potential safety benchmark to obtain the electrochemical potential deviation penalty value. To avoid excessive consumption of industrial electrical energy due to frequent and violent actions of the actuators, the underlying logic calculates the sum of squared increments of the control variables to obtain the control energy consumption penalty value. Finally, the central processing unit performs mathematical addition and merging operations on the hydrogen flux deviation penalty value, the electrochemical potential deviation penalty value, and the control energy consumption penalty value to generate the target optimization function.

[0030] After constructing the multi-objective evaluation system, the central processing unit (CPU) needs to perform nonlinear programming optimization calculations. The control system logically merges the objective optimization function with the nonlinear hard constraints to establish a quadratic programming subproblem mathematical model. The mathematical solver uses the Hessian matrix to evaluate the curvature change characteristics of the objective optimization function in multidimensional space and calculates the gradient search direction for the control variables to approximate the global optimum. In the millisecond-level online degassing coordinated control scenario with extremely high real-time requirements, the control system uses the interior point method to perform multi-step iterative optimization calculations along the gradient search direction within the constraint boundary. The iterative optimization calculations can output a set of optimal control increment sequences that minimize the value of the objective optimization function. The system instruction extraction module performs time-dimensional data slicing on the optimal control increment sequence and extracts the current sampling period increment instruction within the optimal control increment sequence.

[0031] The relative change commands output by the controller need to be converted into absolute target position data recognizable by the underlying hardware driver. The control system performs data parsing and bit manipulation separation processing on the incremental commands of the current sampling period, independently acquiring the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. The arithmetic logic unit of the control system performs a superposition and addition operation on the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency with the actual control values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency stored in registers from the previous sampling period. In scenarios where the algorithm output commands are sent to the fieldbus communication network, the addition arithmetic operation can generate an absolute drive command sequence including the target values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency.

[0032] The final generated absolute drive command sequence needs to be mapped into electrical control signals for specific physical devices via digital-to-analog conversion. The control system outputs digital control words based on the target gas flow rate value within the absolute drive command sequence to adjust the valve opening of the digital mass flow meter, thereby regulating the gas flow rate. The communication conversion module converts the target bias voltage value within the absolute drive command sequence into a duty cycle signal under a pulse width modulation protocol and outputs it to a high-power DC power supply to adjust the bias voltage. The digital-to-analog conversion circuit converts the target stirring speed value within the absolute drive command sequence into an analog voltage signal within a specific range and sends it to the servo inverter to adjust the stirring speed of the rotary motor. The programmable logic device converts the target vibration frequency value within the absolute drive command sequence into a continuously varying frequency-hopping sinusoidal excitation signal to adjust the piezoelectric ceramic oscillator to generate high-frequency mechanical waves and regulate the vibration frequency.

[0033] The hydrogen atom diffusion flux signal is essentially an electrical physical quantity detected in situ by a miniature sensor installed in the core region of the melt within the degassing chamber. Specifically, it manifests as a continuous, weak DC current sequence at the nanoampere or microampere level. The current amplitude is linearly or nonlinearly mapped to the number of hydrogen atoms microscopically removed per unit time through a specific cross-sectional area of ​​the sensor. The electrochemical potential signal is a thermodynamic state characteristic measured relative to a reference electrode array integrated on the sidewall of the refractory material in the degassing chamber. Specifically, it is represented by a millivolt-level voltage waveform reflecting the strength of the redox reaction at the aluminum melt and electrode interface. The fluctuation spectrum of the voltage waveform can indicate whether there is localized severe oxidation or microscopic segregation of chemical components within the melt. The real part of the impedance signal is captured by a broadband impedance spectrum monitoring circuit composed of piezoelectric ceramic elements arranged at the tail of the degassing rotor. Specifically, it is the resistance component representing active power loss in the equivalent circuit model of the melt under a specific AC sweep frequency excitation. The step or abrupt change in the resistance component value is often closely related to the liquid-solid phase transition or latent heat release process occurring within the melt.

[0034] In this scheme, the control variables are explicitly defined as four independent execution parameters that directly participate in regulating the physical field state. Gas flow rate refers to the volumetric flow rate of inert protective gases such as argon or nitrogen injected quantitatively into the degassing rotor via a digital mass flow controller, specifically expressed as liters per minute under standard conditions. Bias voltage is the minute potential difference applied by an externally programmable DC power supply between the graphite electrode and the melt surface, specifically expressed as a DC level value in the range of a few tenths of a volt to several volts, mainly used to construct a polarized electric field at the melt interface to suppress hydrogen reabsorption. Stirring speed is controlled by a servo frequency converter and represents the speed at which the rotating degassing rotor cuts the fluid in the melt to generate shear force, specifically expressed as the number of revolutions per minute. Vibration frequency is the speed of ultrasonic mechanical oscillation generated by the piezoelectric ceramic generator, specifically expressed as a high-frequency digital characteristic with continuous or stepped changes within the kilohertz frequency band.

[0035] The multi-field coupled state tensor is a high-dimensional data structure formed by rearranging and combining heterogeneous multi-source physical signals after spatiotemporal delay compensation. At the algorithmic level, the multi-field coupled state tensor is represented as a mathematical matrix containing time and spatial dimensions. The row vectors of the matrix store the aligned values ​​of hydrogen atom diffusion flux, electrochemical potential, and real part of impedance for the current sampling period and multiple historical sampling periods. The matrix structure completely encapsulates the dynamic evolution trajectory of the melt degassing process within a continuous physical time slice, providing initial boundary condition inputs with time-series characteristics for the subsequent construction of forward-looking state extrapolation algorithms.

[0036] The dynamic cross-coupling Jacobian matrix is ​​a mathematical model built within the controller to characterize the local linearization features of a multivariable, strongly coupled system. The elements of the Jacobian matrix are all partial derivatives calculated from the partial derivatives of each physical state variable with respect to control variables such as gas flow rate, bias voltage, stirring speed, and vibration frequency. These partial derivative values ​​intuitively quantify the extent to which a small disturbance in a single control actuator, under a specific metallurgical operating condition, will couple and cause numerical shifts in several other cross-physical states. This provides key gain adjustment weights for decoupling control algorithms to eliminate mutual physical interference between multiple loops.

