Method and apparatus for generating optimized parameters

Abstract

Embodiments of the present application provide a method and apparatus for generating optimized parameters. The method is suitable for generating optimized parameters related to solder paste printing in surface mount technology. In the method, a cluster parameter model is used to determine whether at least one input parameter deviates from at least one parameter cluster. A global parameter model is used to determine the printing result of a plurality of target parameters in the input parameter. A heuristic algorithm is used to select at least one candidate parameter from the target parameters. An individual parameter model is used to modify the candidate parameter. As a result, the optimized parameters with high operation cycle and low failure rate can be obtained.

Description

Technical Field

[0001] This invention relates to surface mounting technology (SMT), and more particularly to a method and apparatus for generating optimized parameters for solder paste processing. Background Technology

[0002] SMT solder paste printing involves numerous variables that can affect print quality. These variables include stencil type, aperture distribution, operating environment, and printer parameters. Print quality factors include solder paste placement and solder quantity on pads. On the other hand, Solder Paste Inspection (SPI) machines can inspect the solder paste quality on printed circuit boards (PCBs). Due to the high error rate of SPI machines, manual re-inspection is necessary. However, the inspection items for solder paste quality are numerous and complex. Generally, operators need to rely on personal experience to interpret PCB and SPI inspection data, increasing the manpower and time costs of PCB inspection. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide an apparatus and method for generating optimized parameters, which can generate optimal solder paste printing parameter settings, while ensuring that the process results of each pad of the stencil are normal under the optimized parameters.

[0004] The method for generating optimized parameters according to embodiments of the present invention includes (but is not limited to) the following steps: determining whether the input parameters deviate from the parameter group using a cluster parameter model, wherein each input parameter is related to the distribution of one or more apertures on the stencil, solder paste information, and / or process environment. Determining the printing results of multiple target parameters among one or more input parameters using a global parameter model. The global parameter model is trained by a first machine learning algorithm, and each target parameter is related to one or more process parameters of the solder paste operation on the stencil. Selecting candidate parameters from the target parameters using a heuristic algorithm. The process parameters among the target parameters are used as variables in the objective function of the heuristic algorithm. Correcting the candidate parameters using an individual parameter model. The individual parameter model is trained by a second machine learning algorithm based on the process results of the solder paste printing operation and the aperture size. The process results include the results achieved by using those process parameters on the solder pads corresponding to the apertures in the solder paste printing operation.

[0005] The optimization parameter generation apparatus of this invention includes (but is not limited to) a memory and a processor. The memory stores program code. The processor is coupled to the memory. The processor loads and executes the program code configured to determine, through a clustered parameter model, whether input parameters deviate from a parameter group, where each input parameter is related to the distribution of one or more apertures on a stencil, solder paste information, and / or process environment. The overall parameter model determines the printing results of multiple target parameters among the input parameters, a heuristic algorithm filters candidate parameters from the target parameters, and an individual parameter model refines the candidate parameters. The overall parameter model is trained using a first machine learning algorithm, and each target parameter is related to one or more process parameters of the solder paste operation on the stencil. Those process parameters among the target parameters serve as variables in the objective function of the heuristic algorithm. The individual parameter models are trained using a second machine learning algorithm based on the process results of the solder paste printing operation and the aperture size. The process results include the results achieved using those process parameters on the solder pads corresponding to the apertures in the solder paste printing operation.

[0006] Based on the above, the optimization parameter generation apparatus and method according to embodiments of the present invention infer target parameters for the entire steel plate / circuit board and individual hole diameters / pad layers based on multiple parameter models trained by multiple machine learning algorithms, appropriately screen suitable candidate parameters, and correct the candidate parameters. Thus, optimized parameters with high cycle time and low failure rate can be obtained, thereby improving operational efficiency and process yield.

[0007] To make the above features and advantages of the present invention more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings. Attached Figure Description

[0008] Figure 1 This is a block diagram of an apparatus for generating optimization parameters according to an embodiment of the present invention.

[0009] Figure 2 This is a flowchart of a method for generating optimization parameters according to an embodiment of the present invention.

[0010] Figure 3 This is a schematic diagram of clustering according to an embodiment of the present invention.

[0011] Figure 4 This is a schematic diagram of reinforcement learning according to an embodiment of the present invention.

