Parameter optimization control method and system for the entire process of precision electronic surface welding

Abstract

The present invention relates to the technical field of electronic welding and assembly, and discloses a method and system for optimizing and controlling parameters of the entire process of precision electronic surface welding, comprising adjusting preset standard processing parameters based on spatial distribution characteristics and weld position characteristics, processing components to be welded according to real-time processing parameters, obtaining penetration state characteristics during the processing, inputting the penetration state characteristics into a pre-built state classification model to obtain a penetration state category, and finally calculating a defect influence value based on welding defect data and the penetration state category, and optimizing and adjusting the real-time processing parameters based on the defect influence value. In this way, not only defects before processing of components to be welded and defects generated during processing are taken into account, but also different penetration states are considered to affect the adjustment of welding parameters, thereby achieving more accurate welding parameter adjustment. This comprehensive optimization process can effectively reduce the occurrence of welding defects.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic soldering, and more particularly to a method and system for optimizing control of parameters in the entire process of precision electronic surface soldering. Background Art

[0002] The manufacturing process of precision electronic products requires extremely high precision and error control, and laser welding technology is usually used. This technology uses the energy of the laser beam to quickly melt and solidify the material, thereby achieving material connection. As a process that does not require the addition of solder, laser welding can minimize weld deformation while ensuring excellent forming quality and high connection accuracy. Over the past few decades, laser welding technology has developed rapidly, and the welding process level and welding quality have been widely recognized. With the continuous advancement of intelligent welding technology, existing technologies can already judge welding quality through real-time data and dynamically adjust processing parameters accordingly.

[0003] For example, the Chinese patent with publication number CN106270876A provides a method for electron beam brazing welding of aluminum-lithium alloys and titanium alloys. The patented electron beam brazing has the characteristics of concentrated energy, small heat-affected zone, and high power density, which effectively reduces the unevenness of interface reaction. By adjusting the welding angle, groove angle, height difference, preheating method and electron beam parameters of the test plate, the welding process is further optimized. The Chinese patent with publication number CN115319256A provides an electron beam welding method. The patent controls the electron beam to weld the test sample weld, adjusts the electron beam spot parameters and welding parameters according to the quality of the formed weld, and then uses the optimized parameters to weld the workpiece weld.

[0004] Although existing patents can adjust welding parameters in real time during the processing process, due to the defects of the components to be welded before processing and defects generated during the processing, the existing technology usually only optimizes a certain factor. In addition, during the laser welding process, the different penetration states of the surfaces of the components to be welded will also affect the adjustment of welding parameters. However, the existing technology does not take these factors into consideration comprehensively, resulting in the accuracy of the welding parameter adjustment being affected.

[0005] In view of this, the present invention proposes a method and system for optimizing and controlling parameters of the entire process of precision electronic surface welding to solve the above problems. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for optimizing and controlling parameters of the entire process of precision electronic surface welding.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] First, the parameter optimization control method for the entire process of precision electronic surface welding includes:

[0009] Obtain the spatial distribution characteristics of the components to be welded and the weld position characteristics, adjust the preset standard processing parameters based on the spatial distribution characteristics and weld position characteristics, and obtain real-time processing parameters;

[0010] Processing the components to be welded according to real-time processing parameters, obtaining penetration state characteristics during the processing, and inputting the penetration state characteristics into a pre-built state classification model to obtain the penetration state category;

[0011] During the processing, the welding defect data is obtained in real time, the defect impact value is calculated based on the welding defect data and the penetration state category, and the real-time processing parameters are optimized and adjusted according to the defect impact value.

[0012] Furthermore, the method for obtaining spatial distribution characteristics includes:

[0013] Acquire acoustic wave reflection data of the component to be welded, generate a thickness distribution map corresponding to the surface of the component to be welded based on the acoustic wave reflection data, acquire a real-time surface image of the component to be welded, generate a grayscale distribution map corresponding to the surface of the component to be welded based on the real-time surface image, and cascade-fuse the thickness distribution map and the grayscale distribution map to obtain spatial distribution characteristics.

[0014] Furthermore, the method for obtaining weld position features includes:

[0015] Based on the real-time surface image and edge detection algorithm, the first edge point and the second edge point on both sides of the weld are obtained, the first longitudinal distance between the first edge point and any point in the weld is calculated, and the second longitudinal distance between the second edge point and any point in the weld is calculated. The arbitrary point corresponding to the maximum value of the sum of the first longitudinal distance and the second longitudinal distance is taken as the center point of the weld, and the first edge point, the second edge point and the center point of the weld are taken as the weld position features.

[0016] Furthermore, the method of adjusting the preset standard processing parameters to obtain real-time processing parameters includes:

[0017] The spatial distribution characteristics, weld position characteristics and preset standard processing parameters are input into the pre-built parameter adjustment model to obtain the real-time processing parameters output by the parameter adjustment model.

