A method and system for online monitoring of metal hot forming bath metallurgical reactions

Abstract

This invention discloses an online monitoring method and system for metallurgical reactions in a molten pool during metal hot forming. The system includes a metal hot forming unit, a metal oxide flow detection unit, a spectral detection unit, and a control and data processing unit. The control and data processing unit, based on the metal hot forming unit, the metal oxide flow detection unit, and the spectral detection unit, employs an online monitoring method for real-time prediction of the metallurgical reaction state. This includes real-time acquisition of characteristic spectral signals of metal oxide particles generated during the hot forming process, and analysis of the molecular characteristic emission bands of various metal oxides. From the molecular characteristic emission bands, the type, mass percentage content, and real-time content change rate of each metal oxide at the current moment are identified and calculated. Input features are constructed and input into an XGBoost classification model and an XGBoost regression model, respectively, to obtain the type, degree of reaction, and reaction rate of the metallurgical reaction at the current moment. This invention achieves real-time monitoring and quantitative characterization of the entire metallurgical reaction process during metal hot forming.

Description

Technical Field

[0001] This invention relates to the field of metal hot forming technology, and specifically to a method and system for online monitoring of metallurgical reactions in the molten pool during metal hot forming. Background Technology

[0002] During the hot forming process of metal, a series of complex metallurgical reactions occur in the molten pool, such as oxidation reaction, alloy element burn-off reaction, and phase transformation reaction. These metallurgical reactions directly determine the chemical composition, microstructure, and mechanical properties of the formed parts.

[0003] In existing technologies, monitoring of metallurgical reactions in the molten pool during metal hot forming mainly relies on offline detection methods, such as cross-sectional analysis and chemical composition detection of the formed parts. These methods cannot achieve in-situ real-time monitoring, making it difficult to detect metallurgical reaction anomalies and adjust process parameters in a timely manner. Some online monitoring technologies focus on detecting macroscopic parameters such as molten pool temperature and morphology, failing to delve into the essential characteristics of the internal reactions. They cannot deduce the type of metallurgical reaction, quantify the degree of reaction, or the dynamic rate of reaction, making it difficult to accurately reflect the real-time state of the metallurgical reactions within the molten pool. Furthermore, existing detection technologies mostly target element burn-off itself, failing to establish a correlation with the metallurgical reaction through the key intermediate product of metal oxides. This results in low coupling between the monitoring signal and the metallurgical reaction, and significant deviations in the quantification results. Consequently, they cannot provide a reliable basis for the precise control of the metal hot forming process, and cannot meet the demands of high-end equipment manufacturing for high-quality, high-stability, large-scale production of hot-formed parts. Summary of the Invention

[0004] Purpose of the invention: The first purpose of this invention is to provide an online monitoring method for metallurgical reactions in the molten pool during metal hot forming, so as to accurately predict the type, degree, and rate of metallurgical reactions in the molten pool during the metal hot forming process.

[0005] The second objective of this invention is to provide an online monitoring system for metallurgical reactions in a molten pool during hot forming of metals, enabling real-time online detection of the type, degree, and rate of metallurgical reactions within the molten pool.

[0006] Technical solution: To achieve the above objectives, the present invention provides an online monitoring method for metallurgical reactions in a molten pool during hot forming, comprising:

[0007] The characteristic spectral signals of metal oxide particles generated during the hot forming process are acquired in real time, and the molecular characteristic emission bands of various metal oxides generated by the metallurgical reaction are analyzed from the characteristic spectral signals.

[0008] Based on the molecular characteristic emission bands of the various metal oxides, the types, mass percentage content, and real-time content change rate of each metal oxide at the current moment are identified and calculated. These are used as input features for predicting the metallurgical reaction state and are respectively input into the XGBoost classification model and the XGBoost regression model. The category of the metallurgical reaction at the current moment is obtained based on the XGBoost classification model, and the reaction degree value and reaction rate value of each category of the metallurgical reaction at the current moment are obtained based on the XGBoost regression model.