[0012] The attached figures are labeled as follows:

[0013] 100: Optimization parameter generation device

[0014] 110: Storage

[0015] 120: Processor

[0016] S210~S240, S410~S430: Steps

[0017] x, y, z: axes

[0018] G1, G2: Groups

[0019] 301: Outlier

[0020] 302: Non-outlier value Detailed Implementation

[0021] Figure 1 This is a block diagram of an optimization parameter generation device 100 according to an embodiment of the present invention. Please refer to... Figure 1 The optimization parameter generation device 100 includes (but is not limited to) a memory 110 and a processor 120. The optimization parameter generation device 100 may be a desktop computer, a laptop computer, a smartphone, a tablet computer, a server, an image recognition device, a solder paste inspection (SPI) machine, a solder paste printing inspection instrument, or other computing device.

[0022] Storage 110 can be any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or similar component. In one embodiment, storage 110 is used to record program code, software modules, state configurations, data (e.g., input parameters, process parameters, candidate parameters, process results, pad-related parameters, solder paste printing parameters, etc.) or other files, as will be described in detail later.

[0023] Processor 120 is coupled to storage 110. Processor 120 may be a central processing unit (CPU), a graphics processing unit (GPU), or other programmable general-purpose or special-purpose microprocessor, digital signal processor (DSP), programmable controller, field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), neural network accelerator, or other similar components or combinations thereof. In one embodiment, processor 120 is used to perform all or part of the operations of generating apparatus 100, and can load and execute program code, software modules, files, and data recorded in storage 110.

[0024] The method described in this embodiment of the invention will be explained below in conjunction with the various devices, elements, and / or modules in the optimization parameter generation apparatus 100. The various processes of this method may be adjusted according to the implementation situation, and are not limited thereto.

[0025] Figure 2 This is a flowchart of a method for generating optimization parameters according to an embodiment of the present invention. Please refer to... Figure 2 The processor 120 first determines whether one or more input parameters deviate from one or more parameter groups using a clustering parameter model (step S210). Specifically, the clustering parameter model is trained using a machine learning algorithm related to clustering. Clustering methods (also known as grouping algorithms) can be k-means, Gaussian Mixture Model (GMM), Mean-Shift, Hierarchical clustering, Spectral clustering, DBSCAN (Density-based spatial clustering of applications with noise), or other clustering algorithms. Clustering methods can classify parameters and group similar parameters into the same parameter group.

[0026] Each input parameter is related to the pad distribution (or aperture distribution), solder paste information, and / or process environment. In one embodiment, the pad distribution can be an area ratio or a pad density. For example, the area ratio is the quotient obtained by dividing the area of ​​the pad (e.g., the product of its length and width) by the area of ​​the aperture wall (e.g., twice the sum of its length and width by the thickness of its sheet) by the area of ​​the aperture wall. Another example is that the pad density is a specific radius (e.g., a specific radius is set to 5 mm). The processor 120 can determine the skewness and kurtosis values ​​of the aperture volume ratio on the sheet and / or the aperture adjacent density (as aperture distribution) based on the pad distribution. In one embodiment, the aperture adjacent density can be the median of all aperture adjacent densities on the sheet. For example, Table (1) shows the relevant data for the pad distribution:

[0027] Table (1)

[0028]

[0029] The processor 120 can pre-process these data (e.g., length, width, and thickness) to derive volume ratios, densities, or other characteristic variables that can be used by the overall parametric model, the cluster parametric model, and the individual parametric model.

[0030] In addition, solder paste information can include type, supplier, particle size, and / or viscosity. For example, the type is Class 4, and supplier A provides the raw materials. The process environment can be the ambient humidity and temperature. For example, the temperature is 26.8 degrees Celsius, and the humidity is 44.4 percent.

[0031] In one embodiment, the processor 120 preprocesses the input parameters to obtain statistical data on the distribution of multiple apertures on the steel plate. For example, this includes the skewness of the weld pad and aperture distribution, the kurtosis of the area ratio of weld pads to apertures, and the neighborhood density of apertures in the steel plate.

[0032] In one embodiment, the cluster parameter model uses the distribution skewness, kurtosis, and adjacent density of apertures in the steel plate collected in the past or present as training samples. Specifically, the processor 120 statistically analyzes the distribution of the area ratio of the pads to the apertures in the steel plate / circuit board in the input parameters (e.g., area skewness, area skewness, and median neighbor in the steel plate), and uses the cluster parameter model to determine whether the current distribution of the input parameters is an outlier. That is, whether the input parameters belong to any parameter group. For example, Table (2) shows the distribution of aperture-related data in the steel plate:

[0033] Table (2)

[0034]

[0035] In other words, the processor 120 performs outlier detection on the current steel plate data and previously collected steel plate data to identify whether the steel plate under test is an outlier. If the steel plate data is an outlier, it means that the steel plate does not belong to the type of steel plate that can be estimated from the existing steel plate data. Conversely, if the steel plate is not an outlier, it means that the steel plate belongs to the type of steel plate that can be covered and estimated by the overall parameter model. Furthermore, if the values ​​of the steel plate under test are similar to or identical to the collected steel plate data, the processor 120 can estimate the printing condition and related data of the steel plate under test using the collected steel plate data.