[0018] Furthermore, the method for constructing the parameter adjustment model includes:

[0019] Acquire a first sample data set, wherein the first sample data set includes historical spatial distribution characteristics, historical weld position characteristics, preset standard processing parameters, and historical processing parameters;

[0020] Divide the sample data set into a sample training set and a sample test set, and build a regression network;

[0021] The historical spatial distribution characteristics, historical weld position characteristics, and preset standard processing parameters in the sample training set are used as the input data of the regression network, and the historical processing parameters in the sample training set are used as the output data of the regression network. The regression network is trained to obtain an initial regression network for predicting real-time processing parameters.

[0022] The initial regression network is tested using a sample test set, and the output is an initial regression network that satisfies the requirement that the prediction error between the output data and the actual historical processing parameters in the sample test set is less than a preset error value as a parameter adjustment model.

[0023] Furthermore, the method for obtaining the melt penetration state characteristics includes:

[0024] The processing image is acquired, and the melt pool behavior characteristics, melt pool state characteristics and through-hole characteristics are extracted from the processing image. The melt pool behavior characteristics and the melt pool state characteristics are fused to generate the melt pool dynamic characteristics, and the melt pool dynamic characteristics and the through-hole characteristics are used as the melt-through state characteristics.

[0025] Furthermore, the penetration state categories include incomplete penetration, moderate penetration, complete penetration, and excessive penetration. The method for constructing the state classification model includes:

[0026] Acquire Q sets of training data, where Q is a positive integer greater than 1, and the training data includes historical penetration state characteristics and historical penetration state categories;

[0027] The historical penetration state characteristics and historical penetration state categories are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0028] The historical penetration state features in the training set are used as input data, and the historical penetration state categories in the training set are used as output data to train the classifier and obtain an initial classifier;

[0029] The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the state classification model.

[0030] Furthermore, the welding defect data includes amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter and real-time molten pool area, and the defect impact value is determined according to the amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter, real-time molten pool area and penetration state category.

[0031] Furthermore, the method for optimizing and adjusting the real-time processing parameters according to the defect impact value includes:

[0032] S301: Using M real-time processing parameters as an initial population and the defect impact value as a fitness function, the fitness value of each individual in the initial population is calculated;

[0033] S302: Using the tournament selection strategy, randomly select 4 individuals from the population, compare their fitness values, and select the individual with the smallest fitness value as the parent individual;

[0034] S303: Repeat S301-S302 until N parent individuals are selected, then go to S304;

[0035] S304: Select an intersection point on the path of each pair of parent individuals, use the single-point crossover method to operate on all parent individuals, exchange the parts of the parent individuals after the intersection point, and generate two new child individuals;

[0036] S305: Randomly select two nodes for each child individual and exchange the positions of the two nodes, forming a new population from all the child individuals and replacing the original population, ensuring that the size of the new population remains N;

[0037] S306: The convergence threshold of the preset fitness value is CT. If the fitness value change of the optimal individual in the population of consecutive K generations is less than CT, the genetic algorithm terminates.

[0038] S307: When the genetic algorithm terminates, the sub-individual with the smallest fitness value in the population is the optimal processing parameter.

[0039] In the second aspect, a system for optimizing and controlling parameters of the entire process of precision electronic surface welding is provided, which is used to implement the above-mentioned method for optimizing and controlling parameters of the entire process of precision electronic surface welding, including:

[0040] Pre-adjustment module: used to obtain the spatial distribution characteristics of the components to be welded and the weld position characteristics, and adjust the preset standard processing parameters based on the spatial distribution characteristics and weld position characteristics to obtain real-time processing parameters;

[0041] Classification module: used to process the components to be welded according to real-time processing parameters, obtain the penetration state characteristics during the processing, input the penetration state characteristics into the pre-built state classification model, and obtain the penetration state category;

[0042] Optimization module: used to obtain welding defect data in real time during the processing, calculate the defect impact value based on the welding defect data and penetration state category, and optimize and adjust the real-time processing parameters based on the defect impact value.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention first adjusts the preset standard processing parameters based on the spatial distribution characteristics and the weld position characteristics to obtain real-time processing parameters, then processes the components to be welded according to the real-time processing parameters, obtains the penetration state characteristics during the processing, inputs the penetration state characteristics into a pre-built state classification model to obtain the penetration state category, and finally calculates the defect influence value according to the welding defect data and the penetration state category, and optimizes and adjusts the real-time processing parameters according to the defect influence value. In this way, not only the defects of the components to be welded before processing and the defects generated during the processing are taken into account, but also the fact that different penetration states will affect the adjustment of welding parameters, thereby achieving more accurate welding parameter adjustment. This comprehensive optimization process can effectively reduce the occurrence of welding defects and improve the stability and consistency of welding quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of the method for optimizing the control of parameters in the entire process of precision electronic surface welding in the present invention;

[0046] Figure 2 Schematic diagram of data flow in the present invention;

[0047] Figure 3 Schematic diagram of the weld, the first edge point, the second edge point, and the weld center point in the present invention;

[0048] Figure 4 It is a structural diagram of the parameter optimization control system for the entire process of precision electronic surface welding in the present invention. DETAILED DESCRIPTION

[0049] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Example 1

[0051] See also Figure 1 、 Figure 2 As shown, Figure 1 This is a flow chart of the method for optimizing the control of parameters in the entire process of precision electronic surface welding in the present invention. Figure 2 This is a schematic diagram of the data flow in the present invention. This embodiment discloses a method for optimizing and controlling parameters of the entire process of precision electronic surface welding, including:

