Self-adaptive full-band efficiency correction method for LIBS (laser-induced breakdown spectroscopy)

Abstract

The invention discloses a self-adaptive full-band efficiency correction method for an LIBS (laser-induced breakdown spectroscopy). The method comprises the following steps: for each laser pulse, inputting collected spectral data into a trained physical information deep network to obtain a current plasma temperature; when the current plasma temperature is not in a preset ideal temperature interval, predicting by using a physical information deep network to obtain an optimal laser energy adjustment sequence; wherein energy adjustment is performed according to the optimal laser energy adjustment sequence, so that a future temperature sequence can be converged to a preset ideal temperature interval; controlling an energy controller of the laser to emit a next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma; and after the stable plasma is obtained, collecting target spectral data, and outputting a full-band efficiency calibration curve according to the target spectral data. And high-precision and high-robustness full-band efficiency correction can be realized in a complex and changeable real environment.

Description

Technical Field

[0001] The present application relates to the technical field of efficiency correction of breakdown spectroscopy, and in particular to an adaptive full-band efficiency correction method for LIBS laser induced spectrometer. Background Art

[0002] Laser-induced breakdown spectroscopy (hereinafter referred to as "breakdown spectroscopy") technology is a key tool for elemental composition analysis of materials, and precise calibration of the spectrometer's full-band (especially the ultraviolet band) efficiency curve is the cornerstone for achieving accurate quantitative analysis.

[0003] Traditional calibration methods rely on standard light sources, such as halogen lamps for visible and near-infrared calibration, and expensive, life-limited deuterium lamps for UV calibration. This approach not only increases equipment complexity and cost, but the switching and splicing of the two light sources can easily introduce systematic errors.

[0004] As an improvement, a "correction method based on the principle of plasma temperature consistency" was proposed. This method utilizes a core physical assumption: under conditions of local thermodynamic equilibrium (hereinafter referred to as "local thermal equilibrium"), the temperature of the laser-induced plasma is uniform across the entire spectral range. By selecting multiple characteristic spectral lines of specific elements (such as iron) in the visible light band, the "reference temperature" of the plasma is calculated using the Saha-Boltzmann diagram method. Subsequently, based on this single temperature, the theoretical intensities of the spectral lines of these elements in the ultraviolet band are reversely calculated. Finally, by comparing the theoretical intensity of the ultraviolet spectral lines with the actual measured intensity, the efficiency correction coefficient of the spectrometer in this band is indirectly calculated. This method cleverly avoids dependence on a standard ultraviolet light source.

[0005] However, this method is still based on a relatively idealized physical model. In practical applications, developments in related technical fields, such as the application of deep learning in complex signal processing and active sensing and control in precision measurement, have provided new ideas for overcoming the limitations of existing methods. However, these have not yet been systematically integrated into the field of efficiency correction for breakdown spectroscopy. Summary of the Invention

[0006] The adaptive full-band efficiency correction method for LIBS laser-induced spectrometer provided in this application can achieve high-precision and high-robustness full-band efficiency correction in complex and changeable real environments.

[0007] In a first aspect, the present application provides an adaptive full-band efficiency correction method for a LIBS laser induced spectrometer, which includes: for each laser pulse, inputting the collected spectral data into a trained physical information deep network to obtain the current plasma temperature; when the current plasma temperature is not within a preset ideal temperature range, using the physical information deep network to predict and obtain an optimal laser energy adjustment sequence; wherein, energy adjustment according to the optimal laser energy adjustment sequence can enable the future temperature sequence to converge to the preset ideal temperature range; controlling the energy controller of the laser to emit the next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma; after obtaining the stable plasma, collecting target spectral data, and outputting a full-band efficiency calibration curve based on the target spectral data.

[0008] Among them, the energy controller of the laser is controlled to emit the next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma, including: calculating a first control instruction according to the optimal laser energy adjustment sequence; using the first control instruction to control the energy controller of the laser to emit the next laser pulse, executing the step of inputting the collected spectral data into the trained physical information deep network to obtain the current plasma temperature, until a stable plasma is obtained.

[0009] Among them, collecting target spectral data and outputting a full-band efficiency calibration curve based on the target spectral data include: collecting target spectral data; calibrating the visible light region in the target spectral data to obtain a visible light efficiency curve; calibrating the ultraviolet region in the target spectral data to obtain an ultraviolet efficiency curve; splicing the visible light efficiency curve and the ultraviolet efficiency curve to obtain a full-band efficiency curve; performing anomaly detection and correction on the full-band efficiency curve to obtain a full-band efficiency calibration curve.