[0037] The recursive least squares algorithm is deployed in the underlying arithmetic unit of the control system as the core engine for adaptive parameter identification. Because the kinematic viscosity of molten aluminum undergoes nonlinear and dramatic changes during phase transformations or abrupt changes in alloy composition, conventional fixed-parameter models become instantly invalid. The recursive least squares algorithm introduces a forgetting factor between zero and one to continuously input and process a series of incremental control input data and state measurement output data from the past. In the multidimensional parameter space, the recursive least squares algorithm iteratively approximates and updates the internal partial derivative elements of the dynamically cross-coupled Jacobian matrix online, ensuring that the mathematical model always closely follows and conforms to the actual physical fluid dynamics of the melt.

[0038] The system state evolution trajectory is a set of estimated data points generated by forward rolling the predicted state-space equations towards the future time axis. When the controller injects the hypothetical control increment sequence into the updated prediction model in the internal virtual environment, the model simulates the expected waveforms of hydrogen atom diffusion flux and electrochemical potential over the next tens or hundreds of control cycles. The prediction algorithm, by predicting the future system state evolution trajectory in advance, compares the system state evolution trajectory with the ideal process degassing rate and safety anti-oxidation benchmark, using this as a basis for decision-making to adjust the optimal control increment sequence in advance.

[0039] The objective optimization function is a mathematical scalar expression that comprehensively measures various control costs and deviation penalties. Internally, the objective optimization function incorporates three weighted penalty terms: a hydrogen flux deviation penalty to assess the severity of the predicted degassing effect deviating from the target process requirements; an electrochemical potential deviation penalty to assess the probability of irreversible oxidation side reactions occurring on the melt surface; and a control energy consumption penalty, which, by calculating the sum of squares of the changes in control variables, constrains excessively frequent or drastic unnecessary physical adjustments by the actuator. The controller solves a quadratic programming subproblem to find the data point that minimizes the sum of the objective optimization function, thereby calculating the globally optimal control strategy among the three mutually constraining physical objectives of degassing efficiency, melt quality protection, and equipment energy consumption. The multi-parameter collaborative control method for online degassing of aluminum alloy melt is not a purely abstract mathematical derivation and calculation rule divorced from the actual physical environment. Instead, it maps the objective laws of metallurgical thermodynamics and fluid dynamics in the online degassing process of aluminum alloy melt into mathematical matrix characteristics in a discrete state space. The construction of the multi-field coupled state tensor and the solution of the objective optimization function essentially seek a control strategy that minimizes the physical interference energy cost under the objective physical boundary constraints of the industrial equipment's operating limits, based on real physical feedback data collected by multi-dimensional physical sensors. The control system then converts the optimal control increment sequence calculated and output in the digital space into... This is transformed into a sequence of absolute drive commands for specific electrical equipment, which directly alters the protective gas flow rate, DC polarization electric field strength, mechanical shear rate, and ultrasonic cavitation sound field distribution within the degassing chamber. This enables precise physical interference with the removal of hydrogen atoms and the movement of microbubbles within the aluminum alloy melt. From the hardware acquisition of multi-source heterogeneous physical signals and the derivation of mathematical spatial matrix algorithms to the driving of underlying physical electrical components, a complete technical closed loop is constructed. By employing technical means that follow objective natural laws, this effectively solves the technical problem of fluctuating degassing effects caused by sudden changes in complex metallurgical fluid operating conditions.

[0040] On a continuous casting production line, molten aluminum alloy continuously flows through a degassing chamber, and the underlying sensor network continuously outputs a large amount of analog data characterizing the metallurgical physical state. The control system needs to digitally convert the acquired heterogeneous physical signals. The control system acquires the hydrogen atom diffusion flux signal, electrochemical potential signal, and real part impedance signal of the molten aluminum alloy. The hydrogen atom diffusion flux signal is represented as a current data sequence reflecting the microscopic physical rate of hydrogen evolution inside the melt. The electrochemical potential signal is represented as a voltage data sequence characterizing the intensity of the interfacial redox reaction. The real part impedance signal is represented as an AC resistance data sequence indicating the evolution of the internal liquid-solid structure of the crystal phase. Due to the physical spatial installation distance of the measuring probes inside the degassing chamber, there is a time difference in the transmission of data from different physical fields to the control system due to fluid physics. To eliminate this time difference, the control system acquires the kinematic viscosity characteristic value of the molten aluminum alloy. The kinematic viscosity characteristic value is a digital representation of a physical quantity reflecting the internal frictional resistance of the molten metal fluid. A spatial phase compensation delay time constant with time dimension is generated by performing algebraic operations of multiplication and division using a constant proportionality coefficient, stirring speed, and the kinematic viscosity characteristic value. The control system uses this spatial phase compensation delay time constant to perform time shift compensation on the acquired electrochemical potential signal, aligning the electrochemical potential signal with the hydrogen atom diffusion flux signal in the time dimension. The system scheduler concatenates the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the current sampling period into a one-dimensional column vector in the spatial dimension, and extracts a preset number of one-dimensional column vectors from historical continuous sampling periods in the data buffer. Subsequently, the control system arranges the one-dimensional column vector of the current sampling period and the one-dimensional column vectors of the historical continuous sampling periods sequentially according to the time sequence, thereby constructing a multi-field coupled state tensor reflecting the spatiotemporal evolution law of the melt physical state. The preset number is parameterized to 19 to ensure the physical consistency of data memory allocation and matrix iteration operation dimensions.

[0041] In actual production scenarios where a factory production line switches from producing VI-series aluminum alloys to producing VII-series aluminum alloys, abrupt changes in the silicon and copper content within the molten aluminum alloy can cause drastic physical fluctuations in the melt's kinematic viscosity. Conventional independent proportional-integral-derivative (PID) control methods are prone to mutual interference between control variables, leading to system calculation divergence and numerical oscillations. The control system establishes a dynamic cross-coupling Jacobian matrix in the memory address space. To capture the dynamic coupling relationship caused by changes in alloy composition, the control system calculates the difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period by executing subtraction instructions. The internal data comparator uses the difference value as the rate of change of the real part of the impedance signal. In scenarios where a local temperature drop in the melt triggers a precursory eutectic phase transition from liquid to solid in the alloy, microscopic changes in the crystal structure lead to a decrease in the impedance value, i.e., the difference value becomes negative and its absolute value increases. When the difference value is negative and its absolute value exceeds a preset difference threshold for multiple consecutive instruction cycles, the logic judgment of continuously exceeding the preset difference threshold can filter out occasional electromagnetic interference noise signals, thereby triggering a recursive least squares algorithm with a forgetting factor in the underlying controller to update the dynamic cross-coupling Jacobian matrix. At this time, the control system obtains the input increments of gas flow rate, bias voltage, stirring speed, and vibration frequency from the historical data stack for the historical sampling period. The control system simultaneously obtains the state measurement output increments within the multi-field coupled state tensor. The recursive least squares algorithm combines the input increments and the state measurement output increments to perform multiple iterative calculations in the matrix space, refreshing the internal partial derivative elements of the dynamic cross-coupling Jacobian matrix. A predictive model is established by combining the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix. The control system uses the updated dynamic cross-coupling Jacobian matrix to quantify the interference of control variables on the hydrogen atom diffusion flux signal. The control variables include the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency. The iterative calculation operation can quantify and analyze the cross-interference values ​​of the stirring speed and the bias voltage on the hydrogen atom diffusion flux signal.