[0009] The XGBoost classification model and XGBoost regression model are based on the XGBoost algorithm. They use the types, mass percentages, and real-time content change rates of metal oxides continuously collected during the hot forming process of several metals to construct a common input feature matrix. The XGBoost classification model is trained with the metallurgical reaction category vectors that are time-aligned with the input feature matrix as the target features. The XGBoost regression model is trained with the reaction degree vectors and reaction rate vectors of each category that are time-aligned with the input feature matrix as the target features.

[0010] Preferably, the characteristic spectral signal is preprocessed, including abnormal spectrum removal, adaptive baseline correction, multi-scale noise reduction, molecular band feature extraction, and protective gas background subtraction.

[0011] Preferably, the method for identifying the type of each metal oxide at the current time is as follows: the molecular characteristic emission spectrum is compared with the standard characteristic peak position, relative peak intensity ratio, and characteristic wavelength range of each oxide in the standard database of molecular emission spectrum of metal oxides. When the matching degree between a certain molecular characteristic emission spectrum and the standard characteristic peak position is ≥95% and the relative peak intensity ratio deviation is ≤10%, the type of metal oxide corresponding to the molecular characteristic emission spectrum is determined.

[0012] Preferably, the method for calculating the mass percentage content and real-time content change rate of each metal oxide at the current moment is as follows: for the identified metal oxides, extract the characteristic peak intensity of the molecular characteristic emission spectrum of the metal oxide, substitute it into the pre-built calibration curve, and calculate the mass percentage content of the metal oxide at the current moment; based on the difference between the mass percentage content of the metal oxide at the previous moment and the current moment, divide it by the time step to obtain the content change rate of the metal oxide at the current moment.

[0013] Preferably, the constructed common input feature is represented as: X=[X1,X2,X3], where X1 is the sequence feature matrix composed of the metal oxide type codes at each time step, and the matrix elements x 1(i,j)This represents the code value of the j-th metal oxide at time i, where i = 1 ~ I, j = 1 ~ J, and I and J represent the total number of time points and the number of metal oxide types, respectively; when the j-th metal oxide is identified at time i, x 1(i,j) =1, x when not detected 1(i,j) =0;

[0014] X2 is a matrix of metal oxide content values ​​at each time point, with matrix element x... 2(i,j) This represents the mass percentage content of the j-th metal oxide at time i.

[0015] X3 is a matrix representing the rate of change of metal oxide content at each time point, with matrix elements x... 3(i,j) This represents the rate of change of the content of the j-th metal oxide at time i.

[0016] Preferably, the metallurgical reaction category vector is represented as an I*N column vector Y. type Matrix element Y type(i,n) Y represents the state of the nth metallurgical reaction at time i, where n = 1 to N. type(i,n) =1 indicates that at time i, the metallurgical reaction corresponding to that number is occurring in the molten pool; Y type(i,n) =0 indicates that at time i, the metallurgical reaction corresponding to that number in the molten pool has not occurred;

[0017] The degree of reaction vector is represented by an I*N column vector Y. degree The degree of reaction of each metallurgical reaction is quantified by the ratio of the total mass of metal elements that have participated in the oxidation reaction in the molten pool to the total mass of the initial metal elements at different times.

[0018] The reaction rate vector is represented as an I*N column vector Y. rate The rate of change of the content of various metal oxides in the molten pool at different times is quantitatively labeled to indicate the intensity of each metallurgical reaction.

[0019] The present invention discloses an online monitoring system for metallurgical reactions in a molten pool during metal hot forming, comprising a metal hot forming unit for performing metal hot forming operations, a metal oxide flow detection unit for collecting flue gas containing metal oxide particles generated during the metal hot forming operations, a spectral detection unit for acquiring spectral data of the flue gas, and a control and data processing unit for obtaining the type, degree, and rate prediction results of the metallurgical reaction based on the acquired spectral data.

[0020] Preferably, the metal thermoforming unit includes a power supply, a protective gas cylinder, a nozzle, a welding torch, a wire feeding device, a molten pool, and a worktable; the metal oxide flow detection unit is located above the tail of the molten pool and includes a fan and a duct; the spectral detection unit is located on the side of the duct and includes a pulsed laser, a collecting mirror, a spectrometer, and a delay generator.

