Self-adaptive laser processing control method and system based on real-time molten pool monitoring

Abstract

The invention discloses a self-adaptive laser processing control method and system based on real-time molten pool monitoring, and relates to the field of data processing.The method comprises the steps that firstly, processing conditions are initialized based on workpiece parameters, a normal molten pool feature library is constructed, and molten pool images, temperature and spectral data are synchronously collected through multiple sensors during processing; and comprehensive features are obtained through preprocessing and weighted fusion. The real-time features are compared with the feature library, if the real-time features are abnormal, the type is recognized, the adjustment direction is determined, a particle swarm optimization algorithm is adopted to collaboratively optimize laser power and speed parameters, and closed-loop verification is conducted till the molten pool returns to normal. And updating the feature library through incremental learning after the whole-process data is recorded, and realizing self-adaptive adjustment of processing parameters and continuous optimization of the knowledge base. The method has the advantages that the high-precision intelligent control of laser processing is realized by collecting the molten pool data, accurately detecting the abnormity through fusion analysis, adjusting the processing parameters by combining the multi-parameter collaborative optimization model with the particle swarm optimization algorithm, and updating the feature library through incremental learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to an adaptive laser processing control method and system based on real-time molten pool monitoring. BACKGROUND

[0002] Laser processing technology is widely used in high-precision manufacturing field, but the traditional processing process relies on preset parameters, which is difficult to cope with the dynamic changes of molten pool caused by material property fluctuations, heat accumulation effects or environmental disturbances, etc., and is easy to cause uneven weld formation, porosity or spatter and other defects. With the increasing demand of industry on processing quality and efficiency, real-time monitoring and adaptive control become the key to break through the bottleneck. The adaptive laser processing method based on real-time molten pool monitoring emerges as the times require.

[0003] The current laser processing method and system on the market rely on a single sensor to monitor the molten pool, such as only using a high-speed camera, which cannot comprehensively obtain the temperature, spectrum and other information of the molten pool, and it is difficult to accurately judge the state of the molten pool. The data processing capability is relatively weak, and there is a lack of effective preprocessing and fusion of the collected data, resulting in low information utilization rate, inability to form accurate molten pool comprehensive feature data, and large errors in abnormal detection, which easily misses key problems. In response to abnormalities, the traditional method is difficult to quickly determine the adjustment direction of the laser processing parameters once the molten pool appears abnormal, lacks the ability of multi-parameter collaborative optimization, and can only rely on the experience of the operator to adjust manually, which is low in efficiency and poor in adjustment effect, and is difficult to guarantee the processing quality. Moreover, the traditional method usually does not have the function of continuous learning, and cannot record data and update molten pool sample data in real time during processing, which is difficult to adapt to changes of different workpieces and complex working conditions, and has poor adaptability when facing new materials and new processes. SUMMARY

[0004] In order to improve the existing method and system, an adaptive laser processing control method and system based on real-time molten pool monitoring is provided, which synchronously collects molten pool data through multiple source sensors, accurately detects abnormalities through fusion analysis, dynamically adjusts processing parameters through a multi-parameter collaborative optimization model combined with a particle swarm optimization algorithm, and can also update the feature library through incremental learning, realizing high precision, high stability and intelligent control of laser processing.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is: The adaptive laser processing control method based on real-time molten pool monitoring comprises: Obtaining workpiece parameter data and normal molten pool sample data under different processing parameter combinations, and initializing laser processing parameters to construct a normal molten pool feature library; Processing the workpiece based on the initialized laser processing parameters, and collecting molten pool images, temperature data and spectrum data through high-speed cameras, infrared sensors and spectrum sensors; Based on the acquisition data preprocessing, the preprocessed data is fused by a weighted fusion algorithm, and the fused comprehensive feature data of the molten pool is obtained; The obtained comprehensive feature data of the molten pool is compared with the normal molten pool feature library to determine whether the current molten pool feature is within the normal threshold range of the corresponding processing stage. If at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal, and the abnormal type is determined through feature matching; Based on the matching determined abnormal type, the adjustment direction of the laser processing parameters is determined, and a multi-parameter collaborative optimization model is constructed. The processing parameters are iteratively optimized in the adjustment direction by a particle swarm optimization algorithm to obtain the optimal adjustment parameter group. Based on the adjusted molten pool feature data, the molten pool state is compared with the normal molten pool feature library again to determine whether the molten pool state has returned to normal. If the molten pool state is still abnormal, the parameter optimization adjustment process is performed until the molten pool feature returns to the normal threshold range. During laser processing, processing data is recorded in real time to form a processing database. After processing is completed, the normal molten pool sample data in the database is updated and supplemented through an incremental learning algorithm.

