A red copper IGBT heat dissipation substrate cold forging forming process

Abstract

The present application relates to the field of power electronic device heat dissipation, specifically to a red copper IGBT heat dissipation substrate cold forging forming process, the process steps include: the original plate material is obtained by adaptive cutting path optimization to obtain the optimal cutting path, and the blank holding time is adjusted in combination with the BP annealing hardness prediction model; the thickness of the lubricating layer is controlled by PID after annealing, and the cold forging parameters are optimized based on the Box-Behnken experiment and the response surface model to realize one-time forming; after cold forging, the flash and positioning needle residues are positioned by machine vision, the drilling deviation is cut by greedy algorithm and corrected by multiple linear regression, and ultrasonic cleaning is performed; after cleaning, the sand blasting parameters are optimized by the borer beetle algorithm, the burr is detected by the horizontal set algorithm, and the pre-bending angle is calculated according to the thermal expansion difference; finally, the plating thickness is controlled to be greater than or equal to 7 microns by the nickel plating differential equation model, and a complete heat dissipation substrate is obtained. The present application improves material utilization and forming precision, and is suitable for mass production.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of heat dissipation of power electronic devices, and particularly relates to a red copper IGBT heat dissipation substrate cold forging forming process. BACKGROUND

[0002] In the field of power electronic devices, red copper is the core material of the IGBT heat dissipation substrate due to high thermal conductivity, but there are many technical pain points in the traditional cold forging forming process. The unreasonable cutting path of the original plate material leads to low material utilization rate, the annealing hardness depends on experience control and is prone to insufficient softening or excessive softening, the lubricating layer thickness is unstable in the cold forging process, and the pressure and pressure holding time parameters are not well matched, which often causes problems such as poor substrate forming precision and convex surface tolerance exceeding the standard;

[0003] The flash and positioning needle residues after cold forging are removed by manual or fixed path cutting, which is low in efficiency and high in residue risk; the drilling deviation lacks real-time compensation, affecting the assembly precision; the burr residue rate is high due to insufficient optimization of the sand blasting parameters; the warping deformation caused by the difference in the thermal expansion coefficient during welding is difficult to accurately compensate; and the plating thickness control is unstable, affecting the corrosion resistance of the substrate. Therefore, a red copper IGBT heat dissipation substrate cold forging forming process is needed to solve the above problems. SUMMARY

[0004] In order to solve the technical problems in the background art, the application provides a red copper IGBT heat dissipation substrate cold forging forming process.

[0005] The purpose of the application can be achieved by the following technical solutions.

[0006] The application provides a red copper IGBT heat dissipation substrate cold forging forming process, and the specific steps are as follows:

[0007] Step 1: Obtain the original plate material for processing the heat dissipation substrate, add an adaptive cutting path optimization algorithm before cutting the original plate material to obtain the optimal cutting path, and then predict the hardness of the red copper after annealing by using a preset BP annealing hardness prediction model to adjust the holding time of the blank;

[0008] The actual size of the original plate material is obtained by using a laser instrument, the thickness deviation is obtained by subtracting the actual thickness from the target thickness, the thickness deviation needs to be less than or equal to a preset thickness deviation threshold, and then the defect points of the original plate material are obtained by using an ultrasonic flaw detector, the defect points are integrated into the plate defect area, the material utilization rate TH and the cutting efficiency TY are obtained, and the target optimization function F is calculated by weighted summation, and the calculation logic of the material utilization rate is: Where S 胚料 is the total effective area of the qualified blank, S 板材 is the total area of the original plate, and the calculation logic of the cutting efficiency utilization rate is: Where LT 路径LT represents the actual total cutting path length. 路径，max This represents the theoretical maximum path length. It should be noted that the original board is the unprocessed initial material, and the blank is a semi-finished product cut from the original board.

[0009] The original sheet metal is uniformly divided into initial candidate blank positions according to a preset grid size. A random perturbation is applied to each candidate position, and blanks overlapping with defect areas are removed, forming an initial population. The fitness of each individual in the initial population is calculated, and the calculation logic is as follows: Where valid(BG) i Let be the constraint satisfaction function for the i-th blank. If the blank satisfies the preset size, thickness deviation, and non-overlap with defect areas, then valid(BG) is valid. i The value is 1 if any constraint is not met, and -1 if any constraint is not met. I is the total number of blanks. Several individuals are randomly selected from the population. Each individual is a set of blank coordinates. The individual with the highest fitness is selected to enter the next generation. The operation is repeated until M parent individuals are generated. Two parent individuals are randomly selected. The path is divided into the first half and the second half according to the cutting order. The second half of the path is swapped to generate offspring individuals. This process is repeated. When the number of iterations reaches the preset value, the optimal cutting path parameters are output. The optimal cutting path parameters include the number of blanks, coordinates, and cutting order.

[0010] The average temperature during the holding process is obtained through thermocouples in the box furnace and marked as the annealing temperature. When the furnace temperature reaches the set temperature, calculations are triggered, and the effective holding time of the billet is obtained. The initial hardness of the copper billet is collected using a Vickers hardness tester. The annealing temperature, effective holding time, and initial hardness are input into a preset BP annealing hardness prediction model. The predicted annealing hardness value is obtained through the input layer, implicit layer, and output layer. If the predicted annealing hardness value is greater than the first preset hardness value, it indicates insufficient softening. The preset correction temperature and preset correction time are embedded into the original temperature and original time to obtain the additional calibration temperature and additional calibration time. If the predicted annealing hardness value is less than the second preset hardness value... If the value is too low, it indicates over-softening. Subtract the preset correction temperature and preset correction time from the original temperature and original time to obtain the subtracted calibration temperature and subtracted calibration time. It should be noted that the first preset hardness value is greater than the second preset hardness value. The establishment of the preset BP annealing hardness prediction model requires the collection of multiple sets of historical data of copper annealing process. Each set of data includes input parameters and output parameters. Randomly initialize the weights from the input layer to the hidden layer, the hidden layer to the output layer, and the biases of each layer. Divide the dataset into a training set and a validation set in a 7:3 ratio. Stop training when the validation set loss does not decrease for several consecutive iterations. Calculate the weighted output of the input parameters through the network layers to obtain the predicted annealing hardness value.

[0011] Step 2: Lubricate the surface of the annealed billet. During the lubrication process, the thickness of the lubrication layer is controlled within the target range using a PID control algorithm. Then, the cold forging parameters of the cold forging press are optimized using a parameter optimization algorithm to obtain the optimal cold forging pressure, optimal holding time, and optimal die temperature.