[0042] After obtaining the quantified cross-interference characteristics, relying solely on the current data feedback for hysteresis control calculations is insufficient to avoid the generation of microporosity defects within the cast grains. The control system utilizes the updated dynamic cross-coupling Jacobian matrix to establish a predictive state-space equation reflecting the inertial hysteresis characteristics within the melt, thereby constructing a predictive model for forward-looking logic calculations. Mechanical equipment and electrical components in industrial settings face objective physical constraints. The control system acquires the stirring speed limit and bias voltage limit values ​​for the aluminum alloy melt. To prevent excessively high stirring speeds from causing vortices that draw in external air from the melt surface and to prevent excessively high bias voltages from causing localized electrochemical oxidation breakdown, the control system sets the stirring speed limit and bias voltage limit values ​​as nonlinear hard constraints within the solution space of the control algorithm. The control system uses the predictive state-space equations combined with a preset hypothetical input control sequence to deduce the system state evolution trajectory for multiple future steps. The control system extracts the predicted hydrogen flux value and predicted electrochemical potential value within the system state evolution trajectory, calculates the (numerical) difference between the predicted hydrogen flux value and the target degassing rate flux, and obtains the hydrogen flux deviation penalty value. The system's processing unit calculates the numerical difference between the predicted electrochemical potential and the electrochemical potential safety benchmark to obtain the electrochemical potential deviation penalty value. To avoid excessive consumption of industrial electrical energy due to frequent and violent actions of the actuator, the control system calculates the sum of squares of the increments of the control variables to obtain the control energy consumption penalty value. The control system performs mathematical addition and merging operations on the hydrogen flux deviation penalty value, the electrochemical potential deviation penalty value, and the control energy consumption penalty value to generate the objective optimization function. The control system merges the objective optimization function with the nonlinear hard constraint condition to establish a quadratic programming subproblem. The system uses the Hessian matrix to evaluate the curvature change characteristics of the objective optimization function in multidimensional space and calculates the gradient search direction for the control variables to approach the optimal solution. The control system solves the objective optimization function and uses the interior point method to perform multi-step iterative optimization along the gradient search direction within the constraint boundary to obtain the optimal control increment sequence that minimizes the value of the objective optimization function. The control system extracts the current sampling period increment command within the optimal control increment sequence.

[0043] The relative change command output by the controller needs to be converted into absolute target position data executable by the underlying hardware driver. The control system performs data parsing and bit manipulation separation processing on the incremental command of the current sampling period to independently obtain the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. The arithmetic logic unit of the control system performs arithmetic operations on the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency with the actual control values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency of the previous sampling period stored in the register. The arithmetic operation generates an absolute drive command sequence including the target values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. The generated absolute drive command sequence needs to be mapped to the electrical control signals of the physical device through digital-to-analog conversion to adjust the gas flow rate, bias voltage, stirring speed, and vibration frequency. The control system outputs a digital control word according to the target value of gas flow rate in the absolute drive command sequence to adjust the valve opening of the digital mass flow meter, thereby adjusting the gas flow rate. The communication conversion module converts the target bias voltage value in the absolute drive command sequence into a duty cycle signal under a pulse width modulation protocol and outputs it to a high-power DC power supply to adjust the bias voltage. The digital-to-analog converter (DAC) converts the target stirring speed value in the absolute drive command sequence into an analog voltage signal within a specific range and sends it to the servo frequency converter to adjust the rotation speed of the rotary motor. The programmable logic device (PLD) extracts the target vibration frequency value in the absolute drive command sequence and inputs it to the direct digital frequency synthesizer (DFD) circuit module integrated on the control motherboard. The DFD circuit module, in conjunction with the DAC chip, converts the target vibration frequency value into a continuously changing frequency-hopping sine wave signal. The analog amplification drive circuit performs a power amplification operation on the continuously changing frequency-hopping sine wave signal to generate a continuously changing frequency-hopping sine excitation signal. The analog amplification drive circuit sends the continuously changing frequency-hopping sine excitation signal to the piezoelectric sound generator to adjust the vibration frequency. After the optimal control increment sequence is converted into an absolute drive command sequence and sent to the underlying hardware, each execution unit begins a physical-level coordinated response. After receiving a digital control word representing the target gas flow rate, the microcontroller chip of the digital mass flow meter drives an internal stepper motor to adjust the flow cross-sectional area of ​​the throttling valve core, changing the flow rate of the protective gas entering the rotating degassing rotor, thereby regulating the number and size of the fine bubbles dispersed in the melt. Upon receiving the duty cycle signal, the high-power DC power supply's switching transistor array, by changing the on-time ratio of the high-frequency switching transistors, establishes a DC polarized electric field of specific strength between the graphite electrode immersed in the melt and the metal lining of the degassing chamber.The DC polarized electric field repels hydrogen ions adsorbed at the melt interface and inhibits the cathodic reduction reaction of moisture in the air, thus cutting off the path for hydrogen atoms to re-dissolve into the molten aluminum from a physicochemical perspective. After receiving the analog voltage signal, the servo frequency converter changes the frequency of the three-phase power supply output to the AC asynchronous motor, directly altering the mechanical rotational angular velocity of the degassing rotor. This change in mechanical rotational angular velocity causes a proportional change in the hydrodynamic shear force within the melt, shearing and breaking large bubbles into micron-sized, high-specific-surface-area microbubbles, improving the diffusion efficiency of hydrogen atoms into the bubble interior. Upon receiving the frequency-hopping sinusoidal excitation signal, the piezoelectric sound generator generates high-frequency mechanical resonance in its internal piezoelectric ceramic oscillator, radiating this high-frequency mechanical resonance as ultrasonic waves into the surrounding high-temperature molten aluminum. These ultrasonic waves induce a periodic acoustic cavitation effect within the melt, utilizing the micro-jets generated at the moment of cavitation bubble collapse to strip adsorbed hydrogen from the oxide inclusions, while simultaneously breaking down the microstructure of large molecular clusters in the molten aluminum, reducing the apparent kinematic viscosity within the molten aluminum, and accelerating the collision, floating, and separation of microbubbles.