[0052] S10: Acquire spatial distribution characteristics of the components to be welded and weld position characteristics, and adjust preset standard processing parameters based on the spatial distribution characteristics and weld position characteristics to obtain real-time processing parameters;

[0053] In this embodiment, the components to be welded refer to precision electronic components, such as microprocessors, integrated circuit (IC) chips, semiconductor devices, sensors, etc. The manufacture of precision electronic components requires the use of laser welding technology. The preset standard processing parameters include but are not limited to laser power, welding speed, focus position, welding angle, etc. The preset standard processing parameters are pre-set and stored in the database. This embodiment does not go into details about this.

[0054] Methods for obtaining spatial distribution characteristics include:

[0055] Acquire acoustic wave reflection data of the component to be welded, generate a thickness distribution map corresponding to the surface of the component to be welded based on the acoustic wave reflection data, acquire a real-time surface image of the component to be welded, generate a grayscale distribution map corresponding to the surface of the component to be welded based on the real-time surface image, and cascade-fuse the thickness distribution map and the grayscale distribution map to obtain spatial distribution characteristics.

[0056] It should be noted that the method for obtaining the acoustic wave reflection data is to transmit a beam of ultrasonic waves into the component to be welded through an ultrasonic probe (transducer). When the acoustic wave passes through the material, it will be reflected at the interface or defects inside the material. The transducer receives the reflected acoustic wave data. The thickness distribution map refers to an image generated based on the acoustic wave reflection data, which shows the changes in the surface thickness of the component to be welded at different positions. The thickness distribution map can be a two-dimensional or three-dimensional visualization chart. The grayscale distribution map refers to the distribution image of the surface defects of the component to be welded. Cascade fusion refers to the front-to-back splicing of two different feature vectors. In this embodiment, the thickness distribution map and the grayscale distribution map are respectively converted into two different feature vectors, and then spliced ​​to obtain the spatial distribution characteristics.

[0057] This embodiment cascades and fuses the thickness distribution map with the grayscale distribution map to more comprehensively describe the spatial distribution characteristics of the components to be welded. For example, the thickness distribution map reflects the geometric properties of the material, while the grayscale distribution map provides optical properties or defect information on the surface. The combination of the two can help optimize the parameter settings during the laser welding process and ensure welding quality.

[0058] Methods for obtaining weld position features include:

[0059] Based on the real-time surface image and edge detection algorithm, the first edge point and the second edge point on both sides of the weld are obtained, the first longitudinal distance between the first edge point and any point in the weld is calculated, and the second longitudinal distance between the second edge point and any point in the weld is calculated. The arbitrary point corresponding to the maximum value of the sum of the first longitudinal distance and the second longitudinal distance is taken as the center point of the weld, and the first edge point, the second edge point and the center point of the weld are taken as the weld position features.

[0060] It should be noted that if Figure 3 As shown, Figure 3 The weld WDS, the first edge point FEP, the second edge point SEP, the first longitudinal distance FLD, the second longitudinal distance SLD and the weld center point WCP are shown. It can be understood that the above-mentioned arbitrary point refers to any point on the inner wall of the weld WDS, and the arbitrary point must be located between the first edge point FEP and the second edge point SEP.

[0061] Among them, the edge detection algorithm can be a Canny edge detection algorithm or a Sobel edge detection algorithm. It can be understood that in the real-time surface image, the edge of the weld corresponds to the brightness difference between the weld area and the surrounding material area, which is usually where the gradient of the image changes greatly. The purpose of the edge detection algorithm is to identify these significant brightness change areas, that is, the edge of the weld.

[0062] This embodiment calculates the longitudinal distances from the first edge point and the second edge point to any point in the weld, and takes the point with the largest sum of the distances as the center point of the weld. This method can effectively determine the true center of the weld, which can avoid the deviation of the weld center point due to the irregularity of the weld edge or the influence of noise. Moreover, since it is based on the distance calculation of the edge points on both sides of the weld, the determination of the center point does not rely on a single pixel or local feature. This can improve the anti-interference ability of the weld edge irregularity, surface defects or noise, thereby enhancing the robustness and stability of the weld position detection, and calculating the sum of the distances of the edge points on both sides can dynamically adapt to changes in the weld shape, ensuring that the calculation of the center point is always accurate and adaptable to welds of different shapes.

[0063] Methods for adjusting preset standard processing parameters to obtain real-time processing parameters include:

[0064] The spatial distribution characteristics, weld position characteristics and preset standard processing parameters are input into the pre-built parameter adjustment model to obtain the real-time processing parameters output by the parameter adjustment model.

[0065] The construction methods of the parameter adjustment model include:

[0066] Acquire a first sample data set, wherein the first sample data set includes historical spatial distribution characteristics, historical weld position characteristics, preset standard processing parameters, and historical processing parameters;

[0067] Divide the sample data set into a sample training set and a sample test set, and build a regression network;

[0068] The historical spatial distribution characteristics, historical weld position characteristics, and preset standard processing parameters in the sample training set are used as the input data of the regression network, and the historical processing parameters in the sample training set are used as the output data of the regression network. The regression network is trained to obtain an initial regression network for predicting real-time processing parameters.