[0012] The thickness of the lubricating layer is acquired in real time using a laser thickness gauge. The target thickness is obtained and the difference between this and the actual lubricating layer thickness is calculated to obtain the lubricating thickness deviation rc(t). The initial nozzle pressure of the coating equipment is acquired, and the nozzle pressure adjustment amount gc(t) is calculated in real time using a PID formula. The calculation logic is as follows: K p K is a proportionality coefficient that reflects the immediate effect of the deviation. d The differential coefficient is the rate of change of the deviation, which suppresses system oscillations and enhances stability. s The sampling period is rc(t-1), where rc(t-1) represents the lubrication thickness deviation at the previous moment. The nozzle pressure adjustment is embedded into the initial nozzle pressure to obtain the complete nozzle pressure value. This complete nozzle pressure value is then sent to the coating equipment, and lubrication is applied to the surface of the copper billet. Using the Box-Behnken experimental design method, experiments are conducted on several combinations of different cold forging pressures C, holding times FR, and die temperatures R. Each experiment is repeated multiple times, and the quality results of the copper billet after each experiment are collected. The quality results include convexity tolerance and hardness. Their average values ​​are calculated as the convexity response value GT and the hardness response value H. The mapping relationship between the parameters and the quality target is fitted using the least squares method to construct the response surface model δ. * Its construction logic is as follows:

[0013] δ * =β0+β1C+β2FR+β3R+β4C+β5FR+β6R+β7C·FR+β8C·R+β9FR·R, where β0 to β9 are regression coefficients;

[0014] Constraints are set for each parameter. The Lagrange multiplier method combined with grid search is used to select parameter combinations that satisfy the constraints from the response surface model and output the baseline optimal solution. The baseline optimal solution includes the optimal cold forging pressure, the optimal holding time, and the optimal die temperature. The baseline optimal solution is sent to the cold forging press to perform cold forging of the copper billet in one step to obtain the initial blank of the heat dissipation substrate.

[0015] The convex forming height of the initial blank of the heat dissipation substrate is collected in real time by a laser displacement sensor, and the actual pressure C of the press is recorded simultaneously. 实测 and holding time FR 实测 The hardness quality of the initial blank of the heat dissipation substrate is detected by a pre-trained cold deformation grain refinement model. The calculation logic is as follows: H * =H 标准 +k1·(C 实测-3150)+k2·(FR 实测 -FR 标准 ), where H 标准 The standard hardness value is from the process database, k1 is the pressure correction factor, 3150 is the standard pressure value, k2 is the time correction factor, and FR is the standard hardness value. 标准 The standard holding time is set. If the hardness quality is less than the preset hardness threshold, the holding time will be extended or the pressure will be increased.

[0016] Step 3: Remove flash and positioning pin residue from the cold-forged heat dissipation substrate blank. Obtain the contour coordinates of the flash area based on the positioning model, then generate an adaptive path to cut the flash and positioning pins using a greedy algorithm. Output the predicted hole diameter deviation using a multiple linear regression compensation model and adjust the drilling diameter command in real time. Finally, remove residual debris and oil from the drilling holes using ultrasonic cleaning. Acquire images of the heat dissipation substrate blank using an industrial camera, convert the images to grayscale images IU(x,y), expand the grayscale range using histogram equalization to improve the grayscale difference between the flash and the substrate, and set a low threshold T. low and high threshold T high The edge contour E(x,y) is extracted using the Canny operator, and its detection logic is: E(x,y)=Canny(IU(x,y),T low ,T high ), where E(x,y)=1 represents edge pixels, E(x,y)=0 represents non-edge pixels, and edge pixels are pixels in the edge contour. The theoretical edge coordinates of the substrate are loaded, and the theoretical edge coordinates are set as the baseline L0. The edge pixels are traversed, and the continuous contours outside L0 are marked as flash areas. The contour coordinate set of the flash areas is obtained. The actual width of the flash is obtained by machine vision detection. The positioning pin residue is a cylindrical protrusion on the edge of the substrate. Morphological feature matching is used for recognition. Specifically, the grayscale image is binarized to obtain the foreground target, and the circularity CQ of the target area is calculated. The calculation logic is as follows: Where SC is the area and LR is the perimeter. When the roundness is greater than or equal to the preset roundness threshold, it is determined to be a positioning pin residue, and the residue detection coordinates are output. Based on the improved greedy algorithm, the adaptive path is used to cut the burrs and positioning pins. The minimum cutting length and the lowest residue risk are used as objectives to construct an optimization model. The optimization logic is as follows:

[0017] Where LH is the objective function value, x g and y g Let x be the coordinate of the g-th node on the cutting path. g Let y be the horizontal coordinate. g Let G be the vertical coordinate, and let we be the total number of nodes on the cutting path. g2 d represents the actual width of the flash at the g2th detection point. g2Let ω be the cutting depth at the g2th detection point, ω be the penalty coefficient, and G2 be the total number of detection points.

[0018] The first pixel of the edge contour is set as the starting point of the flash, and the last pixel of the edge contour is set as the ending point. An adaptation path is generated according to the path sorting rules. The sorting rules are: priority 1 for positioning pin residual nodes, and priority 2 for connecting the starting point to the ending point according to the natural order of the flash contour. The coordinates of the adaptation path are converted into motion commands of the cutting platform to execute the cutting. After the cutting is completed, a second image is acquired to detect the residual amount. If the residual amount is greater than the preset residual threshold, a supplementary cutting path is generated, and the flash parameters, path data, and residual detection results are stored in the process database.

[0019] Machining drilling refers to the process of machining a substrate after cold forging and trimming using CNC machining equipment. This involves machining assembly holes, positioning holes, or heat dissipation through-holes at predetermined locations on the substrate, ensuring the hole position accuracy meets design requirements. The current drill speed and feed rate are obtained, and the hardness H at the drilling location on the initial substrate blank is measured. * The drill bit rotation speed BE and feed rate VX are input to the preset borehole compensation model, which outputs the predicted borehole diameter deviation value ΔPE. The calculation logic is as follows:

[0020] in to By fitting historical data, the target borehole deviation value is obtained and the difference is calculated with the predicted borehole deviation value to obtain the borehole adjustment command, which is then sent to the CNC machining equipment terminal. Ultrasonic cleaning is used to remove residual debris and oil from the borehole.

[0021] Step 4: The cleaned heat dissipation substrate blank is subjected to sandblasting. The sandblasting parameters are optimized using the beetle whisker search algorithm. After sandblasting, the burr detection is performed using the level set evolution algorithm. If the index exceeds the standard, a second detection is performed and sandblasting is repeated. If it passes, the welding allowance is designed for pre-bending. The required pre-bending angle is obtained by simulating the welding temperature field using a pre-bending model. 120-mesh sandblasting deburring is a surface treatment process that uses 120-mesh abrasive. Compressed air is used to spray fine-grained abrasive at high speed onto the workpiece surface, and then the impact force removes micro-burrs, forming roughness and enhancing the adhesion of the coating. Roughness RA is collected in real time using a laser confocal microscope, oxide layer residue rate KL is monitored using an eddy current thickness gauge, and hardness increase ΔH is detected offline using a microhardness tester. The roughness, oxide layer residue rate, and hardness increase are weighted and summed to obtain the beetle fitness function FC.

