Method for determining the content of impurities in a metal melt

Abstract

The application relates to the technical field of metallurgical detection, and discloses a method for determining impurity content in metal melt detection, which comprises the following steps: acquiring characteristic spectral line profiles of a high-temperature center area and a low-temperature edge area of plasma; measuring the self-reversal wavelength width of the spectral line of the high-temperature center area and the full width at half maximum of the Stark broadening of the low-temperature edge area; calculating the scale ratio of the two widths to determine an optical thickness invariant for representing the self-absorption degree of radiation transport; correcting the light intensity of the characteristic spectral line by using the optical thickness invariant, and determining the impurity content in combination with real-time temperature and reference element intensity; the application establishes a spatial decoupling mechanism based on spectral line type geometry, uses the inherent broadening law generated by the collision of plasma micro-particles to establish a physical invariant as a correction reference, eliminates the interference of plume volume distortion caused by melt surface tumbling, and effectively widens the linear measurement interval of detection.

Description

Technical Field

[0001] This invention relates to a method for determining the impurity content in the detection of molten metal, belonging to the field of metallurgical detection technology. Background Technology

[0002] Current laser-induced breakdown spectroscopy (LAS) uses pulsed lasers to focus on the surface of molten metal to generate plasma, and collects the characteristic spectral lines generated by the plasma to analyze the composition of impurity elements. This technology has the characteristics of fast online monitoring and non-contact signal acquisition, and is the mainstream technical path to realize closed-loop feedback of quality in metallurgical production processes. During the plasma evolution process, the intensity of spectral lines is constrained by the probability of particle energy level transitions and the concentration of matter.

[0003] Under high-temperature metallurgical smelting conditions, the surface of the molten metal pool experiences intense thermodynamic convection and surface turbulence. Transient fluctuations in the liquid surface cause deviations between the laser target and the focusing plane, resulting in an asymmetric geometric distortion and spatially heterogeneous distribution of the excited plasma plume. When the target impurity is in a high-concentration range, photons emitted from within the plasma are reabsorbed by surrounding particles, producing a characteristic spectral self-absorption phenomenon. Conventional techniques attempt to correct this spectral self-absorption effect using spatial difference logic or intensity ratio logic. These methods rely on the assumption that the plasma plume maintains an ideal symmetrical sphere and satisfies the rarefaction assumption of radiative transport. However, in environments with frequent fluctuations on the melt surface, the transient distortion of the plume volume causes the spatial intensity distribution to lose its definite functional correlation. This introduces nonlinear artifacts caused by geometric changes into the intensity correlation model, leading to saturation of measurement data in the high-concentration range. Simply adding pulses... While averaging the number of pulses or attempting to introduce mechanical flow stabilization devices, it is difficult to decouple the interaction between plume geometry and self-absorption effect from the physical essence of plasma radiation transport. For example, Chinese invention patent application CN115753741A discloses a single-sample laser-induced breakdown spectrum self-absorption correction method and system. It defines the relative self-absorption coefficient by establishing the ratio of actual spectral intensity to theoretical intensity, and then performs inversion correction on the elemental content of the sample to be tested. However, under the complex dynamic conditions of metallurgical smelting, the measured value of absolute plasma radiation intensity undergoes drastic random jumps due to liquid surface fluctuations. This model, which relies excessively on the absolute value of light intensity scalar, is prone to misjudging artifacts introduced by geometric distortion as self-absorption evolution. As a result, the correction accuracy is constrained by the strong coupling between excitation energy level and absolute energy density, making it difficult to decouple the self-absorption effect from the physical essence of plume morphology distortion.

[0004] Therefore, how to determine a physical parameter benchmark that can characterize the intrinsic physical state of plasma and is not affected by the geometric distortion of the plume shape, so as to achieve accurate determination of impurity content under complex working conditions, has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A method for determining the impurity content in the detection of molten metal, comprising the following steps: Step S101: Acquire the spatially resolved spectral signal of laser-induced plasma and extract the characteristic spectral lines of the target impurity element from it; Step S102: Determine the high-temperature central region and low-temperature edge region in the plasma, and extract the spectral line self-reversal profile of the characteristic spectral lines in the high-temperature central region and the collision broadening profile in the low-temperature edge region. Step S103: Measure the wavelength width at the inversion depression position in the spectral inversion profile, and measure the full width at half maximum (FWHM) of Stark broadening dominated by the electron density broadening effect in the collision broadening profile. Step S104: Calculate the optical thickness invariant based on the geometric scale ratio of the full width at half maximum (FWHM) of the Stark broadening to the wavelength width at the self-reversing concave position, and use the ratio to offset the nonlinear deviation introduced by plasma volume distortion. Step S105: Obtain the real-time temperature of the molten metal and extract the reference spectral line intensity of the matrix reference element in the plasma; Step S106: Self-absorption correction is performed on the light intensity of the characteristic spectral lines using optical thickness invariant to obtain the corrected light intensity. Using a calibration model that establishes the mapping relationship between material concentration and corrected light intensity, real-time temperature, and reference spectral line intensity, the impurity content in the metal melt is determined.