[0069] The initial regression network is tested using a sample test set, and the output is an initial regression network that satisfies the prediction error between the output data and the actual historical processing parameters in the sample test set is less than a preset error value as a parameter adjustment model. The initial regression network is preferably a deep neural network model.

[0070] It is understood that historical processing parameters are derived from experiments conducted by those skilled in the art and adjustments made to preset standard processing parameters. Exemplarily, the preset standard processing parameters include laser power, welding speed, and welding angle. In the thickness distribution diagram of the components to be welded, the thickness of some areas is significantly increased, while the thickness of other areas is thinner. When the material is thicker, the laser power needs to be increased to ensure that the laser has sufficient energy to penetrate and melt the material to form a strong weld. In the thicker areas, the laser incident angle needs to be adjusted to ensure that the laser beam acts more effectively on the weld area and reduce the laser reflection loss on the surface. If the grayscale distribution diagram shows surface defects such as cracks or pits, the welding speed needs to be reduced to ensure that the laser has more time to act around the defect and perform a more uniform repair. When the distance between the edge points of the weld, that is, the weld width, becomes narrower, reducing the laser power can avoid melting too much material. When the weld width increases, the laser power can be increased to ensure that all areas within the weld are melted. When the weld is no longer a straight line, the welding angle needs to be adjusted accordingly to maintain a vertical angle of incidence of the laser beam. This ensures that the laser energy is evenly distributed throughout each part of the weld.

[0071] In this embodiment, by sequentially acquiring the spatial distribution characteristics of the components to be welded and the weld position characteristics before welding, the spatial distribution characteristics include thickness and grayscale distribution, and the system can determine the optimal processing parameters before welding begins. The advantage of this is that the uncertainties and potential problems in the welding process have been resolved in the parameter adjustment stage, thereby reducing interruptions, failures or welding defects in the welding process and improving the success rate of welding. In addition, all factors affecting the welding quality are analyzed and corrected before welding, which can minimize the uncertainty in the welding process. The analysis of thickness distribution, grayscale distribution and weld position combined with model prediction can eliminate variables caused by material properties or morphological changes. Through early correction and optimization, the impact of uncertainties in the welding process on the results can be reduced.

[0072] S20: Processing the component to be welded according to the real-time processing parameters, obtaining the melt-through state characteristics during the processing, and inputting the melt-through state characteristics into a pre-built state classification model to obtain the melt-through state category;

[0073] Methods for obtaining melt penetration characteristics include:

[0074] The processing image is acquired, and the melt pool behavior characteristics, melt pool state characteristics and through-hole characteristics are extracted from the processing image. The melt pool behavior characteristics and the melt pool state characteristics are fused to generate the melt pool dynamic characteristics, and the melt pool dynamic characteristics and the through-hole characteristics are used as the melt-through state characteristics.

[0075] It can be understood that the processing image refers to the image collected in real time during the laser welding process by a visual sensor or a high-resolution camera. The processing image captures the formation and evolution of the molten pool during the welding process, as well as the visual expression of the penetration state. The molten pool refers to the liquid metal area formed during the laser welding process due to the high energy input of the laser, which causes the material in the local area to be heated to a molten state. The through hole refers to the small hole or through hole formed on the surface or inside the material due to the high energy of the laser beam penetrating the material during the laser welding process.

[0076] It should be added that the molten pool behavior characteristics refer to the dynamic changes of the molten pool during the welding process, including the formation, movement, expansion, contraction and other behavioral characteristics of the molten pool. The molten pool state characteristics refer to the static characteristics of the molten pool at a certain moment in the welding process, which describes the morphology and physical state of the molten pool, including the shape and edge clarity of the molten pool. The shape of the molten pool can be circular, elliptical, etc. The penetration hole characteristics refer to the characteristics of the small holes or penetration phenomena formed inside the components to be welded or on the surface of the components to be welded during the welding process. The penetration hole characteristics are an important indicator for judging the penetration state. An excessively large penetration hole means excessive penetration, and the absence of a penetration hole means incomplete penetration.

[0077] The fusion of the molten pool behavior characteristics and the molten pool state characteristics mentioned above refers to combining the dynamic change process of the molten pool during welding with its static characteristics to form a more comprehensive description of the molten pool. This fusion can be performed in a variety of ways. The main purpose is to combine the dynamic information of the time series with the spatially stable morphological information. Dynamic information refers to behavioral characteristics, and morphological information refers to state characteristics. For example, by capturing the dynamic changes of the molten pool at different time points during welding, that is, behavioral characteristics, combined with the static morphology of the molten pool at a certain moment or multiple moments, that is, state characteristics, a complete spatiotemporal description of the molten pool is formed. This can be achieved by processing spatial features through a convolutional neural network (CNN), and then combining it with a recurrent neural network (RNN) or a long short-term memory (LSTM) network to process the dynamic features of the time series. After the two are fused, the spatiotemporal change pattern of the molten pool can be captured.

