Laser welding element burning loss multiple regression monitoring method based on spectral information

Abstract

The invention discloses a laser welding element burning loss multiple regression monitoring method based on spectral information, and belongs to the field of material processing engineering.The method comprises the steps that a spectrograph is adopted for collecting spectral signals in the welding process, noise components in the spectral signals are automatically reduced through processing methods such as background removal, high-frequency noise filtering and normalization, and the spectral signals are obtained; and constructing an element burning loss multiple regression prediction model database. And then a laser welding element burning loss multiple regression prediction model based on machine learning is constructed. Dimensionality reduction of spectral signals is realized by adopting a feature screening algorithm, and the calculation efficiency and prediction precision of the model are improved. The element burning loss multiple regression prediction model is automatically optimized through algorithms such as grid search and cross validation, and element burning loss prediction in the laser welding process is accurately achieved. The laser welding quality can be improved.

Description

Technical Field

[0001] The invention belongs to the field of material processing engineering, and in particular relates to a laser welding element burnout multivariate regression monitoring method based on spectral information. Background Art

[0002] With the rapid development of industrial intelligent manufacturing, intelligent welding systems have become an important part of modern industry. Intelligent welding not only improves production efficiency, but also significantly improves welding quality, which is of great significance to social and economic development. Especially in the manufacturing fields of automobiles, aerospace, etc., welding quality is directly related to the safety and reliability of products. In the process of welding automation, the traditional post-weld inspection method can no longer meet the requirements of modern manufacturing for real-time and accuracy. Therefore, online welding quality monitoring technology came into being. This technology provides a reliable basis for the prediction and judgment of welding quality by real-time monitoring of physical and chemical changes in the welding process, such as arc spectrum, acoustic spectrum, welding current, voltage and other parameters. The application of online monitoring technology not only ensures the high quality of welds, but also improves the competitiveness and production efficiency of products. With the continuous advancement of technology, online monitoring of welding quality will become an important development direction in the field of intelligent manufacturing in the future, injecting new vitality into industrial development. However, during the laser welding process, the equilibrium vapor pressure of some reinforcing elements in the material to be welded is much higher than that of the matrix alloy elements, resulting in lower melting and boiling points. During the laser welding process, they will evaporate before the matrix alloy elements, resulting in the reinforcing element content in the weld after welding being lower than the alloy element content in the parent material matrix. The loss of reinforcing elements will cause the performance of the weld joint to be lower than that of the parent material, resulting in a decrease in the quality of the parts after welding.

[0003] Therefore, it is necessary to develop an online monitoring system for the quality of the laser welding process, especially a quantitative detection system for element burnout, which will provide a scientific basis for the quantitative evaluation of weld performance. At present, some studies at home and abroad have shown that the loss of elements in the welding process can be monitored. Chinese patent application CN115615939A discloses a monitoring device and method for manganese evaporation in the arc additive process of a high nitrogen steel melting electrode, which collects spectral signals in a stable arc welding process, and manually screens and obtains manganese spectral lines that are not interfered by other characteristic spectra, and then qualitatively judges whether there is manganese element burnout in the arc welding process based on the obtained manganese spectral lines. However, this invention only qualitatively analyzes manganese element burnout by manual means. Chinese patent application CN115060708A discloses a method for detecting trace elements in laser plasma. This method provides a method of using arc to excite laser plasma to ensure that a stable spectral signal can be obtained, but does not provide a semi-quantitative or even quantitative method for element content. Chinese patent application CN107764798A discloses an online quality detection system for metal additive manufacturing, provides a method for collecting spectral signals during the additive process, and proposes a free calibration method to calculate the amount of element burnout, but does not provide a specific method for quantitative calculation of element burnout. In addition, unlike the strict quantitative analysis requirements of laser induced breakdown spectroscopy, in the laser welding process, due to the influence of factors such as the energy density of the equipment laser, the method of spectrum collection, and the influence of environmental noise, the collected spectral signal and element burnout are completely a nonlinear relationship, and the free calibration method for calculating element burnout will be greatly disturbed. Chinese patent CN112017183B discloses an online detection method for welding wire composition that coordinates arc image and arc spectral information. The prediction of welding wire composition is achieved by fusing image and spectral information using a deep learning method, but this method is not proposed for the composition content of the weld after welding.