[0022] Extract the current sandblasting parameter vector XQ, including jet pressure, jet angle, and sand particle velocity. Randomly generate a search direction vector. Extract the initial sensing distance dw of the longhorn beetle's two whiskers. The initial sensing distance is the information acquisition interval of the simulated longhorn beetle's two whiskers. Based on the initial sensing distance, obtain the position XQ of the left whisker. Land right mustache position XQ R The calculation logic for their respective positions is as follows: in The direction vector is used; the fitness value of the longhorn beetle corresponding to the position of the two whiskers is calculated, the difference between the fitness values ​​of the two whiskers is compared, and the current parameter vector is updated. If the fitness of the left whisker is better, it moves in the positive direction along the direction vector; if the fitness of the right whisker is better, it moves in the negative direction. If the fluctuation range of the longhorn beetle fitness value is less than or equal to the set fluctuation threshold during continuous iteration, the process terminates. The optimized sandblasting parameters are output and sent to the sandblasting equipment terminal.

[0023] A level set function ε is defined. If ε is less than 0, the region is considered a burr area; otherwise, it is considered a background area. If ε equals 0, the region is considered a contour line. The initial contour is a rectangular bounding box representing the potential burr area. An energy function is then constructed to converge the contour towards the burr edge. The calculation logic is as follows: Where LS(ε) is the data term that drives the contour to move towards the burr edge, and DA(ε) is the area term that controls the contour's shrinkage and expansion trends. and The preset fixed weights are 0.42 and 0.58, respectively. The level set function is iteratively updated using gradient descent based on the energy function to converge the contour. The converged contour is then subjected to morphological erosion and dilation to remove noise regions without burrs. The closed regions of the contour are extracted and marked as candidate burr regions. The total number of pixels in the candidate burr regions is obtained and mapped to the actual area. The maximum distance along the normal direction of the substrate surface is then obtained to obtain the maximum height. If the actual area and the maximum height are less than the corresponding preset area threshold and height threshold, the result is qualified; otherwise, it is unqualified. The burr detection result is output, including the position, size, and whether it exceeds the standard. If it does not exceed the standard, the sandblasting is repeated.

[0024] The pre-bending forming design welding allowance refers to the small deformation that is pre-set in the pre-bending process before nickel plating of the copper IGBT heat sink substrate, based on the difference in thermal expansion coefficient between the welding material and copper, in order to compensate for the warping deformation caused by thermal stress during the welding process, so that the flatness of the substrate after welding can meet the standard.

[0025] The thermal expansion coefficients of the substrate and solder are obtained, and the difference between them is calculated to obtain the thermal expansion coefficient difference ΔTCP. Then, the peak welding temperature and room temperature are obtained and the difference is calculated to obtain the welding temperature difference ΔYA. Based on the thermal expansion coefficient difference and the welding temperature difference, the warpage W after welding is calculated. weld Its calculation logic is as follows: Where gs is the shape factor, L is the substrate length, and h is the substrate thickness; because welding causes deformation, pre-bending requires reverse deformation, and its calculation logic is: W pre =-gc·W weld Where gc is the compensation allowance coefficient, with a value of 1.05, it converts the pre-bending amount into a pre-bending angle that the equipment can execute. The calculation logic is as follows: Send the pre-bending angle to the corresponding device for adjustment.

[0026] Step 5: By establishing a nickel plating differential equation model, the predicted plating thickness is output, and the complete copper heat dissipation substrate is obtained by controlling the plating thickness.

[0027] The electroplating solution temperature TD, current density j, metal ion concentration cf, and tank stirring rate vx are collected in real time to establish a nickel plating differential equation model. The logic for establishing the model is as follows: in Here, denoted as Caputo fractional derivative, l as reaction rate constant, EC as activation energy, R as gas constant, and e, nd, and r as kinetic exponents, obtained through experimental calibration, reflecting the influence weights of current, concentration, and stirring rate, respectively.

[0028] Test calibration process: Using offline test data, such as orthogonal experiments with 5 sets of current, 3 sets of concentration, and 3 sets of temperature, the parameters are solved using the least squares method. The solution logic is as follows: Where η = [l,e,nd,r], HD(t) z ) represents the actual thickness of the coating obtained by measurement, HS(η,t) z ) represents the numerical solution of the fractional-order model, Z represents the total number of experimental samples, and b represents the experimental sample number;

[0029] The electroplating solution temperature, current density, metal ion concentration, and tank stirring rate are input into the nickel plating differential equation model. Through iterative solution, the predicted plating thickness is output. If the predicted plating thickness is less than 7 μm, an "overly thin" warning is issued, generating a remedial command that is sent to the electroplating equipment terminal to perform a second electroplating on the entire board. Otherwise, manufacturing ends, resulting in a complete copper heat dissipation substrate. Compared with existing technologies, the advantages of this invention are:

[0030] During the material pretreatment stage, the adaptive cutting path optimization algorithm avoids the defect area of ​​the sheet metal, and the heat preservation parameters are dynamically adjusted in combination with the BP annealing hardness prediction model to improve the material utilization rate and ensure the uniformity of the billet hardness, so as not to affect the subsequent cold forging quality due to improper softening.

[0031] During cold forging, the PID control algorithm enables precise control of the lubricant layer thickness. Combined with the parameter optimization model based on experimental design, it ensures the accuracy of the substrate forming in one step, reduces forming defects caused by poor parameter matching, and reduces the need for secondary processing.

[0032] After cold forging, machine vision positioning and intelligent cutting path planning are used to remove burrs and positioning pin residues. Combined with a real-time drilling deviation compensation mechanism, the accuracy of the substrate edge and hole position is improved, and errors caused by manual intervention are reduced. In the surface treatment stage, sandblasting parameter optimization and burr detection closed-loop control ensure that the surface quality meets the standards. The pre-bending process accurately compensates for welding warpage based on the thermal deformation law. The nickel plating thickness prediction model ensures the protective effect of the plating layer, improves the corrosion resistance, weldability and structural integrity of the substrate, and adapts to the needs of mass production. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to illustrate the main idea of ​​the present invention. Figure 1 This is a flowchart of the process steps of the present invention. Detailed Implementation

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of the present invention.

[0035] Please refer to Figure 1 As shown, the present invention provides a cold forging process for a copper IGBT heat dissipation substrate, the specific steps of which are as follows: Step 1: Obtain the original plate material for processing the heat dissipation substrate. Before cutting the original plate material, add an adaptive cutting path optimization algorithm to obtain the optimal cutting path. Then, predict the hardness of the copper after annealing by using a preset BP annealing hardness prediction model, thereby adjusting the heat preservation time of the blank.