[0078] The penetration state categories include incomplete penetration, moderate penetration, complete penetration, and excessive penetration. The construction method of the state classification model includes:

[0079] Acquire Q sets of training data, where Q is a positive integer greater than 1, and the training data includes historical penetration state characteristics and historical penetration state categories;

[0080] The historical penetration state characteristics and historical penetration state categories are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0081] The historical penetration state features in the training set are used as input data, and the historical penetration state categories in the training set are used as output data to train the classifier and obtain an initial classifier;

[0082] The initial classifier is tested using the test set, and the classifier whose classification accuracy meets the preset accuracy is output as the state classification model.

[0083] It should be noted that when the penetration state category is incomplete penetration, the molten pool is small, the dynamic changes are not obvious, and no penetration holes are formed. When the penetration state category is moderate penetration, the molten pool size and shape are moderate, the molten pool is dynamically stable, and the penetration holes are small and stable. When the penetration state category is complete penetration, the molten pool is large and stable, the penetration holes are large and continuous, and the welding strength reaches the optimal state. When the penetration state category is excessive penetration, the molten pool is too large and dynamically unstable, and the penetration hole size exceeds the normal value.

[0084] S30: acquiring welding defect data in real time during the processing, calculating a defect impact value based on the welding defect data and the penetration state category, and optimizing and adjusting real-time processing parameters based on the defect impact value.

[0085] In this embodiment, the welding defect data includes amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter and real-time molten pool area. The amplitude fluctuation value refers to the fluctuation value of the sound amplitude during the processing process. The amplitude fluctuation value and the current fluctuation value are obtained in the same time period. The system will simultaneously measure and record the amplitude fluctuation value and the current fluctuation value in the same fixed time period. The amplitude fluctuation value refers to the amplitude change of the sound, and the current fluctuation value refers to the change of the current. The above-mentioned real-time penetration hole diameter and real-time molten pool area are obtained through machine vision technology. By obtaining the processing process image, using image segmentation and visual measurement technology, the real-time penetration hole diameter and real-time molten pool area are obtained. This embodiment does not go into details about this.

[0086] Methods for calculating defect impact values ​​include:

[0087] DIV= ;

[0088] Where DIV is the defect impact value, is the amplitude fluctuation value, is the current fluctuation value, is the real-time penetration hole diameter, is the standard penetration hole diameter, is the real-time molten pool area, is the standard molten pool area, For the melt-through state category, is the inverse cotangent function, is the inverse tangent function, is the hyperbolic cosine function, 、 are logarithmic functions with bases 2 and 3 respectively. 、 Both are weight factors related to the penetration state category.

[0089] In this embodiment, taking the amplitude fluctuation value and the real-time penetration hole diameter as an example, the larger the amplitude fluctuation value, the more unstable the sound during the processing, reflecting the uneven input of the arc or laser, resulting in poor welding quality. A real-time penetration hole diameter that is too large or too small will affect the welding quality. An overly large penetration hole usually indicates excessive penetration, while a too small or no penetration hole indicates incomplete penetration. Therefore, it can be seen from the above content that the larger the defect impact value, the worse the welding quality during the processing.

[0090] It should be added that 、 The sum is 1, 、 are all related to the melting state category, which means that if the melting state category is different, 、 For example, when the penetration state category is incomplete penetration, the molten pool is small, the dynamic change is not obvious, and no penetration hole is formed. In this case, the system will pay more attention to processing the impact related to the incomplete penetration state, so Much greater than , is 0.8, When the penetration state is 0.2, the molten pool is large and stable, and the penetration hole is large and continuous. At this time, the welding quality reaches the best state. The system will focus more on ensuring this ideal welding effect. To be greater than , is 0.3, is 0.7.

[0091] Methods for optimizing and adjusting real-time processing parameters based on defect impact values ​​include:

[0092] S301: Using M real-time processing parameters as an initial population and the defect impact value as a fitness function, the fitness value of each individual in the initial population is calculated;

[0093] The M real-time processing parameters in the welding process are regarded as individuals in the initial population, such as laser power, welding speed, and focus position. The defect influence value is used to quantify the impact of welding defects on welding quality during the welding process. As a fitness function, the fitness value of each individual measures the welding quality performance corresponding to its processing parameters.

[0094] S302: Using the tournament selection strategy, randomly select 4 individuals from the population, compare their fitness values, and select the individual with the smallest fitness value as the parent individual;

[0095] Among them, the tournament selection strategy is a selection method based on random competition. Each time, 4 individuals are randomly selected from the population. By comparing the fitness values, the individual with the smallest fitness value, that is, the individual with the best welding quality, is selected as the "parent individual".

[0096] S303: Repeat the above selection process until N parent individuals are selected, and then go to S304;

[0097] Among them, by repeating the tournament selection strategy multiple times, the system can gradually select N parent individuals to ensure that there are enough high-quality individuals in the population to participate in the next crossover and mutation operations.

[0098] The mutation operation increases the diversity of the population by randomly selecting two nodes in the child individual and exchanging their positions, thus preventing the algorithm from falling into the local optimal solution. Mutation is a small-scale perturbation aimed at exploring more solution spaces.