[0010] The ultraviolet region in the target spectral data is calibrated to obtain an ultraviolet efficiency curve, including: calculating the theoretical intensity of the iron spectrum line in the ultraviolet region using a reference temperature; obtaining the measured intensity from the ultraviolet region in the target spectral data; and obtaining the ultraviolet efficiency curve based on the measured intensity and the theoretical intensity.

[0011] The UV efficiency curve is obtained according to the measured intensity and the theoretical intensity, including: obtaining an efficiency coefficient according to the measured intensity and the theoretical intensity; and obtaining the UV efficiency curve according to the efficiency coefficient.

[0012] Among them, the full-band efficiency curve is subjected to abnormality detection and correction to obtain the full-band efficiency calibration curve, including: morphological inspection and convergence inspection of the full-band efficiency curve; when an abnormality occurs, the agent optimization model is used to perform local correction on the full-band efficiency curve to obtain the full-band efficiency calibration curve.

[0013] Among them, the physical information deep network is trained in the following way: during training, a number of wavelength points are input into the physical information deep network, and the spectral intensity measured at the wavelength point is used as the target. The network weights are adjusted through the back propagation algorithm to minimize the mean square error between the intensity output by the physical information deep network and the measured intensity; and the physical information deep network is constrained by using a composite loss function and self-supervised learning.

[0014] Among them, the composite loss function includes data loss terms and physical loss terms for fitting the measured data; the physical loss terms include global Saha-Boltzmann consistency loss and temperature uniformity constraint loss; the goal of global Saha-Boltzmann consistency loss is to minimize the variance between the temperature values ​​calculated for all spectral lines and a global reference temperature to be optimized; the temperature uniformity constraint loss is composed of the square of the difference between the visible light temperature and the ultraviolet light temperature.

[0015] Among them, the physical information deep network includes an implicit neural representation network and an autoencoder structure, which is used to become a digital twin of the spectrum after training is completed.

[0016] Among them, the overall optimization goal of the adaptive full-band efficiency correction method for LIBS laser-induced spectrometer is as follows: ; in, represents the physical consistency loss function, Φ(⋅) is the implicit neural representation network, S measured (U) represents the original spectral data sequence actually measured, represents the global plasma reference temperature, η: the full-band efficiency curve function to be optimized, represents the control cost loss function, represents the efficiency curve regularization loss function.

[0017] The beneficial effects of the present application are as follows: Different from the prior art, the present application provides an adaptive full-band efficiency correction method for a LIBS laser induced spectrometer, which includes: for each laser pulse, inputting the collected spectral data into a trained physical information deep network to obtain the current plasma temperature; when the current plasma temperature is not within the preset ideal temperature range, using the physical information deep network for prediction to obtain the optimal laser energy adjustment sequence; wherein, energy adjustment according to the optimal laser energy adjustment sequence can make the future temperature sequence converge to the preset ideal temperature range; according to the optimal laser energy adjustment sequence, the energy controller of the laser is controlled to emit the next laser pulse to obtain a stable plasma; after obtaining the stable plasma, the target spectral data is collected, and the full-band efficiency calibration curve is output according to the target spectral data. In the above manner, high-precision and high-robustness full-band efficiency correction can be achieved in a complex and changeable real environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. Among them: Figure 1 This is a flow chart of an embodiment of a method for adaptive full-band efficiency correction of a LIBS laser induced spectrometer provided by the present application; Figure 2 yes Figure 1 A flow chart of an embodiment of step 13; Figure 3 yes Figure 1 FIG. 1 is a flow chart of an embodiment of step 14 in FIG. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It will be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some, rather than all, structures related to the present application are shown in the drawings. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0020] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] Although innovative, the plasma temperature consistency-based approach still faces a complete and deeply logically related technical challenge in practical application, which limits its accuracy, stability, and scope of application: Core challenge: The existing method is a passive, open-loop calibration process based on ideal physical assumptions. It is unable to cope with and compensate for the dynamic uncertainties and non-idealities of real plasmas, resulting in fundamental constraints on the reliability and accuracy of the calibration results.

[0022] This core challenge can be broken down into three interrelated layers: 1. Fragility of physical assumptions: This method relies heavily on an "ideal and stable" local thermal equilibrium state. However, real laser-induced plasma is an extremely complex transient process. Small fluctuations in laser energy, inhomogeneities on the sample surface, and the dynamic evolution of the plasma plume all cause plasma parameters (such as temperature and electron density) to vary dramatically between different laser pulses (i.e., "between pulses") and within the duration of a single pulse. This dynamic instability makes it difficult to perfectly and continuously maintain the local thermal equilibrium state, resulting in significant errors and fluctuations in the "reference temperature" calculated using the Saha-Boltzmann diagram. This, in turn, directly transfers to the efficiency calculation in the UV band, significantly compromising its accuracy.