[0044] In the application scenario of online degassing process control technology for aluminum alloy melt, the underlying sensor network continuously outputs a large amount of data characterizing the metallurgical physical state. The control system needs to perform precise time-dimensional compensation calculations on the collected heterogeneous physical signals. The control system collects temperature characteristic data of the aluminum alloy melt in real time through thermocouple sensors arranged inside the degassing chamber. The control system extracts aluminum alloy production grade data from external input terminals. Using the temperature characteristic data and the aluminum alloy production grade data, the control system performs interpolation matching addressing operations in a pre-configured metallurgical property association database in the non-volatile memory within the control system. The control system obtains the kinematic viscosity characteristic value of the aluminum alloy melt corresponding to the real-time operating conditions through the interpolation matching addressing operation. The control system synchronously collects the stirring speed of the current sampling period. The control system has a built-in constant proportionality coefficient related to the physical structure of the degassing chamber. The control system uses the constant proportionality coefficient, the stirring speed, and the kinematic viscosity characteristic value to perform multiplication and division algebraic operations to generate a spatial phase compensation delay time constant. The specific calculation formula for generating the spatial phase compensation delay time constant is as follows:

[0045] in, Represents the spatial phase compensation delay time constant. This represents the constant proportionality coefficient. This represents the numerical value of the kinematic viscosity characteristic. The stirring speed represents the current sampling period. The control system uses the calculated spatial phase compensation delay time constant to perform a reverse time shift operation on the time axis of the acquired electrochemical potential signal, eliminating data lag caused by the physical flow process of the melt, thus aligning the shifted and compensated electrochemical potential signal with the hydrogen atom diffusion flux signal acquired at the same physical moment in time. Faced with the nonlinear physical disturbances generated by the aluminum alloy melt during the phase transformation process, the control system needs to use discrete difference calculations to sensitively capture the abrupt change nodes of the melt's microstructure. The control system retrieves the real part of the impedance signal from the previous sampling period stored in the data register in real time. The control system calculates the difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period, and uses this difference as the rate of change of the real part of the impedance signal. The specific calculation formula for extracting the difference value is:

[0046] in, Represents the difference value. The impedance real part signal represents the current sampling period. This represents the real part of the impedance signal from the previous sampling period. When the difference value is negative, and the absolute value of the difference value exceeds a preset difference threshold for multiple consecutive calculations, the data processing logic determines that the melt is at the edge of a eutectic phase transition.

[0047] The physical field coupling relationships within the aluminum alloy melt in its pre-eutectic phase transition state are extremely unstable, leading to severe distortion in conventional static models. The control system triggers a recursive least squares algorithm with a forgetting factor to dynamically update the dynamic cross-coupling Jacobian matrix within the control core at high frequency. The control system acquires input increments of gas flow rate, bias voltage, stirring speed, and vibration frequency from historical sampling periods. Simultaneously, the control system acquires state measurement output increments within the multi-field coupling state tensor. The computation unit uses these input increments and state measurement output increments to iteratively calculate the covariance matrix and Kalman gain matrix in a multi-dimensional matrix space, thereby refreshing the dynamic cross-coupling Jacobian matrix. The arithmetic unit uses the input increment and the state measurement output increment to iteratively calculate the covariance matrix and Kalman gain matrix in a multidimensional matrix space, thereby refreshing the dynamic cross-coupling Jacobian matrix. Before the control system executes the recursive least squares algorithm, the arithmetic unit pre-sets the input and output data matrix structure and initial parameters of the algorithm according to the physical scenario of online degassing of aluminum alloy melt. Specifically, the control system constructs the input increment as a 4-row, 1-column input feature column vector, and the internal elements of the 4-row, 1-column input feature column vector are, in order, the control increment of gas flow rate, the control increment of bias voltage, the control increment of stirring speed, and the control increment of vibration frequency. At the same time, the control system constructs the state measurement output increment as a 3-row, 1-column output feature column vector, the internal elements of which are, in order, the measurement increment of hydrogen atom diffusion flux signal, the measurement increment of electrochemical potential signal, and the measurement increment of the real part of impedance signal. The computational unit sets the initial covariance matrix as the product of a 4x4 unit diagonal matrix and a preset maximal constant, where the specific value of the preset maximal constant is 10000. This data matrix structure establishes the inherent mathematical correlation mechanism between the action amplitude of the underlying physical actuators and the measurement changes of the multi-source physical sensors, ensuring the alignment of matrix multiplication and addition operations in the subsequent iterative equations in the mathematical dimension. The specific algorithmic equations for updating the dynamic cross-coupling Jacobian matrix include:

[0048]

[0049]

[0050] in, The Kalman gain matrix represents the current sampling period. The covariance matrix representing the previous sampling period, Represents the input increment. Represents the forgetting factor, The transpose matrix representing the input increment. The covariance matrix representing the current sampling period, This represents the updated dynamic cross-coupling Jacobian matrix. The dynamic cross-coupling Jacobian matrix representing the previous sampling period. This represents the increment of the state measurement output. The transpose of the Kalman gain matrix for the current sampling period.