[0036] Specifically, blanking is the first step in the manufacturing process, which refers to separating the blank from the raw material by cutting it to meet the dimensional requirements of subsequent processing.

[0037] The actual dimensions of the original sheet material are obtained using a laser instrument. The difference between the actual thickness and the target thickness is used to obtain the thickness deviation, which must be less than or equal to a preset thickness deviation threshold. Then, each defect point of the original sheet material is obtained using an ultrasonic flaw detector. The defect points are integrated into the sheet material defect area, and the material utilization rate TH and cutting efficiency TY are obtained. A weighted sum is then calculated to obtain the target optimization function F. The calculation logic for the material utilization rate is as follows: Where S 胚料 S represents the total effective area of ​​the qualified billet. 板材 Given the total area of ​​the original board material, the calculation logic for its cutting efficiency utilization rate is as follows: Among them LT 路径LT represents the actual total cutting path length. 路径，max This represents the theoretical maximum path length. It should be noted that the original board is the unprocessed initial material, and the blank is a semi-finished product cut from the original board.

[0038] The original sheet metal is uniformly divided into initial candidate blank positions according to a preset grid size. A random perturbation is applied to each candidate position, and blanks overlapping with defect areas are removed, forming an initial population. The fitness of each individual in the initial population is calculated, and the calculation logic is as follows: Where valid(BG) i Let be the constraint satisfaction function for the i-th blank. If the blank satisfies the preset size, thickness deviation, and non-overlap with defect areas, then valid(BG) is valid. i The value is 1 if any constraint is not met, and -1 if any constraint is not met. I is the total number of blanks. Several individuals are randomly selected from the population. Each individual is a set of blank coordinates. The individual with the highest fitness is selected to enter the next generation. The operation is repeated until M parent individuals are generated. Two parent individuals are randomly selected. The path is divided into the first half and the second half according to the cutting order. The second half of the path is swapped to generate offspring individuals. This process is repeated. When the number of iterations reaches the preset value, the optimal cutting path parameters are output. The optimal cutting path parameters include the number of blanks, coordinates, and cutting order.

[0039] In the cold forging process of copper IGBT heat dissipation substrate, annealing softening is achieved by placing the cut billet in a box furnace and holding it at 400-450℃ for 2-3 hours, and then slowly cooling it to room temperature in the furnace. This allows the deformed grains inside the copper to be restored to a regular arrangement, eliminates the internal stress generated by cold working, and reduces the hardness of the material.

[0040] The average temperature during the holding process is obtained through thermocouples in a box furnace and marked as the annealing temperature. When the furnace temperature reaches the set temperature, calculations are triggered, and the effective holding time of the billet is obtained. The initial hardness of the copper billet is collected using a Vickers hardness tester. The annealing temperature, effective holding time, and initial hardness are input into a preset BP annealing hardness prediction model. The predicted annealing hardness value is obtained through the input layer, implicit layer, and output layer. If the predicted annealing hardness value is greater than the first preset hardness value, it indicates insufficient softening. The preset correction temperature and preset correction time are embedded into the original temperature and original time to obtain the additional calibration temperature and additional calibration time. If the predicted annealing hardness value is less than the second preset hardness value, it indicates insufficient softening. To prevent over-softening, the preset correction temperature and time are subtracted from the original temperature and time to obtain the subtracted calibration temperature and time. It should be noted that the first preset hardness value is greater than the second preset hardness value. It should also be noted that the establishment of the preset BP annealing hardness prediction model requires the collection of multiple sets of historical data on copper annealing processes. Each set of data includes input parameters and output parameters. The weights from the input layer to the hidden layer, from the hidden layer to the output layer, and the biases of each layer are randomly initialized. The dataset is divided into a training set and a validation set in a 7:3 ratio. Training is stopped when the loss of the validation set does not decrease for several consecutive iterations. The weighted output of the input parameters through the network layers is calculated to obtain the predicted annealing hardness value.

[0041] Step 2: Lubricate the surface of the annealed billet. During the lubrication process, the thickness of the lubrication layer is controlled within the target range using a PID control algorithm. Then, the cold forging parameters of the cold forging press are optimized using a parameter optimization algorithm to obtain the optimal cold forging pressure, optimal holding time, and optimal die temperature.

[0042] Specifically, in the cold forging process of copper IGBT heat dissipation substrates, surface lubrication treatment refers to coating the surface of the copper billet with a lubricating medium and controlling parameters before cold forging. By forming a uniform and stable lubricating layer, the interface state between the billet and the mold is optimized, avoiding direct friction between the billet and the mold due to insufficient thickness or forming accuracy deviation due to excessive thickness. The thickness of the lubricating layer is collected in real time by a laser thickness gauge, and the target thickness is obtained. The difference between the target thickness and the lubricating layer thickness is used to obtain the lubrication thickness deviation rc(t). The initial nozzle pressure of the coating equipment is obtained, and the nozzle pressure adjustment amount gc(t) is calculated in real time using a PID formula. The calculation logic is as follows: K p K is a proportionality coefficient that reflects the immediate effect of the deviation. d The differential coefficient is the rate of change of the deviation, which suppresses system oscillations and enhances stability. sThe sampling period is rc(t-1), where rc(t-1) is the lubrication thickness deviation at the previous moment. The nozzle pressure adjustment is embedded into the initial nozzle pressure to obtain the complete nozzle pressure value. This complete nozzle pressure value is then sent to the coating equipment, and lubrication is sprayed onto the surface of the copper billet. 3150T cold forging in one-time forming refers to a single precision forging operation using a 3150-ton cold forging press at room temperature, directly processing the copper billet into a heat dissipation substrate with a pre-set convex center structure, without subsequent secondary forming processing. The Box-Behnken experimental design method is used to conduct experiments on several level combinations of different cold forging pressures C, holding times FR, and die temperatures R. Each experiment is repeated multiple times, and the quality results of the copper billet after each experiment are collected. The quality results include convex tolerance and hardness. Their average values ​​are calculated as the convex response value GT and the hardness response value H. The mapping relationship between the parameters and the quality target is fitted using the least squares method to construct the response surface model δ. * Its construction logic is as follows:

[0043] δ * =β0+β1C+β2FR+β3R+β4C+β5FR+β6R+β7C·FR+β8C·R+β9FR·R, where β0 to β9 are regression coefficients;

[0044] Constraints are set for each parameter. The Lagrange multiplier method combined with grid search is used to select parameter combinations that satisfy the constraints from the response surface model and output the baseline optimal solution. The baseline optimal solution includes the optimal cold forging pressure, the optimal holding time, and the optimal die temperature. The baseline optimal solution is sent to the cold forging press to perform cold forging of the copper billet in one step to obtain the initial blank of the heat dissipation substrate.