[0023] 2. Limitations of Information Utilization: Traditional methods select only a few "clean," self-absorption-free iron spectral lines from massive amounts of spectral data for temperature calculation. This "manual screening" approach is not only highly subjective but also significantly wastes the rich information contained in the spectrum. Spectral line profiles, background bremsstrahlung, and spectral lines from other elements all contain valuable information about the plasma state. Relying solely on the intensity ratios of a few spectral lines makes this method extremely sensitive to line selection, the accuracy of the spectral database, and misjudgment of self-absorption effects, resulting in a severely limited robustness of the system.

[0024] 3. Passive nature of the calibration process: The entire calibration process is one-way and passive. The system first passively collects spectral data, then performs a series of calculations to ultimately produce a calibration result. It relies entirely on the assumption that the plasma is in ideal conditions at the moment the data is collected. If the plasma is not in optimal condition, the system cannot detect it, let alone take any action to intervene or compensate. It lacks real-time quality control and feedback correction, unable to proactively maintain the plasma in the optimal "sweet spot" for temperature consistency calibration, nor can it self-correct when potential deviations in the calibration results are detected.

[0025] Therefore, the present invention aims to address the aforementioned core technical challenge: how to transform the calibration process from a passive, open-loop calculation relying on ideal assumptions into an adaptive closed-loop system capable of real-time perception, active control, and intelligent correction. This approach enables high-precision, robust, full-band efficiency calibration in complex and changing real-world environments. See the following examples for details.

[0026] See Figure 1 , Figure 1 This is a flow chart of an embodiment of a method for adaptive full-band efficiency correction of a LIBS laser induced spectrometer provided in this application. The method for adaptive full-band efficiency correction of a LIBS laser induced spectrometer includes: Step 11: For each laser pulse, the collected spectral data is input into the trained physical information deep network to obtain the current plasma temperature.

[0027] In some embodiments, the physical information deep network is trained in the following manner: during training, a number of wavelength points are input into the physical information deep network, and the spectral intensity measured at the wavelength points is used as the target. The network weights are adjusted through the back propagation algorithm to minimize the mean square error between the intensity output by the physical information deep network and the measured intensity; and the physical information deep network is constrained using a composite loss function and self-supervised learning.

[0028] Among them, the composite loss function includes data loss terms and physical loss terms for fitting the measured data; the physical loss terms include global Saha-Boltzmann consistency loss and temperature uniformity constraint loss; the goal of global Saha-Boltzmann consistency loss is to minimize the variance between the temperature values ​​calculated for all spectral lines and a global reference temperature to be optimized; the temperature uniformity constraint loss is composed of the square of the difference between the visible light temperature and the ultraviolet light temperature.

[0029] Among them, the physical information deep network includes an implicit neural representation network and an autoencoder structure, which is used to become a digital twin of the spectrum after training is completed.

[0030] In some embodiments, the present application provides a four-stage closed-loop intelligent calibration system.

[0031] Phase 1: Plasma Full-Spectrum Implicit Neural Representation (High-Fidelity Data Foundation).

[0032] 1. Core Objectives: This phase addresses the low information utilization and sensitivity to raw data quality (e.g., noise) issues inherent in traditional methods. The goal is to transform discrete, noisy raw spectral data into a continuous, smooth, high-fidelity "full-spectrum digital twin," providing the highest-quality data foundation for subsequent physical analysis.

[0033] 2. Technical Core: Deeply integrate the ideas of Implicit Neural Representation (INR) and Autoencoder.

[0034] Implicit Neural Representation (INR): Instead of viewing the spectrum as an array of discrete data points, we construct a deep neural network (typically a multilayer perceptron (MLP)) that learns a continuous function intensity = f(wavelength). This function itself is a complete mathematical representation of the spectrum.

[0035] Autoencoder structure: The training process is similar to that of an autoencoder. The goal of the network is to input a series of wavelength coordinates and output intensity values ​​that can perfectly reconstruct (decode) the original measured spectrum.

[0036] 3. Implementation process: (a) Data acquisition: Collect raw, full-band breakdown spectrum data generated by a high-purity iron target under laser bombardment.

[0037] (b) Network construction: Design an MLP network whose input layer receives wavelength values ​​and whose output layer outputs a single intensity value.