[0099] S304: Select an intersection point on the path of each pair of parent individuals, use the single-point crossover method to operate on all parent individuals, exchange the parts of the parent individuals after the intersection point, and generate two new child individuals;

[0100] S305: Randomly select two nodes for each child individual and exchange the positions of the two nodes, forming a new population from all the child individuals and replacing the original population, ensuring that the size of the new population remains N;

[0101] S306: The convergence threshold of the preset fitness value is CT. If the fitness value change of the optimal individual in the population of consecutive K generations is less than CT, the genetic algorithm terminates.

[0102] S307: When the algorithm terminates, the sub-individual with the smallest fitness value in the population is the optimal processing parameter.

[0103] When the algorithm terminates, the individual with the smallest fitness value in the population is considered to be the optimal solution, which represents the best combination of processing parameters when welding defects are minimized.

[0104] In this embodiment, the penetration state is detected in real time during the welding process through step S20, so that potential problems in welding can be identified in advance, and the welding parameters can be optimized and adjusted in time in combination with the defect data obtained during the welding process in step S30. This not only avoids the accumulation of welding defects, but also can correct the welding parameters in real time, thereby ensuring the continuous stability of welding quality. This "detect first and then adjust" sequence can effectively reduce the uncertainty in the welding process and improve the overall welding quality. The penetration state classification result of step S20 is input as feedback to step S30, allowing step S30 to optimize the processing parameters more specifically. This precise feedback control can ensure that the parameter adjustment in each welding process is based on the actual detected penetration state and defect data, rather than simply based on a preset model or standard. This makes the welding process more intelligent and personalized, and helps to improve the consistency and quality of welding.

[0105] In this embodiment, the preset standard processing parameters are first adjusted based on the spatial distribution characteristics and the weld position characteristics to obtain real-time processing parameters, and then the components to be welded are processed according to the real-time processing parameters. During the processing, the penetration state characteristics are obtained, and the penetration state characteristics are input into a pre-built state classification model to obtain the penetration state category. Finally, the defect influence value is calculated according to the welding defect data and the penetration state category, and the real-time processing parameters are optimized and adjusted according to the defect influence value. This not only takes into account the defects of the components to be welded before processing and the defects generated during the processing, but also takes into account that different penetration states will also affect the adjustment of welding parameters, thereby achieving more accurate welding parameter adjustment. This comprehensive optimization process can effectively reduce the occurrence of welding defects and improve the stability and consistency of welding quality.

[0106] Example 2

[0107] See also Figure 4 As shown, based on the same inventive concept, this embodiment discloses a parameter optimization control system for the entire process of precision electronic surface welding. For details not provided in this embodiment, please refer to the description of the relevant parts in Example 1. The system includes:

[0108] Pre-adjustment module: used to obtain the spatial distribution characteristics of the components to be welded and the weld position characteristics, and adjust the preset standard processing parameters based on the spatial distribution characteristics and weld position characteristics to obtain real-time processing parameters;

[0109] In this embodiment, the components to be welded refer to precision electronic components, such as microprocessors, integrated circuit (IC) chips, semiconductor devices, sensors, etc. The manufacture of precision electronic components requires the use of laser welding technology. The preset standard processing parameters include but are not limited to laser power, welding speed, focus position, welding angle, etc. The preset standard processing parameters are pre-set and stored in the database. This embodiment does not go into details about this.

[0110] Methods for obtaining spatial distribution characteristics include:

[0111] Acquire acoustic wave reflection data of the component to be welded, generate a thickness distribution map corresponding to the surface of the component to be welded based on the acoustic wave reflection data, acquire a real-time surface image of the component to be welded, generate a grayscale distribution map corresponding to the surface of the component to be welded based on the real-time surface image, and cascade-fuse the thickness distribution map and the grayscale distribution map to obtain spatial distribution characteristics.

[0112] Methods for obtaining weld position features include:

[0113] Based on the real-time surface image and edge detection algorithm, the first edge point and the second edge point on both sides of the weld are obtained, the first longitudinal distance between the first edge point and any point in the weld is calculated, and the second longitudinal distance between the second edge point and any point in the weld is calculated. The arbitrary point corresponding to the maximum value of the sum of the first longitudinal distance and the second longitudinal distance is taken as the center point of the weld, and the first edge point, the second edge point and the center point of the weld are taken as the weld position features.

[0114] Methods for adjusting preset standard processing parameters to obtain real-time processing parameters include:

[0115] The spatial distribution characteristics, weld position characteristics and preset standard processing parameters are input into the pre-built parameter adjustment model to obtain the real-time processing parameters output by the parameter adjustment model.

[0116] The construction methods of the parameter adjustment model include:

[0117] Acquire a first sample data set, wherein the first sample data set includes historical spatial distribution characteristics, historical weld position characteristics, preset standard processing parameters, and historical processing parameters;

[0118] Divide the sample data set into a sample training set and a sample test set, and build a regression network;

[0119] The historical spatial distribution characteristics, historical weld position characteristics, and preset standard processing parameters in the sample training set are used as the input data of the regression network, and the historical processing parameters in the sample training set are used as the output data of the regression network. The regression network is trained to obtain an initial regression network for predicting real-time processing parameters.

