Surface mounting process parameter adaptive optimization method based on automatic optical detection feedback

Abstract

The invention discloses a surface mounting process parameter self-adaptive optimization method based on automatic optical detection feedback, relates to a surface mounting process parameter automatic optimization method, and belongs to the field of electric appliance technology and electrical engineering. The objective of the invention is to solve the problems that an existing surface mounting process parameter optimization method depends on manual adjustment, cannot effectively, dynamically and adaptively regulate and control various parameters, and cannot guarantee the regulation and control effect, so that the mounting precision and efficiency are reduced, and mounting defects and manual repair work are increased. The invention discloses a surface mounting process parameter adaptive optimization method based on automatic optical detection feedback, and the method comprises the steps: carrying out the adjustment and optimization of the mounting height based on a Bayesian optimization method, carrying out the adjustment and optimization of the blowing time delay based on a particle swarm optimization method, and carrying out the automatic compensation of the mounting offset based on a moving average method. A self-adaptive optimization method based on automatic optical detection feedback is completely given; and various parameters are adjusted, so that defects in the mounting process are reduced, and the mounting precision and efficiency are improved.

Description

Technical Field

[0001] The invention relates to an automatic optimization method for surface mounting process parameters, and belongs to the field of electrical technology and electrical engineering. Background Art

[0002] In modern life and production, various electrical devices are widely used in daily life and industrial production. The functions of these devices depend on circuit boards, which are equipped with a variety of electronic components and chips, and achieve coordinated work of various components through electrical connections. In order to efficiently assemble fully functional circuit boards, surface mounting technology has been widely used, among which the key equipment is the placement machine. These placement machines can quickly and accurately pick up various components and quickly place them on the target position on the surface of the circuit board. To simplify the description, the subsequent text will collectively refer to electronic components, electronic devices and chips as "components", and the process of picking and placing will be referred to as "pick and place".

[0003] Figure 2 The placement machine in the machine uses a dual-drive gantry motion platform, in which the X-axis and Y-axis motors drive the placement carrier to move along the X-axis and Y-axis guide rails. The feeder base is fixed with a feeder that can automatically supply components and wait for the suction rod to pick them up. At the same time, the conveyor belt transports the circuit board to the designated position for placement.

[0004] exist Figure 3 Inside the SMT carrier, the vacuum rod is controlled by the Z-axis drive motor to achieve vertical movement so that its end can accurately contact the component surface. The end of the rod is equipped with a nozzle, which can reliably pick up and place components by generating suction and blowing. In order to accurately rotate the picked component to the required placement angle, the θ-axis motor is installed behind the Z-axis motor to facilitate rotation around the central axis. In order to adapt to various types of components, matching nozzles need to be selected, and the replacement of nozzles is achieved through an automatic nozzle changer.

[0005] Figure 4 The automatic optical inspection equipment in the system uses a dual-drive gantry motion platform. The X-axis motor and Y-axis motor of the platform drive a high-resolution automatic optical inspection camera to check the placement results of each component on the circuit board. This equipment evaluates the difference between the actual placement result and the standard placement result based on the template matching method, obtains the placement offset, and detects various types of defects based on the deep learning classification method, thereby identifying the source of the error, such as whether the defect is caused by too small a placement height.

[0006] In traditional surface mount production, technicians can use their professional knowledge to adjust the surface mount process parameters of the placement machine based on the test results of the automatic optical inspection equipment. For each placement point, the surface mount process parameters include: X-axis offset compensation value, Y-axis offset compensation value, θ-axis deflection compensation value, placement height compensation value and placement blowing delay. These parameters have a significant impact on the placement yield, placement accuracy and placement efficiency of components.

[0007] In order to analyze the causes of various placement defects, we must first understand the inherent deviations in the mechanical structure. The machine itself has machining and assembly errors, so it is necessary to compensate for the offset of the X-axis placement coordinates, Y-axis placement coordinates, and theta-axis rotation angle of the placement point. As the placement machine runs continuously for a long time, dynamic changes such as thermal deformation will occur, so it is necessary to perform real-time online compensation, which is not possible with current manual compensation.

[0008] On a micron scale, any assembly deviation can significantly affect the contact between the circuit board surface and the suction rod tip. Figure 5 The assembly deviation angles of the suction rod are shown, where α and β represent the deviation angles of the Z-axis and θ-axis motors in three-dimensional space, respectively. These angular deviations cause the end point of the suction rod to no longer remain stationary when the θ-axis motor rotates 360 degrees, but to form an approximately elliptical trajectory in a non-horizontal plane. This not only indicates the presence of linear and angular deviations, but also means that there is a height difference Δz between the suction rod's highest and lowest points. Therefore, different mounting height compensations are required for different mounting angles. Previous mounting height adjustments relied on a lot of manual trial and error. In addition, the heights of different components differ from the thickness of the circuit board, and the circuit board may fluctuate due to warping, and traditional adjustment methods are difficult to adapt to these changes.

[0009] When the end of the suction rod reaches the specified placement position and height, the blowing process begins. This action stops the suction of the vacuum nozzle, causing the component to fall and complete the placement. The duration of the blowing process directly affects the reliability of placement. Although extending the blowing time can improve the placement results, too long a time will reduce the overall circuit board placement efficiency. On the contrary, if the blowing time is too short, the components that have been placed in place may be sucked up again due to the residual vacuum suction, resulting in placement deviation or even placement defects. As the blowing delay approaches the critical threshold, increasing the delay time can improve the accuracy of placement, which can be used as an indicator for adjusting the blowing delay. Usually, a longer blowing time is set first, and then the blowing time is gradually shortened. In the past, the blowing delay was set manually, which had a rough adjustment effect and it was difficult to find the shortest blowing delay to ensure the placement effect.

[0010] By leveraging IIoT technology, placement machines and automatic optical inspection equipment can operate seamlessly as a unified system, significantly improving placement quality and efficiency. Ethernet for control automation technology (EtherCAT), designed for real-time control applications in the automation field, is used to integrate the two machines. This integration enables the registration information of components in the placement machine to be directly transmitted to the automatic optical inspection equipment, simplifying and accelerating the data preparation process. By associating and storing the placement accuracy and defect type of each placement point with its corresponding process parameters, it is expected that the parameters of each placement point can be adjusted independently, achieving personalized adjustment under various production factors, thereby improving placement accuracy and placement yield.

[0011] The main drawback of the current research is that the existing surface mount process parameter optimization method relies on manual adjustment, cannot effectively and dynamically adaptively adjust the parameters, and cannot guarantee the adjustment effect, which leads to reduced mounting accuracy and efficiency, and an increase in mounting defects and manual repair work. Summary of the invention

[0012] The purpose of the present invention is to solve the problem that the existing surface mounting process parameter optimization method relies on manual adjustment, cannot effectively and dynamically adaptively adjust various parameters, and cannot guarantee the adjustment effect, thereby resulting in reduced mounting accuracy and efficiency, increased mounting defects and manual repair work, and proposes a surface mounting process parameter adaptive optimization method based on automatic optical detection feedback.