[0006] Preferably, the normal molten pool feature library is constructed by obtaining the workpiece parameter data and normal molten pool sample data under different processing parameter combinations, and initializing the laser processing parameters, which specifically includes: The workpiece parameter data includes material properties, thickness size and processing target; The normal molten pool sample data includes geometric features, temperature features and spectral features; The laser processing parameters include laser initial power, initial scanning speed, initial defocusing amount and protective gas flow; The normal molten pool feature library is constructed by training the normal molten pool sample data through a machine learning algorithm. The feature library contains molten pool feature threshold ranges corresponding to different processing stages.

[0007] Preferably, the workpiece is processed based on the initialized laser processing parameters, and the molten pool images, temperature data and spectral data collected by the high-speed camera, infrared sensor and spectral sensor specifically include: The high-speed camera collects real-time images of the molten pool at a frame rate of not less than 1000 fps; The infrared sensor collects temperature distribution data of the molten pool area at a sampling frequency of not less than 500 Hz; The spectral sensor collects plasma spectral data of the molten pool in the wavelength range of 200-1100 nm; The three collected data are aligned in time by a synchronous trigger signal to obtain the molten pool image, temperature data and spectral data corresponding to the same processing time.

[0008] Preferably, the collected data is preprocessed, and the preprocessed data is fused by a weighted fusion algorithm to obtain the fused molten pool comprehensive feature data, which specifically includes: The collected molten pool image data is processed by a Gaussian filter algorithm to remove high-frequency noise, histogram equalization to improve the contrast between the molten pool and the background, and edge detection to obtain a clear molten pool contour image and calculate geometric feature parameters. The infrared temperature data is processed by a 3σ criterion to remove data points outside the reasonable temperature range and smoothed to obtain the molten pool center temperature and temperature gradient data. The spectral data is baseline corrected and feature spectral line extracted to obtain feature spectral line intensity and spectral line half-width data. The weight values of each collected data are calculated based on the analytic hierarchy process, and the geometric feature data, temperature feature data and spectral feature data are fused by a weighted fusion algorithm to obtain the molten pool comprehensive feature data.

[0009] Preferably, the obtained molten pool comprehensive feature data is compared with the normal molten pool feature library to determine whether the current molten pool feature is within the normal threshold range of the corresponding processing stage. If at least one molten pool feature data is outside the normal threshold range, it is determined that the current molten pool state is abnormal, and the abnormal type is determined by feature matching, which specifically includes: From the normal molten pool feature library, the reference data matching the current processing stage is selected, and each dimension feature of the current molten pool is compared with the normal feature range of the corresponding processing stage. If all the molten pool comprehensive feature data is within the normal threshold range, it is determined that the current molten pool state is normal, and the processing continues. If at least one molten pool feature data is outside the normal threshold range, it is determined that the current molten pool state is abnormal. Based on the feature dimension and deviation direction outside the normal range, the abnormal type is identified, and the abnormal type includes molten pool overheating, molten pool overcooling, molten pool size being too large, molten pool size being too small and plasma interference.

[0010] Preferably, the adjustment direction of the laser processing parameters is determined based on the matched abnormal type, and a multi-parameter collaborative optimization model is constructed. The processing parameters are iteratively optimized in the adjustment direction by a particle swarm optimization algorithm to obtain the optimal adjustment parameter group, which specifically includes: The adjustment direction of the laser processing parameters is obtained based on the matched abnormal type. The objective function is to return the molten pool characteristics to the normal threshold range, and a multi-parameter collaborative optimization model is constructed by combining the coupling relationship between laser power, scanning speed, and defocusing amount. Based on the particle swarm optimization algorithm, an initial parameter adjustment scheme is generated, and the state change of the molten pool under this scheme is simulated. If the prediction result shows that the molten pool characteristics do not meet the standard, the parameters are fine-tuned based on the coupling relationship to generate a new adjustment scheme. The processing parameters are repeatedly iteratively optimized until the parameter combination that returns the molten pool characteristics to the normal range is obtained, and the optimal adjustment parameters are determined.

[0011] Preferably, the adjusted molten pool characteristic data is compared again with the normal molten pool characteristic library to determine whether the molten pool state has returned to normal. If the molten pool state is still abnormal, the parameter optimization adjustment process is repeated until the molten pool characteristics return to the normal threshold range, which specifically includes: The molten pool characteristic data adjusted based on the optimal adjustment parameter set is compared again with the normal molten pool characteristic library to determine whether the molten pool state has returned to normal. If all dimensional characteristics are within the normal threshold range and the continuous frame fluctuation meets the stability requirement, it is determined that the molten pool state has returned to normal. If any dimensional characteristic still exceeds the normal threshold or the continuous frame fluctuation is still unstable, it is determined that the molten pool state is still abnormal, and the historical record of the last parameter adjustment is retrieved. The historical record data is excluded in the repeated parameter optimization and adjustment process until the molten pool characteristics return to the normal threshold range. If the molten pool state has not returned to normal after three consecutive parameter adjustments, an alarm mechanism is triggered, the laser processing is suspended, and an abnormal prompt information is output.