[0120] The initial regression network is tested using a sample test set, and the output of the initial regression network that satisfies a preset error value is used as a parameter adjustment model. The initial regression network is preferably a deep neural network model.

[0121] Classification module: used to process the components to be welded according to real-time processing parameters, obtain the penetration state characteristics during the processing, input the penetration state characteristics into the pre-built state classification model, and obtain the penetration state category;

[0122] Methods for obtaining melt penetration characteristics include:

[0123] The processing image is acquired, and the melt pool behavior characteristics, melt pool state characteristics and through-hole characteristics are extracted from the processing image. The melt pool behavior characteristics and the melt pool state characteristics are fused to generate the melt pool dynamic characteristics, and the melt pool dynamic characteristics and the through-hole characteristics are used as the melt-through state characteristics.

[0124] It can be understood that the processing image refers to the image collected in real time during the laser welding process by a visual sensor or a high-resolution camera. The processing image captures the formation and evolution of the molten pool during the welding process, as well as the visual expression of the penetration state. The molten pool refers to the liquid metal area formed during the laser welding process due to the high energy input of the laser, which causes the material in the local area to be heated to a molten state. The through hole refers to the small hole or through hole formed on the surface or inside the material due to the high energy of the laser beam penetrating the material during the laser welding process.

[0125] The penetration state categories include incomplete penetration, moderate penetration, complete penetration, and excessive penetration. The construction method of the state classification model includes:

[0126] Acquire Q sets of training data, where Q is a positive integer greater than 1, and the training data includes historical penetration state characteristics and historical penetration state categories;

[0127] The historical penetration state characteristics and historical penetration state categories are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed;

[0128] The historical penetration state features in the training set are used as input data, and the historical penetration state categories in the training set are used as output data to train the classifier and obtain an initial classifier;

[0129] The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the state classification model.

[0130] Optimization module: used to obtain welding defect data in real time during the processing, calculate the defect impact value based on the welding defect data and penetration state category, and optimize and adjust the real-time processing parameters based on the defect impact value.

[0131] In this embodiment, the welding defect data includes amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter and real-time molten pool area. The amplitude fluctuation value refers to the fluctuation value of the sound amplitude during the processing process. The amplitude fluctuation value and the current fluctuation value are obtained in the same time period. The system will simultaneously measure and record the amplitude fluctuation value and the current fluctuation value in the same fixed time period. The above-mentioned real-time penetration hole diameter and real-time molten pool area are obtained through machine vision technology. By obtaining the processing process image, using image segmentation and visual measurement technology, the real-time penetration hole diameter and real-time molten pool area are obtained. This embodiment does not go into details about this.

[0132] Methods for calculating defect impact values ​​include:

[0133] DIV= ;

[0134] Where DIV is the defect impact value, is the amplitude fluctuation value, is the current fluctuation value, is the real-time penetration hole diameter, is the standard penetration hole diameter, is the real-time molten pool area, is the standard molten pool area, For the melt-through state category, is the inverse cotangent function, is the inverse tangent function, is the hyperbolic cosine function, is a logarithmic function with base 2, 、 Both are weight factors related to the penetration state category.

[0135] In this embodiment, taking the amplitude fluctuation value and the real-time penetration hole diameter as an example, the larger the amplitude fluctuation value, the more unstable the sound during the processing, reflecting the uneven input of the arc or laser, resulting in poor welding quality. A real-time penetration hole diameter that is too large or too small will affect the welding quality. An overly large penetration hole usually indicates excessive penetration, while a too small or no penetration hole indicates incomplete penetration. Therefore, it can be seen from the above content that the larger the defect impact value, the worse the welding quality during the processing.

[0136] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0137] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for optimizing and controlling parameters of the entire process of precision electronic surface welding, characterized in that: include: Obtain the spatial distribution characteristics of the components to be welded and the weld position characteristics, and adjust the preset standard processing parameters based on the spatial distribution characteristics and weld position characteristics to obtain real-time processing parameters; the preset standard processing parameters include laser power, welding speed and welding angle; Processing the components to be welded according to real-time processing parameters, obtaining penetration state characteristics during the processing, and inputting the penetration state characteristics into a pre-built state classification model to obtain the penetration state category; Acquire welding defect data in real time during the processing, calculate the defect impact value based on the welding defect data and penetration status category, and optimize and adjust the real-time processing parameters based on the defect impact value; The method for acquiring the spatial distribution characteristics includes: Acquire acoustic wave reflection data of the component to be welded, generate a thickness distribution map corresponding to the surface of the component to be welded based on the acoustic wave reflection data, acquire a real-time surface image of the component to be welded, generate a grayscale distribution map corresponding to the surface of the component to be welded based on the real-time surface image, and cascade-fuse the thickness distribution map and the grayscale distribution map to obtain spatial distribution characteristics; the thickness distribution map refers to the image generated based on the acoustic wave reflection data; the grayscale distribution map refers to the distribution image of surface defects of the component to be welded; if the grayscale distribution map shows surface defects, reduce the welding speed; when the distance between the edge points of the weld, i.e., the weld width, becomes narrower, reduce the laser power; and when the weld width increases, increase the laser power; if the weld is no longer a straight line, adjust the welding angle accordingly; The method for obtaining the weld position feature includes: Based on the real-time surface image and the edge detection algorithm, a first edge point and a second edge point on both sides of the weld are obtained, a first longitudinal distance between the first edge point and any point in the weld is calculated, and a second longitudinal distance between the second edge point and any point in the weld is calculated, an arbitrary point corresponding to the maximum value of the sum of the first longitudinal distance and the second longitudinal distance is used as the weld center point, and the first edge point, the second edge point, and the weld center point are used as weld position features; The welding defect data includes amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter and real-time molten pool area. The defect impact value is determined based on the amplitude fluctuation value, current fluctuation value, real-time penetration hole diameter, real-time molten pool area and penetration state category; the amplitude fluctuation value refers to the fluctuation value of the sound amplitude during the processing process.