[0006] Preferably, step S103 further includes the following sub-steps: step S1031, separating the Stark broadening component in the collision broadening profile of the characteristic spectral lines of the low-temperature edge region using the Lorentz fitting method; step S1032, extracting the half-width at half maximum (WHM) of the Lorentz fitting curve, using it as the full width at half maximum (WHM) of the Stark broadening curve, and using the full width at half maximum (WHM) of the Stark broadening curve to characterize the microscopic particle interaction characteristics of the plasma, thereby offsetting the macroscopic volume change of the plasma caused by the fluctuation of the molten metal surface.

[0007] Preferably, step S103 further includes the following sub-steps: step S1033, by searching for peaks in the characteristic spectral lines of the high-temperature central region, the wavelength position corresponding to the minimum point of the inversion intensity at the center of the inversion profile of the spectral lines is determined; step S1034, the wavelength spacing between the symmetrical intensity points on both sides of the minimum point of the inversion intensity is measured, and this is used as the wavelength width of the inversion depression position.

[0008] Preferably, the method further includes step S107, which involves determining the theoretical emission intensity evolution characteristics of the characteristic spectral lines at real-time temperature based on the Boltzmann distribution law, and using the theoretical emission intensity evolution characteristics to perform temperature compensation on the optical thickness invariant.

[0009] Preferably, the method further includes step S108, which normalizes the characteristic spectral lines using the intensity of the reference spectral lines to reduce the influence of laser-induced energy pulse fluctuations on the intensity of the characteristic spectral lines.

[0010] Preferably, the method further includes step S109, which involves measuring the broadening characteristics of the matrix reference element, monitoring in real time whether the low-temperature edge region is in an optically rarefied state, and filtering the spectral data of the current sampling frame when the low-temperature edge region leaves the optically rarefied state.

[0011] Preferably, step S102 further includes sub-step S1021, obtaining the intensity gradient factors of the plasma plume in the height and radial directions, and dynamically adjusting the boundary distribution weights of the high-temperature center region and the low-temperature edge region based on the geometric symmetry deviation value determined by the intensity gradient factors.

[0012] Preferably, the calibration model in step S106 is constructed through the following sub-step S1061: using the impurity content gradient determined by the standard metal sample set, a nonlinear mapping database is established using a multidimensional regression algorithm, which includes optical thickness invariants, the corrected intensity of characteristic spectral lines and real-time temperature as inputs, and impurity content as output.

[0013] Preferably, after determining the impurity content in the molten metal, the method further includes step S110, adjusting the amount of chemical reagents added to the refining equipment according to the deviation between the impurity content and the preset process parameters, thereby completing the closed-loop control of the metal smelting process.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. In the determination of impurity content in molten metal detection, the difference in spectral distribution between the central and peripheral regions of the plasma is used to construct a spatial intensity gradient factor, revealing the physical barrier state of photon escape within the plasma. This mechanism uses the optical boundary layer of the plasma itself as an endogenous reference, transforming the analysis of spectral signals from a simple intensity numerical correlation to a characterization of the physical evolution of optical thickness. This spatial dimension-based signal decoupling method can automatically cancel out radiation intensity artifacts caused by severe fluctuations on the surface of the molten metal and laser energy instability, maintaining the linear mapping fidelity between the characteristic spectral lines of the target element and the mass fraction under complex metallurgical conditions.

[0015] 2. By replacing the traditional intensity scalar ratio logic with spectral line shape geometric morphology analysis, optical thickness invariant is established by measuring the self-reversal concavity width of characteristic spectral lines in the central region and the broadening characteristics of the edge region. Since the physical broadening law of spectral line shape is controlled by the collision mechanism of plasma microparticles, its geometric scale has natural anti-interference characteristics against external excitation energy drift and plasma macroscopic volume distortion. Even in extreme environments where the molten liquid surface rolls and causes transient shift of the laser focusing target point, this mechanism can still achieve accurate physical correction of the self-absorption effect by extracting stable spectral line profile features, effectively broadening the linear measurement range of optical detection in the analysis of high-concentration impurities.

[0016] 3. By coupling multidimensional data such as real-time temperature of the molten metal, intensity of matrix reference elements, and spatial intensity gradient factor, an adaptive correction model covering thermodynamic distribution and dynamic fluctuations is constructed. Real-time temperature compensation ensures that the model can dynamically adjust to follow the Boltzmann distribution changes of the initial excitation state of the melt. The matrix reference elements realize real-time normalization of the absolute fluctuations of laser-induced energy. This synergistic superposition of multiple features and mechanisms makes the final output mass fraction not only include the composition information of the target element, but also integrate the physical constraints of the plasma excitation environment, improving the repeatability and reliability of the detection results in high-temperature, strong convection metallurgical environments. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method for determining the impurity content in the detection of metal melts according to the present invention; Figure 2 This is a logic block diagram of the method for determining the impurity content in the detection of molten metal according to the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] A method for determining the impurity content in the detection of molten metal includes the following steps: Step S101: Acquire the spatially resolved spectral signal of laser-induced plasma and extract the characteristic spectral lines of the target impurity element from it; Step S102: Determine the high-temperature central region and low-temperature edge region in the plasma, and extract the spectral line self-reversal profile of the characteristic spectral lines in the high-temperature central region and the collision broadening profile in the low-temperature edge region. Step S103: Measure the wavelength width at the inversion depression position in the spectral inversion profile, and measure the full width at half maximum (FWHM) of Stark broadening dominated by the electron density broadening effect in the collision broadening profile. Step S104: Calculate the optical thickness invariant based on the geometric scale ratio of the full width at half maximum (FWHM) of the Stark broadening to the wavelength width at the self-reversing concave position, and use the ratio to offset the nonlinear deviation introduced by plasma volume distortion. Step S105: Obtain the real-time temperature of the molten metal and extract the reference spectral line intensity of the matrix reference element in the plasma; Step S106: Self-absorption correction is performed on the light intensity of the characteristic spectral lines using optical thickness invariant to obtain the corrected light intensity. Using a calibration model that establishes the mapping relationship between material concentration and corrected light intensity, real-time temperature, and reference spectral line intensity, the impurity content in the metal melt is determined.