[0021] Preferably, the control and data processing unit includes a spectral signal acquisition module, a data preprocessing module, a metallurgical reaction state inference module, and a metallurgical reaction state display module;

[0022] Spectral signal acquisition module: used to acquire the characteristic spectral signals of metal oxide particles generated during metal hot forming;

[0023] Data preprocessing module: used to extract the molecular characteristic emission bands of various metal oxides generated by metallurgical reactions from characteristic spectral signals, and based on the molecular characteristic emission bands of each metal oxide, identify the types of metal oxides at the current time, calculate the content and content change rate of each metal oxide at the current time, and further construct the input features for metallurgical reaction state analysis.

[0024] Metallurgical reaction state inference module: includes a metallurgical reaction state prediction model, which is used to predict the metallurgical reaction state at the current moment based on input features, including metallurgical reaction type, degree of metallurgical reaction, and metallurgical reaction rate.

[0025] Metallurgical reaction status display module: used to display the predicted metallurgical reaction status at the current moment in real time on the human-computer interaction interface.

[0026] Preferably, the metallurgical reaction state prediction model includes an XGBoost classification model and an XGBoost regression model. The XGBoost classification model and the XGBoost regression model are based on the XGBoost algorithm and use the types, mass percentage contents, and real-time content change rates of metal oxides continuously collected during the hot forming process of several metals to construct a common input feature matrix. The XGBoost classification model is trained with the metallurgical reaction category vector that is time-aligned with the input feature matrix as the target feature. The XGBoost regression model is trained with the reaction degree vector and reaction rate vector of each category that are time-aligned with the input feature matrix as the target features.

[0027] Beneficial effects: The present invention has the following advantages: The present invention establishes a correlation between metal oxides, a key intermediate product, and metallurgical reactions, thereby achieving real-time and high-precision monitoring of the type, degree, and rate of metallurgical reactions during the metal hot forming process. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the online monitoring system structure in Example 1;

[0029] Figure 2 This is a schematic diagram of the control and data processing unit framework in Example 1;

[0030] Figure 3This is a schematic diagram of the online monitoring method in Example 2;

[0031] The components include: 1. Metal hot forming unit; 1-1. Power supply; 1-2. Protective gas cylinder; 1-3. Nozzle; 1-4. Welding torch; 1-5. Wire feeding device; 1-6. Molten pool; 1-7. Workpiece; 1-8. Worktable.

[0032] 2. Metal oxide flow detection unit; 2-1 Fan; 2-2 Conduit;

[0033] 3. Spectroscopic detection unit; 3-1. Pulsed laser; 3-2. Collecting mirror; 3-3. Spectrometer; 3-4. Delay generator;

[0034] 4. Control and data processing unit. Detailed Implementation

[0035] The technical solution of the present invention will be described in detail below with reference to the embodiments and accompanying drawings.

[0036] Example 1

[0037] like Figure 1 As shown, this embodiment provides an online monitoring system for metallurgical reactions in a molten pool during hot forming of metal, including a metal hot forming unit 1, a metal oxide flow detection unit 2, a spectral detection unit 3, and a control and data processing unit 4.

[0038] The metal hot forming unit 1 is deployed in an open working environment and includes a power supply 1-1, a welding torch 1-4, a wire feeder 1-5, and coaxially arranged nozzles 1-3. Upon system startup, high-purity pure argon protective gas is first blown into the predetermined forming area through nozzles 1-3. The protective gas flow rate needs to be dynamically matched according to the power of the energy source and the scanning speed to ensure the formation and maintenance of a local inert atmosphere in the molten pool 1-6. This avoids interference from external air on the metallurgical reaction and provides a pure environment for the precise collection of metal oxides.

[0039] The metal oxide flow detection unit 2 is installed 15mm directly above the molten pool 1-6 of the metal hot forming unit 1, and includes a fan 2-1 and a duct 2-2. When the fan 2-1 is working, it provides a directional airflow speed of 1.2m / s. After the protective gas carries the oxide particles and completes the spectral detection in the duct, it is naturally discharged from the tail end.