[0012] Preferably, during the laser processing, the processing data is recorded in real time to form a processing process database. After the processing is completed, the normal molten pool sample data in the database is updated and supplemented by an incremental learning algorithm, which specifically includes: During the laser processing, the processing time, workpiece position, molten pool characteristic data, adjusted processing parameters, and processing quality detection results are recorded in real time to form a processing process database. After the processing is completed, the normal molten pool samples that can be used to update the characteristic library are selected to supplement the database. Based on the initial normal molten pool characteristic library, new samples are integrated in an incremental learning manner, and the normal threshold of each dimension is fine-tuned and optimized after updating the characteristic library.

[0013] Further, an adaptive laser processing control system based on real-time molten pool monitoring is proposed, which includes: Laser processing module: The module is used to output a laser beam and process a workpiece. The molten pool monitoring module: the module synchronously collects molten pool images, temperature and spectral data through a high-speed camera, an infrared sensor and a spectral sensor; The data preprocessing module: the module filters, denoises, aligns and feature extracts the original data to generate geometric, temperature and spectral feature parameters; The feature fusion module: the module weightedly fuses multi-source feature data based on an analytic hierarchy process to output a comprehensive molten pool feature vector; The anomaly detection module: the module compares real-time molten pool features with feature library thresholds to identify molten pool overheating and size anomaly fault types; The parameter optimization module: the module constructs a multi-parameter collaborative optimization model and generates an optimal adjustment scheme of laser power and speed parameters by using a particle swarm optimization algorithm; The real-time control module: the module dynamically adjusts laser processing parameters and feeds back molten pool states after adjustment to the anomaly detection module for closed-loop verification; The data storage and learning module: the module is used for recording processing data and updating normal molten pool feature library threshold ranges through incremental learning; The processor: the processor is used for processing calculation processes of formulas and construction and calculation processes of models.

[0014] Compared with the prior art, the advantages of the present application are that: Multi-source synchronous data collection through a high-speed camera, an infrared sensor and a spectral sensor, combined with preprocessing and weighted fusion algorithms, can accurately obtain comprehensive molten pool features and lay a reliable foundation for state judgment; secondly, comparison and feature matching with a normal molten pool feature library can quickly identify various abnormal types and avoid the expansion of processing defects; thirdly, a multi-parameter collaborative optimization model combined with a particle swarm optimization algorithm can efficiently iterate optimal adjustment parameters, and through closed-loop verification, it can ensure that the molten pool returns to normal and guarantee processing quality; finally, the combination of a processing process database and an incremental learning algorithm can continuously update the feature library, improve the adaptability of the method to different workpieces and working conditions, and overall realize high-precision, high-stability and intelligent control of laser processing. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 The method proposed in the present application is shown in the figure; Figure 2 The normal molten pool feature library constructed in the present application is shown in the figure; Figure 3 The data acquisition in the present application is shown in the figure; Figure 4 The comprehensive molten pool feature data acquisition in the present application is shown in the figure; Figure 5 The determination of abnormal types in the present application is shown in the figure; Figure 6 The schematic diagram for obtaining the optimal adjustment parameter set proposed by the present application is shown in the figure; Figure 7 The schematic diagram for optimizing the adjustment of the molten pool characteristic data proposed by the present application is shown in the figure; Figure 8 The schematic diagram for updating the database proposed by the present application is shown in the figure. DETAILED DESCRIPTION

[0016] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought of by those skilled in the art.

[0017] The adaptive laser processing control system based on real-time molten pool monitoring comprises: A laser processing module: the module is used for outputting a laser beam and processing a workpiece; A molten pool monitoring module: the module synchronously collects molten pool images, temperature and spectral data through a high-speed camera, an infrared sensor and a spectral sensor; A data preprocessing module: the module filters, denoises, aligns and extracts features of the original data, and generates geometric, temperature and spectral feature parameters; A feature fusion module: the module weightedly fuses multi-source feature data based on an analytic hierarchy process, and outputs a molten pool comprehensive feature vector; An anomaly detection module: the module compares the real-time molten pool features with the threshold of the feature library, and identifies the molten pool overheating and size anomaly fault types; A parameter optimization module: the module constructs a multi-parameter collaborative optimization model, and generates an optimal adjustment scheme of laser power and speed parameters by using a particle swarm optimization algorithm; A real-time control module: the module dynamically adjusts the laser processing parameters, and feeds back the molten pool state after adjustment to the anomaly detection module for closed-loop verification; A data storage and learning module: the module is used for recording the data in the whole processing process, and updating the threshold range of the normal molten pool feature library through incremental learning; A processor: the processor is used for processing the calculation process of each formula and the construction and calculation process of each model.