[0004] Current research has shown that the use of spectral signals can be used to monitor the burning of elements during the welding process, but related invention methods all focus on the arc or the qualitative analysis of the burning of elements. Summary of the invention

[0005] In order to overcome the shortcomings of the prior art, the present invention proposes a laser welding element burnout multivariate regression monitoring method based on spectral information. During the arc welding process, the elements evaporated from the welding material will be confined inside the arc, and the acquired spectral signal will be relatively more stable. The present invention proposes a method for the plasma that generates spectral signals during laser welding, which will have a process of eruption, expansion, and dissipation. This process is extremely unstable and the acquired signal contains noise components caused by factors such as equipment, human, environment, and materials. In order to solve more nonlinear factors between the quantitative analysis of element burnout and the spectral signal caused by these factors as much as possible, the present invention proposes a burnout element multivariate regression monitoring method based on machine learning. Compared with calibration methods, free calibration methods, etc., the multivariate regression monitoring method based on multivariate can fully extract and learn the spectral signal characteristics during the welding process, and accurately and automatically select the characteristic spectral lines related to the burnout elements according to the importance of the characteristics, reduce monitoring errors, and realize accurate monitoring of the burnout elements of laser welding joints.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme: A laser welding element burnout multivariate regression monitoring method based on spectral information comprises the following steps: S1: In order to obtain spectral signals of different intensities and corresponding element burnout laser welding tests of different degrees, laser welding experiments with different welding process parameters are designed. The process parameters mainly include: laser power and welding speed. Then, a laser welding test is carried out, and spectral signal acquisition parameters are set to collect spectral signals under different welding tests; at the same time, welds under different element burnout states are obtained, and the element content detection method is used to detect the element content of weld burnout after welding to obtain the element content of weld burnout after welding; S2: In order to minimize the noise impact caused by equipment, environment, human and base value effects, the spectral signals obtained by S1 are subjected to denoising preprocessing and corresponding normalization processing in turn; and the spectral lines of the burned elements in the spectral signals after normalization are selected to perform linear fitting with the burned element content in the weld after welding, and a univariate fitting model between the spectral lines of the burned elements and the burned elements in the weld after welding under different normalization methods is established. Then, according to the determination coefficient R of the univariate fitting model, the linear fitting model is constructed. 2 , select the optimal normalization method; S3: Use feature screening algorithm to screen the normalized spectral signal: first calculate the correlation coefficient or importance score between the normalized spectral signals of different wavelengths and the post-weld weld burnout element content; then set different correlation coefficients or importance scores as feature screening thresholds, discard the normalized spectral signals of the corresponding wavelengths less than the feature screening threshold, retain the normalized spectral signals of the remaining wavelengths, and construct a multivariate regression monitoring model data set, and divide the data set into a training set, a test set, and a validation set; then construct a multivariate regression prediction model of the screened spectral signals and the post-weld weld burnout elements obtained by measurement, and then automatically optimize the multivariate regression prediction model with different feature screening thresholds, and calculate the accuracy of the optimized multivariate regression prediction model with different feature screening thresholds; select the feature screening threshold corresponding to the multivariate regression prediction model with the maximum accuracy as the optimal feature screening threshold, and use the multivariate regression prediction model with the maximum accuracy as the multivariate regression monitoring model for element burnout of laser welding joints.

[0007] Preferably, in S1, a collimator is used to focus the spectral signal of the laser plasma at a distance greater than about 1 mm above the welding area, because the plasma continuous spectrum decays faster than the characteristic spectrum, thus reducing the interference of the continuous spectrum on the characteristic spectrum at a distance greater than 1 mm.