[0040] The spectral detection unit 3 precisely corresponds to the particle detection end of the metal oxide conduction detection unit 2, and includes a pulsed laser 3-1, a collecting mirror 3-2, a spectrometer 3-3, and a delay generator 3-4. The output energy of the pulsed laser 3-1 is set to 120 mJ, which can effectively excite metal oxide particles. The delay generator 3-4 controls the trigger time difference between the pulsed laser 3-1 and the spectrometer 3-3 to be 5~30 μs, matching the formation window of free radicals in metal oxide molecules during the plasma cooling stage, ensuring the acquisition of high-intensity, high signal-to-noise ratio molecular characteristic emission band signals. The detection wavelength range of the spectrometer 3-3 is 200-800 nm, and the detection frequency is set to 80 Hz, which can quickly acquire and transmit spectral data.

[0041] The control and data processing unit 4 is connected to the spectral detection unit 3 and the metal thermoforming unit 1 via wired communication, and has pre-stored an argon characteristic peak database and a standard database of metal oxide molecular emission bands; for example Figure 2 As shown, the system includes a built-in spectral signal acquisition module, a data preprocessing module, a metallurgical reaction state inversion module, and a metallurgical reaction state display module. The spectral signal acquisition module acquires the characteristic spectral signals of metal oxide particles generated during metal hot forming. The data preprocessing module analyzes the characteristic emission bands of various metal oxides generated by the metallurgical reaction from the characteristic spectral signals, identifies the types of metal oxides at the current moment, calculates the content and content change rate of each metal oxide at the current moment, and constructs input features for metallurgical reaction state analysis based on these data. The metallurgical reaction state inversion module includes a metallurgical reaction state prediction model, specifically an XGBoost classification model and an XGBoost regression model. Based on the input features, the XGBoost classification model and XGBoost regression model predict the type of metallurgical reaction, the degree of metallurgical reaction, and the metallurgical reaction rate at the current moment, respectively.

[0042] The metallurgical reaction status display module can display the predicted metallurgical reaction status at the current moment in real time on the human-computer interaction interface, including the metallurgical reaction type, the degree of metallurgical reaction, and the metallurgical reaction rate.

[0043] The system described in this embodiment, based on a metal hot forming unit 1, a metal oxide flow detection unit 2, a spectral detection unit 3, and a control and data processing unit 4, can achieve real-time monitoring and quantitative characterization of the entire metallurgical reaction process during metal hot forming. Figure 3 As shown, the specific usage process is as follows:

[0044] S1: System Initialization: Start the nozzle of the metal hot forming unit and blow pure argon at the preset flow rate. Simultaneously start the fan of the metal oxide flow guiding detection unit to form a stable directional airflow at the inlet of the collection chamber, ensuring that the airflow field covers the entire area above the molten pool.

[0045] S2: Metal hot forming and oxide flow: Start the power supply and wire feeding device to begin the metal hot forming operation. The metal material in the molten pool undergoes a series of metallurgical reactions at high temperature. The generated metal oxide particles are all guided into the guide tube by the directional airflow of the metal oxide flow detection unit.

[0046] S3: Spectral detection and data preprocessing: The pulsed laser of the spectral signal acquisition module in the control and data processing unit is continuously focused on the metal oxide particles in the conduit, exciting the particles to generate characteristic spectra. The collecting mirror converges the spectral signal to the spectrometer, and the spectrometer acquires spectral data in real time and transmits it to the data preprocessing module.

[0047] S4: Metallurgical Reaction Backpropagation and Monitoring: The data preprocessing module in the control and data processing unit extracts the types, contents, and real-time change rates of oxides from the spectral data and inputs them into the metallurgical reaction state prediction model trained in the metallurgical reaction state backpropagation module. The model outputs the type, degree assessment value, and dynamic change trend of the metallurgical reaction inside the molten pool, realizing real-time monitoring of the metallurgical reaction.