[0018] Referring to Figure 1 The adaptive laser processing control method based on real-time molten pool monitoring comprises: Step one: obtaining workpiece parameter data and normal molten pool sample data under different processing parameter combinations, and initializing laser processing parameters to construct a normal molten pool feature library; Step two: processing the workpiece based on the initialized laser processing parameters, and collecting molten pool images, temperature data and spectral data through a high-speed camera, an infrared sensor and a spectral sensor; Step three: based on the collected data for preprocessing, through the weighted fusion algorithm for fusion after the data preprocessing, get the fusion of the molten pool comprehensive feature data; Step four: the molten pool comprehensive feature data obtained is compared with the normal molten pool feature library, whether the current molten pool feature is in the normal threshold range of the corresponding processing stage is judged, if at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal, and the abnormal type is determined through feature matching; Step five: based on the matching determined abnormal type, the adjustment direction of the laser processing parameter is determined, and a multi-parameter collaborative optimization model is constructed, and the processing parameters are iteratively optimized in the adjustment direction through the particle swarm optimization algorithm, and the optimal adjustment parameter group is obtained; Step six: based on the adjusted molten pool feature data, the molten pool feature library is compared again, whether the molten pool state is normal is judged, if the molten pool state is still abnormal, the parameter optimization adjustment process is carried out, until the molten pool feature returns to the normal threshold range; Step seven: in the laser processing process, the processing data is recorded in real time, the processing process database is formed, after the processing is completed, the normal molten pool sample data in the database is updated and supplemented through the incremental learning algorithm.

[0019] Referring to Figure 2 As shown, the parameter data of the workpiece to be processed and the normal molten pool sample data under different processing parameter combinations are obtained, and the laser processing parameters are initialized, and the normal molten pool feature library is constructed, specifically including: The parameter data of the workpiece to be processed includes material properties, thickness size and processing target; The normal molten pool sample data includes geometric features, temperature features and spectral features; The laser processing parameters include laser initial power, initial scanning speed, initial defocusing amount and protective gas flow; The normal molten pool sample data is trained through the machine learning algorithm, and the normal molten pool feature library is constructed, and the feature library contains the molten pool feature threshold range corresponding to different processing stages.

[0020] Referring to Figure 3 As shown, based on the initialized laser processing parameters, the workpiece is processed, and the molten pool image, temperature data and spectral data collected by the high-speed camera, infrared sensor and spectral sensor specifically include: The high-speed camera collects real-time images of the molten pool at a frame rate of not less than 1000fps; The infrared sensor collects the temperature distribution data of the molten pool area at a sampling frequency of not less than 500Hz; The spectral sensor collects the plasma spectral data of the molten pool in the wavelength range of 200-1100nm; The three collected data are aligned in time by a synchronous trigger signal to obtain the molten pool image, temperature data and spectral data corresponding to the same processing time.

[0021] Referring to Figure 4 As shown, the collected data is preprocessed, and the preprocessed data is fused by a weighted fusion algorithm to obtain the fused molten pool comprehensive feature data, which specifically includes: The collected molten pool image data is processed by a Gaussian filter algorithm to remove high-frequency noise, histogram equalization to improve the contrast between the molten pool and the background, and edge detection to obtain a clear molten pool contour image, and geometric feature parameters are calculated; The infrared temperature measurement data is processed by removing data points outside the reasonable temperature range using the 3σ criterion, and smoothing to obtain the molten pool center temperature and temperature gradient data; The spectral data is baseline corrected and feature spectral line extracted to obtain feature spectral line intensity and spectral line half-width data; The weight values of each collected data are calculated based on the analytic hierarchy process, and the geometric feature data, temperature feature data and spectral feature data are fused by a weighted fusion algorithm to obtain the molten pool comprehensive feature data.

[0022] Specifically, in the original spectral data, in addition to the characteristic spectral signals generated by the molten pool reaction, there is also baseline drift. A baseline correction algorithm such as a polynomial fitting baseline method is used to identify the baseline component in the spectrum, and then the baseline signal is subtracted from the original spectral data through mathematical operation to eliminate the influence of baseline drift, obtaining corrected spectral data containing only the characteristic spectrum of the molten pool; The corrected spectral data is analyzed, and the corresponding feature spectral line is located and extracted in the spectral curve according to the known characteristic spectral line wavelength of the element involved in the molten pool reaction. For example, if the molten pool contains iron, the spectral line peak at the wavelength position can be found according to the characteristic spectral line wavelength of iron; For each extracted feature spectral line, its characteristic parameters are calculated, wherein the feature spectral line intensity is determined by measuring the maximum gray value of the spectral line peak, representing the strength of the feature spectral line; the spectral line half-width is the spectral line width at half the height of the spectral line peak, reflecting the sharpness of the spectral line, and the feature spectral line intensity and spectral line half-width data are recorded; A hierarchical structure model is constructed to clearly define the target, criteria and scheme layers of the analytic hierarchy. The target layer is to determine the weight of each collected data, which is used for molten pool comprehensive feature data fusion. The criterion layer is the key factor affecting the importance of the data, such as the accuracy of the data reflecting the molten pool state, the stability of the data collection, and the relevance of the data to the core features of the molten pool. The scheme layer is the three types of data whose weights need to be determined, namely, the molten pool geometric feature data, the molten pool temperature feature data, and the molten pool spectral feature data. Experts in the field of invitation are invited to compare the relative importance of the three types of data in the scheme layer according to the importance of each factor in the criterion layer. A 1-9 scale method is used to construct a pairwise comparison judgment matrix to ensure that the elements in the matrix meet the consistency requirements. Hierarchical single ordering and consistency checking: Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector. After normalization processing of the eigenvector, the weight vector of the three types of data in the scheme layer relative to each factor in the criterion layer is obtained. At the same time, the consistency index, random consistency index and consistency ratio are calculated. If the consistency ratio is less than 0.1, it is considered that the judgment matrix meets the consistency requirements and the weight distribution is reasonable. If it does not meet the requirements, experts need to adjust the judgment matrix again until it meets the consistency requirements.