[0008] Preferably, in S1, a collimator is used to convert the divergent spectral signal emitted by the welding plasma into a parallel light beam, and then the spectral signal is transmitted to a spectrometer via an optical fiber coupled to the collimator, and then the acquisition parameters of the spectrometer are set to record the welding spectral signals at different welding times online.

[0009] Furthermore, in the actual spectral signal acquisition process, in order to prevent the spectral signal intensity from being oversaturated, it is necessary to use an attenuation plate with a corresponding transmittance, set the integration time and number of integration times of the signal acquisition, etc.

[0010] Preferably, in S1, the post-weld element acquisition first uses wire cutting or CNC processing equipment to make a weld metallographic part perpendicular to the weld, and then uses sandpaper to perform grinding and other operations successively, and then uses component detection technology to measure the content of elements burned in the weld after welding, but is not limited to: XRF, EDS, LIBS and other component detection technologies.

[0011] Preferably, in S2, during the welding process, due to the presence of composite radiation and inverse bremsstrahlung, the collected spectral signal has continuous spectral lines with a certain intensity. These signals are more of background noise, which will seriously affect the spectral signal intensity of the wavelength related to the element to be predicted, and it is necessary to use background correction to remove them, wherein the background correction includes but is not limited to: least squares method, polynomial fitting and other algorithms; at the same time, due to factors such as environment, equipment, signal acquisition method, etc., the acquired signal has a large amount of high-frequency noise, and it is also necessary to use a corresponding high-frequency filtering and noise reduction algorithm to remove it, wherein the high-frequency filtering and noise reduction algorithm is not limited to: Fourier transform, wavelet transform, wavelet packet transform and other methods; then the normalization method can be used to further remove the interference caused by equipment, human factors and base value effect, including but not limited to maximum and minimum value normalization, background normalization, internal standard normalization, area normalization, etc., and then the spectral line of the burned element in the normalized spectral signal is selected to perform single variable linear fitting with the measured content of the burned element in the weld after welding, and the line with the largest correlation determination coefficient R is selected. 2 The normalization method is taken as the optimal normalization method for spectral signals.

[0012] Furthermore, there are a large number of signals with low correlation with the burnt elements in the spectral signals actually obtained. The signals in these bands will seriously interfere with the calculation speed and accuracy of the multivariate regression monitoring model, which not only affects the monitoring accuracy of element burnt in the actual welding process, but also causes a certain delay in monitoring. Therefore, it is necessary to perform feature screening to remove spectral signals that are not related to the burnt elements. By calculating the characteristic correlation coefficient or importance score between the spectral signal and the weld burnt elements after welding in the training set, validation set, and test set, the closeness between the burnt elements and the spectral signal is characterized. The method for calculating the characteristic correlation coefficient or importance score is not limited to: variance threshold, correlation coefficient method, principal component analysis, model-based feature selection, etc., and then by setting the characteristic correlation coefficient or importance score as the feature screening threshold, the wavelength spectral signal less than the threshold is discarded, and the wavelength spectral signal greater than the threshold is retained; by setting different feature screening thresholds, the spectral signal is screened to construct a data set with different feature screening thresholds; then each spectral signal data set is divided into a training set, a validation set, and a test set.

[0013] Furthermore, a multivariate regression prediction model for elemental burnout based on machine learning is constructed, and the spectral signals of the training set and the validation set after screening by different thresholds are used as input, respectively, and the measured weld burnout element content after welding is used as the output of the model. Subsequently, the model hyperparameters are automatically optimized under different feature screening thresholds through the multivariate regression prediction model optimization algorithm, wherein the multivariate regression prediction model optimization algorithm is not limited to: Bayesian optimization, grid search method; the accuracy of the optimized multivariate regression monitoring model with different feature screening thresholds is calculated, and the threshold corresponding to the maximum accuracy is selected as the optimal feature screening threshold for spectral signal screening, and the model with the maximum prediction accuracy is used as the optimal laser welding joint elemental burnout multivariate regression monitoring model.