[0036] For example, Figure 3 This is a schematic diagram of grouping according to an embodiment of the present invention. Please refer to... Figure 3 Each point represents statistical data on the distribution of apertures corresponding to multiple weld pads of a steel plate model. The x-axis represents the skewness of the area ratio of weld pads to apertures, the y-axis represents the kurtosis of the volume ratio of weld pads to apertures, and the z-axis represents the aperture density of the steel plate. The processor 120 uses a cluster parameter model trained with the DBSCAN algorithm to group the distribution density, kurtosis, and aperture density of the collected steel plate data into parameter groups (e.g., groups G1 and G2). If the data of the steel plate to be tested is input into the cluster parameter model and the distribution of the steel plate is found to be an outlier (301), it means that the steel plate to be tested belongs to a steel plate that cannot be inferred from the existing steel plate data. On the other hand, if the data of the steel plate to be tested is input into the cluster parameter model and the aperture condition of the steel plate to be tested belongs to a non-outlier (302, belonging to group G1) that can be inferred from the existing data, it means that the steel plate to be tested is the same as or similar to the existing data and can be inferred using the overall parameter model.

[0037] It is worth noting that the aperture sizes on different types of steel plates are all different. Therefore, in this stage (also known as the outlier identification stage), the collected historical steel plate data (i.e., steel plate aperture data) is used as training and comparison data. Outlier or outlier identification results are provided through a clustering parameter model. That is, the clustering parameter model based on the clustering algorithm can confirm that the steel plate under test belongs to a type of steel plate that can be covered by a model derived from existing steel plate related data (e.g., a global parameter model or an individual parameter model detailed below). In other words, even if the existing steel plate data (i.e., historical steel plate data) does not contain the same distribution as the steel plate under test, the clustering parameter model can still confirm that the steel plate under test has a high similarity to the existing steel plate data (i.e., it belongs to the outlier category), thus confirming that the process results or other process-related results of the steel plate under test can be inferred from historical steel plate data.

[0038] Please refer to Figure 2 The processor 120 determines the printing result of multiple target parameters for one or more input parameters using an overall parameter model (step S220). Specifically, this embodiment of the invention mainly targets solder paste printing parameters related to surface mount technology (SMT). The processor 120 can train the overall parameter model using a first machine learning algorithm. That is, the overall parameter model is related to the solder paste printing operation.

[0039] Machine learning algorithms analyze training samples to derive patterns, thereby predicting unknown data based on these patterns. The parametric model is the machine learning model constructed after learning, and it is used to infer from the data to be evaluated. The first machine learning algorithm of this invention can be a regression analysis algorithm, an extreme gradient boosting (XGboost) algorithm, a lightweight gradient boosting machine (LightGBM), a bootstrap aggregating (Bagged) algorithm, a neural network algorithm, a random forest algorithm, or other algorithms.

[0040] In one embodiment, the overall parameter model uses one or more process parameters related to solder paste printing operations on the stencil collected in the past or present as training samples. These parameters include, for example, squeegee pressure and speed, stencil demolding speed and distance, stencil cleaning frequency, the distribution and density of apertures in the stencil, solder paste particle size and viscosity, ambient temperature and humidity, and the corresponding printing results. These process parameters are all variables affecting the printing results. Therefore, the trained overall parameter model can be used to judge whether the printing results obtained using the process parameters are normal or abnormal (or successful or unsuccessful). That is, the overall parameter model is a binary classifier.

[0041] In one embodiment, the processor 120 can filter process parameters as training samples (i.e., process parameters) through feature extraction. Feature extraction constructs informative and non-redundant derived values ​​(or features) from the initially collected / extracted data parameters. Feature extraction can assist subsequent learning and rule induction processes and provide a better interpretation of the initial data parameters. In other words, feature extraction can simplify or compute the input parameters into a feature set, and the feature set can be directly used to perform subsequent tasks (e.g., model training, parameter analysis, printing result inference, etc.).