[0045] The convex forming height of the initial blank of the heat dissipation substrate is collected in real time by a laser displacement sensor, and the actual pressure C of the press is recorded simultaneously. 实测 and holding time FR 实测 The hardness quality of the initial blank of the heat dissipation substrate is detected by a pre-trained cold deformation grain refinement model. The calculation logic is as follows: H * =H 标准 +k1·(C 实测 -3150)+k2·(FR 实测 -FR 标准 ), where H 标准 The standard hardness value is from the process database, k1 is the pressure correction factor, 3150 is the standard pressure value, k2 is the time correction factor, and FR is the standard hardness value. 标准 The standard holding time is set. If the hardness quality is less than the preset hardness threshold, the holding time will be extended or the pressure will be increased.

[0046] Step 3: Remove flash and positioning pin residue from the initial blank of the cold-forged heat dissipation substrate. Obtain the contour coordinates of the flash area based on the positioning model, and then generate an adaptive path to cut the flash and positioning pins using a greedy algorithm. Output the predicted value of the hole diameter deviation using a multiple linear regression compensation model, and adjust the drilling diameter command in real time. Finally, remove residual debris and oil stains from the drilling through ultrasonic cleaning. Specifically, during the cold forging process, some materials may form excess protrusions beyond the design size on the edge of the substrate due to uneven die gap or pressure distribution. The flash on the edge of the substrate after cold forging, as well as the positioning pin traces or protrusions left during the cold forging positioning process, are removed by a punching die.

[0047] Images of the initial heat dissipation substrate blank are acquired using an industrial camera, and the images are converted into grayscale images IU(x,y). Histogram equalization is used to expand the grayscale range and improve the grayscale difference between the flash and the substrate. A low threshold T is set. low and high threshold T high The edge contour E(x,y) is extracted using the Canny operator, and its detection logic is: E(x,y)=Canny(IU(x,y),T low ,T high ), where E(x,y)=1 represents edge pixels, E(x,y)=0 represents non-edge pixels, and edge pixels are pixels in the edge contour. The theoretical edge coordinates of the substrate are loaded, and the theoretical edge coordinates are set as the baseline L0. The edge pixels are traversed, and the continuous contours outside L0 are marked as flash areas. The contour coordinate set of the flash areas is obtained. The actual width of the flash is obtained by machine vision detection. The positioning pin residue is a cylindrical protrusion on the edge of the substrate. Morphological feature matching is used for recognition. Specifically, the grayscale image is binarized to obtain the foreground target, and the circularity CQ of the target area is calculated. The calculation logic is as follows: Where SC is the area and LR is the perimeter. When the roundness is greater than or equal to the preset roundness threshold, it is determined to be a positioning pin residue, and the residue detection coordinates are output. Based on the improved greedy algorithm, the adaptive path is used to cut the burrs and positioning pins. The minimum cutting length and the lowest residue risk are used as objectives to construct an optimization model. The optimization logic is as follows:

[0048] Where LH is the objective function value, x g and y g Let x be the coordinate of the g-th node on the cutting path. g Let y be the horizontal coordinate. g Let G be the vertical coordinate, and let we be the total number of nodes on the cutting path. g2 d represents the actual width of the flash at the g2th detection point. g2 Let ω be the cutting depth at the g2th detection point, ω be the penalty coefficient, and G2 be the total number of detection points.

[0049] The first pixel of the edge contour is set as the starting point of the flash, and the last pixel of the edge contour is set as the ending point. An adaptation path is generated according to the path sorting rules, with priority 1 for positioning pin residual nodes and priority 2 for connecting the starting point to the ending point according to the natural order of the flash contour. The coordinates of the adaptation path are converted into motion commands for the cutting platform to execute the cutting. After the cutting is completed, a second image acquisition is performed to detect the residual amount. If the residual amount is greater than the preset residual threshold, a supplementary cutting path is generated, and the flash parameters, path data, and residual detection results are stored in the process database. It should be noted that the theoretical edge coordinates of the substrate refer to the ideal position parameters of the substrate edge in the two-dimensional coordinate system, which are predefined according to design requirements and process standards. It is a benchmark for measuring whether the actual processed edge meets the accuracy requirements.

[0050] Machining drilling refers to the process of machining a substrate after cold forging and trimming using CNC machining equipment. This involves machining assembly holes, positioning holes, or heat dissipation through-holes at predetermined locations on the substrate, ensuring the hole position accuracy meets design requirements. The current drill speed and feed rate are obtained, and the hardness H at the drilling location on the initial substrate blank is measured. * The drill bit rotation speed BE and feed rate VX are input to the preset borehole compensation model, which outputs the predicted borehole diameter deviation value ΔPE. The calculation logic is as follows:

[0051] in to By fitting historical data, the target borehole deviation value is obtained and the difference is calculated with the predicted borehole deviation value to obtain the borehole adjustment command, which is then sent to the CNC machining equipment terminal. Ultrasonic cleaning is used to remove residual debris and oil from the borehole.

[0052] Step 4: Sandblasting is performed on the cleaned heat dissipation substrate blank. The sandblasting parameters are optimized using the beetle whisker search algorithm. After sandblasting, the burr detection is performed using the horizontal set evolution algorithm. If the index exceeds the standard, a second detection is performed and sandblasting is repeated. If it passes, the welding allowance is designed by pre-bending and the welding temperature field is simulated using the pre-bending model to obtain the required pre-bending angle.

[0053] Specifically, 120-mesh sandblasting deburring is a surface treatment process that uses 120-mesh abrasive to blast the surface of the workpiece. Compressed air is used to spray fine-grained abrasive at high speed onto the surface of the workpiece, and then the impact force removes the micro-burrs, thus improving the roughness and enhancing the adhesion of the coating.

[0054] Roughness RA is acquired in real time using a laser confocal microscope, oxide layer residue rate KL is monitored using an eddy current thickness gauge, and hardness increase ΔH is detected offline using a microhardness tester. The roughness, oxide layer residue rate and hardness increase are weighted and summed to obtain the longhorn beetle fitness function FC.

[0055] Extract the current sandblasting parameter vector XQ, including jet pressure, jet angle, and sand particle velocity. Randomly generate a search direction vector. Extract the initial sensing distance dw of the longhorn beetle's two whiskers. The initial sensing distance is the information acquisition interval of the simulated longhorn beetle's two whiskers. Based on the initial sensing distance, obtain the position XQ of the left whisker. L and right mustache position XQ R The calculation logic for their respective positions is as follows: in The direction vector is used; the fitness value of the longhorn beetle corresponding to the position of the two whiskers is calculated, the difference between the fitness values ​​of the two whiskers is compared, and the current parameter vector is updated. If the fitness of the left whisker is better, it moves in the positive direction along the direction vector; if the fitness of the right whisker is better, it moves in the negative direction. If the fluctuation range of the longhorn beetle fitness value is less than or equal to the set fluctuation threshold during continuous iteration, the process terminates. The optimized sandblasting parameters are output and sent to the sandblasting equipment terminal.