[0014] Furthermore, the model is trained through cross-validation, and then the performance of the model is verified using a test set. The prediction accuracy of the constructed multivariate regression monitoring model is calculated to verify the prediction accuracy and generalization ability of the model.

[0015] The present invention can realize online real-time monitoring of weld burnout elements after laser welding, and has the following beneficial effects compared with the traditional element measurement method: 1. The present invention applies LIBS-related technical theories to the welding process, avoiding the reuse of a complete set of LIBS non-destructive testing equipment during the welding process. Therefore, the method and device are a simpler, more integrated device and non-destructive testing method.

[0016] 2. The present invention is a real-time, online, non-destructive detection method that can realize in-welding monitoring and avoid excessive post-welding process detection methods, which can significantly reduce production costs and related detection processes.

[0017] 3. The present invention is a multivariate regression monitoring method for element burnout based on machine learning. Since machine learning has strong nonlinear fitting capabilities, interference caused by factors such as equipment, human factors, and matrix effects can be largely ignored. In addition, there is no need for too much professional knowledge and experience in the field of nondestructive testing and spectrum-related. Optimization algorithms such as denoising and grid search are used to autonomously and fully capture spectral signals of bands strongly correlated with the predicted burnout elements, without the need to manually query and search for characteristic spectral lines of the burnout elements, and automatically optimize the multivariate regression prediction model for element burnout, thereby realizing quantitative monitoring of element burnout during welding. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a multivariate regression monitoring method for laser welding element burnout based on spectral information of the present invention; Figure 2 It is a schematic diagram of a welding plasma spectrum information acquisition device of the present invention; Figure 3Schematic diagram of the spectral signal obtained by the present invention and the measurement of magnesium element in the weld after welding; wherein a is the spectral signal collected after welding, b is Figure 3 a is the average value of the spectral signal in a window with a width of 0.5 mm, c is a schematic diagram of magnesium element collection in the weld after welding, and d is a schematic diagram of EDS point scanning of magnesium element content in the weld after welding; Figure 4 It is a schematic diagram of the spectral signal denoising process and results of the present invention; Figure 5 This is a graph showing the fitting results of the single variable loss monitoring of magnesium element using different normalization methods in the present invention; Figure 6 It is a vertical line diagram of the importance scores of spectral signals of different wavelengths of the present invention; Figure 7 This is the model fitting result corresponding to the feature screening using different thresholds in the present invention.

[0019] In the figure, the reference numerals are: 1, industrial computer; 2, spectrometer; 3, display; 4, optical fiber; 5, collimator; 6, neutral attenuation plate; 7, protective lens; 8, plasma; 9, welding workbench; 10, laser head; 11, laser. DETAILED DESCRIPTION

[0020] In order to make the features, technical solutions and advantages of the present invention more understandable, the present invention is further described in detail in conjunction with the accompanying drawings and specific implementation schemes. In addition, the specific implementation schemes described herein are only used to explain the main points of the present invention and are not used to limit the present invention. The present invention can be appropriately improved according to specific actual use conditions.

[0021] like Figure 1 As shown, the laser welding element burnout multivariate regression monitoring method based on spectral information of the present invention comprises the following steps: S1: In order to obtain spectral signals of different intensities and different degrees of elemental burnout laser welding tests, laser welding experiments with different welding process parameters are designed. The different welding process parameters include: welding power and welding speed. Then, laser welding is performed, and spectral signal acquisition parameters are set to collect spectral signals under different laser welding tests; at the same time, the corresponding welds under different elemental burnout states are obtained, and the elemental content detection method is used to detect the elemental content of the weld after welding to obtain the elemental content of the weld after welding.