[0048] Example 2

[0049] Taking the hot forming metallurgical process of Q235 steel as an example, this embodiment provides an online monitoring method for the metallurgical reaction of the molten pool during metal hot forming, including the following:

[0050] 1. Collect the characteristic spectral signals of metal oxide particles generated during the hot forming process of Q235 steel, and preprocess them. From the preprocessed characteristic spectral signals, extract the molecular characteristic emission bands of various metal oxides generated by the metallurgical reaction. Each molecular characteristic emission band is a time sequence composed of emission intensity data of characteristic wavelength range corresponding to the same metal oxide at continuous sampling time with a fixed sampling time step.

[0051] The preprocessing includes abnormal spectrum removal, adaptive baseline correction, multi-scale noise reduction, and molecular band feature extraction. If pure argon protective gas is introduced into the working area during the metal hot forming process, the argon background needs to be further subtracted.

[0052] Furthermore, for the construction of a sample library for Q235 steel, the temperature, oxygen partial pressure, and reaction kinetics of the molten pool metallurgical reaction can be changed by adjusting the hot forming process parameters. This allows for the control of the types, contents, and formation rates of metal oxides generated by the metallurgical reaction, enabling the collection of a large number of labeled sample data under multiple working conditions. The process parameters include welding current, voltage, wire feed speed, and shielding gas flow rate.

[0053] 2. Match the molecular characteristic emission bands of each metal oxide with the standard database of metal oxide molecular emission bands to identify the types of metal oxides at each time point. Compare the measured molecular characteristic emission bands with the standard characteristic peak positions, relative peak intensity ratios, and characteristic wavelength ranges of each oxide in the database. When the matching degree of the characteristic peak positions of the measured molecular characteristic emission bands with a certain standard oxide band is ≥95% and the deviation of the relative peak intensity ratio is ≤10%, it is determined that the corresponding type of metal oxide has been detected at that time point.

[0054] For the identified metal oxides, the characteristic peak intensities of their corresponding molecular emission bands are extracted and substituted into a pre-established calibration curve to calculate the mass percentage (wt%) of the metal oxide at that time. The pre-established calibration curve is a linear fitting equation established in advance through standard sample experiments, relating the integrated intensity of the characteristic peak of the molecular emission band of the metal oxide to the mass percentage content of the metal oxide. By substituting the measured characteristic peak intensities as independent variables into the equation, the real-time content of the corresponding oxide can be directly obtained.

[0055] Based on the content of various metal oxides at each time point, the real-time content change rate of each metal oxide is calculated using the difference method. Using a fixed sampling time step Δt (in seconds) as the time interval, the content of the same metal oxide at time i and time i-1 is calculated using the difference method. The formula for calculating the content change rate of the metal oxide at time i is:

[0056]

[0057] In the formula, v(i) is the rate of change of the content of the metal oxide at time i, in wt% / s; ω(i) is the measured content of the oxide at time i, and ω(i-1) is the measured content of the oxide at the previous time.

[0058] 3. Construct a supervised learning dataset for metallurgical reaction state analysis during the hot forming process of Q235 steel, including the input feature matrix X and the output label matrix Y.

[0059] Based on the time series data of metal oxides obtained in step 2, the input feature matrix X=[X1,X2,X3] is constructed.

[0060] Wherein, X1 is a sequence feature matrix composed of metal oxide type codes at each time point, that is, the metal oxide type codes contained in the metal oxide particles produced during the hot forming process of Q235 steel (such as One-Hot codes), reflecting the metal oxide categories present at different times; in the X1 matrix, the rows are time points, the columns are the code values ​​of metal oxides, and the matrix elements x 1(i,j) Let x represent the code value of the j-th metal oxide at time i, where i = 1 ~ I, j = 1 ~ J, and I and J represent the total number of time points and the number of metal oxide types, respectively. When the j-th metal oxide is detected at time i, x 1(i,j) =1, x when not detected 1(i,j) =0, which reflects the types of metal oxides present in the molten pool at different times. If FeO, MnO, and SiO2 are detected at time i, but Fe2O3 and Fe3O4 are not detected, then the X1 row vector corresponding to that time is [1,0,0,1,1].