[0013] The specific process of the surface mount process parameter adaptive optimization method based on automatic optical inspection feedback is as follows:

[0014] Step 1: Obtain the data of the circuit board to be mounted, initialize the surface mounting process parameters and the search range of the surface mounting process parameters, initialize the parameters of the mounting height adaptive optimization method based on the Bayesian optimization algorithm, the parameters of the mounting alignment adaptive optimization method based on the moving average method, and initialize the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm; the specific process is:

[0015] Step 11, obtaining a data file of a circuit board to be mounted, the data file of the circuit board to be mounted including the serial number, X-axis coordinate, Y-axis coordinate, and θ-axis angle of each mounting point;

[0016] Step 12, set the total number K of circuit boards to be mounted;

[0017] Step 13: Initialize surface mount process parameters;

[0018] The surface mounting process parameters include the X-axis offset compensation value δX, the Y-axis offset compensation value δY, the θ-axis deflection compensation value δθ, the mounting height compensation value δZ, the mounting blowing delay T b ;

[0019] Step 14: Initialize the search range of surface mount process parameters. The search range of δX is SS δX =[-100,100], the search range of δY is SS δY =[-100,100], the search range of δθ is SS δθ =[-1,1], the search range of δZ is SS δZ =[-500,500], T b The search range is SS Tb =[5,50];

[0020] SS δX The unit is μm, SS δY The unit is μm, SS δθ The unit is °, SS δZ The unit is μm, SS Tb The unit is ms;

[0021] Step 15: For the placement height adaptive optimization method based on the Bayesian optimization algorithm, initialize the exploration-utilization parameter ξ=3, the initial placement height compensation value δZ=0; initialize the number of placement height observation points OB=30: SZ=linspace(SS δZ ,OB);

[0022] OB is the total number of observation points of mounting height in the Bayesian optimization algorithm, SZ is the observation point array, linspace(SS δZ ,OB) refers to the δZ Generate an array of OB elements at equal intervals within the range;

[0023] Step 16: For the placement alignment adaptive optimization method based on the moving average method, initialize the size of the moving average window W = 20; the X-axis offset accumulated value ε X =0, Y-axis offset accumulated value ε Y =0, θ axis offset accumulated value ε θ =0;

[0024] Step 17: For the adaptive optimization method of blowing delay based on particle swarm optimization algorithm, initialize the number of particles N = 5, randomly select N mounting points for blowing delay test, particle counting variable n = 1, individual learning factor c1 = 1 for each particle, social learning factor c2 = 1 for each particle, inertia factor w = 0.5, in the search space SS Tb In the example, randomly initialize the position pb of each particle n and speed vb n ;

[0025] Step 2: Based on the initialized surface mounting process parameters, the first circuit board is mounted, and automatic optical inspection is performed on the mounted circuit board. According to the automatic optical inspection results, the mounting height adaptive optimization method parameters based on the Bayesian optimization algorithm are updated, the mounting alignment adaptive optimization method parameters based on the moving average method are updated, the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value are updated, and the blowing delay adaptive optimization method parameters based on the particle swarm optimization algorithm are updated;

[0026] Step 3, determine whether the circuit board counting variable k is less than or equal to the total number of circuit boards K;

[0027] If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5.

[0028] If not, it means that all circuit boards have been mounted and production has stopped;

[0029] Step 4: Execute the mounting height adaptive optimization method based on the Bayesian optimization algorithm to update the mounting height compensation value;

[0030] Step 5: Execute the blowing delay adaptive optimization method based on the particle swarm optimization algorithm to update the placement blowing delay;

[0031] Step 6: Based on the surface mounting process parameters updated in step 4 and step 5, mount the next circuit board, and perform automatic optical inspection on the mounted circuit board;

[0032] Step 7: Update the parameters of the mounting height adaptive optimization method according to the automatic optical detection result of step 6;

[0033] Step 8: updating the parameters of the blowing delay adaptive optimization method according to the automatic optical detection result of step 6;

[0034] Step nine: according to the offset of each mounting point detected by automatic optical detection in step six, execute the mounting alignment adaptive optimization method based on the moving average method to update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value.

[0035] The beneficial effects of the present invention are:

[0036] The present invention discloses a method for adaptive optimization of surface mounting process parameters based on automatic optical detection feedback, proposes a complete sequential optimization framework, regulates surface mounting process parameters from high to low importance, realizes the integrated operation of the placement machine and the automatic optical detection equipment, and thus improves the overall quality and efficiency of circuit board placement. Within the optimization framework, the placement height is optimized based on the Bayesian optimization method, the blowing delay is optimized based on the particle swarm optimization method, and the placement offset is automatically compensated based on the moving average method, and a complete adaptive optimization method based on automatic optical detection feedback is given. Adjustments are made to various parameters, thereby achieving a significant improvement in defect reduction, placement accuracy and efficiency during the placement process.

[0037] Specifically, the present invention has the following two innovative features:

[0038] The present invention relates the mounting quality to the surface mounting process parameters, performs accurate calculation and adaptive adjustment, realizes the integrated operation of the chip mounter and the automatic optical detection equipment, and improves the mounting quality and efficiency of the circuit board.

[0039] The present invention proposes a complete set of sequential optimization frameworks, and introduces an adaptive optimization method guided by automatic optical detection feedback in the framework. For the regulation of various parameters, adaptation is performed based on Bayesian optimization and particle swarm optimization methods, respectively, thereby achieving a reduction in defects in the mounting process and an improvement in mounting accuracy and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the method for self-adaptive optimization of surface mount process parameters based on automatic optical detection feedback of the present invention;

[0041] Figure 2 This is a schematic diagram of a chip mounter;

[0042] Figure 3 Schematic diagram of the patch carrier;

[0043] Figure 4 It is a schematic diagram of an automatic optical inspection device;

[0044] Figure 5 Schematic diagram of the assembly deviation angle of the suction rod. DETAILED DESCRIPTION

[0045] Specific implementation method 1: The specific process of the surface mounting process parameter adaptive optimization method based on automatic optical detection feedback in this implementation method is as follows:

[0046] Step 1: Obtain the data of the circuit board to be mounted, initialize the surface mounting process parameters and the search range of the surface mounting process parameters, initialize the parameters of the mounting height adaptive optimization method based on the Bayesian optimization algorithm, the parameters of the mounting alignment adaptive optimization method based on the moving average method, and initialize the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm; the specific process is:

[0047] Step 11, obtain the data file of the circuit board to be mounted, the data file of the circuit board to be mounted includes the serial number, X-axis coordinate, Y-axis coordinate, and θ-axis angle of each mounting point (the X-axis coordinate, Y-axis coordinate, and θ-axis of the existing dual-drive gantry motion platform, the X-axis coordinate, Y-axis coordinate, and θ-axis are also introduced at the bottom of the first page); as shown in Table 1 circuit board data file;

[0048] Step 12, set the total number K of circuit boards to be mounted;

[0049] Step 13: Initialize surface mount process parameters;

[0050] The surface mounting process parameters include the X-axis offset compensation value δX, the Y-axis offset compensation value δY, the θ-axis deflection compensation value δθ, the mounting height compensation value δZ, the mounting blowing delay T b ; As shown in Table 2, the surface mount process parameter file;

[0051] Step 14: Initialize the search range of surface mount process parameters. The search range of δX is SS δX =[-100,100], the search range of δY is SS δY =[-100,100], the search range of δθ is SS δθ =[-1,1], the search range of δZ is SS δZ =[-500,500], T b The search range is SS Tb =[5,50];