[0022] S2: In order to minimize the noise impact caused by equipment, environment, human and base value effects, the spectral signals obtained by S1 were de-noised and normalized in turn, and the burnt element spectrum lines in the spectral signals after normalization were selected to perform linear fitting with the burnt element content in the weld after welding, and a univariate fitting model between the burnt element spectrum lines under different normalization methods and the burnt element content in the weld after welding was established. Then, according to the determination coefficient R of the univariate fitting model, the linear fitting model was constructed. 2 , select the optimal normalization method.

[0023] S3: Use feature screening algorithm to screen the normalized spectral signal: first calculate the correlation coefficient between the normalized spectral signals of different wavelengths and the content of the weld burnt elements after welding, or calculate the importance scores of different wavelengths; then set different correlation coefficients or importance scores as feature screening thresholds, discard the spectral signals with wavelengths smaller than the feature screening threshold, retain the normalized spectral signals of the remaining wavelengths, and construct a multivariate regression monitoring model data set, and divide the data set into a training set, a test set, and a validation set; then construct a multivariate regression prediction model of the screened spectral signals and the weld burnt elements obtained by measurement, and then automatically optimize the multivariate regression prediction model with different feature screening thresholds, and calculate the accuracy of the optimized multivariate regression prediction model with different feature screening thresholds; select the feature screening threshold corresponding to the multivariate regression prediction model with the maximum accuracy as the optimal feature screening threshold, and use the multivariate regression prediction model with the maximum accuracy as the multivariate regression monitoring model for the burnt elements of laser welding joints.

[0024] Preferably, in S1, a collimator 5 is used to focus the spectral signal of the laser plasma about 1 mm above the welding area. Because the continuous spectrum of the plasma 8 decays faster than the characteristic spectrum, collecting the spectral signal at a distance greater than 1 mm can reduce the interference of the continuous spectrum on the characteristic spectrum.

[0025] Preferably, in S2, a collimator 5 is used to convert the divergent spectral signal emitted by the welding plasma into a parallel light beam, and then the spectral signal is transmitted to the spectrometer 2 via the optical fiber 4 coupled to the collimator, and then the acquisition parameters of the spectrometer 2 are set to record the welding spectral signals at different welding times online.

[0026] Furthermore, in the actual spectral signal acquisition process, in order to prevent the spectral signal intensity from being oversaturated, it is necessary to use a neutral attenuation plate 6 with a corresponding transmittance, and set the integration time and number of integrations of the spectrometer 2 during the signal acquisition process.

[0027] Preferably, in S2, the post-weld element acquisition first adopts wire cutting or CNC processing equipment to make a weld metallographic part perpendicular to the weld, and then uses sandpaper to grind it successively, and then measures the post-weld weld burnt element content but is not limited to: X-ray fluorescence spectrometry analysis (XRF), energy dispersive X-ray spectrometry analysis (EDS), laser induced breakdown spectroscopy (LIBS) and other methods.

[0028] Preferably, in S3, during the welding process, due to the presence of composite radiation and inverse bremsstrahlung, the collected spectral signal has continuous spectral lines with a certain intensity. These signals are more of background noise, which will seriously affect the spectral signal intensity of the wavelength related to the element to be predicted, and it is necessary to use background correction to remove them, wherein the background correction includes but is not limited to: least squares method, polynomial fitting and other algorithms; at the same time, due to factors such as environment, equipment, signal acquisition method, etc., the acquired signal has a large amount of high-frequency noise, and it is also necessary to use a corresponding high-frequency filtering noise reduction algorithm to remove it, wherein the high-frequency filtering noise reduction algorithm is not limited to: Fourier transform, wavelet transform, wavelet packet transform and other methods; then the normalization method can be used to further remove the interference caused by equipment, human factors and base value effect, including but not limited to maximum and minimum value normalization, background normalization, internal standard normalization, area normalization, etc., and then the burnt element spectral line in the normalized spectral signal is selected to perform single variable linear fitting with the measured post-weld weld burnt element content, and the maximum correlation determination coefficient R is selected. 2 The normalization method is taken as the optimal normalization method for spectral signals.