[0061] X2 is a matrix of metal oxide content values ​​at various times. In the X2 matrix, the rows represent the time, and the columns represent the metal oxide content values. Matrix elements x 2(i,j) This represents the content (in wt%) of the j-th metal oxide at time i. A matrix-matched dual internal standard calibration method is used, with the argon characteristic spectral line as the first internal standard and the characteristic spectral lines of the major metal elements as the second internal standard. Based on the calibration curve, the net peak area of ​​the corresponding oxide molecular band in the sample is converted into the real-time mass content.

[0062] X3 is a matrix representing the rate of change of metal oxide content at various time points. In the X3 matrix, rows represent time points, and columns represent the rate of change of metal oxide content. Matrix elements x 3(i,j) This represents the rate of change of the content of the j-th metal oxide at time i (in wt% / s).

[0063] Further construct the output label matrix Y corresponding to the input feature matrix X, Y = [Y type , Y degree ,Y rate In this matrix, the rows of matrix Y correspond one-to-one with the rows of matrix X, and the columns consist of three output labels:

[0064] Y type This is an I*N column vector representing the types of metallurgical reactions occurring in the molten pool at each time step, where N is the total number of monitored metallurgical reactions. Matrix element Y type(i,n) This represents the state of the nth metallurgical reaction at time i, where n = 1 to N, and:

[0065] Y type(i,n) =1: This means that at time i, the metallurgical reaction corresponding to that number is taking place in the molten pool;

[0066] Ytype(i,n) =0: This means that at time i, the metallurgical reaction corresponding to that number in the molten pool did not occur.

[0067] Table 1 shows the types of metallurgical reactions and quantitative judgment conditions during the hot forming process of Q235 steel.

[0068] Table 1. Types and Quantitative Judgment Criteria of Metallurgical Reactions During Hot Forming of Q235 Steel

[0069]

[0070] Y degree It is an I*N column vector representing the degree of reaction of different metallurgical reactions in the molten pool at each time. It is quantified by the ratio of the total mass of alloying elements that have participated in the oxidation reaction to the total mass of the initial alloying elements in the molten pool at different times.

[0071] Y rate It is an I*N column vector representing the reaction rate of different metallurgical reactions in the molten pool at each time. It is quantified by the rate of change of the content of various metal oxides in the molten pool at different times, thus characterizing the intensity of the metallurgical reaction.

[0072] 4. Using the dataset, two XGBoost models were trained to obtain prediction models for the metallurgical reaction state of Q235 steel, including a classification model and a regression model.

[0073] (1) XGBoost classification model: used for predicting metallurgical reaction types, with advantages such as handling high-dimensional features, automatic feature selection, and preventing overfitting. Model structure: uses 30 decision trees with a maximum depth of 8, a learning rate of 0.1, a subsampling rate of 0.8, and regularization parameters λ=1 and γ=0.1. Training process: the input feature X is used as the model input, and the classification label Y is used as the model input. type As the prediction target, the model is iteratively trained by minimizing the multi-label binary cross-entropy loss function to establish a nonlinear mapping relationship between input features and metallurgical reaction types. Through multi-label classification output, the corresponding metallurgical reaction type encoding is obtained, thereby achieving accurate identification of metallurgical reaction types.

[0074] (2) XGBoost Regression Model: Used to simultaneously output two regression tasks: response degree and response rate. The model structure is consistent with the classification model to ensure model synergy. Training process: Input features X are used as model input, and regression labels Y are used to train the model. degree Y rate As a prediction target, the model is iteratively trained by minimizing the mean squared error (MSE) loss function to establish a nonlinear mapping relationship between input features and reaction degree and reaction rate, and to simultaneously output reaction degree and reaction rate values, thereby achieving synchronous quantitative output of the two regression tasks.

[0075] To further optimize training performance, the objective functions of both models are unified to include a regularization term Ω. Ω is the regularization term for the XGBoost model, containing a leaf node number penalty and a leaf node weight L2 regularization term, used to control model complexity and prevent overfitting. Specifically, the objective function of the XGBoost classification model primarily uses a multi-label binary cross-entropy loss function, while the objective function of the XGBoost regression model primarily uses a mean squared error loss function. For small-sample abnormal reactions such as deoxygenation and secondary oxidation, the loss weights of the XGBoost classification model are increased to alleviate the sample imbalance problem. Finally, two well-trained XGBoost prediction models are obtained.