[0021] Preferably, step S103 further includes the following sub-steps: step S1031, separating the Stark broadening component in the collision broadening profile of the characteristic spectral lines of the low-temperature edge region using the Lorentz fitting method; step S1032, extracting the half-width at half maximum (WHM) of the Lorentz fitting curve, using it as the full width at half maximum (WHM) of the Stark broadening curve, and using the full width at half maximum (WHM) of the Stark broadening curve to characterize the microscopic particle interaction characteristics of the plasma, thereby offsetting the macroscopic volume change of the plasma caused by the fluctuation of the molten metal surface.

[0022] Preferably, step S103 further includes the following sub-steps: step S1033, by searching for peaks in the characteristic spectral lines of the high-temperature central region, the wavelength position corresponding to the minimum point of the inversion intensity at the center of the inversion profile of the spectral lines is determined; step S1034, the wavelength spacing between the symmetrical intensity points on both sides of the minimum point of the inversion intensity is measured, and this is used as the wavelength width of the inversion depression position.

[0023] Preferably, the method further includes step S107, which involves determining the theoretical emission intensity evolution characteristics of the characteristic spectral lines at real-time temperature based on the Boltzmann distribution law, and using the theoretical emission intensity evolution characteristics to perform temperature compensation on the optical thickness invariant.

[0024] Preferably, the method further includes step S108, which normalizes the characteristic spectral lines using the intensity of the reference spectral lines to reduce the influence of laser-induced energy pulse fluctuations on the intensity of the characteristic spectral lines.

[0025] Preferably, the method further includes step S109, which involves measuring the broadening characteristics of the matrix reference element, monitoring in real time whether the low-temperature edge region is in an optically rarefied state, and filtering the spectral data of the current sampling frame when the low-temperature edge region leaves the optically rarefied state.

[0026] Preferably, step S102 further includes sub-step S1021, obtaining the intensity gradient factors of the plasma plume in the height and radial directions, and dynamically adjusting the boundary distribution weights of the high-temperature center region and the low-temperature edge region based on the geometric symmetry deviation value determined by the intensity gradient factors.

[0027] Preferably, the calibration model in step S106 is constructed through the following sub-step S1061: using the impurity content gradient determined by the standard metal sample set, a nonlinear mapping database is established using a multidimensional regression algorithm, which includes optical thickness invariants, the corrected intensity of characteristic spectral lines and real-time temperature as inputs, and impurity content as output.

[0028] Preferably, after determining the impurity content in the molten metal, the method further includes step S110, adjusting the amount of chemical reagents added to the refining equipment according to the deviation between the impurity content and the preset process parameters, thereby completing the closed-loop control of the metal smelting process.

[0029] Example 1: In the online composition monitoring scenario for continuous smelting of high alloy steel, the method for determining the impurity content in the molten metal is operated on the surface of the molten metal maintained at 1600℃. The thermal convection and surface tumbling on the surface of the molten metal pool cause anisotropic deformation of the plume morphology of the laser-induced plasma. This causes measurement deviations due to the inability of the acquisition scheme relying solely on light intensity values ​​to identify noise introduced by the sudden change in plasma volume. By acquiring the spatially resolved spectral signal of the laser-induced plasma and extracting the characteristic spectral lines of the target impurity element, the high-temperature central region and the low-temperature edge region in the plasma are determined. The self-reversal profile of the characteristic spectral line in the high-temperature central region and the collision broadening profile in the low-temperature edge region are extracted. The spatially resolved spectral signal of the central region is analyzed and the self-reversal concave profile at the top of the characteristic spectral line of the target impurity element is located.