[0038] (c) Model training: The collected spectral data is used as the training set. During training, a large number of wavelength points are input to the network, and the measured spectral intensities at these wavelength points are used as targets. The network weights are adjusted through the backpropagation algorithm to minimize the mean squared error between the network output intensity and the measured intensity.

[0039] (d) Output and Effect: After training, the neural network itself becomes a "digital twin" of the spectrum. It naturally has the following advantages: Denoising: When the network learns the overall distribution, it naturally ignores high-frequency random noise.

[0040] Continuity: The intensity value at any wavelength position can be queried to achieve "super-resolution" interpolation.

[0041] Completeness: Completely characterizes the entire spectrum from ultraviolet to near-infrared, including all spectral lines, background and other information.

[0042] 4. Key Role: This stage is a high-quality data preprocessor for the entire system. It refines the raw measurement data, which is full of uncertainty, into a deterministic, mathematically perfect function model, providing ideal noise-free and complete input for the second stage of physical constraint solving.

[0043] Phase 2: Physics-Constrained State Solving 1. Core Objectives: The core challenge of addressing the fragility of physical assumptions is to determine the unique and most reliable global plasma "reference temperature" from a high-fidelity "full-spectrum digital twin." This temperature must simultaneously satisfy the physical laws across the entire spectrum.

[0044] 2. Technical Core: The core concept of Physics-Informed Neural Network (PINN) is introduced to perform self-supervised learning.

[0045] Composite loss function: In the training network's loss function, in addition to the L_data term, which fits the measured data, we also creatively incorporate the L_physics term, which penalizes deviations from physical laws. The total loss function is L_total = L_data + λ L_physics, where λ is a balancing factor.

[0046] Self-supervised learning: The physical laws themselves (rather than external labeled data) provide supervisory signals for network training, guiding the network to converge to a physically valid state while fitting the data.

[0047] 3. Implementation process: (a) Define the physical loss term L_Physics: This is the essence of this stage and consists of multiple sub-terms: Global Saha-Boltzmann consistency loss: Extract all spectral lines of iron in the visible and ultraviolet regions from the spectrum represented by the network. Substitute the intensities and transition probabilities of these spectral lines into the Saha-Boltzmann equation to calculate the corresponding temperatures.

[0048] The goal of this loss term is to minimize the variance between the temperature values ​​calculated for all spectral lines and a global "reference temperature T" to be optimized. This forces the network to find a single temperature solution that can uniformly explain the intensity of all spectral lines.

[0049] Temperature uniformity constraint loss: More directly, the average temperatures T_visible and T_ultraviolet are calculated based on the spectral line sets in the visible and ultraviolet regions, respectively. This loss term is defined as (T_visible - T_ultraviolet)², which directly drives the temperature solutions of the two bands to converge to the same.

[0050] (b) Joint optimization: During the training process, the back-propagation algorithm not only adjusts the network weights to fit the spectral data, but also optimizes the global “reference temperature T” as a trainable parameter.

[0051] (c) Output and effect: After training convergence, the system outputs two core results: a more realistic "full-spectrum digital twin" that has been "corrected" by physical laws, and a high-precision and highly robust global reference temperature T_ref.

[0052] 4. Key role: This stage upgrades the neural network from a "data fitter" to a "physical solver." It utilizes full spectrum information and physical prior knowledge to resolve the uncertainty of plasma states and provide the most solid physical benchmark for efficiency calibration.

[0053] Step 12: When the current plasma temperature is not within the preset ideal temperature range, the physical information deep network is used for prediction to obtain the optimal laser energy adjustment sequence; wherein, energy adjustment according to the optimal laser energy adjustment sequence can make the future temperature sequence converge to the preset ideal temperature range.

[0054] Step 13: Control the laser's energy controller to emit the next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma.

[0055] In some embodiments, see Figure 2 , step 13 can be the following process: Step 131: Calculate the first control instruction according to the optimal laser energy adjustment sequence.

[0056] Step 132: Use the first control instruction to control the energy controller of the laser to emit the next laser pulse, and execute the step of inputting the collected spectral data into the trained physical information deep network to obtain the current plasma temperature until a stable plasma is obtained.

[0057] The third stage of the four-stage closed-loop intelligent calibration system: active sensing and model predictive control (Active Environment Stabilization) 1. Core Objectives: Solve the passive problem of the calibration process. The goal of this stage is to actively intervene in the experimental conditions before conducting formal efficiency calibration, driving and locking the actual plasma state into the most ideal and stable local thermal equilibrium "sweet spot".

[0058] 2. Technical Core: Deep fusion of information-optimal sensing (Information-Optimal Sensing) and model predictive control (MPC) strategies.