[0076] 5. For the hot forming process of Q235 steel, using spectral acquisition parameters completely consistent with those of the sample acquisition, the characteristic emission spectral signals of metal oxide particles generated in the hot forming metallurgical reaction zone are acquired in real time. Further analysis yields the current-time feature vectors X1 (metal oxide type), X2 (content), and X3 (content change rate), forming a feature matrix X that meets the model input requirements. X is then input into two pre-trained XGBoost models in real time: ① Input to the XGBoost classification model, which outputs the metallurgical reaction type of the Q235 steel hot forming process at the current time; ② Input to the XGBoost regression model, which simultaneously outputs the degree of metallurgical reaction and the metallurgical reaction rate at the current time. Combining the outputs of the two models, real-time monitoring and quantitative characterization of the entire metallurgical reaction process during hot forming are achieved.

Claims

1. A method for online monitoring of metallurgical reactions in a molten pool during hot forming of metals, characterized in that, include: The characteristic spectral signals of metal oxide particles generated during the hot forming process are acquired in real time, and the molecular characteristic emission bands of various metal oxides generated by the metallurgical reaction are analyzed from the characteristic spectral signals. Based on the molecular characteristic emission bands of the various metal oxides, the types, mass percentage content, and real-time content change rate of each metal oxide at the current moment are identified and calculated. These are used as input features for predicting the metallurgical reaction state and are respectively input into the XGBoost classification model and the XGBoost regression model. The category of the metallurgical reaction at the current moment is obtained based on the XGBoost classification model, and the reaction degree value and reaction rate value of each category of the metallurgical reaction at the current moment are obtained based on the XGBoost regression model. The XGBoost classification model and XGBoost regression model are based on the XGBoost algorithm. They use the types, mass percentages, and real-time content change rates of metal oxides continuously collected during the hot forming process of several metals to construct a common input feature matrix. The XGBoost classification model is trained with the metallurgical reaction category vectors that are time-aligned with the input feature matrix as the target features. The XGBoost regression model is trained with the reaction degree vectors and reaction rate vectors of each category that are time-aligned with the input feature matrix as the target features.

2. The online monitoring method according to claim 1, characterized in that, The characteristic spectral signals are preprocessed, including abnormal spectrum removal, adaptive baseline correction, multi-scale noise reduction, molecular band feature extraction, and protective gas background subtraction.

3. The online monitoring method according to claim 1, characterized in that, The method for identifying the types of metal oxides at the current time is as follows: the molecular characteristic emission spectrum is compared with the standard characteristic peak position, relative peak intensity ratio, and characteristic wavelength range of each oxide in the standard database of molecular emission spectrum of metal oxides. When the matching degree between a certain molecular characteristic emission spectrum and the standard characteristic peak position is ≥95% and the relative peak intensity ratio deviation is ≤10%, the type of metal oxide corresponding to the molecular characteristic emission spectrum is determined.

4. The online monitoring method according to claim 1, characterized in that, The method for calculating the mass percentage content and real-time content change rate of each metal oxide at the current moment is as follows: For the identified metal oxides, extract the characteristic peak intensity of the molecular characteristic emission spectrum of the metal oxide, substitute it into the pre-built calibration curve, and calculate the mass percentage content of the metal oxide at the current moment; based on the difference between the mass percentage content of the metal oxide at the previous moment and the current moment, divide it by the time step to obtain the content change rate of the metal oxide at the current moment.