[0030] The first wavelength width of the self-reversing concave profile was measured. The Stark broadening profile is measured in the edge region to obtain the second half-wave width. Optical thickness invariants are constructed using two geometric parameters based on spectral profile morphology. ,in ,in, For optical thickness invariant, The first wavelength width to be measured, To determine the second half-wavelength, based on the spectral growth curve theory in the radiation transport equation, under local thermodynamic equilibrium, the width of the characteristic spectral line's self-reversal depression is jointly determined by the plasma absorption coefficient and microscopic collision broadening. In the actual radiation transport process, the plasma plume excited by the pulsed laser can be equivalent to a non-uniform radiator composed of a central high-temperature, high-density emission source and an outer low-temperature, thin layer. The intrinsic emission line shape of the target element generated in the central high-temperature region is mainly modulated by Stark broadening caused by microscopic particle collisions, while the ground-state atoms of the same type in the outer cold flow layer perform resonant self-absorption on the radiated photons at the central wavelength of the spectral line, thereby peeling off a geometric depression at the top of the macroscopic spectral line. When the surface of the molten metal churns violently, causing macroscopic volume distortion in the plasma plume, this distortion is physically equivalent to an anisotropic scaling of the macroscopic radiation conduction path length of the plasma along the probe optical axis. According to the linear wavefront approximation of the radiation conduction equation, this macroscopic variation in path length will serve as a common-mode scaling factor, simultaneously modulating the width of the self-reversal depression in the central high-temperature region in the time domain with the same linear attenuation weight. Stark widens half-height and full width at the edge of the low-temperature zone. Therefore, the ratio is constructed by performing algebraic division on the two at the same sampling time. The transient fluctuation factor of macroscopic spatial geometric distance is completely canceled out at both ends of the fraction, thus reducing the dimension of the composite spectral signal to a thermodynamic physical invariant determined only by the microscopic atomic transition cross section and the intrinsic energy level particle number density. This achieves closed-loop isolation between macroscopic geometric distortion and microscopic intrinsic state in radiation conduction theory. By extracting the ratio of the first wavelength width to the second half-wave width dominated by the electron density, the optical thickness invariant is calculated to directly characterize the intrinsic photon absorption characteristics of the cold flow layer around the plasma, offsetting the measurement interference introduced by the random distortion of the total plasma volume caused by local microscopic electron density fluctuations and macroscopic fluctuations of the liquid surface. Furthermore, the macroscopic geometric distortion of the plasma volume will induce a proportional scaling of the photon escape radiation path length in terms of physical morphology. According to radiation conduction theory, this macroscopic path scaling process will apply the same variable to the width of the self-reversing concave in the central high-temperature region and the full width at half maximum (FWHM) of the Stark broadening in the edge low-temperature region. The linear modulation amplification effect of the direction of change and equal attenuation weights is considered. Therefore, by constructing the algebraic division ratio of the two at the same moment, the common-mode interference factor characterizing the fluctuation of macroscopic spatial geometric distance is directly canceled from the underlying mathematical logic. This enables the extracted optical thickness invariant to achieve physical dimensionality reduction and be transformed into a pure thermodynamic function that is only related to the transition absorption cross section and ground-state atomic volume density of the target element itself. Within the scope of physics, a closed-loop isolation and elimination mechanism for macroscopic fluid distortion noise is established. This mechanism uses the inherent physical broadening law of the spectral line shape as a correction benchmark to transform the liquid surface fluctuation and excitation energy variation into a self-cancelling geometric proportional factor. Combined with the real-time melt temperature and the matrix reference element intensity input into the preset radiation transport inversion model, the linear measurement range of the system under high-concentration impurity conditions is broadened, and the repeatability standard deviation of the component determination is maintained below 2%, providing data support for the closed-loop control of the metallurgical process.

[0031] Example 2: In a 1600℃ induction furnace test environment simulating continuous melting of high-alloy steel, the method for determining impurity content in molten metal detection is applied to the surface of molten steel, which is subjected to violent surface turbulence driven by electromagnetic induction. Specifically, a Gaussian white noise with a signal-to-noise ratio of 20dB is actively injected into the spectral acquisition circuit via a signal generator to simulate electromagnetic background interference in the metallurgical environment. During actual processing, to eliminate the random amplitude distortion of the original spectral signal waveform caused by the injected 20dB Gaussian white noise, the spectral data must be sequentially passed through passive hardware impedance filtering before geometric contour feature extraction. A cascaded noise reduction architecture consisting of a network and a software-based digital low-pass sliding window smoothing algorithm is employed. The software smoothing algorithm establishes a sliding time window covering five spectral sampling channels, centered on the wavelength of the current pixel. It performs high-order Savitzky-Golay polynomial least-squares fitting filtering on the original radiant intensity data within the window. This smoothed and denoised spectral profile completely eliminates spurious noise, thus recovering the physical lineform with deterministic analytical boundaries. Based on this, the processor can then perform Lorentz curve fitting for peak finding and bilateral symmetrical extrapolation extraction, ensuring the width of the first wavelength extracted under strong noise interference. With the second half-wave width It maintains the true microscopic physical linear geometry and eliminates noise interference on the base of the depression depth and the half-width at half-maximum of the wing. Simultaneously, the output fluctuation of the laser pulse energy is set within the range of ±5% to generate unsteady drift of the radiation intensity. The selection of the sampling frequency balances the transient capture accuracy of the molten pool surface morphology with the computational load of the data processing unit. At a sampling frequency set to 10Hz, the original spatially resolved spectral signal of chromium at 425.4nm is obtained. Under this initial environment, the displacement of the geometric center of the plasma plume caused by the tumbling of the melt surface is observed, resulting in an 8.5% measurement deviation in the traditional control group scheme that relies on absolute light intensity scalars.