[0056] A level set function ε is defined. If ε is less than 0, the region is considered a burr area; otherwise, it is considered a background area. If ε equals 0, the region is considered a contour line. The initial contour is a rectangular bounding box representing the potential burr area. An energy function is then constructed to converge the contour towards the burr edge. The calculation logic is as follows: Where LS(ε) is the data term that drives the contour to move towards the burr edge, and DA(ε) is the area term that controls the contour's shrinkage and expansion trends. and The preset fixed weights are 0.42 and 0.58, respectively. The level set function is iteratively updated using gradient descent based on the energy function to converge the contour. The converged contour is then subjected to morphological erosion and dilation to remove noise regions without burrs. The closed regions of the contour are extracted and marked as candidate burr regions. The total number of pixels in the candidate burr regions is obtained and mapped to the actual area. The maximum distance along the normal direction of the substrate surface is then obtained to obtain the maximum height. If the actual area and the maximum height are less than the corresponding preset area threshold and height threshold, the result is qualified; otherwise, it is unqualified. The burr detection result is output, including the position, size, and whether it exceeds the standard. If it does not exceed the standard, the sandblasting is repeated.

[0057] The pre-bending forming design welding allowance refers to the small deformation that is pre-set in the pre-bending process before nickel plating of the copper IGBT heat sink substrate, based on the difference in thermal expansion coefficient between the welding material and copper, in order to compensate for the warping deformation caused by thermal stress during the welding process, so that the flatness of the substrate after welding can meet the standard.

[0058] The thermal expansion coefficients of the substrate and solder are obtained, and the difference between them is calculated to obtain the thermal expansion coefficient difference ΔTCP. Then, the peak welding temperature and room temperature are obtained and the difference is calculated to obtain the welding temperature difference ΔYA. Based on the thermal expansion coefficient difference and the welding temperature difference, the warpage W after welding is calculated. weld Its calculation logic is as follows: Where gs is the shape factor, L is the substrate length, and h is the substrate thickness; because welding causes deformation, pre-bending requires reverse deformation, and its calculation logic is: W pre =-gc·W weld Where gc is the compensation allowance coefficient, with a value of 1.05. It should be noted that to completely offset welding deformation, the pre-bending amount only needs to be W. pre =-W weld However, in actual practice, the predicted value of welding deformation may deviate from the actual value. Adding a compensation allowance coefficient ensures that even with errors, the final welding deformation can still be controlled within the target range. The pre-bending amount is converted into a pre-bending angle that the equipment can execute; the calculation logic is as follows: Send the pre-bending angle to the corresponding device for adjustment.

[0059] Step 5: By establishing a nickel plating differential equation model, the predicted plating thickness is output, and the complete copper heat dissipation substrate is obtained by controlling the plating thickness.

[0060] Specifically, nickel plating ≥7μm means that when electroplating a nickel layer on the surface of a copper IGBT heat sink substrate, the nickel layer thickness needs to be no less than 7 micrometers. This indicator is obtained by controlling parameters such as the temperature and current density of the electroplating solution, and must also consider corrosion resistance and heat dissipation efficiency. Real-time data collection of the electroplating solution temperature TD, current density j, metal ion concentration cf, and tank stirring rate vx is used to establish a differential equation model for nickel plating. The logic for establishing this model is as follows: in Here, denoted as Caputo fractional derivative, l as reaction rate constant, EC as activation energy, R as gas constant, and e, nd, and r as kinetic exponents, obtained through experimental calibration, reflecting the influence weights of current, concentration, and stirring rate, respectively.

[0061] Test calibration process: Using offline test data, such as orthogonal experiments with 5 sets of current, 3 sets of concentration, and 3 sets of temperature, the parameters are solved using the least squares method. The solution logic is as follows: Where η = [l,e,nd,r], HD(t) z ) represents the actual thickness of the coating obtained by measurement, HS(η,t) z ) represents the numerical solution of the fractional-order model, Z represents the total number of experimental samples, and b represents the experimental sample number;

[0062] The electroplating solution temperature, current density, metal ion concentration, and tank stirring rate are input into the nickel plating differential equation model. The model iteratively solves the model and outputs a predicted plating thickness. If the predicted thickness is less than 7 μm, an "over-thin" warning is issued, generating a remedial command that is sent to the electroplating equipment terminal to perform a second electroplating on the entire board. Otherwise, manufacturing ends, resulting in a complete copper heat dissipation substrate. All calculation formulas involved in this process are dimensionless, eliminating dimensional influences through standardization and other methods, retaining only numerical values ​​for calculation. The specific dimensionless method can be flexibly selected according to the actual scenario and will not be elaborated here. All formulas are derived from extensive process data through software simulation and optimization, accurately reflecting the actual forming process. The preset parameters in the formulas need to be specifically set by those skilled in the art in conjunction with production conditions and material characteristics.

[0063] The above-described process embodiments can be implemented through software, hardware, firmware, or any combination thereof. If implemented in software, it can be presented entirely or partially as a computer program product, containing one or more computer instructions or programs. When loaded or executed on a computer (including general-purpose computers, special-purpose computers, computer networks, and other programmable devices), it can generate all or part of the corresponding process flow and functions. Computer instructions can be stored in a computer-readable storage medium or transmitted between computer-readable storage media via wired (e.g., network transmission), wireless (e.g., infrared, microwave, wireless communication), or other means. The computer-readable storage medium can be any available medium accessible to a computer, or data storage devices such as servers and data centers containing magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), semiconductor media (e.g., solid-state drives). It should be noted that the sequence numbers of the processes in the embodiments of this application are for convenience only and do not represent the execution order. The actual execution order should be determined according to the functional logic, and the sequence number does not constitute any limitation on the process implementation.

[0064] Those skilled in the art will recognize that the units and algorithm steps involved in the embodiments can be implemented through electronic hardware or a combination of computer software and electronic hardware. The specific implementation method depends on the application scenario and design constraints of the technical solution. Professionals can flexibly choose the implementation method for different applications, but its implementation should not exceed the protection scope of this application.

[0065] In the embodiments of this application, the disclosed systems, apparatuses, and methods can be implemented in other forms. For example, the division of apparatus units is only a logical functional division, and can be adjusted according to actual needs. For instance, multiple units or components can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or communication connection between the various devices and units can be indirect coupling or communication through interfaces, and the forms include electrical, mechanical, or other adaptation forms.

[0066] The units described as separate components may or may not be physically separated; the components shown as units may be physical units or distributed among multiple network units, and some or all units may be selected to achieve the purpose of this embodiment according to actual needs.