[0051] After obtaining the updated dynamic cross-coupling Jacobian matrix, the computation module needs to establish a forward-looking state extrapolation mechanism oriented towards the future physical timeline. The control system uses the updated dynamic cross-coupling Jacobian matrix to establish a predictive state-space equation and construct a predictive model. The control system inputs the multi-field coupled state tensor as the initial state into the predictive state-space equation, and, combined with a series of future control action sequences generated by internal system assumptions, recursively calculates the future multi-step system state evolution trajectory. The discrete state-space equation for extrapolating the system state evolution trajectory is:

[0052] in, Represents the prediction of the future number in the current sampling period. The system state evolution trajectory of the step, The system state transition constant matrix represents the inertial hysteresis characteristics within the melt. Represents the prediction of the future number in the current sampling period. The system state evolution trajectory of the step, This represents the updated dynamic cross-coupling Jacobian matrix. Represents the prediction of the future number in the current sampling period. The control system assumes a control increment sequence. When constructing the predicted state-space equations, the control system pre-defines the mathematical dimensions and element mapping relationships between the system state transition constant matrix and the dynamic cross-coupling Jacobian matrix based on the data structure of the multi-field coupled state tensor and control variables. Since the system state evolution trajectory includes three physical characteristic dimensions—hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal—the control system constructs the system state transition constant matrix as a 3x3 square matrix. The main diagonal elements of the 3x3 square matrix represent the autoregressive inertial hysteresis constant of a single physical state dimension, while the off-diagonal elements represent the internal coupling influence constants between the three physical state dimensions. Since the control variables include four physical execution dimensions—gas flow rate, bias voltage, stirring speed, and vibration frequency—the control system constructs the dynamic cross-coupling Jacobian matrix as a 3x4 mapping transformation matrix. Each column of the 3x4 mapping transformation matrix corresponds to the local cross-partial derivative values ​​of the single control variable with respect to the three independent physical characteristic dimensions in the same time slice. The aforementioned mathematical dimension settings establish the cross-connection relationship between the underlying physical execution quantities and the multi-source system state variables within the prediction model. This ensures that the algebraic operation dimensions are absolutely aligned when the predicted state space equations perform forward discrete iterative matrix multiplication and matrix addition operations, guaranteeing that the output system state evolution trajectory always maintains a 3x1 column vector structure. To select the optimal control strategy for degassing quality and equipment energy consumption from a massive number of possible control sequences, the control system constructs a multi-objective cost evaluation system within the microprocessor. The control system extracts the predicted hydrogen flux and predicted electrochemical potential values ​​within the system state evolution trajectory. The computational logic pre-processes the data from different physical dimensions using a maximum-minimum mapping rule to eliminate dimensional differences. During this normalization process, the control system pre-sets the multi-dimensional global boundary reference parameters required by the algorithm based on the physical properties of the online degassing process for aluminum alloy melt. Specifically, the control system sets the global minimum boundary reference parameter for the hydrogen flux dimension to 0 microamps and the global maximum boundary reference parameter for the hydrogen flux dimension to 200 microamps; the control system sets the global minimum boundary reference parameter for the electrochemical potential dimension to -1200 millivolts and the global maximum boundary reference parameter for the electrochemical potential dimension to 0 millivolts. The computational logic extracts the original value of the single dimension to be processed, performs algebraic subtraction by subtracting the corresponding global minimum boundary reference parameter from the original value of the single dimension to obtain a first algebraic difference, and performs algebraic subtraction by subtracting the corresponding global minimum boundary reference parameter from the global maximum boundary reference parameter of the same dimension to obtain a second algebraic difference. The computational logic divides the first algebraic difference by the second algebraic difference to calculate a dimensionless floating-point number distributed in the interval between 0 and 1, and outputs the dimensionless floating-point number as the feature input data of the algorithm model.The aforementioned data mapping steps and boundary parameter settings clarify the computational path for transforming multi-source heterogeneous physical quantities in the metallurgical substrate into the dimensionless digital space of the optimization algorithm. This demonstrates the inherent correlation between the algorithm's input raw data and output normalized data, avoiding numerical truncation and divergence caused by differences in the absolute magnitude of the data during the Hessian matrix differentiation operation in the quadratic programming algorithm. The computational logic calculates the squared error between the normalized predicted hydrogen flux and the normalized target degassing flux, assigning penalty weights accordingly. It also calculates the squared error between the normalized predicted electrochemical potential and the normalized electrochemical potential safety benchmark, assigning penalty weights as well. Simultaneously, it calculates the sum of the squares of the normalized control increment sequence action amplitudes as a penalty for equipment energy consumption. The control system performs a mathematical summation of all penalty terms to generate the target optimization function. The specific formula for calculating the target optimization function is: [Formula omitted].

[0053] in, Represents the objective optimization function. Represents the number of prediction time-domain steps. This represents the predicted hydrogen flux value. This represents the target degassing flux. The weight matrix representing the hydrogen flux deviation penalty is... This represents the predicted electrochemical potential value. This represents the electrochemical potential safety benchmark. Represents the electrochemical potential bias penalty weight matrix. Represents the number of control time-domain steps. This represents the control increment sequence. This represents the control energy consumption penalty weight matrix. When establishing and solving the objective optimization function, the control system pre-sets the hyperparameters of the prediction model and the structure of each penalty weight matrix based on the physical residence time characteristics of the fluid inside the online degassing chamber of the aluminum alloy melt. The control system sets the specific value of the prediction time-domain steps to 30 steps and the specific value of the control time-domain steps to 5 steps. Since both the hydrogen atom diffusion flux signal and the electrochemical potential signal are single-dimensional feature quantities, the computational logic sets both the hydrogen flux deviation penalty weight matrix and the electrochemical potential deviation penalty weight matrix as constant scalars. The control system sets the specific value of the hydrogen flux deviation penalty weight matrix to 0.6 and the specific value of the electrochemical potential deviation penalty weight matrix to 0.3. Since the control increment sequence includes four independent physical execution quantities in four dimensions—gas flow rate, bias voltage, stirring speed, and vibration frequency—the control system constructs the control energy consumption penalty weight matrix as a 4x4 positive definite diagonal matrix. The computational logic sets the diagonal elements of the 4x4 positive definite diagonal matrix to have values ​​of 0.05, 0.02, 0.01, and 0.02 respectively, and sets the off-diagonal elements to have values ​​of 0. These parameter and matrix structure settings establish an intrinsic correlation between the algorithm's prediction timescale and the actual physical response process of metallurgical fluid dynamics, ensuring the objective mathematical convergence of the nonlinear programming solver within a finite control period.

[0054] Since the extremum solution of the objective optimization function cannot be separated from the mechanical and electrical boundary constraints of the physical execution equipment, the control system obtains the stirring speed limit and bias voltage limit values ​​of the aluminum alloy melt, and sets these limits as nonlinear hard constraints. The control system merges the objective optimization function and the nonlinear hard constraints to establish a quadratic programming subproblem. The underlying solver uses the Hessian matrix to approximate the second-order partial derivative information of the objective function in local space, and calculates the gradient search direction for the control variable iterating in the direction of numerical decrease. The underlying approximate mathematical model for calculating the gradient search direction and the optimal solution is expressed as follows:

[0055] in, This represents the mathematical operation process of finding the minimum value of a sequence of control variables. The transpose of the first-order Jacobian gradient vector of the objective function is given by [matrix]. This represents the gradient search direction. The transpose matrix representing the gradient search direction. The Hessian matrix is ​​represented by the interior-point method. The control system performs multi-point iterative optimization calculations along the gradient search direction using the interior-point method, and finally obtains the optimal control increment sequence that makes the objective optimization function converge to the global minimum. The optimal control increment sequence is then sent to the underlying communication driver module according to the data bus protocol.