[0032] The experimental group analyzed the spectral signals of the high-temperature central region of the plasma and located the self-reversing concave profile at the top of the characteristic spectral lines of the target impurity element, and measured the first wavelength width of the self-reversing concave profile. The wavelength is 0.045 nm. The processor first locates the absolute minimum intensity of the self-reversal intensity within the self-reversal concave profile, and simultaneously acquires the theoretical peak intensity of the spectral line before self-reversal. The arithmetic mean of these two values ​​is calculated as the target symmetric intensity threshold. Subsequently, the processor performs a bidirectional linear search along both short and long wavelengths, centered on the wavelength position corresponding to the absolute minimum, to find two discrete spectral sampling points where the intensity first reaches the target symmetric intensity threshold. These two sampling points are the symmetric intensity points on either side, and the wavelength interval between them is used as the wavelength width value at the self-reversal concave position. Simultaneously, characteristic spectral lines are extracted in the low-temperature edge region, and the electron density-dominated Stark broadening profile is measured to obtain the second half-wavelength. The value is 0.021 nm. The optical thickness invariant is calculated using the formula. It is 2.14, of which, For optical thickness invariant, The first wavelength width to be measured, The ratio for the second half-wavelength, as measured, remained constant as the laser energy increased from 80 mJ to 120 mJ, while the partially missing control group used only the first wavelength width. When used as a correction factor, its output fluctuates positively with laser energy drift. The experimental group constructed a self-cancellation mechanism using two geometric parameters based on physical linear laws to keep the optical thickness invariant. It is decoupled from plume volume distortion only by the mass fraction of the target element.

[0033] The performance of determining impurity mass fractions was examined within a gradient range of 0.05% to 10.0%. When the chromium mass fraction was at the nominal point of 2.0%, the measured mass fraction output by the experimental group was 2.03%, and the repeatability standard deviation remained below 1.5%. As the mass fraction increased to near the upper threshold of 12.0%, [further details were observed]. The growth slope decreases and a performance inflection point appears. This phenomenon corresponds to the optical thickness saturation effect in the radiation transport process. When the self-reversing indentation depth exceeds the upper limit of the detector's dynamic range, the feature extraction accuracy of the geometric profile decreases. This confirms that within the mass fraction range defined by this invention, there is a stable monotonic mapping relationship between the optical thickness invariant and the impurity content. This effectively broadens the linear measurement range under high concentration conditions and suppresses the measurement noise introduced by the tumbling of the molten pool.

[0034] Example 3: In the online composition monitoring scenario of special stainless steel refining, the method for determining the impurity content in the molten metal is operated in a turbulent melt environment accompanied by electromagnetic stirring. The system acquires the spatially resolved spectral signal of laser-induced plasma and, before extracting the characteristic spectral lines of the target impurity element, first uses a quantization calibration program to locate the plume's physical region, which serves as the benchmark for subsequent geometric parameter extraction. This program uses the characteristic spectral intensity distribution of the matrix element iron at 438.3 nm as a spatial location reference. The processor calculates the intensity gradient diffusing outward from the plasma's central axis and determines the spatial coordinate position where the intensity drops to 15% of the axial peak intensity as the physical boundary between the high-temperature central region and the low-temperature edge region. Within the determined high-temperature central region, the self-reversing concave profile of chromium at 425.4 nm is acquired, and the first wavelength width is extracted. Simultaneously, characteristic spectral lines are extracted in the low-temperature edge region outside the boundary line, and the Stark broadening profile is measured to obtain the second half-wavewidth. The intensity gradient factors of the plasma plume in the height and radial directions are obtained by measuring the light intensity difference between adjacent spatial pixels in the vertical axis and horizontal radial directions and dividing it by the spatial displacement. After obtaining the height gradient factor and the radial gradient factor, the ratio of the absolute difference between the two to their arithmetic mean is calculated. This ratio is defined as the geometric symmetry deviation value. When the geometric symmetry deviation value is greater than the preset 0.15, it indicates that the plasma plume has undergone severe geometric distortion and asymmetric deformation. At this time, the system initiates dynamic adjustment of the boundary distribution weight. Specifically, the system multiplies the geometric symmetry deviation value by a fixed fine-tuning coefficient of 0.5 and subtracts the product from the original 15% light intensity reduction threshold, thereby reducing the light intensity boundary ratio of the high-temperature center region and increasing the sampling spatial weight of the low-temperature edge region, so that the boundary division moves towards the high-density center to compensate for the spatial spectral non-uniformity caused by the plume shape distortion.