[0067] If the function is implemented as a software functional unit and sold or used as an independent product, it can be stored on a computer-readable storage medium. Based on this, the contribution of this technical solution to the prior art or its core functions can be embodied in a software product, which is stored on a storage medium (such as a USB flash drive, external hard drive, ROM, RAM, magnetic disk, optical disk, etc.) and contains several instructions to cause a computer device to execute all or part of the steps of the process methods of the various embodiments of this application.

[0068] The scope of protection of this application shall be determined by the appended claims. Any variations or substitutions that can be easily conceived by a person skilled in the art within the scope of the technology disclosed in this application shall be included within the scope of protection.

Claims

1. A cold forging process for a copper IGBT heat dissipation substrate, characterized in that, Includes the following steps: Step 1: Obtain the original board material for processing the heat dissipation substrate. Before cutting the original board material, add an adaptive cutting path optimization algorithm to obtain the optimal cutting path. Then, use a preset BP annealing hardness prediction model to predict the hardness of the copper after annealing, and adjust the heat preservation time of the billet accordingly. Step 2: Lubricate the surface of the annealed billet. During the lubrication process, the thickness of the lubrication layer is controlled within the target range using a PID control algorithm. Then, the cold forging parameters of the cold forging press are optimized using a parameter optimization algorithm to obtain the optimal cold forging pressure, optimal holding time, and optimal die temperature. Step 3: Remove the flash and positioning pin residue from the cold-forged heat dissipation substrate blank. Obtain the contour coordinates of the flash area based on the positioning model. Then, generate an adaptive path to cut the flash and positioning pin using a greedy algorithm. Output the hole diameter deviation prediction value using a multiple linear regression compensation model and adjust the drilling diameter command in real time. Finally, remove the residual debris and oil stains from the drilling through ultrasonic cleaning. Step 4: Sandblasting is performed on the cleaned heat dissipation substrate blank. The sandblasting parameters are optimized using the beetle whisker search algorithm. After sandblasting, the burr detection is performed using the horizontal set evolution algorithm. If the index exceeds the standard, a second detection is performed and sandblasting is repeated. If it passes, the welding allowance is designed by pre-bending and the welding temperature field is simulated using the pre-bending model to obtain the required pre-bending angle. Step 5: By establishing a nickel plating differential equation model, the predicted plating thickness is output, and by controlling the plating thickness, a complete copper heat dissipation substrate is obtained.

2. The cold forging process for a copper IGBT heat dissipation substrate according to claim 1, characterized in that, Based on step three, which involves removing burrs and positioning pin residues from the cold-forged heat dissipation substrate blank, obtaining the contour coordinates of the burr area based on the positioning model, and then generating an adaptive path to cut the burrs and positioning pins using a greedy algorithm, the specific steps are as follows: Images of the initial heat dissipation substrate blank are acquired using an industrial camera, the images are converted into grayscale images, the grayscale range is expanded by histogram equalization, low and high thresholds are set, and edge contours are extracted using the Canny operator. The theoretical edge coordinates of the substrate are loaded, the theoretical edge coordinates are set as the baseline, the edge pixels are traversed, and the continuous contours outside the baseline are marked as the flash area. The set of contour coordinates of the flash area is obtained. The actual width of the burr is obtained by machine vision detection. The grayscale image is binarized to obtain the foreground target. The roundness of the target area is calculated. If the roundness is greater than or equal to the preset roundness threshold, it is determined to be a positioning pin residue. The residue detection coordinates are output. Based on the improved greedy algorithm, the appropriate path is obtained to cut the burr and positioning pin. The minimum cutting length and the lowest residue risk are taken as the objectives to build an optimization model. The first pixel of the edge contour is set as the starting point of the flash, and the last pixel of the edge contour is set as the ending point. An adaptation path is generated according to the path sorting rules. The sorting rules are: priority 1 for positioning pin residual nodes, and priority 2 for connecting the starting point to the ending point according to the natural order of the flash contour. The coordinates of the adaptation path are converted into motion commands of the cutting platform to execute the cutting. After the cutting is completed, a second image is acquired to detect the residual amount. If the residual amount is greater than the preset residual threshold, a supplementary cutting path is generated, and the flash parameters, path data, and residual detection results are stored and sent to the process database.

3. The cold forging process for a copper IGBT heat dissipation substrate according to claim 2, characterized in that, The borehole diameter deviation is predicted using a multiple linear regression compensation model, and the borehole diameter command is adjusted in real time. Finally, ultrasonic cleaning is used to remove residual debris and oil from the borehole. Specifically: The current drill bit speed and feed rate are obtained. The hardness of the drilling position of the heat dissipation substrate blank, the drill bit speed and feed rate are input into the preset drilling compensation model. The hole diameter deviation prediction value is output. The target hole deviation value is obtained and the difference with the hole diameter deviation prediction value is calculated to obtain the hole adjustment command, which is sent to the CNC machining equipment terminal. The residual debris and oil stains in the drilling are removed by ultrasonic cleaning.

4. The cold forging process for a copper IGBT heat dissipation substrate according to claim 1, characterized in that, Based on step two, the surface of the annealed billet is lubricated. During the lubrication process, a PID control algorithm is used to control the thickness of the lubrication layer within the target range. Specifically: The thickness of the lubricating layer is collected in real time by a laser thickness gauge. The target thickness is obtained and the difference between the target thickness and the lubricating layer thickness is obtained to obtain the lubricating thickness deviation. The initial nozzle pressure of the coating equipment is obtained, and the nozzle pressure adjustment amount is calculated in real time by a PID formula. The nozzle pressure adjustment amount is embedded into the initial nozzle pressure to obtain the complete nozzle pressure value. The complete nozzle pressure value is sent to the coating equipment, and lubrication spraying is performed on the surface of the copper billet.

5. The cold forging process for a copper IGBT heat dissipation substrate according to claim 4, characterized in that, Then, the cold forging parameters of the cold forging press are optimized using a parameter optimization algorithm to obtain the optimal cold forging pressure, optimal holding time, and optimal die temperature, specifically as follows: The Box-Behnken experimental design method was used to conduct experiments on several level combinations of different cold forging pressure, holding time and die temperature. Each group of experiments was repeated multiple times, and the quality results of the copper billet after each experiment were collected. The quality results include the convex surface tolerance and hardness. The average values ​​were calculated as the convex surface response value GT and the hardness response value H. The mapping relationship between the parameters and the quality target was fitted by the least squares method to construct the response surface model. Constraints are set for each parameter. The Lagrange multiplier method combined with grid search is used to select parameter combinations that satisfy the constraints from the response surface model and output the baseline optimal solution. The baseline optimal solution includes the optimal cold forging pressure, the optimal holding time, and the optimal die temperature. The baseline optimal solution is sent to the cold forging press to perform cold forging of the copper billet in one step to obtain the initial blank of the heat dissipation substrate. The convex forming height of the heat dissipation substrate blank is collected in real time by a laser displacement sensor, and the actual pressure and holding time of the press are recorded simultaneously. The hardness quality of the heat dissipation substrate blank is detected by a pre-trained cold deformation grain refinement model. If the hardness quality is less than the preset hardness threshold, the holding time is extended or the pressure is increased.