[0059] Real-time prediction model: The physical information network trained in the second stage is used as a fast and accurate "forward prediction model" in this stage. It can infer the plasma state in real time based on the current spectrum and predict the future evolution of the plasma state after adjusting the laser parameters.

[0060] Model Predictive Control (MPC): It is an advanced control strategy that uses a model to predict the system behavior for a period of time in the future during each control cycle, and optimizes a series of control operations based on this. It then executes only the first operation and repeats this process in the next cycle.

[0061] 3. Implementation process: (a) Real-time diagnostics: The system enters "pre-calibration" mode. For each laser pulse, the acquired spectrum is immediately fed into the solidified physical information network, which calculates the current plasma temperature, T_current, within milliseconds.

[0062] (b) State evaluation and control triggering: Compare T_current with the preset ideal temperature range [T_min, T_max]. If T_current is outside this range, the MPC controller is triggered.

[0063] (c) Prediction and Optimization: The MPC controller activates. Using the physical information network as a prediction engine, it performs rapid internal simulations to answer questions such as, "In the next five laser pulses, if the laser energy sequence is [ΔE1, ΔE2, ΔE3, ΔE4, ΔE5], what will the temperature sequence be?" The controller uses optimization algorithms (such as particle swarm optimization) to find the optimal laser energy adjustment sequence, with the goal of returning the future temperature sequence to the ideal range as quickly and smoothly as possible.

[0064] (d) Execution and loop: The MPC sends the first control instruction (e.g., ΔE1) of the calculated optimal sequence to the laser's energy controller for execution. After the next laser pulse, the entire closed-loop process (a) to (d) repeats.

[0065] 4. Key Role: This stage stabilizes the plasma state of the system. It ensures that the spectral data used for the final efficiency calculation is collected under the most stable conditions and closest to ideal physical assumptions, fundamentally guaranteeing the reliability and repeatability of the calibration results.

[0066] Step 14: After obtaining a stable plasma, collect target spectrum data and output a full-band efficiency calibration curve based on the target spectrum data.

[0067] In some embodiments, see Figure 3 , step 14 can be the following process: Step 141: Collect target spectrum data.

[0068] Step 142: Calibrate the visible light region in the target spectrum data to obtain a visible light efficiency curve.

[0069] Step 143: Calibrate the ultraviolet region in the target spectrum data to obtain an ultraviolet efficiency curve.

[0070] The theoretical intensity of the iron spectrum line in the ultraviolet region is calculated using a reference temperature. The measured intensity in the ultraviolet region of the target spectrum data is obtained. The ultraviolet efficiency curve is obtained based on the measured intensity and the theoretical intensity. For example, the efficiency coefficient is obtained based on the measured intensity and the theoretical intensity. The ultraviolet efficiency curve is obtained based on the efficiency coefficient.

[0071] Step 144: Combine the visible light efficiency curve and the ultraviolet efficiency curve to obtain a full-band efficiency curve.

[0072] Step 145: Perform abnormality detection and correction on the full-band efficiency curve to obtain a full-band efficiency calibration curve.

[0073] In some embodiments, the full-band efficiency curve is subjected to morphology check and convergence check; when an anomaly occurs, the full-band efficiency curve is partially corrected using the proxy optimization model to obtain a full-band efficiency calibration curve.

[0074] The fourth stage of the four-stage closed-loop intelligent calibration system: dual-mode error correction and efficiency curve generation (Quality Assurance & Self-Correction) 1. Core Objectives: On the basis of generating the final efficiency curve, a quality assurance and self-correction link is added to ensure the physical rationality and high precision of the output results.

[0075] 2. Technical Core: Drawing on the design philosophy of Physics-Driven Anomaly Detection and Correction.

[0076] 3. Implementation process: (a) Main Mode - Efficiency Calculation and Stitching: Visible light region calibration: Use a standard halogen lamp to complete the efficiency calibration above 400nm to obtain η_visible.

[0077] UV Calibration: After stabilizing the plasma in the third stage, spectra are collected. Using the high-precision reference temperature T_ref output from the second stage, the theoretical intensity of the iron line in the UV region, I_theoretical_UV, is calculated. The measured intensity, I_measured_UV, is then read from the spectrum. Finally, the efficiency coefficient η_UVPoint = I_measured_UV / I_theoretical_UV, is calculated for multiple discrete wavelength points in the UV region.

[0078] Curve Generation: Using methods such as cubic spline interpolation or polynomial fitting, these discrete η_UV points are connected to form a smooth UV efficiency curve η_UV. Finally, η_Visible and η_UV are smoothly spliced ​​together to form a full-band efficiency curve.