2. The method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to claim 1 is characterized in that: The method of adjusting the preset standard processing parameters to obtain real-time processing parameters includes: The spatial distribution characteristics, weld position characteristics and preset standard processing parameters are input into the pre-built parameter adjustment model to obtain the real-time processing parameters output by the parameter adjustment model.

3. The method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to claim 2 is characterized in that: The construction method of the parameter adjustment model includes: Acquire a first sample data set, wherein the first sample data set includes historical spatial distribution characteristics, historical weld position characteristics, preset standard processing parameters, and historical processing parameters; Divide the sample data set into a sample training set and a sample test set, and build a regression network; The historical spatial distribution characteristics, historical weld position characteristics, and preset standard processing parameters in the sample training set are used as the input data of the regression network, and the historical processing parameters in the sample training set are used as the output data of the regression network. The regression network is trained to obtain an initial regression network for predicting real-time processing parameters. The initial regression network is tested using a sample test set, and the output is an initial regression network that satisfies the requirement that the prediction error between the output data and the actual historical processing parameters in the sample test set is less than a preset error value as a parameter adjustment model.

4. The method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to claim 1 is characterized in that: The method for obtaining the melt penetration state characteristics includes: The processing image is acquired, and the melt pool behavior characteristics, melt pool state characteristics and through-hole characteristics are extracted from the processing image. The melt pool behavior characteristics and the melt pool state characteristics are fused to generate the melt pool dynamic characteristics, and the melt pool dynamic characteristics and the through-hole characteristics are used as the melt-through state characteristics.

5. The method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to claim 4 is characterized in that: The penetration state categories include incomplete penetration, moderate penetration, complete penetration and excessive penetration, and the method for constructing the state classification model includes: Acquire Q sets of training data, where Q is a positive integer greater than 1, and the training data includes historical penetration state characteristics and historical penetration state categories; The historical penetration state characteristics and historical penetration state categories are used as sample sets, the sample sets are divided into training sets and test sets, and a classifier is constructed; The historical penetration state features in the training set are used as input data, and the historical penetration state categories in the training set are used as output data to train the classifier and obtain an initial classifier; The initial classifier is tested using the test set, and the classifier that meets the preset accuracy is output as the state classification model.

6. The method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to claim 1 is characterized in that: The method for optimizing and adjusting real-time processing parameters according to the defect impact value includes: S301: Using M real-time processing parameters as an initial population and the defect impact value as a fitness function, the fitness value of each individual in the initial population is calculated; S302: Using the tournament selection strategy, randomly select 4 individuals from the population, compare their fitness values, and select the individual with the smallest fitness value as the parent individual; S303: Repeat S301-S302 until N parent individuals are selected, then go to S304; S304: Select an intersection point on the path of each pair of parent individuals, use the single-point crossover method to operate on all parent individuals, exchange the parts of the parent individuals after the intersection point, and generate two new child individuals; S305: Randomly select two nodes for each child individual and exchange the positions of the two nodes, forming a new population with all the child individuals and replacing the original population, ensuring that the size of the new population remains N; S306: The convergence threshold of the preset fitness value is CT. If the fitness value change of the optimal individual in the population of consecutive K generations is less than CT, the genetic algorithm terminates. S307: When the genetic algorithm terminates, the sub-individual with the smallest fitness value in the population is the optimal processing parameter.

7. A system for optimizing and controlling parameters of the entire process of precision electronic surface welding, which is used to implement the method for optimizing and controlling parameters of the entire process of precision electronic surface welding according to any one of claims 1 to 6, characterized in that: include: A pre-adjustment module is used to obtain the spatial distribution characteristics of the components to be welded and the weld position characteristics, and adjust the preset standard processing parameters based on the spatial distribution characteristics and the weld position characteristics to obtain real-time processing parameters; A classification module is used to process the components to be welded according to real-time processing parameters, obtain the characteristics of the penetration state during the processing, and input the penetration state characteristics into a pre-built state classification model to obtain the penetration state category; The optimization module is used to obtain welding defect data in real time during the processing, calculate the defect impact value based on the welding defect data and the penetration state category, and optimize and adjust the real-time processing parameters based on the defect impact value.

Images