[0035] Construct optical thickness invariants based on physical geometric proportions. ,in ,in, For optical thickness invariant, The first wavelength width to be measured, To determine the second half-wavelength, the radiative transport inversion model completes data transfer in the following way: using the intensity ratio of reference element spectral lines and combining it with the Saha equation to calculate the transient electron temperature of the plasma. In actual operation, to overcome the physical disruption of local thermodynamic equilibrium caused by the violent churning of the macroscopic liquid surface of the molten metal, the system locks and executes a specific microsecond-level integral delay time window in the acquisition front-end hardware. This time window triggering logic is based on the mathematical isolation between the nanosecond-level relaxation time of plasma microparticle collision energy transfer and the millisecond-level period of macroscopic fluid mechanical convection fluctuations. By stripping away the slowly evolving fluid dynamic interference in the time domain, a transient static cooling slice is extracted where sufficient collision energy exchange has been completed between microparticles. This ensures that the electron temperature and excited-state excitation temperature of the sampling region tend to be consistent within this extremely short microscopic moment. Thus, a strictly valid prerequisite physical boundary is constructed for the application of the Saha equation and the Boltzmann distribution model in an extremely unsteady fluid environment, based on the invariant optical thickness. The real-time value is retrieved from the preset self-absorption gain correction coefficient table to find the corresponding compensation operator, and the corrected characteristic spectral intensity is calculated. ; Electronic temperature Compared with the corrected characteristic spectral intensity The system inputs a linear decision equation, calculates the mass fraction of chromium, and establishes a nonlinear mapping model between the feature input and the mass fraction using a partial least squares regression algorithm. It acquires the real-time macroscopic temperature of the molten metal surface continuously collected by an infrared thermometer and uses this real-time macroscopic temperature as the physical boundary condition input to the model. This determines the thermodynamic mass baseline of the initial plasma ablation plume. The system then calculates the weighting of the microscopic transient electron temperature on the optical thickness change and iteratively fits the principal components within the defined thermodynamic mass baseline range. This suppresses the nonlinear residuals of the conversion from macroscopic temperature fluctuations to microscopic emission intensity, and outputs a mass fraction value that eliminates thermal convection bias. The specific mapping execution logic is as follows: To eliminate the dimensional conflict between the macroscopic Celsius temperature, the microscopic dimensionless optical thickness invariant, and the corrected spectral intensity preserved in a counting unit table, the system performs a unified scaling homogenization transformation. The system uses the equilibrium spectral data of a standard metal sample at a reference temperature of 1600℃ as the reference denominator, and calculates the real-time temperature, optical thickness invariant, and corrected spectral intensity separately. Dividing by their respective benchmark values, all physical quantities are transformed into dimensionless relative deviation scalars, unifying the logical dimensions. During the iterative fitting process of the partial least squares latent feature space, for the nonlinear residuals generated when converting macroscopic temperature fluctuations to microscopic emission intensity, the system sets an absolute convergence criterion for the principal components: when the absolute value of the difference in the variance contribution rates of the principal components extracted in two consecutive iterations is less than 0.0001, and the rate of change of the sum of squares of the residual matrix is ​​less than 0.05%, the feature matrix is ​​considered to have converged, the iteration stops, and the final normalized principal components are output, thus closing the adaptive dimensionality reduction mapping path for multi-source heterogeneous data. The processor constructs the input feature vector from three scalar dimensions: the optical thickness invariant extracted from the same sampling frame, the corrected feature spectral intensity, and the real-time temperature characterizing the boundary conditions. By calculating the projection score of this input vector in the partial least squares latent feature space, the first principal component matrix covering most of the variance information is extracted. The nonlinear mapping database internally uses macroscopic real-time temperature as the basis for its analysis. The system employs a chained hash index architecture with primary keys. Each temperature node carries a set of coefficient matrices trained and solidified in the partial least squares latent feature space using a multidimensional regression algorithm. In the system's offline calibration process, the generation mechanism of the multidimensional regression coefficient matrix is ​​as follows: multiple sets of input feature vectors under different mass fraction gradients are collected using a standard metal sample set to construct a standard calibration source matrix. Orthogonal decomposition is then performed using a partial least squares algorithm to calculate the weighted regression coefficient vector mapping the independent variable space to the dependent variable concentration space. The data is archived and stored in a database according to temperature gradients for real-time matching during online operation. A cross-validation method is used, with the objective function being minimizing the root mean square prediction error of the known calibrated impurity concentration of the target sample. This method iteratively solves for the final linear weighted regression coefficient vector from multiple feature dimensions to the component space. During real-time measurement, the input feature vector is subjected to a matrix inner product operation with the solidified regression coefficient vector, completing a closed-loop mapping from multi-source input adaptive dimensionality reduction to a single impurity content numerical output. This processing logic transforms the spatial description into coordinate determination based on light intensity gradients and decomposes the physical inversion into a deterministic algebraic operation process. The system faces the melt surface... When the height fluctuation is ±2mm, the absolute error of chromium measurement remains within 0.04%. The detection results show a decoupled state from the molten pool tumbling noise. After determining the output chromium mass fraction, the boundary value of the preset process component control range is extracted. The concentration difference between the current measured mass fraction and the boundary value of the control range is calculated. The concentration difference is input into the proportional-integral-derivative controller, which calculates and outputs a compensation component deviation adjustment command. The adjustment command is converted into a standard industrial voltage signal and sent to the refining addition equipment drive control board to change the servo motor feed speed and directly control the physical feeding amount of the deoxidizing chemical reagent into the molten pool, thus executing the metallurgical component closed-loop adjustment action.