[0056] When the control system executes the recursive least squares algorithm, the microprocessor memory allocates a dedicated matrix operation buffer. The microprocessor extracts the covariance matrix of the previous sampling period and the currently acquired input increment, and uses the hardware multiply-accumulate unit to perform high-frequency multiplication and accumulation operations to calculate the Kalman gain matrix, which reflects the system error weights. After obtaining the Kalman gain matrix, the logic operation unit combines the difference between the state measurement output increment and the system's historical prediction value, and uses the Kalman gain matrix to perform step-by-step correction on the partial derivative elements of the dynamic cross-coupling Jacobian matrix. During the prediction time-domain extrapolation process, the prediction state-space equation serves as a discrete mathematical mapping model characterizing the internal state of the aluminum alloy melt. The controller, according to the preset sampling step size, uses the current actual initial physical state as the starting point for calculation, and continuously substitutes the hypothetical control commands generated in the sequence into the system state transition constant matrix for forward iterative accumulation, thereby calculating a hydrogen flux prediction curve and an electrochemical potential prediction curve for the next tens of seconds in the memory space. For solving the quadratic programming subproblem, the construction of the Hessian matrix relies on a second-order Taylor expansion of the objective optimization function. The controller uses the finite difference method to approximate the second-order partial derivatives in each control variable direction. During the finite difference operation, the controller applies a preset displacement perturbation step size to each independent physical execution quantity in the control increment sequence, extracts the characteristic values ​​of the objective optimization function before and after the perturbation in the corresponding direction, and uses the central difference formula to divide by twice the displacement perturbation step size to obtain the second-order partial derivative elements in each control variable direction, which are then arranged to construct the Hessian matrix. The control system sets the specific value of the displacement perturbation step size to one-thousandth of the limit value of the corresponding physical execution quantity. The interior-point solver, based on the curvature information provided by the Hessian matrix and the spatial polyhedral boundary formed by the nonlinear hard constraints, introduces a logarithmic barrier function to transform the nonlinear hard constraints into boundary exclusion penalty terms in the computational space of the objective optimization function. In each optimization iteration, the search step size is dynamically adjusted to ensure that the gradient search direction always points to the legal control interval where the total penalty cost decreases most steeply and does not exceed the boundary. The algorithm terminates when the cost difference between two adjacent iterations is less than the convergence minimum set by the system, or when the current optimization iteration step of the interior-point solver reaches the preset maximum iteration count criterion. The specific value of the preset maximum iteration count criterion set by the control system is 50 iterations. Setting the maximum iteration count to 50 limits the longest physical computation time of the nonlinear programming solution algorithm to be absolutely lower than the single sampling control cycle of the online degassing system for aluminum alloy melt. The extracted instruction sequence at this time is the optimal control increment sequence that brings the entire degassing process to a comprehensive optimal state under multi-objective constraints.

[0057] Before establishing a predictive model, the control system needs to pre-construct the system state transition constant matrix characterizing the inertial hysteresis characteristics inside the melt and obtain the initial dynamic cross-coupling Jacobian matrix. When conducting open-loop step response tests on the aluminum alloy degassing equipment under no-load fluid circulation conditions, the control system applies step excitation signals with amplitudes of 10% of the operating range to the control variables, and the underlying sensor network synchronously collects and records the dynamic response data of the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal within the corresponding time period.

[0058] Subsequently, the logic operation unit uses the subspace state-space system identification algorithm to perform offline data fitting operations on the collected dynamic response data. During the offline data fitting operation, the logic operation unit arranges the step excitation signal sequence in chronological order to construct the input block Hankel matrix, arranges the synchronously collected dynamic response data sequences of hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal in chronological order to construct the output block Hankel matrix, and performs orthogonal projection operation on the output block Hankel matrix using the input block Hankel matrix to obtain the projection matrix characterizing the system state.

[0059] Furthermore, the logic operation unit performs singular value decomposition on the projection matrix, extracts singular values ​​whose values ​​are greater than the preset system noise value, determines the total number of extracted singular values ​​as the feature dimension of the system state variables, and extracts the corresponding left singular vectors to construct the observable subspace matrix. Combining the observable subspace matrix and the translated observable subspace matrix shifted by a single sampling period in the time dimension, a system of linear algebraic equations is established. The fixed element constant values ​​inside the system state transition constant matrix are calculated by performing least squares fitting operation on the system of linear algebraic equations.

[0060] Simultaneously, the logic operation unit extracts the changes in the input control variables and the changes in the output state variables at steady state time. It calculates the initial local linear partial derivatives by dividing the changes in the output state variables by the changes in the input control variables, and arranges the initial local linear partial derivatives according to the correspondence between the control variables and the state variables, thereby constructing the initial dynamic cross-coupling Jacobian matrix.

[0061] Finally, during the real-time online control process, the control system mathematically combines the pre-identified system state transition constant matrix with the real-time updated dynamic cross-coupling Jacobian matrix to construct a complete discrete state-space equation as a prediction model.

[0062] The preset differential threshold value set in the internal memory of the control system is 0.5 ohms per second. When the differential value of the real part of the impedance signal is negative and the absolute value of the differential value exceeds 0.5 ohms per second, it indicates that a drastic phase transition has occurred in the microstructure of the aluminum alloy melt. The forgetting factor configured in the recursive least squares algorithm executed by the control system is 0.98, and this forgetting factor value of 0.98 enables the computational logic to quickly filter out computational interference caused by outdated historical data. The target degassing rate flux set by the controller is 50 microamps, corresponding to an ideal degassing rate of 0.02 ml of hydrogen per minute per 100 grams of molten aluminum inside the aluminum alloy melt. The electrochemical potential safety benchmark set by the safety monitoring module is -850 millivolts. A voltage value lower than -850 millivolts will cause a drastic and irreversible electrochemical oxidation side reaction at the interface of the aluminum alloy melt. The mathematical solver sets a convergence minimum of 0.001 when iteratively searching for the minimum of the objective optimization function using the interior-point method. A difference in penalty cost between two adjacent iterations of the objective optimization function is less than 0.001, causing the nonlinear programming optimization to terminate promptly. The control system sets a stirring speed limit of 600 revolutions per minute. A stirring speed exceeding 600 revolutions per minute will cause severe air vortices to form on the surface of the molten aluminum alloy. Simultaneously, the bias voltage limit is set at 5 volts. A bias voltage exceeding 5 volts across the graphite electrodes will physically break down the anti-oxidation film on the surface of the molten aluminum alloy.