[0029] Furthermore, after normalization, there are a large number of signals with low correlation with element burnout in the spectral signal. The signals in these bands will seriously interfere with the calculation speed and accuracy of the multivariate regression monitoring model, which not only affects the monitoring accuracy of element burnout in the actual welding process, but also causes a certain delay in monitoring. Therefore, it is necessary to perform feature screening to remove spectral signals that are not related to the burnout elements. By calculating the characteristic correlation coefficient or importance score between the spectral signal and the weld burnout element after welding, the closeness between the burnout element and the spectral signal is characterized. The method for calculating the characteristic correlation coefficient or importance score is not limited to: variance threshold, correlation coefficient method, principal component analysis, model-based feature selection, etc., and then by setting the characteristic correlation coefficient or importance score as the feature screening threshold, the wavelength spectral signal less than the threshold is discarded, and the wavelength spectral signal greater than the threshold is retained; by setting different feature screening thresholds, the spectral signal is screened to construct a data set with different feature screening thresholds; then each spectral signal data set is divided into a training set, a validation set, and a test set at a ratio of 7:2:1.

[0030] Furthermore, in S3, a multivariate regression prediction model of burnout elements based on machine learning is constructed, the spectral signals of the training set and the validation set are used as inputs of the multivariate regression prediction model, and the measured post-weld weld burnout element content is used as output of the multivariate regression prediction model.

[0031] Subsequently, the model hyperparameters are automatically optimized under different feature screening thresholds through the model optimization algorithm, wherein the multivariate regression prediction model optimization algorithm is not limited to: Bayesian optimization, grid search method; at the same time, the accuracy of the optimized multivariate regression monitoring model with different feature screening thresholds is calculated, and the threshold corresponding to the maximum accuracy is selected as the optimal feature screening threshold for spectral signal screening, and the model with the maximum prediction accuracy is used as the optimal multivariate regression monitoring model for burnt element burnt of laser welding joints.

[0032] Furthermore, the model is trained through cross-validation, and then the performance of the model is verified using a test set. The prediction accuracy of the constructed multivariate regression monitoring model is calculated to verify the prediction accuracy and generalization ability of the model.

[0033] Example: A laser welding element burnout multivariate regression monitoring method based on spectral information in this embodiment includes: First, Al-Mg alloy with the grade of 5251 was selected for laser welding experiment. During the welding process, a collimating lens was used to convert the divergent beam into a parallel beam, and then it was transmitted to the 314nm-418nm channel of the 8-channel spectrometer through an optical fiber to collect and obtain the spectral signal; a protective lens 7 was installed in turn in front of the collimating lens to prevent splash damage to the lens, and a 20% neutral attenuation plate 6 of the entire band was used to prevent the spectral signal from being oversaturated. The integration time was set to 5ms, and the number of integrations was set to 2, such as Figure 2 As shown. The laser welding work is completed by the industrial computer 1, the display 3, the welding workbench 9, the laser head 10, and the laser 11. The industrial computer 1 and the laser 11 are connected by a signal line to control the laser process parameters, etc. The laser 11 and the laser head 10 are connected by optical fiber to realize the transmission of the laser beam from the laser 11 to the laser head 10. At the same time, the industrial computer 1 is connected to the welding workbench 9 by a signal line to control the welding workbench; the collection and analysis of the spectral signal during the welding process is completed by the industrial computer 1, the spectrometer 2, the display 3, the optical fiber 4, the collimator 5, the neutral attenuation sheet 6, the protective lens 7, and the welding plasma 8. The protective lens 7 and the neutral attenuation sheet 6 can be fastened to the collimator 5 in sequence by threads, and then the collimator and the spectrometer 2 are coupled together by the optical fiber 4, and the spectral signal emitted by the plasma 8 is transmitted to the spectrometer 2, and recorded in the industrial computer 1, and displayed on the display 3 at the same time. Then, the analysis of the laser welding element burnout monitoring task is carried out on the industrial computer 1.