6. The cold forging process for a copper IGBT heat dissipation substrate according to claim 1, characterized in that, Based on step four, the cleaned heat dissipation substrate blank undergoes a sandblasting process. The sandblasting parameters are optimized using a beetle whisker search algorithm. After sandblasting, a level set evolution algorithm is used to perform burr detection. If the indicators exceed the standards, a second inspection is performed and sandblasting is repeated. Specifically: Roughness is collected in real time using a laser confocal microscope, oxide layer residue rate is monitored using an eddy current thickness gauge, and hardness increase is detected offline using a microhardness tester. The roughness, oxide layer residue rate and hardness increase are weighted and summed to obtain the longhorn beetle fitness function. Extract the current sandblasting parameter vector, including spray pressure, spray angle, and sand flow velocity. Randomly generate a search direction vector. Extract the initial sensing distance of the longhorn beetle's two whiskers. Based on the initial sensing distance, obtain the positions of the left and right whiskers. Calculate the longhorn beetle fitness value corresponding to the whisker positions. Compare the difference in the whisker fitness values ​​and update the current parameter vector. If the left whisker has better fitness, move forward along the direction vector; if the right whisker has better fitness, move backward. If the fluctuation range of the longhorn beetle fitness value is less than or equal to the set fluctuation threshold during continuous iteration, the process terminates. Output the optimized sandblasting parameters and send them to the sandblasting equipment terminal. Set the level set function ,like A value less than 0 indicates a burr area, while a value less than 0 indicates a background area. If the value is 0, it represents the outline. The initial outline is a rectangular box representing the potential burr area. An energy function is constructed to converge the outline towards the burr edge. The level set function is iteratively updated using gradient descent based on the energy function to converge the outline. Morphological erosion and dilation are performed on the converged outline to remove noise areas without burrs. The closed regions of the outline are extracted and marked as candidate burr regions. The total number of pixels in the candidate burr regions is obtained and mapped to the actual area. The maximum distance along the normal direction of the substrate surface is then obtained to obtain the maximum height. If the actual area and maximum height are less than the corresponding preset area threshold and height threshold, it is considered qualified; otherwise, it is considered unqualified. The burr detection result is output, including the location, size, and whether it exceeds the standard. If it does not exceed the standard, the sandblasting is repeated.

7. The cold forging process for a copper IGBT heat dissipation substrate according to claim 6, characterized in that, The required pre-bending angle is obtained by designing the welding allowance through pre-bending and simulating the welding temperature field using a pre-bending model. The specific steps are as follows: The thermal expansion coefficients of the substrate and solder are obtained, and the difference between the two is obtained to get the thermal expansion coefficient difference. The welding peak temperature and room temperature are obtained and the difference is obtained to get the welding temperature difference. The warpage after welding is calculated based on the thermal expansion coefficient difference and the welding temperature difference. The pre-bending amount is obtained based on the warpage after welding and the pre-bending amount is converted into a pre-bending angle that the equipment can execute.

8. The cold forging process for a copper IGBT heat dissipation substrate according to claim 1, characterized in that, Based on step five, the predicted plating thickness is obtained by establishing a differential equation model for nickel plating. By controlling the plating thickness, a complete copper heat dissipation substrate is obtained. The specific steps are as follows: The electroplating solution temperature, current density, metal ion concentration, and tank stirring rate are collected in real time. A nickel plating differential equation model is established. The electroplating solution temperature, current density, metal ion concentration, and tank stirring rate are input into the nickel plating differential equation model. The predicted plating thickness is output through iterative solution. If the predicted plating thickness is less than 7μm, it is judged as an over-thin warning. A remedial command is generated and sent to the electroplating equipment terminal to perform a second electroplating on the whole board. Otherwise, the manufacturing process ends and a complete copper heat dissipation substrate is obtained.

9. The cold forging process for a copper IGBT heat dissipation substrate according to claim 1, characterized in that, Based on step one, which obtains the original board material for processing the heat dissipation substrate, an adaptive cutting path optimization algorithm is added before cutting the original board material to obtain the optimal cutting path, specifically as follows: The actual dimensions of the original sheet material are obtained using a laser instrument. The difference between the actual thickness and the target thickness is used to obtain the thickness deviation, which must be less than or equal to a preset thickness deviation threshold. Then, each defect point of the original sheet material is obtained using an ultrasonic flaw detector. These defect points are integrated into the sheet material defect area, and the material utilization rate and cutting efficiency are obtained. A weighted summation is then performed to calculate the target optimization function. The original sheet metal is evenly divided into initial candidate blank positions according to a preset grid size. Random perturbation is applied to each candidate position, and blanks that overlap with defect areas are removed to form an initial population. The fitness of each individual in the initial population is calculated. Several individuals are randomly selected from the population, and each individual is a set of blank coordinates. The individual with the highest fitness is selected to enter the next generation. The operation is repeated until M parent individuals are generated. Two parent individuals are randomly selected, and the path is divided into the first half and the second half according to the cutting order. The second half of the path is swapped to generate offspring individuals. This process is repeated. When the number of iterations reaches a preset value, the optimal cutting path parameters are output. The optimal cutting path parameters include the number of blanks, coordinates, and cutting order.

10. The cold forging process for a copper IGBT heat dissipation substrate according to claim 9, characterized in that, The hardness of the annealed copper is then predicted using a preset BP annealing hardness prediction model, and the holding time of the billet is adjusted accordingly. The specific steps are as follows: The average temperature during the heat preservation process is obtained through thermocouples in the box furnace and marked as the annealing temperature. When the furnace temperature reaches the set temperature, the calculation is triggered, and the effective heat preservation time of the billet is obtained. The initial hardness of the copper billet is collected through a Vickers hardness tester. The annealing temperature, effective heat preservation time, and initial hardness are input into the preset BP annealing hardness prediction model. The predicted annealing hardness value is obtained through the input layer, hidden layer, and output layer. If the predicted annealing hardness value is greater than the first preset hardness value, the preset correction temperature and preset correction time are embedded into the original temperature and original time to obtain the increased calibration temperature and increased calibration time. If the predicted annealing hardness value is less than the second preset hardness value, the preset correction temperature and preset correction time are subtracted from the original temperature and original time to obtain the reduced calibration temperature and reduced calibration time.

Images