[0052] SS δX The unit is μm, SS δY The unit is μm, SS δθ The unit is °, SS δZ The unit is μm, SS Tb The unit is ms;

[0053] Step 15: For the placement height adaptive optimization method based on the Bayesian optimization algorithm, initialize the exploration-utilization parameter ξ=3, the initial placement height compensation value δZ=0; initialize the number of placement height observation points OB=30: SZ=linspace(SS δZ ,OB);

[0054] Note: The parameters set in step 15 will be applied to the placement height adaptive optimization method based on the Bayesian optimization algorithm in step 4;

[0055] OB is the total number of observation points of mounting height in the Bayesian optimization algorithm, SZ is the observation point array, linspace(SS δZ ,OB) refers to the δZ Generate an array of OB elements at equal intervals within the range;

[0056] Step 16: For the placement alignment adaptive optimization method based on the moving average method, initialize the size of the moving average window W = 20; the X-axis offset accumulated value ε X =0, Y-axis offset accumulated value ε Y =0, θ axis offset accumulated value ε θ =0;

[0057] Note: The parameters set in step 16 will be applied to the placement alignment adaptive optimization method based on the moving average method in step 7;

[0058] Step 17: For the air blowing delay adaptive optimization method based on the particle swarm optimization algorithm, the number of particles N is initialized to 5, N mounting points for air blowing delay test are randomly selected (referred to as "air blowing test points"), the particle counting variable n is 1, the individual learning factor c1 of each particle is 1, the social learning factor c2 of each particle is 1, the inertia factor w is 0.5, and the search space SS is Tb In the example, randomly initialize the position pb of each particle n and speed vb n ;

[0059] Note: The parameters set in step 17 will be applied to the blowing delay adaptive optimization method based on particle swarm optimization algorithm in step 5.

[0060] Step 2: Based on the initialized surface mounting process parameters, the first circuit board is mounted, and automatic optical inspection is performed on the mounted circuit board. According to the automatic optical inspection results, the mounting height adaptive optimization method parameters based on the Bayesian optimization algorithm are updated, the mounting alignment adaptive optimization method parameters based on the moving average method are updated, the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value are updated, and the blowing delay adaptive optimization method parameters based on the particle swarm optimization algorithm are updated;

[0061] Step 3, determine whether the circuit board counting variable k is less than or equal to the total number of circuit boards K;

[0062] If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5.

[0063] If not, it means that all circuit boards have been mounted and production has stopped;

[0064] Step 4: Execute the mounting height adaptive optimization method based on the Bayesian optimization algorithm to update the mounting height compensation value;

[0065] Step 5: Execute the blowing delay adaptive optimization method based on the particle swarm optimization algorithm to update the placement blowing delay;

[0066] Step 6: Based on the surface mounting process parameters updated in step 4 and step 5, mount the next circuit board, and perform automatic optical inspection on the mounted circuit board;

[0067] Step 7: Update the parameters of the mounting height adaptive optimization method according to the automatic optical detection result of step 6;

[0068] Step 8: updating the parameters of the blowing delay adaptive optimization method according to the automatic optical detection result of step 6;

[0069] Step nine: according to the offset of each mounting point detected by automatic optical detection in step six, execute the mounting alignment adaptive optimization method based on the moving average method to update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value.

[0070] Specific implementation method 2: This implementation method is different from specific implementation method 1 in that, in step 2, based on the initialized surface mounting process parameters, the first circuit board is mounted, and the mounted circuit board is subjected to automatic optical inspection, and the mounting height adaptive optimization method based on the Bayesian optimization algorithm is updated according to the automatic optical inspection result, the mounting alignment adaptive optimization method parameters based on the moving average method are updated, the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value are updated, and the blowing delay adaptive optimization method parameters based on the particle swarm optimization algorithm are updated;

[0071] The specific process is:

[0072] Step 21. Initialize the surface mount process parameters, including: initial mount height compensation value δZ = 0, initialize X-axis offset compensation value δX = 0, Y-axis offset compensation value δY = 0, θ-axis deflection compensation value δθ = 0, set the blowing delay of each blowing test point to pb n The air blowing delay of other mounting points (other mounting points are all mounting points except the air blowing test points) is set to a safe value T ba =SSTb (2); Initialize the mounting height history value array δz′=δZ;

[0073] The SS Tb (2) Blowing delay T for mounting b The upper limit of the search range; n∈[1:N];

[0074] Step 22: Based on the initialized surface mount process parameters, complete the mounting of the first circuit board; set the mounted circuit board count variable k=1, perform automatic optical inspection on the mounted circuit board, and obtain the result of the automatic optical inspection;

[0075] The results of automatic optical inspection include the offset of each placement point∈ X ,∈ Y ,∈ θ , total number of mounting defects D 0 , defect sources; defect sources are divided into: defects caused by too high mounting height, mounting defects caused by too short blowing delay, and mounting defects caused by too large mounting offset;

[0076] Note: Automatic optical inspection is realized by an automatic optical inspection machine, which can measure the position offset and angle offset of each mounting point, classify the existing mounting defects, and determine the source of the defects;

[0077] Step 2 and 3: Update the parameters of the placement height adaptive optimization method based on the Bayesian optimization algorithm according to the automatic optical detection results; the specific process is:

[0078] Step 231: According to the automatic optical inspection results, determine whether the defect is caused by the placement height being too high. If so, update the upper limit of the placement height δZ search range and assign SS δZ (2) = δZ; if not, update the lower limit of the search range of the mounting height δZ and assign SS δZ (1) = δZ;

[0079] Among them, SS δZ (2) The search range for the mounting height δZ is SS δZ Upper limit, SS δZ (1) The search range for the mounting height δZ is SS δZ The lower limit of

[0080] Step 232: Calculate the defect rate R = D 0 / P;

[0081] Initialize defect rate history value array r=R;

[0082] Initialize the number of placement defects D of the previous circuit board l =D0 ;

[0083] Where P is the total number of mounting points on the circuit board;

[0084] Step 24: Update the parameters of the placement alignment adaptive optimization method based on the moving average method according to the automatic optical detection results, and update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value; the specific process is:

[0085] Step 241: Add the offset of each mounting point to the cache array: B X =[B X (∈ X +δX)], B Y =[B Y (∈ Y +δY)], B θ =[B θ (∈ θ +δθ)];

[0086] Wherein, δX represents the X-axis offset compensation value, δY represents the Y-axis offset compensation value, and δθ represents the θ-axis deflection compensation value;

[0087] ∈ X ,∈ Y ,∈ θ Indicates the offset of each mounting point; B X Represents the cache array for storing the historical data of the X-axis placement offset when not compensated, B Y Represents the cache array for storing the historical data of the Y-axis placement offset when not compensated, B θ Represents a cache array for storing historical data of theta-axis mounting offset when not compensated;

[0088] Step 242: Update the X-axis offset accumulated value ε X =ε X +∈ X +δX, Y axis offset accumulated value ε Y =ε Y +∈ Y +δY, θ axis offset accumulated value ε θ =ε θ +∈ θ +δθ;

[0089] Step 243: Get the cache array length n B =length(B X );

[0090] Among them, length() means getting the number of elements in the array in brackets;

[0091] Step 244: Update X-axis offset compensation value δX=εX / n B , Y-axis offset compensation value δY=ε Y / n B 、θ axis deflection compensation value δZ=ε θ / n B ;