[0063] Embodiment 1 of this invention: After the control system is started, it collects the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the aluminum alloy melt through induction arrays arranged at different positions inside the degassing chamber. The hydrogen atom diffusion flux signal reflects the hydrogen removal rate inside the melt, the electrochemical potential signal indicates the redox reaction state at the interface, and the impedance real part signal is used to monitor the fluid microstructure. Due to physical deviations in the sensor installation positions, the control system acquires the kinematic viscosity characteristic value and stirring speed of the aluminum alloy melt, and uses the stirring speed and kinematic viscosity characteristic value to generate a spatial phase compensation delay time constant. Subsequently, the spatial phase compensation delay time constant is used to perform time-shift compensation on the collected electrochemical potential signal, so that the compensated electrochemical potential signal is aligned with the hydrogen atom diffusion flux signal in the time dimension. The computation module concatenates the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal to construct a multi-field coupled state tensor, establishes a dynamic cross-coupling Jacobian matrix, and uses the rate of change of the impedance real part signal to trigger the update of the dynamic cross-coupling Jacobian matrix. The system combines the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model. Then, the multi-field coupled state tensor is input into the prediction model to deduce the system state evolution trajectory for future multiple steps. Finally, the system uses the system state evolution trajectory to calculate the objective optimization function, obtains the optimal control increment sequence by solving the objective optimization function, and synchronously adjusts the gas flow rate, bias voltage, stirring speed, and vibration frequency according to the optimal control increment sequence. This processing logic substantially solves the technical problem of unstable degassing effect caused by independent parameter control.

[0064] Embodiment 2 of the present invention: When the production line switches from a VI series aluminum alloy to a VII series aluminum alloy, the sudden change in the chemical composition inside the aluminum alloy melt will cause a violent physical fluctuation in the kinematic viscosity of the melt. The control system calculates the difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period as the rate of change of the real part of the impedance signal; when the difference value is negative and the absolute value of the difference value continuously exceeds the preset difference threshold, the data processing logic determines that the aluminum alloy melt is a precursor to a eutectic phase transition. The aforementioned logic determines that the recursive least squares algorithm with a forgetting factor is used to update the dynamic cross-coupled Jacobian matrix. At this time, the underlying controller obtains the input increment of the control variables in the historical sampling period and the state measurement output increment in the multi-field coupled state tensor; the recursive least squares algorithm iteratively calculates the internal partial derivative elements of the dynamic cross-coupled Jacobian matrix, quantifying the cross-interference value of the stirring speed and bias voltage on the hydrogen atom diffusion flux signal under the state of drastic change in kinematic viscosity. Subsequently, the operational logic uses the updated dynamic cross-coupling Jacobian matrix to correct the prediction model and re-deduces the system state evolution trajectory; the control system uses the system state evolution trajectory to adjust the optimal control increment sequence in real time, and finally issues the optimal control increment sequence to drive the underlying execution components to adjust the gas flow and bias voltage. The aforementioned dynamic update process solves the technical problem of poor adaptability to production processes.

[0065] Embodiment 3 of this invention: In the degassing process of aluminum alloy melt, improving degassing efficiency often comes with the objective risk of increased surface oxidation of the melt. The control system uses a predictive model to deduce the system state evolution trajectory for multiple future steps, extracts the predicted hydrogen flux value and the predicted electrochemical potential value within the system state evolution trajectory, and calculates the difference between the predicted hydrogen flux value and the target degassing rate flux to obtain the hydrogen flux deviation penalty value; simultaneously, the computing unit calculates the sum of squared errors between the predicted electrochemical potential value and the electrochemical potential safety benchmark to obtain the electrochemical potential deviation penalty value, and obtains the control energy consumption penalty value by calculating the sum of squared increments of the control variables. The processor adds the hydrogen flux deviation penalty value, the electrochemical potential deviation penalty value, and the control energy consumption penalty value to generate the objective optimization function, and obtains the stirring speed limit value and the bias voltage limit value of the aluminum alloy melt as nonlinear hard constraints. The low-level solver merges the objective optimization function and the nonlinear hard constraints to establish a quadratic programming subproblem, uses the Hessian matrix to approximate the gradient search direction of the control variables, and uses the interior point method to iteratively optimize along the gradient search direction to obtain the optimal control increment sequence that minimizes the objective optimization function value. The control system performs data analysis on the optimal control increment sequence to obtain the absolute drive command sequence of gas flow rate, bias voltage, stirring speed and vibration frequency, and coordinates the underlying actuator according to the absolute drive command sequence. The aforementioned coordinated control mechanism solves the quality fluctuation problem caused by the lack of coordinated optimization based on the final degassing quality.

[0066] This invention also provides an application embodiment, conducting a comparative experiment on melt degassing of 7075 aluminum alloy on an aluminum alloy continuous casting production line, with the experiment lasting 120 minutes. The control equipment used a traditional single-loop proportional-integral-derivative (PID) independent control method for the first 60 minutes, and switched to a multi-parameter collaborative control method for the next 60 minutes. The system recorded the outlet hydrogen content of the aluminum alloy melt in real time using an online hydrogen analyzer. The data recorded by the monitoring instrument showed that under the traditional single-loop PID independent control method, the average outlet hydrogen content of the aluminum alloy melt was 0.18 ml per 100 grams of aluminum, and the fluctuation range of the outlet hydrogen content reached 0.06 ml per 100 grams of aluminum; while after the control equipment adopted the multi-parameter collaborative control method, the average outlet hydrogen content of the aluminum alloy melt decreased to 0.11 ml per 100 grams of aluminum, and the fluctuation range of the outlet hydrogen content narrowed to 0.015 ml per 100 grams of aluminum. Furthermore, ultrasonic flaw detection equipment was used to inspect the 7075 aluminum alloy ingots produced in the final casting. The detection results showed that the microporosity defect rate inside the ingot was 4.2% under the traditional single-loop proportional-integral-derivative independent control method, while the microporosity defect rate was reduced to 0.8% under the multi-parameter collaborative control method. Objective test data proves that the multi-parameter collaborative control method effectively reduces the mutual physical interference between control variables and significantly improves the degassing quality stability of the continuous production line under complex operating conditions.