[0036] Example 4: In the offline calibration process of the supporting standard material gradient library, the method for determining the impurity content in metal melt detection improves the detection accuracy by constructing a self-absorption gain correction coefficient table for retrieval of the radiation transport inversion model. The system acts on the surface of a standard alloy sample containing a specific impurity content gradient. Under a controlled experimental atmosphere, the stage is driven to complete vertical displacement changes to simulate the optical path evolution caused by fluctuations in the height of the molten pool. The laser output energy is simultaneously adjusted to vary within the range of 60mJ to 140mJ. For each set of physical conditions, the optical thickness invariant is measured. The response values ​​were obtained and the self-absorption loss of the corresponding spectral lines were acquired. An algebraic mapping method was used to establish... The functional relationship between the gain compensation operator and the self-absorption gain correction coefficient table is stored as a two-dimensional discrete data array matrix in non-volatile memory cells. Its vertical index represents the offline calibrated standard material mass fraction gradient, and its horizontal index represents the calculated optical thickness invariant. The discrete value range is from 1.0 to 5.0, with a step size of 0.1; the fixed values ​​within the matrix elements are the corresponding self-absorption dimensionless gain compensation operators. During the online monitoring phase, the processing unit calculates the current sampling frame in real time. The measured values ​​are first determined by two adjacent vertical index nodes in the location coefficient table. Then, an algebraic approximation is performed along the horizontal one-dimensional axis using a linear piecewise interpolation formula. This approximation is used to retrieve and calculate the gain compensation operator that accurately corresponds to the current operating condition. The operator is then multiplied by the original spectral intensity to complete the correction. The generated discrete data array is stored as a reference parameter in the system's non-volatile storage unit, serving as the physical basis for correcting self-absorption deviation during the online monitoring stage. When the system faces metal smelting conditions with different matrix materials, in order to determine the physical boundary between the high-temperature central region and the low-temperature edge region during the spatial resolution spectral signal extraction process, the processor reads the emission spectral profile of the matrix element of the current melt under test in the preset band and calculates the spatial evolution of the intensity gradient. It analyzes the radiation attenuation slope during the outward diffusion process of the plasma plume's central axis and sets the judgment threshold under this gradient in combination with the effective focusing diameter of the laser pulse, so that the geometric feature extraction result of the high-temperature core region matches the physical state under the current alloy system. This pre-program is used to eliminate the regional division offset caused by the overlap of matrix element spectral lines or the difference in background radiation intensity, so that the calculation benchmark of the optical thickness invariant is anchored at the current physical equilibrium position.

[0037] Example 5: In a deployment scenario targeting the refining of high-purity aluminum liquid, the method for determining impurity content in molten metal detection uses a spatial position calibration program to offset measurement errors caused by fluctuations in plasma plume geometry due to molten pool level. This program acquires the radiation intensity profile of the characteristic spectral line of magnesium element at a wavelength of 285.2 nm along the axial direction, calculates the first derivative of light intensity with respect to spatial displacement to determine the gradient field distribution within the plasma, and the processing unit reads the spatial coordinates corresponding to the maximum gradient value and determines the judgment threshold based on the root mean square noise of the plasma background radiation. The system acquires continuous background bands with no elemental emission lines near the characteristic spectral lines, collects spectral intensity samples within the continuous background bands of the current frame, calculates the standard deviation of light intensity, multiplies the standard deviation of light intensity by a preset signal-to-noise ratio (SNR) decision coefficient to generate the background noise tolerance for the current sampling period, establishes a sliding time window containing several sampling periods, filters the average background noise tolerance values ​​accumulated in each period within the window, and outputs an updated decision threshold. To compensate for the time-varying drift in background continuous radiation intensity caused by multiphase slag tumbling and steam dust during metallurgical smelting operations, the preset signal-to-noise ratio (SNR) determination coefficient follows the mathematical and statistical principles of instrument detection limits in analytical chemistry. In this scenario, this coefficient constant is fixed at an integer of 3. This constant value originates from offline background radiation tests of early standard pure metal samples under an absolutely stable argon atmosphere. Multiplying this coefficient by the standard deviation accurately generates an envelope boundary containing 99.7% random Gaussian noise fluctuations, thus ensuring the background noise tolerance determined by this product and the derived judgment threshold. It possesses true physical photon discrimination capability, ensuring that the final delineated boundary precisely delineates the extreme point where the plasma characteristic signal and the continuous bremsstrahlung radiation of the substrate intersect, when the local light intensity gradient exceeds the judgment threshold. At that time, the coordinates were determined as the physical boundary between the high-temperature central region and the low-temperature edge region; the system locked the geometric center of the high-temperature central region as the starting coordinate for obtaining the self-reversing concave profile, and established a sampling window in the low-temperature edge region outside the physical boundary line for extracting the Stark broadening feature.