5. The online monitoring method according to claim 1, characterized in that, The constructed common input feature is represented as: X = [X1, X2, X3], wherein X1 is a sequence feature matrix composed of the encoding of each metal oxide species at each time point, the matrix element x 1(i,j) represents the encoding value of the jth metal oxide at the ith time point, i = 1 ~ I, j = 1 ~ J, I and J represent the total number of time points and the number of metal oxide species respectively; when the jth metal oxide at the ith time point is identified, x 1(i,j) = 1, and when it is not detected, x 1(i,j) = 0; X2 is a matrix of metal oxide content values ​​at each time point, with matrix element x... 2(i,j) This represents the mass percentage content of the j-th metal oxide at time i. X3 is a matrix representing the rate of change of metal oxide content at each time point, with matrix elements x... 3(i,j) This represents the rate of change of the content of the j-th metal oxide at time i.

6. The online monitoring method according to claim 5, characterized in that, The metallurgical reaction category vector is represented by an I*N column vector Y. type Matrix element Y type(i,n) Y represents the state of the nth metallurgical reaction at time i, where n = 1 to N. type(i,n) =1 indicates that at time i, the metallurgical reaction corresponding to that number is occurring in the molten pool; Y type(i,n) =0 indicates that at time i, the metallurgical reaction corresponding to that number in the molten pool has not occurred; The degree of reaction vector is represented by an I*N column vector Y. degree The degree of reaction of each metallurgical reaction is quantified by the ratio of the total mass of metal elements that have participated in the oxidation reaction in the molten pool to the total mass of the initial metal elements at different times. The reaction rate vector is represented as an I*N column vector Y. rate The rate of change of the content of various metal oxides in the molten pool at different times is quantitatively labeled to indicate the intensity of each metallurgical reaction.

7. An online monitoring system for metallurgical reactions in a molten pool during hot forming of metals, characterized in that, It includes a metal hot forming unit for performing metal hot forming operations, a metal oxide flow detection unit for collecting flue gas containing metal oxide particles generated during metal hot forming operations, a spectral detection unit for acquiring spectral data of the flue gas, and a control and data processing unit for obtaining the type, degree, and rate prediction results of metallurgical reactions based on the acquired spectral data.

8. The online monitoring system according to claim 7, characterized in that, The metal hot forming unit includes a power supply (1-1), a protective gas cylinder (1-2), a nozzle (1-3), a welding torch (1-4), a wire feeding device (1-5), a molten pool (1-6), and a worktable (1-8); the metal oxide flow detection unit is located above the tail of the molten pool (1-6) and includes a fan (2-1) and a guide tube (2-2); the spectral detection unit (3) is located on the side of the guide tube (2-2) and includes a pulsed laser (3-1), a collecting mirror (3-2), a spectrometer (3-3), and a delay generator (3-4).

9. The online monitoring system according to claim 7, characterized in that, The control and data processing unit includes a spectral signal acquisition module, a data preprocessing module, a metallurgical reaction state inference module, and a metallurgical reaction state display module. Spectral signal acquisition module: used to acquire the characteristic spectral signals of metal oxide particles generated during metal hot forming; Data preprocessing module: used to extract the molecular characteristic emission bands of various metal oxides generated by metallurgical reactions from characteristic spectral signals, and based on the molecular characteristic emission bands of each metal oxide, identify the types of metal oxides at the current time, calculate the content and content change rate of each metal oxide at the current time, and further construct the input features for metallurgical reaction state analysis. Metallurgical reaction state inference module: includes a metallurgical reaction state prediction model, which is used to predict the metallurgical reaction state at the current moment based on input features, including metallurgical reaction type, degree of metallurgical reaction, and metallurgical reaction rate. Metallurgical reaction status display module: used to display the predicted metallurgical reaction status at the current moment in real time on the human-computer interaction interface.

10. The online monitoring system according to claim 9, characterized in that, The metallurgical reaction state prediction model includes an XGBoost classification model and an XGBoost regression model. The XGBoost classification model and XGBoost regression model are based on the XGBoost algorithm and use the types, mass percentages, and real-time content change rates of metal oxides continuously collected during the hot forming process of several metals to construct a common input feature matrix. The XGBoost classification model is trained with the metallurgical reaction category vectors that are time-aligned with the input feature matrix as the target features. The XGBoost regression model is trained with the reaction degree vectors and reaction rate vectors of each category that are time-aligned with the input feature matrix as the target features.

Images