[0034] Furthermore, the collected spectral signals are Figure 3 As shown in a. Then, a wire cutting machine is used to obtain the longitudinal section metallographic part of the weld along the center of the weld, and then the metallographic part is ground to obtain the metallographic part as shown in a. Figure 3 The longitudinal section metallographic part of the weld shown in c; then the energy dispersive X-ray spectroscopy (EDS) point scanning method was used, as shown in Figure 3 As shown in d; then, the magnesium content of multiple points in a window with a width of 0.5 mm is measured at intervals of 5 mm along the weld and the average is taken. From the initial time T0 to the final time T1, the magnesium content of multiple window areas is measured and obtained, as shown in Figure 3 As shown in c, it is used as the output of the subsequent multivariate regression prediction model; accordingly, the average spectral signal in the same interval and width window is obtained, as shown in Figure 3 As shown in (b), it serves as the input of the subsequent multivariate regression prediction model.

[0035] Furthermore, the acquired spectral signal is subjected to denoising, and the process and results are as follows: Figure 4 The polynomial fitting algorithm and the wavelet packet algorithm are used in turn to obtain the background and high-frequency noise components, and then the noise components are subtracted from the original signal to obtain the denoised spectral signal.

[0036] Furthermore, the spectral signal obtained after denoising is subjected to different normalization processing. In this embodiment, no normalization processing, area normalization processing, and background normalization processing are compared; area normalization processing: the integral definition algorithm is used to obtain the total area of ​​the spectral signal within the wavelength range of 314nm-418nm, and then the spectral intensity of different wavelengths is compared with the total integrated area; background normalization processing: the intensity of the spectral signal at each wavelength after denoising is compared with the intensity of the background noise at the corresponding wavelength obtained during denoising; then the magnesium spectrum line at 383.829nm after normalization is linearly fitted with the element content obtained by EDS measurement to construct a univariate fitting model, and the results are shown as follows. Figure 5 As shown. Figure 5 It can be seen that the area normalization method can effectively improve the determination coefficient R of the univariate model. 2 Therefore, the acquired spectral signals were area normalized to construct a data set for the multivariate regression prediction model of weld magnesium burnout.

[0037] Furthermore, a magnesium burnout monitoring model based on LightGBM (lightweight gradient boosting machine learning) machine learning was constructed. Then, based on the principle of feature screening using the embedding method, the LightGBM model was used to automatically count the relationship between spectral signals of different wavelengths and element content, and calculate the importance of different wavelengths. The results are as follows: Figure 6 shown.

[0038] Furthermore, by setting different importance thresholds, feature screening of different wavelengths is performed, wavelengths below the set threshold are discarded, and wavelengths greater than or equal to the set threshold are retained. The spectral signal obtained by screening is used as the input of the LightGBM model, and the magnesium content in the weld after welding is measured as the output of the model. Then, the grid search algorithm is used to search for the optimal model parameters, including: boosting_type (model boosting algorithm), colsample_bytree (sampling rate), learning_rate (learning rate), max_depth (maximum tree depth), n_estimators (number of evaluators), num_leaves (number of leaf nodes), reg_alpha (L1 regularization coefficient), reg_lambda (L2 regularization coefficient), and finally the optimal model under different importance thresholds is obtained, and the accuracy of the optimal model under different thresholds is calculated. The results are as follows: Figure 7 As shown. Figure 7 It can be seen that when the model importance threshold is set to 0.05 in this embodiment, the constructed weld element burnout multivariate regression prediction model is optimal.