[0092] Step 25: Update the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm according to the automatic optical detection results; the specific process is:

[0093] Step 251: Initialize particle count variable n = 1, set penalty weight value V = 10 90 ;

[0094] Note: In the particle swarm optimization algorithm, "particle" represents a candidate solution;

[0095] Step 252: determine whether n is less than N, if so, execute step 253, otherwise execute step 257;

[0096] Step 253: According to the automatic optical inspection results, determine whether the nth blowing test point has a mounting defect caused by a short blowing delay or a mounting defect caused by a large mounting offset. If so, set ψ n =1; if not, then ψ n =0;

[0097] Among them, ψ n is a binary variable, indicating whether there is a mounting defect at the nth blowing test point;

[0098] Step 254: Calculate the fitness value of the particle fit n =pb n *ψ n *V+pb n ;

[0099] Among them, * is the multiplication sign;

[0100] Note: Fitness refers to the degree of performance of a solution in the optimization process, usually expressed as a numerical value, namely the fitness value;

[0101] Note: The smaller the fitness value given in step 264, the better. This calculation function uses the penalty weight V to punish the situation where defects occur due to too small blowing delay;

[0102] Step 255: Update the optimal fitness value fo of particle n n =fit n , update the individual optimal position po of particle n n =pb n ;

[0103] Step 256: Update the particle count variable n=n+1, and return to step 252;

[0104] Step 257: Get the optimal fitness value OF and the particle number j that reaches the optimal fitness value, [OF, j] = min(fit);

[0105] The min(·) refers to finding the minimum value of the array in the brackets and the element number of the minimum value in the array;

[0106] Step 258: Determine whether OF is greater than V. If so, it means that all the currently tested air blowing delays will cause defects, and the optimal air blowing delay is set to the safe value OT. b =T ba If not, set the optimal blowing delay OT b =pb j ;

[0107] Among them, Pb j is the position of particle j.

[0108] The other steps and parameters are the same as those in the first embodiment.

[0109] Specific implementation method three: This implementation method is different from specific implementation methods one or two in that in step three, it is determined whether the circuit board counting variable k is less than or equal to the total number of circuit boards K;

[0110] If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5.

[0111] If not, it means that all circuit boards have been mounted and production has stopped;

[0112] The specific process is:

[0113] Step 31: Update the mounted circuit board count variable k=k+1;

[0114] Step 32: determine whether the circuit board counting variable k is less than or equal to the total number of circuit boards K;

[0115] If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5.

[0116] If yes, execute step 33;

[0117] Step 3: All circuit boards have been mounted and production is stopped.

[0118] The other steps and parameters are the same as those in the first or second embodiment.

[0119] Specific implementation method 4: This implementation method is different from any one of specific implementation methods 1 to 3 in that in step 4, a mounting height adaptive optimization method based on a Bayesian optimization algorithm is executed to update the mounting height compensation value; the specific process is:

[0120] Step 41: According to the observation results, the mounting height historical value array δz′ and the defect rate historical value array r are fitted using the square exponential check to obtain the Gaussian process regression model gprm=GPRM(δz′,r);

[0121] Among them, GPRM is the Gaussian process regression model, gprm is the Gaussian process regression model;

[0122] The observation result refers to the defect rate r when the mounting height is δz′;

[0123] Kernel functions are used to describe the correlation between data points. The squared exponential kernel is the most commonly used kernel function, which assumes that data points close to each other are highly correlated, and as the distance increases, the correlation decays exponentially.

[0124] Note: Gaussian Process regression Model (GPRM), Gaussian process is a powerful non-parametric machine learning method, commonly used in regression and classification tasks; it predicts new data points by using the correlation (covariance function) between data points; the Gaussian process model can not only predict the mean of the observation point, but also give the standard deviation of this prediction;

[0125] Step 42: According to the Gaussian process regression model gprm obtained by fitting in step 41, predict the predicted mean μ(ob) and standard deviation v(ob) at each observation point SZ(ob);

[0126] Where ob∈[1:OB] is the element index in the observation point array SZ;

[0127] Step 43: Calculate rb = min(r), d = (rb - μ - ξ). / v;

[0128] The “. / ” means that the elements in the array before and after the symbol are divided one by one, and the quotient is stored as an array;

[0129] Among them, rb is the minimum defect rate in the search defect rate history value array, r is the defect rate history value array, d is an array (used to guide the improvement amount of the next decision of the Bayesian optimization algorithm, reflecting the possible improvement degree at the current position compared with the current known optimal solution;), ξ is the exploration-utilization parameter, which is a parameter that weighs the exploration of new solutions (exploration) and the utilization of known optimal solutions (utilization);

[0130] μ is the predicted mean array, v is the standard deviation array;

[0131] Step 44. Calculate the expected improvement EI = d*Φ(d)+v*φ(d);

[0132] Where Φ(d) is the cumulative distribution function of the standard normal distribution, and φ(d) is the probability density function of the standard normal distribution;

[0133] Note: The cumulative distribution function Φ(d) of the standard normal distribution describes the cumulative probability that the random variable is less than or equal to each point in the array d under the standard normal distribution;

[0134] The probability density function φ(d) of the standard normal distribution describes the probability density of each point in the array d under the standard normal distribution;

[0135] Step 4 and 5: Weight the EI within the range of the optimal value; the specific process is:

[0136] Note: Steps 4 and 5 are intended to increase the probability of selecting observation points within the potential optimal value region, thereby more efficiently locating the optimal placement height;

[0137] Step 451: Initialize observation point index array obs = [1:OB];

[0138] obs=[1:OB] is an array used to refer to all observation points in the process of searching for placement height, where the elements are 1:2:…:OB, ob is the index variable of the placement height observation point, and ob∈[1:OB] is a value in the range of 1:2:…:OB;

[0139] Step 452: Get the minimum index outside the range of the optimal value ob1 = find (SZ <SS δZ (1)) and the maximum index ob2 = find(SZ>SS δZ (2));

[0140] The optimal value is the optimal mounting height that the optimization algorithm aims to achieve, and the range of the optimal value is the interval where the optimal mounting height may exist;

[0141] Among them, SZ is the observation point array, SS δZ(2) The search range for the mounting height δZ is SS δZ Upper limit, SS δZ (1) The search range for the mounting height δZ is SS δZ The lower limit of

[0142] The find(·) function indicates obtaining the element numbers in the array in the following brackets that satisfy the conditions in the brackets.

[0143] Step 453: Delete the index outside the range of the optimal value obs([ob1:ob2])=[];

[0144] The [] is an empty array, and the ob1 and ob2 elements in the array are assigned to [], which means clearing the ob1 and ob2 elements;

[0145] Step 454: Weight the EI within the range of the optimal value, EI(obs)=EI(obs)*V;

[0146] The index of the element in the range is obs, and EI(obs) is weighted. The weighted EI also includes other elements besides obs, but those elements remain unchanged, and only the elements indexed by obs are weighted;

[0147] EI=d*Φ(d)+v(obs)*φ(d), EI is an array, use obs as the index to get the elements corresponding to these indexes in EI;

[0148] Step 46: Select the next placement height value δZ from the observation point corresponding to the maximum expected improvement; the specific process is:

[0149] Step 461: Obtain the maximum value of the expected improvement mei=max(EI);

[0150] The max(·) means obtaining the maximum value in the array in brackets;

[0151] EI represents an array, all elements are present, including weighted EI (obs) and other unweighted EI ();

[0152] Step 462, obtain the index array iδzs=find(EI==mei) of all elements equal to the maximum value mei of the expected improvement;

[0153] Among them, == represents a relational operator, which is used to compare whether two values ​​are equal;

[0154] Step 463, obtain the number of elements in the array iδzs nδzs = length(iδzs);

[0155] Step 464, randomly select the index of the next observation point iδz=ceil(rand(1)*nδzs);

[0156] rand(1) means to randomly get a real number in (0,1); ceil(·) means to round up the array in the brackets;

[0157] Step 465: Update the mounting height compensation value δZ=δz′(iδz).