[0023] Referring to Figure 5 The acquired comprehensive molten pool feature data is compared with the normal molten pool feature library to determine whether the current molten pool feature is within the normal threshold range of the corresponding processing stage. If at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal, and the abnormal type is determined through feature matching. Specifically, it includes: From the normal molten pool feature library, the reference data matching the current processing stage is selected, and the dimensional features of the current molten pool are compared with the normal feature range of the corresponding processing stage; If all the comprehensive molten pool feature data is within the normal threshold range, it is determined that the current molten pool state is normal, and the processing continues; If at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal; Based on the feature dimension exceeding the normal range and the deviation direction, the abnormal type is identified, including molten pool overheating, molten pool overcooling, molten pool size being too large, molten pool size being too small, and plasma interference.

[0024] Specifically, through the control system of the processing equipment, process flow record or real-time monitoring data such as processing time, completed work amount, process parameter setting, the specific processing stage of the current molten pool is determined; According to the determined current processing stage, the reference data corresponding to this stage is selected from the normal molten pool feature library, including the normal threshold range of the geometric feature, temperature feature and spectral feature of the molten pool in this stage, to form a list of reference features in the current stage; The acquired comprehensive molten pool feature data is decomposed into geometric features, temperature features and spectral features, and the specific parameters of each dimension are compared one by one according to the threshold range in the reference feature list of the current stage: Geometric feature comparison: Check whether the contour length, area, equivalent diameter and aspect ratio of the current molten pool are within the normal range of the geometric feature of the processing stage; Temperature feature comparison: check whether the center temperature and temperature gradient of the current molten pool are within the normal range of the temperature feature of the current processing stage; Spectrum feature comparison: check whether the intensity and half-width of each characteristic spectrum line of the current molten pool are within the normal range of the spectrum feature of the current processing stage; If the comparison result shows that all characteristic parameters of the current molten pool in all dimensions are within the normal threshold range corresponding to the reference feature list of the current stage, and none of the parameters exceeds the range, it is determined that the current molten pool state is normal, and no intervention is required, and the processing equipment continues to process according to the original process parameters. If the comparison result shows that at least one characteristic parameter of the current molten pool exceeds the corresponding normal threshold range, it is directly determined that the current molten pool state is abnormal, and the abnormal type identification process needs to be started immediately, and intervention measures are taken according to the subsequent results.

[0025] Referring to Figure 6 As shown, the adjustment direction of the laser processing parameters is determined based on the matched abnormal type, and a multi-parameter collaborative optimization model is constructed, and the particle swarm optimization algorithm is used to iteratively optimize the processing parameters in the adjustment direction to obtain the optimal adjustment parameter group, which specifically includes: The adjustment direction of the laser processing parameters is obtained based on the matched abnormal type. A multi-parameter collaborative optimization model is constructed by taking the regression of the molten pool features to the normal threshold range as the objective function, and combining the coupling relationship among the laser power, scanning speed and defocusing amount. Based on the particle swarm optimization algorithm, an initial parameter adjustment scheme is generated, and the state change of the molten pool under the scheme is simulated. If the prediction result shows that the molten pool features do not meet the requirements, the parameters are fine-tuned based on the coupling relationship to generate a new adjustment scheme. The processing parameters are repeatedly iteratively optimized until the parameter combination that makes the molten pool features return to the normal range is obtained, and the optimal adjustment parameter group is determined.