[0039] The present invention does not elaborate on some of the common technologies of those skilled in the art. The above-described embodiments are only descriptions of preferred implementations of the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementations described. Without departing from the design spirit of the present invention, various modifications and improvements made by ordinary technicians in this field to the technical solution of the present invention should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A multivariate regression monitoring method for laser welding element burnout based on spectral information, characterized in that: The following steps are involved: S1: Design laser welding experiments with different welding process parameters, including laser power and welding speed; then perform laser welding, set spectrum signal acquisition parameters, and collect spectrum signals under different laser welding tests; at the same time, obtain welds with different degrees of element burnout, and use element content detection methods to detect the element content of weld burnout after welding, and obtain the content of weld burnout elements after welding; S2: The spectral signal obtained by S1 is subjected to denoising preprocessing and corresponding normalization processing, and the spectral line of the burned element in the normalized spectral signal is selected to perform linear fitting with the content of the burned element in the weld after welding, and a univariate fitting model between the spectral line of the burned element under different normalization methods and the burned element in the weld after welding is established. Then, according to the determination coefficient R of the univariate fitting model, 2 , select the optimal normalization method; S3: Use feature screening algorithm to screen the normalized spectral signal: first calculate the correlation coefficient or importance score between the normalized spectral signals of different wavelengths and the content of the weld burnt elements after welding; then set different correlation coefficients or importance scores as feature screening thresholds, discard the spectral signals with wavelengths smaller than the feature screening threshold, and retain the normalized spectral signals of the remaining wavelengths to construct a multivariate regression prediction model data set, and divide the data set into a training set, a test set, and a validation set; then construct a multivariate regression prediction model of the screened spectral signals and the weld burnt elements obtained by measurement, and then automatically optimize the multivariate regression prediction model with different feature screening thresholds, and calculate the accuracy of the optimized models with different feature screening thresholds; select the feature screening threshold corresponding to the multivariate regression prediction model with the maximum accuracy as the optimal feature screening threshold, and use the multivariate regression prediction model with the maximum accuracy as the multivariate regression prediction model for laser welding element burnt elements.

2. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 1 is characterized in that: In S1, during the welding process, the spectral signal of the laser plasma at a position greater than 1 mm above the welding area is collected to reduce the interference of the continuous spectrum on the characteristic spectrum; a collimator is used to convert the divergent spectral signal emitted by the welding plasma into a parallel beam, and then the spectral signal is transmitted to the spectrometer via an optical fiber coupled to the collimator, and then the acquisition parameters of the spectrometer are set to record the welding spectral signals at different welding times online.

3. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 2 is characterized in that: In the process of spectral signal acquisition, in order to prevent the spectral signal intensity from being oversaturated, a neutral attenuation plate with corresponding transmittance is used, and the integration time and number of integration times collected by the spectrometer are set.

4. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 1 is characterized in that: As described in S2, the spectral signal denoising preprocessing includes the removal of background and high-frequency noise, and then the determination coefficients of the established single variable linear fitting models of the burnt elements are compared, and the normalization method with the maximum determination coefficient is selected as the optimal normalization method.

5. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 1 is characterized in that: In the S3, a multivariate regression prediction model for element burnout of welds after welding based on machine learning is constructed; the spectral signals filtered by different feature screening thresholds are respectively used as the input of the multivariate regression prediction model, and the element content of weld burnout after welding is used as the output of the multivariate regression prediction model; then, the model hyperparameters under different feature screening thresholds are automatically optimized through the model optimization algorithm, and then the accuracy of the optimized model under different feature screening thresholds is calculated, and the threshold corresponding to the maximum accuracy is selected as the optimal threshold for spectral signal screening, and the model with the maximum prediction accuracy is used as the optimal multivariate regression prediction model for element burnout.

6. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 1 is characterized in that: In S3, the feature screening algorithm includes a variance threshold method, a correlation coefficient method, a principal component analysis method, a recursive feature elimination method, and a model-based feature selection method.

7. The laser welding element burnout multivariate regression monitoring method based on spectral information according to claim 5 is characterized in that: Model optimization algorithms include Bayesian optimization and grid search methods.

Images