[0158] The other steps and parameters are the same as those in Specific Embodiments 1 to 3.

[0159] Specific implementation method 5: This implementation method is different from any one of specific implementation methods 1 to 4 in that in step 5, a blowing delay adaptive optimization method based on a particle swarm optimization algorithm is executed to update the placement blowing delay; the specific process is:

[0160] Step 51, initialize the particle counting variable n=1;

[0161] Step 52: Determine whether n is less than N, if so, execute step 53, otherwise execute step 6;

[0162] Step 53: Update the position and velocity of the nth particle; the specific process is:

[0163] Step 531: Determine whether n is equal to j, if so, execute step 54, otherwise execute step 532;

[0164] Note: Considering the uncertainty in the placement process, the optimal blowing delay for each round remains unchanged in order to more fully and thoroughly evaluate the placement effect of the blowing delay;

[0165] Step 532: Update particle velocity vb n =w*vb n +c1*rand(1)*(po n -pb n )+c2*rand(1)*(OT b -pb n );

[0166] Step 533: Update particle position pb n =pb n +vb n ;

[0167] Step 54: Update the blowing delay of the nth blowing test point to pb n ;

[0168] Step 55: Update the particle count variable n=n+1, and return to step 52.

[0169] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0170] Specific implementation method 6: This implementation method is different from specific implementation methods 1 to 5 in that: in step 6, based on the surface mounting process parameters updated in steps 4 and 5, the next circuit board is mounted, and the mounted circuit board is subjected to automatic optical inspection; the specific process is:

[0171] Based on the surface mounting process parameters updated in step 4 and step 5, the mounting of the kth circuit board is completed; automatic optical inspection is performed on the mounted circuit board; and the result of the automatic optical inspection is obtained;

[0172] The results of automatic optical inspection include: Total number of placement defects D 0 , the offset of each mounting point∈ X ,∈ Y ,∈ θ , defect sources;

[0173] The sources of defects are divided into: defects caused by too high mounting height, mounting defects caused by too short blowing delay, and mounting defects caused by too large mounting offset.

[0174] The other steps and parameters are the same as those in Specific Implementation Methods 1 to 5.

[0175] Specific implementation method 7: This implementation method is different from specific implementation methods 1 to 6 in that: in step 7, the parameters of the mounting height adaptive optimization method based on the Bayesian optimization algorithm are updated according to the automatic optical detection results of step 6; the specific process is:

[0176] Step 71: According to the automatic optical inspection results, determine whether the defect is caused by too high a mounting height. If so, update the upper limit of the mounting height δZ search range and assign SS δZ (2) = δZ; if not, update the lower limit of the search range of the mounting height δZ and assign SS δZ (1) = δZ;

[0177] Step 7.2: Calculate the defect rate R = D 0 / D l ; Update the number of mounting defects D of the previous circuit board l =D 0 ;

[0178] Step 73: Update the mounting height history value array δz′=[δz′δZ];

[0179] Step 74: Update the defect rate history value array r = [r R].

[0180] The other steps and parameters are the same as those in Specific Embodiments 1 to 6.

[0181] Specific implementation eight: This implementation differs from specific implementations one to six in that: in step eight, the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm are updated according to the automatic optical detection results of step six; the specific process is:

[0182] Step 81, initialize the particle counting variable n=1;

[0183] Step 82, determine whether n is less than or equal to N, if so, execute step 83, otherwise execute step 87;

[0184] Step 83: According to the automatic optical inspection results, determine whether the nth blowing test point has a mounting defect caused by a short blowing delay or a mounting defect caused by a large mounting offset. If so, set ψ n =1; if not, set ψ n =0;

[0185] Step 84: Calculate the fitness value of the particle fit n =pb n *ψ n *V+pb n ;

[0186] Step 85: Determine fit n Is it less than fo n , if the optimal fitness value of particle n is updated fo n =fit n , update the individual optimal position po of particle n n =pb n , otherwise it will not be updated;

[0187] Step 86: Update the particle count variable n=n+1, and return to step 82;

[0188] Step 87: Get the optimal fitness value OF and the particle number j that reaches the optimal fitness value, [OF, j] = min(fit);

[0189] Step 88: Determine whether OF is greater than V. If so, it means that all the currently tested air blowing delays will cause defects, and the optimal air blowing delay is set to the safe value OT. b =T ba If not, set the optimal blowing delay OT b =pb j ;

[0190] Step 89: Reduce the blowing delay T b Search range SS Tb .

[0191] The other steps and parameters are the same as those in Specific Embodiments 1 to 7.

[0192] Specific implementation method 9: This implementation method is different from specific implementation methods 1 to 8 in that: the blowing delay T is shortened in steps 89 b Search range SS Tb ; The specific process is:

[0193] Step 891: Obtain the test point index array ns=find(ψ==1) where the blowing delay is too short;

[0194] Among them, ψ represents the array of whether there are mounting defects at all air blowing test points;

[0195] Step 892: Update search range T b Nether SS Tb (1) = max([pb ns SS Tb (1)]);

[0196] Among them, pb ns Expressed as the position of the particle in ns, SS Tb (1) indicates T b The lower bound of

[0197] Step 14 gives SS Tb , is an array of two values, where the first element is updated, and the first element represents the lower bound of the blowing delay search;

[0198] Step 893: Determine whether OF is less than V. If so, update the search range T. b The upper bound of SS Tb (2) = pb j ; If not, update SS Tb (2) = SS Tb (2)+1;

[0199] Among them, SS Tb (2) indicates T b The upper bound of .

[0200] Step 14 gives SS Tb , is an array of two values, where the second element is updated, and the second element represents the upper bound of the blowing delay search.

[0201] The other steps and parameters are the same as those in Specific Embodiments 1 to 8.