[0026] Specifically, according to the logic of the particle swarm optimization algorithm, the optimal adjustment parameter group is gradually obtained through the processes of generating an initial scheme, simulating and verifying, fine-tuning parameters, and repeated iteration. The particle position update formula in the particle swarm optimization algorithm is:

[0027] wherein, is the velocity of the i-th particle in the d-th dimension, is the position of the i-th particle in the d-th dimension, is the inertia weight, , is the learning factor, , is a random number in the range of [0, 1], The individual optimal position of the i th particle, The global optimal position; According to the pre-determined molten pool abnormal type, combined with the correlation law of laser processing parameters and molten pool characteristics, the corresponding parameter adjustment direction of each abnormal type is determined, including the adjustment direction corresponding to the molten pool overheating, the adjustment direction corresponding to the molten pool undercooling, the adjustment direction corresponding to the molten pool size being too large, the adjustment direction corresponding to the molten pool size being too small, and the adjustment direction corresponding to the plasma interference; The objective function is defined as the minimization of the sum of the deviations of the current molten pool characteristic parameters from the corresponding normal threshold range, and the constraint conditions are set: the laser power needs to be within the rated power range of the equipment, the scanning speed needs to meet the processing efficiency requirements, and the defocusing amount needs to be within the adjustable range of the equipment; Coupling of power and scanning speed: when the power is increased, if the scanning speed is also increased synchronously, part of the heat accumulation can be offset (to avoid molten pool overheating); when the power is reduced, if the scanning speed is also reduced synchronously, heat loss can be reduced, and the two need to be adjusted coordinately to balance the heat input; Coupling of power and defocusing amount: high power combined with small defocusing amount is easy to cause energy to be too concentrated, so the defocusing amount needs to be increased synchronously; low power combined with large defocusing amount is easy to cause energy deficiency, so the defocusing amount needs to be reduced synchronously, and the two need to be matched to control the energy density; Coupling of scanning speed and defocusing amount: high scanning speed combined with large defocusing amount is easy to cause the molten pool size to be too small due to short action time and low energy density, so the defocusing amount needs to be reduced synchronously; low scanning speed combined with small defocusing amount is easy to cause the molten pool to be overheated due to long action time and energy concentration, so the defocusing amount needs to be increased synchronously, and the two need to be coordinated to control the balance between molten pool diffusion and heat accumulation.

[0028] Referring to Figure 7 As shown, the molten pool characteristic data adjusted based on the adjusted parameters is compared with the normal molten pool characteristic library again to determine whether the molten pool state has returned to normal, and if the molten pool state is still abnormal, the adjustment process of parameter optimization is repeated until the molten pool characteristics return to the normal threshold range, which specifically includes: The molten pool characteristic data adjusted based on the optimal adjustment parameter group is compared with the normal molten pool characteristic library again to determine whether the molten pool state has returned to normal; If all dimensional characteristics are within the normal threshold range, and the continuous frame fluctuation meets the stability requirement, it is determined that the molten pool state has returned to normal; If any dimensional characteristic still exceeds the normal threshold or the continuous frame fluctuation is still unstable, it is determined that the molten pool state is still abnormal, the historical record of the last parameter adjustment is retrieved, and the historical record data is excluded in the repeated parameter optimization and adjustment process until the molten pool characteristics return to the normal threshold range; If the molten pool state has not returned to normal after three times of parameter adjustment, an alarm mechanism is triggered, the laser processing is suspended, and abnormal prompt information is output.

[0029] Specifically, from the normal molten pool feature library, the reference data corresponding to the current processing stage is retrieved again, and the newly collected adjusted molten pool feature data is compared with the normal threshold range in the reference data in a dimension-by-dimension comparison manner, and the specific value of each dimension feature parameter and whether it is within the normal range are recorded; If the adjusted molten pool state does not meet the normal determination standard, the parameter optimization iteration process is started, the last parameter optimization history record is retrieved from the device parameter adjustment log, including the last abnormal type, initial parameter combination, adjustment direction, and all generated parameter schemes, and a historical parameter adjustment list is arranged to clearly indicate the parameter range and adjustment direction that have been tried to avoid repeated use of the same or similar parameter schemes in subsequent iterations and improve optimization efficiency; Based on the current existing abnormal type, in combination with the historical parameter adjustment list, when constructing a new multi-parameter collaborative optimization model, the parameter combination range that has been tried is excluded to generate a new initial parameter adjustment scheme; According to the process of constructing a multi-parameter collaborative optimization model-particle swarm optimization algorithm iteration optimization-determining the optimal adjustment parameter group, the newly generated parameter scheme is simulated and verified, fine-tuned and optimized to determine the optimal adjustment parameter group for the next round, and then the device parameters are adjusted again, the molten pool feature data is collected, compared with the normal feature library, and the above cycle is repeated until all dimension features of the molten pool return to the normal threshold range and the frame fluctuation is stable.

[0030] Referring to Figure 8 As shown in the figure, during the laser processing, the processing data is recorded in real time to form a processing process database, and after the processing is completed, the normal molten pool sample data in the database is updated and supplemented by an incremental learning algorithm, specifically including: During the laser processing, the processing time, workpiece position, molten pool feature data, adjusted processing parameters, and processing quality detection results are recorded in real time to form a processing process database; After the processing is completed, normal molten pool samples that can be used to update the feature library are selected to supplement the database; Based on the initial normal molten pool feature library, new samples are integrated in an incremental learning manner, and after updating the feature library, the normal threshold of each dimension is fine-tuned and optimized.