[0079] (b) Anomaly Detection: The system’s built-in “physical consistency check module” is activated to evaluate the generated curve: Morphology check: Check the curve for violent oscillations, spikes, or discontinuities that do not conform to physical meaning.

[0080] Convergence check: Perform multiple calibrations to check whether the output efficiency curve is stable and convergent. If there is a large fluctuation for a long time, it is considered abnormal.

[0081] (c) Correction Mode - Proxy Optimization: Once an anomaly is detected, the system activates a lightweight "proxy optimization model." Instead of re-performing the complex global optimization, this model performs rapid local fine-tuning based on the solution of the master physical information network. Its goal is to find a set of minimal corrections to the original solution that minimizes the detected physical inconsistency (e.g., smoothing a curve).

[0082] (d) Final output: Outputs a highly reliable full-band efficiency calibration curve after final verification and intelligent correction.

[0083] 4. Key Role: This phase is the system's quality assurance and feedback loop. It empowers the system with self-review and self-correction capabilities, preventing unexpected output errors and serving as the final line of defense for the system's high reliability.

[0084] In some embodiments, the overall optimization goal of the adaptive full-band efficiency correction method for LIBS laser induced spectrometer is as follows: ; in, represents the physical consistency loss function, Φ(⋅) is the implicit neural representation network, S measured (U) represents the original spectral data sequence actually measured, represents the global plasma reference temperature, η: the full-band efficiency curve function to be optimized, represents the control cost loss function, represents the efficiency curve regularization loss function.

[0085] This formula describes a multi-objective, multi-stage joint optimization problem. Below, we'll break down each part of the formula in detail and explain how it embodies the four-stage closed-loop intelligent calibration concept.

[0086] Part I: Physical consistency loss function Lphysics.

[0087] This is the cornerstone of the entire optimization framework, embodying the core ideas of the first stage (data characterization) and the second stage (physical constraint resolution). It measures the extent to which the efficiency-corrected spectral data conforms to the underlying physical laws.

[0088]

[0089] Parameter details: S measured (U): This is a function that represents the raw spectral data sequence actually measured under the system control strategy U. This reflects that in the third stage, the spectral data is the result of active control.

[0090] Φ(⋅): represents the implicit neural representation network of the first stage. It transforms the original, noisy spectral data S 实测 Mapped into a continuous, smooth, high-fidelity "full-spectrum digital twin".

[0091] T 参考 : Global plasma reference temperature. This is no longer a pre-calculated fixed value, but a variable to be solved for during the overall optimization.

[0092] η : Full-band efficiency curve function to be optimized.

[0093] : The efficiency-corrected spectral intensity is defined as .

[0094] A ( λ ), g ( λ ), E ( λ ) are wavelengths λ The transition probability of the spectral line at , the upper energy level degeneracy and the upper energy level energy are known physical constants.

[0095] k B : Boltzmann constant.

[0096] : This is a creative extension of the traditional Saha-Boltzmann graph expression. It no longer does linear fitting based on a few points, but requires all spectral lines The resulting Saha-Boltzmann relations must all refer to the same reference temperature. The goal of this loss term is to minimize the deviation of all spectral lines from this ideal linear relationship. This embodies the second stage of full spectrum information utilization and physical mechanism constraints.

[0097] : This is a more advanced physical constraint. T 参考 is the temperature calculated based on the neutral atomic spectral lines, and is the temperature calculated based on the ion spectrum. This constraint requires that both must be equal. This is stronger than the simple Saha–Boltzmann constraint because it also implicitly satisfies the Saha–Eggert ionization equilibrium equation, thus more comprehensively constraining the plasma state. γ is its weight factor.

[0098] Core idea: This part will be the efficiency curve η and reference temperature T 参考 Deep coupling means that they are no longer calculated step by step, but are solved jointly under a unified physical framework to maximize global physical consistency.

[0099] Part II: Control Cost Loss Function L Control This part embodies the core idea of ​​the third stage (active sensing and model predictive control). It ensures that the spectral data used for calibration are obtained under the optimal and most stable conditions.

[0100]

[0101] Parameter details: U: This is a time series, representing the time from the start to the end of pre-calibration T 预校准 The system control strategy, for example, the laser energy variation sequence u( t ).

[0102] : This is the predicted value of the future plasma temperature by the MPC controller using the physical information network.

[0103] : Preset ideal plasma temperature best suited for calibration.

[0104] : Penalizes the term that deviates from the predicted temperature to the ideal temperature.