[0202] Specific implementation method ten: This implementation method is different from specific implementation methods one to eight in that: in step nine, according to the offset of each mounting point detected by automatic optical detection in step six, a mounting alignment adaptive optimization method based on a moving average method is executed to update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value; the specific process is:

[0203] Step 91: Get the cache array length n B =length(B X );

[0204] Step 92: Determine n B Is it equal to W? If so, update the X-axis offset accumulated value ε X =ε X -B X (1) Y-axis offset accumulated value ε Y =ε Y -B Y (1) The accumulated value of the θ-axis offset ε θ =ε θ -B θ (1) Delete the first element B in the cache array X (1) = [], B Y (1) = [], B θ (1) = []; otherwise no update is performed;

[0205] Among them, B X (1) indicates the first one in the history data of X-axis placement offset when no compensation is stored, B Y (1) indicates the first one in the history data of Y-axis placement offset when no compensation is stored, B θ (1) indicates the first one of the historical data of the θ-axis mounting offset when not compensated, wherein [] is an empty array, and assigning certain elements in the array to [] indicates clearing these elements;

[0206] Step 93: Add the offset of each mounting point to the cache array: B X =[B X (∈ X +δX)], B Y =[B Y (∈ Y +δY)], B θ =[B θ (∈ θ +δθ)];

[0207] Among them, B X =[B X (∈ X +δX)] means to convert (∈ X +δX) into array B XAs the last element; B Y =[B Y (∈ Y +δY)] means to convert (∈ Y +δY) into array B Y As the last element; B θ =[B θ (∈ θ +δθ)] means to convert (∈ θ +δθ) into array B θ as the last element;

[0208] Step 94: Update the X-axis offset accumulated value ε X =ε X +∈ X , Y-axis offset accumulated value ε Y =ε Y +∈ Y , θ axis offset accumulated value ε θ =ε θ +∈ θ ;

[0209] Step 95: Update X-axis offset compensation value δX=ε X / n B , Y-axis offset compensation value δY=ε Y / n B 、θ axis deflection compensation value δZ=ε θ / n B ;

[0210] Step 96: Return to step 31.

[0211] The other steps and parameters are the same as those in Specific Embodiments 1 to 9.

[0212] The following examples are used to verify the beneficial effects of the present invention:

[0213] Embodiment 1:

[0214] The surface mount process parameter adaptive optimization method based on automatic optical detection feedback in this embodiment is specifically prepared according to the following steps:

[0215] The example considers the process of optimizing the surface mounting process parameters in collaboration between a placement machine and an automatic optical inspection device. The components on the circuit board used for the placement test use a 0201 package, which has the smallest component size. When the placement machine runs at the highest speed, it is more likely to cause misalignment and defects when assembling such components. Gong conducted 10 placement experiments and calculated the average value to evaluate the optimization effect.

[0216] The present invention is used to adaptively optimize surface mounting process parameters based on automatic optical detection feedback. Experimental results show that the method of the present invention makes an average of 5.7 adjustments in the process of adjusting the mounting height to eliminate defects, during which a total of 88 defects are caused; the traditional manual adjustment method makes an average of 6.5 adjustments in the process of adjusting the mounting height to eliminate defects, during which a total of 192 defects are caused; in the method of improving mounting accuracy, the method of the present invention is used to increase the Cpk value (the larger the better) from 0.5 to 1.87; without causing a decrease in mounting performance, the method of the present invention shortens the mounting blowing delay from 17.5ms to 15.9ms.

[0217] The present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

[0218] Table 1 Circuit board data files

[0219] Serial number X-axis coordinate (mm) Y axis coordinate (mm) θ axis angle (°) 1 12 12 0 2 14.5 12 90 3 17.75 12 180 4 21.75 12 270 … … … …

[0220] Table 2 Surface mount process parameters

[0221]

[0222] The present invention may also have many other embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art may make various corresponding changes and modifications based on the present invention, but these corresponding changes and modifications should all fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for adaptive optimization of surface mount process parameters based on automatic optical inspection feedback, characterized in that: The specific process of the method is: Step 1: Obtain the data of the circuit board to be mounted, initialize the surface mounting process parameters and the search range of the surface mounting process parameters, initialize the parameters of the mounting height adaptive optimization method based on the Bayesian optimization algorithm, the parameters of the mounting alignment adaptive optimization method based on the moving average method, and initialize the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm; the specific process is: Step 11, obtaining a data file of a circuit board to be mounted, the data file of the circuit board to be mounted including the serial number, X-axis coordinate, Y-axis coordinate, and θ-axis angle of each mounting point; Step 12, set the total number K of circuit boards to be mounted; Step 13: Initialize surface mount process parameters; The surface mounting process parameters include the X-axis offset compensation value δX, the Y-axis offset compensation value δY, the θ-axis deflection compensation value δθ, the mounting height compensation value δZ, the mounting blowing delay T b ; Step 14: Initialize the search range of surface mount process parameters. The search range of δX is SS δX =[-100,100], the search range of δY is SS δY =[-100,100], the search range of δθ is SS δθ =[-1,1], the search range of δZ is SS δZ =[-500,500], T b The search range is SS Tb =[5,50]; SS δX The unit is μm, SS δY The unit is μm, SS δθ The unit is °, SS δZ The unit is μm, SS Tb The unit is ms; Step 15: For the placement height adaptive optimization method based on the Bayesian optimization algorithm, initialize the exploration-utilization parameter ξ=3, the initial placement height compensation value δZ=0; initialize the number of placement height observation points OB=30: SZ=linspace(SS δZ ,OB); OB is the total number of observation points of mounting height in the Bayesian optimization algorithm, SZ is the observation point array, linspace(SS δZ ,OB) refers to the δZ Generate an array of OB elements at equal intervals within the range; Step 16: For the placement alignment adaptive optimization method based on the moving average method, initialize the size of the moving average window W = 20; the X-axis offset accumulated value ε X =0, Y-axis offset accumulated value ε Y =0, θ axis offset accumulated value ε θ =0; Step 17: For the adaptive optimization method of blowing delay based on particle swarm optimization algorithm, initialize the number of particles N = 5, randomly select N mounting points for blowing delay test, particle counting variable n = 1, individual learning factor c1 = 1 for each particle, social learning factor c2 = 1 for each particle, inertia factor w = 0.5, in the search space SS Tb In the example, randomly initialize the position pb of each particle n and speed vb n ; Step 2: Based on the initialized surface mounting process parameters, the first circuit board is mounted, and automatic optical inspection is performed on the mounted circuit board. According to the automatic optical inspection results, the mounting height adaptive optimization method parameters based on the Bayesian optimization algorithm are updated, the mounting alignment adaptive optimization method parameters based on the moving average method are updated, the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value are updated, and the blowing delay adaptive optimization method parameters based on the particle swarm optimization algorithm are updated; Step 3, determine whether the circuit board counting variable k is less than or equal to the total number of circuit boards K; If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5. If not, it means that all circuit boards have been mounted and production has stopped; Step 4: Execute the mounting height adaptive optimization method based on the Bayesian optimization algorithm to update the mounting height compensation value; Step 5: Execute the blowing delay adaptive optimization method based on the particle swarm optimization algorithm to update the placement blowing delay; Step 6: Based on the surface mounting process parameters updated in step 4 and step 5, mount the next circuit board, and perform automatic optical inspection on the mounted circuit board; Step 7: Update the parameters of the mounting height adaptive optimization method according to the automatic optical detection result of step 6; Step 8: updating the parameters of the blowing delay adaptive optimization method according to the automatic optical detection result of step 6; Step nine: according to the offset of each mounting point detected by automatic optical detection in step six, execute the mounting alignment adaptive optimization method based on the moving average method to update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value.