Claims

1. A multi-parameter coordinated control method for online degassing of aluminum alloy melt, characterized in that, include: Step 1: Collect the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the aluminum alloy melt. Spatiotemporally align the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal. Then, splice the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal to construct a multi-field coupled state tensor. Step 2: Establish a dynamic cross-coupling Jacobian matrix, calculate the rate of change of the real part of the impedance signal, and use the rate of change of the real part of the impedance signal to trigger the update of the dynamic cross-coupling Jacobian matrix. Combine the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model. Use the updated dynamic cross-coupling Jacobian matrix to quantify the degree of interference of control variables on the hydrogen atom diffusion flux signal. The control variables include gas flow rate, bias voltage, stirring speed, and vibration frequency. Step 3: Input the multi-field coupled state tensor into the prediction model to deduce the system state evolution trajectory for the next multiple steps, use the system state evolution trajectory to calculate the objective optimization function, and solve the objective optimization function to obtain the optimal control increment sequence; Step 4: Adjust the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency according to the optimal control increment sequence.

2. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 1, characterized in that, The spatiotemporal alignment of the hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal includes: The kinematic viscosity characteristic value of the aluminum alloy melt and the constant proportionality coefficient related to the physical structure of the degassing chamber are obtained. The spatial phase compensation delay time constant is generated by performing multiplication and division algebraic operations on the constant proportionality coefficient, the stirring speed and the kinematic viscosity characteristic value. The spatial phase compensation delay time constant is used to perform time shift compensation on the collected electrochemical potential signal so that the electrochemical potential signal and the hydrogen atom diffusion flux signal are aligned in the time dimension.

3. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 2, characterized in that, The step of concatenating the aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal to construct a multi-field coupled state tensor includes: The aligned hydrogen atom diffusion flux signal, electrochemical potential signal, and impedance real part signal of the current sampling period are concatenated in space to form a one-dimensional column vector. A preset number of one-dimensional column vectors of historical continuous sampling periods are obtained. The one-dimensional column vector of the current sampling period and the one-dimensional column vectors of the historical continuous sampling periods are arranged sequentially according to the time sequence to construct the multi-field coupled state tensor.

4. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 1, characterized in that, The calculation of the rate of change of the real part of the impedance signal, and the use of the rate of change of the real part of the impedance signal to trigger the update of the dynamic cross-coupled Jacobian matrix, includes: The difference between the real part of the impedance signal in the current sampling period and the real part of the impedance signal in the previous sampling period is calculated as the rate of change of the real part of the impedance signal. When the difference value is negative and the absolute value of the difference value continuously exceeds a preset difference threshold, the recursive least squares algorithm with a forgetting factor is triggered to update the dynamic cross-coupled Jacobian matrix.

5. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 4, characterized in that, The recursive least squares algorithm with a forgetting factor is used to update the dynamically cross-coupled Jacobian matrix, including: The system acquires the input increments of the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency for historical sampling periods; acquires the state measurement output increment within the multi-field coupled state tensor; iteratively calculates the internal partial derivative elements of the dynamic cross-coupled Jacobian matrix using the recursive least squares algorithm in combination with the input increments and the state measurement output increments; and obtains the cross-interference values ​​generated by the stirring speed and the bias voltage on the hydrogen atom diffusion flux signal.

6. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 5, characterized in that, The step of combining the pre-identified system state transition constant matrix with the updated dynamic cross-coupling Jacobian matrix to establish a prediction model includes: The system state transition constant matrix, which is a pre-identified representation of the inertial hysteresis characteristics inside the melt, is combined with the updated dynamic cross-coupling Jacobian matrix to establish a predictive state space equation and construct the predictive model. The stirring speed limit value and the bias voltage limit value of the aluminum alloy melt are obtained, and the stirring speed limit value and the bias voltage limit value are set as nonlinear hard constraints.

7. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 6, characterized in that, The calculation of the objective optimization function using the system state evolution trajectory includes: The predicted state-space equation is used to deduce the system state evolution trajectory for multiple future steps. The predicted hydrogen flux and predicted electrochemical potential values ​​within the system state evolution trajectory are extracted and dimensionless normalization mapping is performed. The sum of squared errors between the normalized predicted hydrogen flux and the normalized target degassing rate flux is calculated to obtain the hydrogen flux deviation penalty value. The sum of squared errors between the normalized predicted electrochemical potential and the normalized electrochemical potential safety benchmark is calculated to obtain the electrochemical potential deviation penalty value. The sum of squared increments of the normalized control variables is calculated to obtain the control energy consumption penalty value. The hydrogen flux deviation penalty value, the electrochemical potential deviation penalty value, and the control energy consumption penalty value are added together to generate the target optimization function.

8. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 7, characterized in that, Solving the objective optimization function to obtain the optimal control increment sequence includes: The objective optimization function and the nonlinear hard constraint are combined to establish a quadratic programming subproblem. The second-order partial derivative of the objective optimization function is calculated using the finite difference method to construct the Hessian matrix. The gradient search direction of the control variable is calculated using the Hessian matrix. The optimal control increment sequence that minimizes the value of the objective optimization function is obtained by iterative optimization along the gradient search direction using the interior point method. The current sampling period increment command is extracted from the optimal control increment sequence.

9. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 8, characterized in that, The step of adjusting the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency according to the optimal control increment sequence includes: The incremental command of the current sampling period is parsed to obtain the incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency. The incremental values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency are added to the actual control values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency of the previous sampling period to generate an absolute drive command sequence including the target values ​​of gas flow rate, bias voltage, stirring speed, and vibration frequency.

10. The multi-parameter coordinated control method for online degassing of aluminum alloy melt according to claim 9, characterized in that, The adjustment of the gas flow rate, the bias voltage, the stirring speed, and the vibration frequency includes: The gas flow rate is adjusted according to the target gas flow rate value in the absolute drive command sequence, the bias voltage target value in the absolute drive command sequence is converted into a duty cycle signal to adjust the bias voltage, the stirring speed target value in the absolute drive command sequence is converted into an analog voltage signal to adjust the stirring speed, and the vibration frequency target value in the absolute drive command sequence is converted into a frequency-hopping sinusoidal excitation signal to adjust the vibration frequency.

Images