[0105] : Penalizes the item with drastic changes in the control action (such as laser energy) to ensure the smoothness of the control process.

[0106] The core idea is: The optimization goal of this part is to find a control strategy U that can quickly and smoothly maintain the plasma in an ideal state, thereby providing the best quality input data for the first part.

[0107] Part III: Efficiency Curve Regularization Loss Function L Regularization This part reflects the idea of ​​ensuring the quality of the final result in the fourth stage (dual-mode error correction). It imposes physical prior knowledge on the solution of the efficiency curve.

[0108]

[0109] Parameter details: η ( λ ) : Full-band efficiency curve function to be optimized.

[0110] : Smoothness constraint. It penalizes the second-order derivative of the efficiency curve, that is, it requires that the curve cannot have sharp, meaningless bends or oscillations.

[0111] Monotonicity constraint (in certain wavelength bands). Physically, the efficiency of a spectrometer typically varies slowly with wavelength, exhibiting monotonicity in certain regions (e.g., near the blaze wavelength). This penalty curve exhibits an unexpected downward trend.

[0112] Core idea: This part ensures that even in the case of insufficient data or systematic deviations, the final efficiency curve is physically "reasonable" and "beautiful", avoiding overfitting and meaningless solutions, and embodies the quality control and correction concept of the fourth stage. Summary: Creativity and integrity of the formula The creativity of this overall optimization objective function lies in: 1. Dimensionality upgrade from "solving" to "joint optimization": It elevates the traditional step-by-step solution of the Saha-Boltzmann equation to a unified optimization problem that includes control strategies, system states, and measurement results.

[0113] 2. Deep integration of physics and data: Through the PINN concept, physical laws (Saha-Boltzmann, ionization equilibrium) are directly encoded as part of the loss function, achieving a seamless combination of data-driven and physical mechanisms.

[0114] 3. Mathematical expression of closed-loop thinking: By taking the control strategy as the optimization variable and influencing it (through S 实测 (U)) is passed to the physical loss term, which perfectly expresses the closed-loop idea of ​​"active control to serve precise measurement" in mathematics.

[0115] Holistic: This formula organically connects the core logic of the four stages: L control corresponds to the third stage, ensuring high-quality data source S 实测 Φ(⋅) corresponds to the first stage, purifying the data; L physics corresponds to the second stage, applying physical laws to solve the core parameters; L regularization corresponds to the fourth stage, imposing quality constraints on the final results. All these components work together to achieve a common goal, forming a logically rigorous and indivisible whole.

[0116] The unexpected technical effects are as follows: Through the organic combination of the above technical means, the solution of the present invention not only solves the preset technical problems, but also brings a series of non-obvious technical effects that go beyond the scope of traditional calibration methods, constituting the creativity of the solution.

[0117] 1. A paradigm shift from "passive calculation" to "active creation of conditions": Traditional methods passively accept measurement data, and their accuracy relies entirely on the uncontrollable plasma state. This invention overturns this paradigm. Through active sensing and model predictive control, it proactively "creates" and "maintains" the ideal physical conditions for calibration. The system is no longer a passive user of data, but an active builder of the ideal measurement environment. This transforms high-precision calibration from a "low-probability event" into a "normalized capability."

[0118] 2. Dramatic Improvements in Robustness and Accuracy: Because the calibration process is based on a deep fusion of full-spectrum information and physical mechanisms, and is safeguarded by an active control system, this invention's tolerance to interference factors such as laser energy fluctuations, sample surface defects, background noise, and spectral self-absorption is orders of magnitude greater. Calibration results no longer fluctuate significantly due to the "luck" of a single measurement, achieving high repeatability and stability, with accuracy approaching that of methods using high-precision standard light sources.

[0119] 3. The system combines both calibration and advanced diagnostics: The deep physical information network constructed by this invention is itself a powerful real-time online plasma diagnostic tool. While completing calibration tasks, it continuously outputs core plasma parameters (temperature, electron density) with high accuracy and timeliness. This provides an unprecedentedly powerful tool for studying the mechanisms of laser-matter interactions, optimizing breakdown spectroscopy experimental conditions, and even online monitoring of material ablation processes—an added value that traditional calibration methods simply cannot provide.

[0120] 4. Significantly Expands the Technology's Applicability: Because this system proactively responds to and compensates for suboptimal operating conditions, it significantly reduces the requirements for laser system stability and sample uniformity. This means that this calibration method can be deployed in lower-cost, more complex breakdown spectroscopy systems. Whether for online quality control in industrial sites or in-situ geological exploration in the field, laboratory-level calibration accuracy can be achieved, significantly promoting the widespread adoption and application of breakdown spectroscopy technology.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical functional division. In actual implementation, other division methods may be used, such as combining or integrating multiple units or components into another system, or ignoring or not implementing certain features.