2. The method for adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 1, characterized in that: In the step 2, based on the initialized surface mounting process parameters, the first circuit board is mounted, the mounted circuit board is subjected to automatic optical detection, and the mounting height adaptive optimization method based on the Bayesian optimization algorithm is updated according to the automatic optical detection result, the mounting alignment adaptive optimization method parameters based on the moving average method are updated, the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value are updated, and the blowing delay adaptive optimization method parameters based on the particle swarm optimization algorithm are updated; the specific process is: Step 21. Initialize the surface mount process parameters, including: initial mount height compensation value δZ = 0, initialize X-axis offset compensation value δX = 0, Y-axis offset compensation value δY = 0, θ-axis deflection compensation value δθ = 0, set the blowing delay of each blowing test point to pb n , the blowing delay of other mounting points is set to a safe value T ba =SS Tb (2); Initialize the mounting height history value array δz′=δZ; The SS Tb (2) Blowing delay T for mounting b The upper limit of the search range; n∈[1:N]; Step 22: Based on the initialized surface mount process parameters, complete the mounting of the first circuit board; set the mounted circuit board count variable k=1, perform automatic optical inspection on the mounted circuit board, and obtain the result of the automatic optical inspection; The results of automatic optical inspection include the offset of each placement point∈ X ,∈ Y ,∈ θ , total number of mounting defects D 0 , defect sources; defect sources are divided into: defects caused by too high mounting height, mounting defects caused by too short blowing delay, and mounting defects caused by too large mounting offset; Step 2 and 3: Update the parameters of the placement height adaptive optimization method based on the Bayesian optimization algorithm according to the automatic optical detection results; the specific process is: Step 231: According to the automatic optical inspection results, determine whether the defect is caused by the placement height being too high. If so, update the upper limit of the placement height δZ search range and assign SS δZ (2) = δZ; if not, update the lower limit of the search range of the mounting height δZ and assign SS δZ (1) = δZ; Among them, SS δZ (2) The search range for the mounting height δZ is SS δZ Upper limit, SS δZ (1) The search range for the mounting height δZ is SS δZ The lower limit of Step 232: Calculate the defect rate R = D 0 / P; Initialize defect rate history value array r=R; Initialize the number of placement defects D of the previous circuit board l =D 0 ; Where P is the total number of mounting points on the circuit board; Step 24: Update the parameters of the placement alignment adaptive optimization method based on the moving average method according to the automatic optical detection results, and update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value; the specific process is: Step 241: Add the offset of each mounting point to the cache array: B X =[B X (∈ X +δX)], B Y =[B Y (∈ Y +δY)], B θ =[B θ (∈ θ +δθ)]; Wherein, δX represents the X-axis offset compensation value, δY represents the Y-axis offset compensation value, and δθ represents the θ-axis deflection compensation value; ∈ X ,∈ Y ,∈ θ Indicates the offset of each mounting point; B X Represents the cache array for storing the historical data of the X-axis placement offset when not compensated, B Y Represents the cache array for storing the historical data of the Y-axis placement offset when not compensated, B θ Represents a cache array for storing historical data of theta-axis mounting offset when not compensated; Step 242: Update the X-axis offset accumulated value ε X =ε X +∈ X +δX, Y axis offset accumulated value ε Y =ε Y +∈ Y +δY, θ axis offset accumulated value ε θ =ε θ +∈ θ +δθ; Step 243: Get the cache array length n B =length(B X ); Among them, length() means getting the number of elements in the array in brackets; Step 244: Update X-axis offset compensation value δX=ε X / n B , Y-axis offset compensation value δY=ε Y / n B 、θ axis deflection compensation value δZ=ε θ / n B ; Step 25: Update the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm according to the automatic optical detection results; the specific process is: Step 251: Initialize particle count variable n = 1, set penalty weight value V = 10 90 ; Step 252: determine whether n is less than N, if so, execute step 253, otherwise execute step 257; Step 253: According to the automatic optical inspection results, determine whether the nth blowing test point has a mounting defect caused by a short blowing delay or a mounting defect caused by a large mounting offset. If so, set ψ n =1; if not, then ψ n =0; Among them, ψ n is a binary variable, indicating whether there is a mounting defect at the nth blowing test point; Step 254: Calculate the fitness value of the particle fit n =pb n *ψ n *V+pb n ; Among them, * is the multiplication sign; Step 255: Update the optimal fitness value fo of particle n n =fit n , update the individual optimal position po of particle n n =pb n ; Step 256: Update the particle count variable n=n+1, and return to step 252; Step 257: Get the optimal fitness value OF and the particle number j that reaches the optimal fitness value, [OF, j] = min(fit); The min(·) is to find the minimum value of the array in the brackets and the element number of the minimum value in the array; Step 258: Determine whether OF is greater than V. If so, it means that all the currently tested air blowing delays will cause defects, and the optimal air blowing delay is set to the safe value OT. b =T ba If not, set the optimal blowing delay OT b =pb j ; Among them, Pb j is the position of particle j.

3. The method for adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 2, characterized in that: In the step 3, it is determined whether the circuit board counting variable k is less than or equal to the total number of circuit boards K; If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5. If not, it means that all circuit boards have been mounted and production has stopped; The specific process is: Step 31: Update the mounted circuit board count variable k=k+1; Step 32: determine whether the circuit board counting variable k is less than or equal to the total number of circuit boards K; If yes, determine the automatic optical inspection result. If the automatic optical inspection result is a mounting defect caused by too high a mounting height, execute step 4. If the automatic optical inspection result is a mounting defect caused by too short a blowing delay or too large a mounting offset, execute step 5. If yes, execute step 33; Step 3: All circuit boards have been mounted and production is stopped.

4. The method for adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 3, characterized in that: In the step 4, a mounting height adaptive optimization method based on a Bayesian optimization algorithm is executed to update the mounting height compensation value; the specific process is as follows: Step 41: According to the observation results, the mounting height historical value array δz′ and the defect rate historical value array r are fitted using the square exponential check to obtain the Gaussian process regression model gprm=GPRM(δz′,r); Among them, GPRM is the Gaussian process regression model, gprm is the Gaussian process regression model; The observation result refers to the defect rate r when the mounting height is δz′; Step 42: According to the Gaussian process regression model gprm obtained by fitting in step 41, predict the predicted mean μ(ob) and standard deviation v(ob) at each observation point SZ(ob); Where ob∈[1:OB] is the element index in the observation point array SZ; Step 43: Calculate rb = min(r), d = (rb - μ - ξ). / v; The . / means that the elements in the array before and after the symbol are divided one by one, and the quotient is stored as an array; Among them, rb is the minimum defect rate in the search defect rate history value array, r is the defect rate history value array, d is an array, and ξ is the exploration-exploitation parameter; μ is the predicted mean array, v is the standard deviation array; Step 44. Calculate the expected improvement EI = d*Φ(d)+v*φ(d); Where Φ(d) is the cumulative distribution function of the standard normal distribution, and φ(d) is the probability density function of the standard normal distribution; Step 4 and 5: Weight the EI within the range of the optimal value; the specific process is: Step 451: Initialize observation point index array obs = [1:OB]; Step 452: Get the minimum index outside the range of the optimal value ob1 = find (SZ <SS δZ (1)) and the maximum index ob2 = find(SZ>SS δZ (2)); Among them, SZ is the observation point array, SS δZ (2) The search range for the mounting height δZ is SS δZ Upper limit, SS δZ (1) The search range for the mounting height δZ is SS δZ The lower limit of The find(·) function indicates obtaining the element numbers in the array in the following brackets that satisfy the conditions in the brackets. Step 453: Delete the index outside the range of the optimal value obs([ob1:ob2])=[]; The [] is an empty array; Step 454: Weight the EI within the range of the optimal value, EI(obs)=EI(obs)*V; Step 46: Select the next placement height value δZ from the observation point corresponding to the maximum expected improvement; the specific process is: Step 461: Obtain the maximum value of the expected improvement mei=max(EI); The max(·) means obtaining the maximum value in the array in brackets; Step 462, obtain the index array iδzs=find(EI==mei) of all elements equal to the maximum value mei of the expected improvement; Step 463, obtain the number of elements in the array iδzs nδzs = length(iδzs); Step 464, randomly select the index of the next observation point iδz=ceil(rand(1)*nδzs); rand(1) means to randomly get a real number in (0,1); ceil(·) means to round up the array in the brackets; Step 465: Update the mounting height compensation value δZ=δz′(iδz).