[0031] Specifically, after all new samples are integrated into the model, the distribution of the molten pool features of each processing stage is re-counted based on the updated model output, the original sample data of each processing stage is replaced, and the normal threshold range of each dimension feature is fine-tuned based on the new statistical results. If there is a small deviation between the statistical range of a certain feature parameter of the new sample and the initial threshold value, the threshold value is adjusted in the form of the intersection of the initial threshold value and the statistical range of the new sample + reasonable expansion. If the distribution of the feature parameters of the new sample is significantly different from the initial threshold value, it is necessary to combine the processing quality detection results and the opinions of process experts to determine whether the normal range is deviated due to the change of equipment state or material, and then determine the adjustment range of the threshold value, so as to ensure that the adjusted threshold value not only conforms to the current processing actual situation, but also can effectively distinguish the normal and abnormal molten pool. After the threshold value is fine-tuned, the updated normal molten pool feature library is verified: samples in the library are randomly extracted, and it is checked whether the feature parameters are within the adjusted threshold range. At the same time, combined with the historical processing quality data, it is confirmed that the threshold value can accurately correspond to the qualified processing quality. After verification, the updated feature library is saved for subsequent molten pool state judgment of laser processing.

[0032] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes a specific embodiment of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or may be advantageous.

[0033] Each embodiment in the present specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the differences from other embodiments.

[0034] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of adaptive laser processing control based on real-time molten pool monitoring, characterized in that, The application relates to a laser processing parameter adjustment method and device. The method comprises the following steps: acquiring workpiece parameter data and normal molten pool sample data under different processing parameter combinations, and initializing laser processing parameters to construct a normal molten pool feature library; processing the workpiece based on the initialized laser processing parameters, and collecting molten pool images, temperature data and spectrum data through a high-speed camera, an infrared sensor and a spectrum sensor; preprocessing the collected data, fusing the preprocessed data through a weighted fusion algorithm, and acquiring fused molten pool comprehensive feature data; comparing the acquired molten pool comprehensive feature data with the normal molten pool feature library, judging whether the current molten pool features are within the normal threshold range of the corresponding processing stage, and if at least one molten pool feature data is out of the normal threshold range, determining that the current molten pool state is abnormal, and determining the abnormal type through feature matching; determining the adjustment direction of the laser processing parameters based on the matched abnormal type, constructing a multi-parameter collaborative optimization model, and iteratively optimizing the processing parameters in the adjustment direction through a particle swarm optimization algorithm to acquire optimal adjustment parameters; comparing the adjusted molten pool feature data with the normal molten pool feature library again, judging whether the molten pool state returns to normal, and if the molten pool state is still abnormal, adjusting the parameter optimization process until the molten pool features return to the normal threshold range; 2. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, during the laser processing, processing data is recorded in real time to form a processing process database, and after the processing is completed, the normal molten pool sample data in the database is updated and supplemented through an incremental learning algorithm. The acquired workpiece parameter data and normal molten pool sample data under different processing parameter combinations, and the initialized laser processing parameters, construct a normal molten pool feature library, and the workpiece parameter data comprises material properties, thickness size and processing targets. The normal molten pool sample data comprises geometric features, temperature features and spectrum features. The laser processing parameters comprise laser initial power, initial scanning speed, initial defocusing amount and protective gas flow. The normal molten pool sample data is trained through a machine learning algorithm to construct a normal molten pool feature library, and the feature library contains molten pool feature threshold ranges corresponding to different processing stages.

3. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, The high-speed camera collects real-time molten pool images at a frame rate of not less than 1000 fps. The infrared sensor collects temperature distribution data of the molten pool area at a sampling frequency of not less than 500 Hz. The spectrum sensor collects molten pool plasma spectrum data in a waveband range of 200-1100 nm. The three kinds of collected data are time-aligned through a synchronous trigger signal to acquire molten pool images, temperature data and spectrum data corresponding to the same processing moment. The preprocessing of the collected data, the fusion of the preprocessed data through a weighted fusion algorithm, and the acquisition of fused molten pool comprehensive feature data specifically comprise the following steps:

4. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, ​ The collected molten pool image data is processed by a Gaussian filtering algorithm to remove high-frequency noise, histogram equalization to improve the contrast between the molten pool and the background, and edge detection to obtain a clear molten pool contour image, and geometric feature parameters are calculated; The infrared temperature measurement data is processed by removing data points outside the reasonable temperature range using the 3σ criterion and smoothing to obtain molten pool center temperature and temperature gradient data; Baseline correction and characteristic spectral line extraction are performed on the spectral data to obtain characteristic spectral line intensity and spectral line half-width data; The weight values of each collected data are calculated based on the analytic hierarchy process, and the geometric feature data, temperature feature data, and spectral feature data are fused by a weighted fusion algorithm to obtain comprehensive molten pool feature data.

5. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, The obtained molten pool comprehensive feature data is compared with the normal molten pool feature library to determine whether the current molten pool features are within the normal threshold range of the corresponding processing stage. If at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal, and the abnormal type is determined through feature matching, which specifically includes: From the normal molten pool feature library, the reference data matching the current processing stage are selected, and the dimensions of the current molten pool are compared with the normal feature range of the corresponding processing stage; If all the comprehensive feature data of the molten pool is within the normal threshold range, it is determined that the current molten pool state is normal, and the processing continues; If at least one molten pool feature data exceeds the normal threshold range, it is determined that the current molten pool state is abnormal; Based on the feature dimensions and deviation directions that exceed the normal range, the abnormal type is identified. The abnormal types include molten pool overheating, molten pool overcooling, molten pool size being too large, molten pool size being too small, and plasma interference.

6. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, The adjustment direction of the laser processing parameters is determined based on the matched abnormal type, and a multi-parameter collaborative optimization model is constructed. The processing parameters are iteratively optimized in the adjustment direction by a particle swarm optimization algorithm to obtain the optimal adjustment parameter set, which specifically includes: Based on the matched abnormal type, the adjustment direction of the laser processing parameters is obtained; A multi-parameter collaborative optimization model is constructed by taking the regression of the molten pool features to the normal threshold range as the objective function and considering the coupling relationship between the laser power, scanning speed, and defocusing amount; Based on the particle swarm optimization algorithm, an initial parameter adjustment scheme is generated, and the state change of the molten pool under this scheme is simulated. If the prediction result shows that the molten pool features do not meet the standards, the parameters are fine-tuned based on the coupling relationship to generate a new adjustment scheme; The processing parameters are repeatedly iteratively optimized until the parameter combination that returns the molten pool features to the normal range is obtained, and the optimal adjustment parameter set is determined.

7. The adaptive laser processing control method based on real-time molten pool monitoring according to claim 1, characterized in that, The molten pool feature data adjusted based on the adjusted molten pool feature data is compared with the normal molten pool feature library again to determine whether the molten pool state has returned to normal. If the molten pool state is still abnormal, the parameter optimization adjustment process continues until the molten pool features return to the normal threshold range, which specifically includes: The molten pool feature data adjusted based on the optimal adjustment parameter set is compared with the normal molten pool feature library again to determine whether the molten pool state has returned to normal. If all the dimensional features are within the normal threshold range and the continuous frame fluctuations meet the stability requirements, it is determined that the molten pool state has returned to normal. If any dimension feature still exceeds the normal threshold or the continuous frame fluctuation is still unstable, it is determined that the molten pool state is still abnormal, the historical record of the last parameter adjustment is called, and the historical record data is excluded in the repeated parameter optimization and adjustment process until the molten pool feature returns to the normal threshold range; If the molten pool state still does not return to normal after three times of parameter adjustment, an alarm mechanism is triggered, the laser processing is suspended, and an abnormal prompt information is output.

8. The adaptive laser processing control method based on real-time molten pool monitoring of claim 1, wherein, The processing data is recorded in real time during the laser processing to form a processing process database, and after the processing is completed, the normal molten pool sample data in the database is updated and supplemented by an incremental learning algorithm, which specifically includes: During the laser processing, the processing time, workpiece position, molten pool feature data, adjusted processing parameters and processing quality detection results are recorded in real time to form a processing process database; After the processing is completed, the normal molten pool samples that can be used to update the feature library are screened to supplement the database; Based on the initial normal molten pool feature library, new samples are integrated in an incremental learning manner, and after updating the feature library, the normal threshold of each dimension is fine-tuned and optimized.

9. Adaptive laser processing control system based on real-time molten pool monitoring for implementing the adaptive laser processing control method based on real-time molten pool monitoring according to any one of claims 1 to 8, characterized in that It includes: A laser processing module: the module is used to output a laser beam and process a workpiece; A molten pool monitoring module: the module synchronously collects molten pool image, temperature and spectrum data through a high-speed camera, an infrared sensor and a spectrum sensor; A data preprocessing module: the module filters, denoises, aligns and extracts features of the original data to generate geometric, temperature and spectrum feature parameters; A feature fusion module: the module weightedly fuses multi-source feature data based on an analytic hierarchy process to output a molten pool comprehensive feature vector; An abnormality detection module: the module compares real-time molten pool features with feature library thresholds to identify molten pool overheating and size abnormality fault types; A parameter optimization module: the module constructs a multi-parameter collaborative optimization model and generates an optimal adjustment scheme of laser power and speed parameters by using a particle swarm optimization algorithm; A real-time control module: the module dynamically adjusts laser processing parameters and feeds back the molten pool state after adjustment to the abnormality detection module for closed-loop verification; A data storage and learning module: the module is used to record processing process data and update normal molten pool feature library threshold ranges by incremental learning; A processor: the processor is used to process the calculation process of each formula and the construction and calculation process of each model.

Images