[0122] If the integrated units in the other embodiments described above are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0123] The above description is only an implementation method of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the description and drawings of this application, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for adaptive full-band efficiency correction of LIBS laser induced spectrometer, characterized in that: The LIBS laser induced spectrometer adaptive full-band efficiency correction method includes: For each laser pulse, the collected spectral data is input into the trained physical information deep network to obtain the current plasma temperature; When the current plasma temperature is not within a preset ideal temperature range, the physical information deep network is used to perform prediction to obtain an optimal laser energy adjustment sequence; wherein energy adjustment according to the optimal laser energy adjustment sequence can make the future temperature sequence converge to the preset ideal temperature range; An energy controller of the laser is controlled to emit a next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma; After the stable plasma is obtained, target spectrum data is collected, and a full-band efficiency calibration curve is output according to the target spectrum data.

2. The method according to claim 1, characterized in that The energy controller of the laser is controlled to emit the next laser pulse according to the optimal laser energy adjustment sequence to obtain a stable plasma, comprising: Calculating a first control instruction according to the optimal laser energy adjustment sequence; The first control instruction is used to control the energy controller of the laser to emit the next laser pulse, and the collected spectral data is input into the trained physical information deep network to obtain the current plasma temperature until a stable plasma is obtained.

3. The method according to claim 1, characterized in that The collecting target spectral data and outputting a full-band efficiency calibration curve according to the target spectral data include: Collect target spectral data; Calibrate the visible light region in the target spectral data to obtain a visible light efficiency curve; Calibrate the ultraviolet region in the target spectrum data to obtain an ultraviolet efficiency curve; splicing the visible light efficiency curve and the ultraviolet efficiency curve to obtain a full-band efficiency curve; Anomaly detection and correction are performed on the full-band efficiency curve to obtain the full-band efficiency calibration curve.

4. The method according to claim 3, characterized in that The calibrating the ultraviolet region in the target spectrum data to obtain an ultraviolet efficiency curve includes: The theoretical intensity of the iron spectrum in the ultraviolet region is calculated using the reference temperature; Obtaining measured intensity from the ultraviolet region of the target spectrum data; The ultraviolet efficiency curve is obtained according to the measured intensity and the theoretical intensity.

5. The method according to claim 4, characterized in that The step of obtaining the UV efficiency curve according to the measured intensity and the theoretical intensity includes: Obtaining an efficiency coefficient based on the measured intensity and the theoretical intensity; The ultraviolet efficiency curve is obtained according to the efficiency coefficient.

6. The method according to claim 3, characterized in that The abnormality detection and correction of the full-band efficiency curve to obtain the full-band efficiency calibration curve includes: Performing morphology inspection and convergence inspection on the full-band efficiency curve; When an abnormality occurs, the full-band efficiency curve is partially corrected using the proxy optimization model to obtain the full-band efficiency calibration curve.

7. The method according to claim 1, characterized in that The physical information deep network is trained in the following way: During training, a number of wavelength points are input to the physical information deep network, and the spectral intensity measured at these wavelength points is used as the target. The network weights are adjusted through the back-propagation algorithm to minimize the mean square error between the intensity output by the physical information deep network and the measured intensity. And the physical information deep network is constrained by using a composite loss function and self-supervised learning.

8. The method according to claim 7, characterized in that The composite loss function includes a data loss term and a physical loss term for fitting measured data; The physical loss term includes global Saha-Boltzmann consistency loss and temperature uniformity constraint loss; The goal of the global Saha-Boltzmann consistency loss is to minimize the variance between the temperature values ​​calculated for all spectral lines and a global reference temperature to be optimized; The temperature uniformity constraint loss is composed of the square of the difference between the visible light temperature and the ultraviolet light temperature.

9. The method according to claim 7, characterized in that The physical information deep network includes an implicit neural representation network and an autoencoder structure, which is used to become a digital twin of the spectrum after training is completed.

10. The method according to any one of claims 1 to 9, characterized in that The overall optimization goal of the adaptive full-band efficiency correction method for LIBS laser-induced spectrometer is as follows: ； in, represents the physical consistency loss function, Φ(⋅) is the implicit neural representation network, S measured (U) represents the original spectral data sequence actually measured, represents the global plasma reference temperature, η: the full-band efficiency curve function to be optimized, represents the control cost loss function, represents the efficiency curve regularization loss function.

Images