[0038] The correction parameters in the radiation transport inversion model were obtained through offline testing of five sets of standard alloy samples with mass fraction gradients ranging from 0.1% to 5.0%. The system calculates the optical thickness invariant for each mass fraction. The baseline response value, where ,in, For optical thickness invariant, The first wavelength width to be measured, To determine the second half-wavelength, during the online inversion stage, the processing unit follows an algebraic operation path, that is, it calculates the optical thickness invariant under the current operating conditions based on the real-time acquired spectral signal. The system uses a linear piecewise interpolation method to retrieve the corresponding compensation operator from a pre-set self-absorption gain correction coefficient table. If the peak intensity of the spectral line of the target impurity element reaches the upper limit of the detector's saturation level, i.e., 65535 count units, the system automatically switches to a geometric width analysis method based on the spectral line wing morphology to maintain detection linearity. Finally, the mass fraction is output by combining the corrected spectral line intensity with the intensity of the reference element. This calibration procedure transforms the spatial region division from a qualitative description to a coordinate determination based on the light intensity gradient, and breaks down the construction of the inversion model into offline data filling and online retrieval processes. In the continuous 24-hour monitoring of aluminum liquid composition, the deviation between the measured mass fraction of impurity elements and the laboratory standard value is within 3%. After eliminating the nonlinear interference caused by plume volume distortion and detector saturation, the response stability of the detection system in high-concentration impurity scenarios is improved.

[0039] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for determining the impurity content in the detection of molten metal, characterized in that, Includes the following steps: Step S101: Acquire the spatially resolved spectral signal of laser-induced plasma and extract the characteristic spectral lines of the target impurity element from it; Step S102: Determine the high-temperature central region and low-temperature edge region in the plasma, and extract the spectral line self-reversal profile of the characteristic spectral lines in the high-temperature central region and the collision broadening profile in the low-temperature edge region. Step S103: Measure the wavelength width at the inversion depression position in the spectral inversion profile, and measure the full width at half maximum (FWHM) of Stark broadening dominated by the electron density broadening effect in the collision broadening profile. Step S104: Calculate the optical thickness invariant based on the geometric scale ratio of the full width at half maximum (FWHM) of the Stark broadening to the wavelength width at the self-reversing concave position, and use the ratio to offset the nonlinear deviation introduced by plasma volume distortion. Step S105: Obtain the real-time temperature of the molten metal and extract the reference spectral line intensity of the matrix reference element in the plasma; Step S106: Self-absorption correction of the light intensity of the characteristic spectral line is performed using the optical thickness invariant to obtain the corrected light intensity. The impurity content in the metal melt is determined by using a calibration model that establishes the mapping relationship between material concentration and corrected light intensity, real-time temperature, and reference spectral line intensity. Step S102 further includes sub-step S1021, which obtains the intensity gradient factors of the plasma plume in the height and radial directions, and dynamically adjusts the boundary distribution weights of the high-temperature center region and the low-temperature edge region based on the geometric symmetry deviation value determined by the intensity gradient factors.

2. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, Step S103 further includes the following sub-steps: Step S1031, the Stark broadening component in the collision broadening profile is separated from the characteristic spectral lines of the low-temperature edge region by the Lorentz fitting method; Step S1032, the half-width at half maximum (WHM) of the Lorentz fitting curve is extracted and used as the full width at half maximum (WHM) of the Stark broadening curve. The full width at half maximum (WHM) of the Stark broadening curve is used to characterize the microscopic particle interaction characteristics of the plasma and to counteract the macroscopic volume change of the plasma caused by the fluctuation of the molten metal surface.

3. The method for determining impurity content in metal melt detection according to claim 1, characterized in that, Step S103 further includes the following sub-steps: Step S1033, by searching for peaks in the characteristic spectral lines of the high-temperature central region, the wavelength position corresponding to the minimum point of the inversion intensity at the center of the inversion profile of the spectral lines is determined; Step S1034, the wavelength spacing between the symmetrical intensity points on both sides of the minimum point of the inversion intensity is measured, and this is used as the wavelength width of the inversion depression position.

4. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, It also includes step S107, which determines the theoretical emission intensity evolution characteristics of the characteristic spectral lines at real-time temperature based on the Boltzmann distribution law, and uses the theoretical emission intensity evolution characteristics to perform temperature compensation on the optical thickness invariant.

5. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, It also includes step S108, which normalizes the characteristic spectral lines using the intensity of the reference spectral lines to reduce the influence of laser-induced energy pulse fluctuations on the intensity of the characteristic spectral lines.

6. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, It also includes step S109, which measures the broadening characteristics of the matrix reference element, monitors in real time whether the low-temperature edge region is in an optically rarefied state, and filters the spectral data of the current sampling frame when the low-temperature edge region leaves the optically rarefied state.

7. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, In step S106, the calibration model is constructed through the following sub-step S1061: using the impurity content gradient determined by the standard metal sample set, a nonlinear mapping database is established using a multidimensional regression algorithm, which includes optical thickness invariants, the corrected intensity of characteristic spectral lines and real-time temperature as inputs, and impurity content as output.

8. The method for determining the impurity content in the detection of molten metal according to claim 1, characterized in that, After determining the impurity content in the molten metal, the process further includes step S110, which adjusts the amount of chemical reagents added to the refining equipment based on the deviation between the impurity content and the preset process parameters, thereby completing the closed-loop control of the metal smelting process.

Images