5. The method for self-adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 4, characterized in that: In the step 5, the blowing delay adaptive optimization method based on the particle swarm optimization algorithm is executed to update the placement blowing delay; the specific process is: Step 51, initialize the particle counting variable n=1; Step 52: Determine whether n is less than N, if so, execute step 53, otherwise execute step 6; Step 53: Update the position and velocity of the nth particle; the specific process is: Step 531: Determine whether n is equal to j, if so, execute step 54, otherwise execute step 532; Step 532: Update particle velocity vb n =w*vb n +c1*rand(1)*(po n -pb n )+c2*rand(1)*(OT b -pb n ); Step 533: Update particle position pb n =pb n +vb n ; Step 54: Update the blowing delay of the nth blowing test point to pb n ; Step 55: Update the particle count variable n=n+1, and return to step 52.

6. The method for self-adaptive optimization of surface mount process parameters based on automatic optical inspection feedback according to claim 5, characterized in that: In step 6, based on the surface mounting process parameters updated in step 4 and step 5, the next circuit board is mounted, and automatic optical inspection is performed on the mounted circuit board; the specific process is: Based on the surface mounting process parameters updated in step 4 and step 5, the mounting of the kth circuit board is completed; and automatic optical inspection is performed on the mounted circuit board; Obtaining results of automated optical inspection; The results of automatic optical inspection include: Total number of placement defects D 0 , the offset of each mounting point∈ X ,∈ Y ,∈ θ , defect sources; The sources of defects are divided into: defects caused by too high mounting height, mounting defects caused by too short blowing delay, and mounting defects caused by too large mounting offset.

7. The method for self-adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 6, characterized in that: In the step 7, the parameters of the mounting height adaptive optimization method based on the Bayesian optimization algorithm are updated according to the automatic optical detection result of step 6; the specific process is: Step 71: According to the automatic optical inspection results, determine whether the defect is caused by too high a mounting height. If so, update the upper limit of the mounting height δZ search range and assign SS δZ (2) = δZ; if not, update the lower limit of the search range of the mounting height δZ and assign SS δZ (1) = δZ; Step 7.2: Calculate the defect rate R = D 0 / D l ; Update the number of mounting defects D of the previous circuit board l =D 0 ; Step 73: Update the mounting height history value array δz′=[δz′δZ]; Step 74: Update the defect rate history value array r = [r R].

8. The method for self-adaptive optimization of surface mount process parameters based on automatic optical detection feedback according to claim 7, characterized in that: In the step eight, the parameters of the blowing delay adaptive optimization method based on the particle swarm optimization algorithm are updated according to the automatic optical detection results of step six; the specific process is: Step 81, initialize the particle counting variable n=1; Step 82, determine whether n is less than or equal to N, if so, execute step 83, otherwise execute step 87; Step 83: According to the automatic optical inspection results, determine whether the nth blowing test point has a mounting defect caused by a short blowing delay or a mounting defect caused by a large mounting offset. If so, set ψ n =1; if not, set ψ n =0; Step 84: Calculate the fitness value of the particle fit n =pb n *ψ n *V+pb n ; Step 85: Determine fit n Is it less than fo n , if the optimal fitness value of particle n is updated fo n =fit n , update the individual optimal position po of particle n n =pb n , otherwise it will not be updated; Step 86: Update the particle count variable n=n+1, and return to step 82; Step 87: Get the optimal fitness value OF and the particle number j that reaches the optimal fitness value, [OF, j] = min(fit); Step 88: Determine whether OF is greater than V. If so, it means that all the currently tested air blowing delays will cause defects, and the optimal air blowing delay is set to the safe value OT. b =T ba If not, set the optimal blowing delay OT b =pb j ; Step 89: Reduce the blowing delay T b Search range SS Tb .

9. The method for self-adaptive optimization of surface mount process parameters based on automatic optical inspection feedback according to claim 8, characterized in that: In the step eight or nine, the blowing delay T is shortened b Search range SS Tb ; The specific process is: Step 891: Obtain the test point index array ns=find(ψ==1) where the blowing delay is too short; Among them, ψ represents the array of whether there are mounting defects at all air blowing test points; Step 892: Update search range T b Nether SS Tb (1) = max([pb ns SS Tb (1)]); Among them, pb ns Expressed as the position of the particle in ns, SS Tb (1) indicates T b The lower bound of Step 893: Determine whether OF is less than V. If so, update the search range T. b The upper bound of SS Tb (2) = pb j ; If not, update SS Tb (2) = SS Tb (2)+1; Among them, SS Tb (2) indicates T b The upper bound of .

10. The method for self-adaptive optimization of surface mount process parameters based on automatic optical inspection feedback according to claim 9, characterized in that: In the step nine, according to the offset of each mounting point detected by automatic optical detection in step six, the mounting alignment adaptive optimization method based on the moving average method is executed to update the X-axis offset compensation value, the Y-axis offset compensation value and the θ-axis deflection compensation value; the specific process is: Step 91: Get the cache array length n B =length(B X ); Step 92: Determine n B Is it equal to W? If so, update the X-axis offset accumulated value ε X =ε X -B X (1) Y-axis offset accumulated value ε Y =ε Y -B Y (1) The accumulated value of the θ-axis offset ε θ =ε θ -B θ (1) Delete the first element B in the cache array X (1) = [], B Y (1) = [], B θ (1) = []; otherwise no update is performed; Among them, B X (1) indicates the first one in the history data of X-axis placement offset when no compensation is stored, B Y (1) indicates the first one in the history data of Y-axis placement offset when no compensation is stored, B θ (1) indicates the first one of the historical data of the θ-axis mounting offset when not compensated, and the [] is an empty array; Step 93: Add the offset of each mounting point to the cache array: B X =[B X (∈ X +δX)], B Y =[B Y (∈ Y +δY)], B θ =[B θ (∈ θ +δθ)]; Among them, B X =[B X (∈ X +δX)] means to convert (∈ X +δX) into array B X As the last element; B Y =[B Y (∈ Y +δY)] means to convert (∈ Y +δY) into array B Y As the last element; B θ =[B θ (∈ θ +δθ)] means to convert (∈ θ +δθ) into array B θ as the last element; Step 94: Update the X-axis offset accumulated value ε X =ε X +∈ X , Y-axis offset accumulated value ε Y =ε Y +∈ Y , θ axis offset accumulated value ε θ =ε θ +∈ θ ; Step 95: Update X-axis offset compensation value δX=ε X / n B , Y-axis offset compensation value δY=ε Y / n B 、θ axis deflection compensation value δZ=ε θ / n B ; Step 96: Return to step 31.

Images