[0042] The target parameters are related to one or more process parameters of the stencil for solder paste printing. These target parameters can be input parameters, solder paste particle size, solder paste viscosity, ambient temperature, ambient humidity, solder paste stencil printing parameters, squeegee pressure, squeegee speed, stencil demolding speed, stencil demolding distance, and stencil cleaning frequency, etc. For example, Table (3) is an example illustrating the target parameters:

[0043] Table (3)

[0044]

[0045] It is worth noting that each circuit board has thousands or even tens of thousands of solder pads, and different solder pads may correspond to different apertures. Therefore, in this stage (also known as the circuit board stage), the circuit board or steel plate as a whole is initially evaluated. One or more target parameters suitable for this aperture distribution of the steel plate are initially provided through the overall parameter model. That is, printing operations based on the target parameters output by the overall parameter model can yield a result of normal or successful printing (i.e., printing result).

[0046] Processor 120 uses a heuristic algorithm to select candidate parameters from the target parameters (step S230). The heuristic algorithm can be a genetic algorithm, hill-climbing algorithm, simulated annealing, Tabu search algorithm, Great Deluge algorithm, or particle swarm optimization algorithm. A heuristic algorithm is an algorithm used to find the optimal solution in a function.

[0047] Each candidate parameter is related to the serial number, squeegee pressure, squeegee speed, slab demolding speed, slab demolding distance, slab cleaning frequency, pore size distribution in the slab, pore size density in the slab, solder paste particle size, solder paste particle viscosity, ambient temperature, ambient humidity, and / or printing results.

[0048] In one embodiment, the processor 120 uses process parameters from the target parameters as variables in the objective function of this heuristic algorithm. The process parameters are related to any set parameters in the solder paste printing operation. For example, process parameters may be squeegee pressure, squeegee speed, stencil demolding speed, stencil demolding distance, and / or stencil cleaning frequency.

[0049] In one embodiment, the processor 120 uses the stencil aperture distribution, stencil aperture density, solder paste particle size, solder paste particle viscosity, ambient temperature, and / or ambient humidity as fixed parameters. The processor 120 dynamically updates the process parameters, which are variables in its objective function, using a heuristic algorithm to derive a trend towards the optimal solution. Furthermore, the processor 120 inputs the fixed parameters and the updated / initial process parameters into an overall parameter model to determine the corresponding printing result (success or failure). If the overall parameter model determines the printing result to be a failure, the processor 120 will try other parameters. If the overall parameter model determines the printing result to be a success, the processor 120 will continue to adjust the parameters until the optimal solution is obtained. Accordingly, the processor 120 can filter parameters with successful printing results as candidate parameters.

[0050] In one embodiment, the heuristic algorithm is a particle swarm optimization (PSO) algorithm. The processor 120 can use the PSO algorithm to determine one or more optimal solutions to the objective function and use these optimal solutions as candidate parameters. For example, Table (4) illustrates the candidate parameters as an example:

[0051] Table (4)

[0052]

[0053] In the example in Table (4), the solder paste information and ambient temperature and humidity are fixed parameters. For example, the solder paste type is Class 4, the solder paste viscosity is 600, the ambient temperature is 26.2 degrees Celsius, and the ambient humidity is 45.2 percent.

[0054] Please refer to Figure 2 The processor 120 then refines the candidate parameters using an individual parameter model (step S240). Specifically, the processor 120 trains an individual parameter model using a second machine learning algorithm based on the process results of the solder paste printing operation and the aperture size. The second machine learning algorithm can be the same as or different from the first machine learning algorithm, and can be determined by the user according to their needs. Unlike the overall parameter model, the training samples of the individual parameter model take into account more the aperture size. For example, the area ratio of the pad to the aperture, the length and width of the aperture, and the aperture density. These parameters are all variables affecting the process results of the solder paste printing operation. The process results are the result caused by the process parameters on the solder pads corresponding to the apertures during the solder paste printing operation.

[0055] In one embodiment, the process results include whether the amount of solder paste for each aperture of each pad in the stencil printing is normal, excessive, or insufficient. For example, if these results are distinguished by error rate, 1.9% is normal, 38% is insufficient, and 59.2% is excessive. Notably, the processor 120 ensures, through an individual parameter model, that the process result for each aperture of each pad in the stencil corresponding to each candidate parameter is a normal amount of solder paste. Therefore, in this stage (also known as the pad stage), the processor 120 observes the pads and further confirms that the process result caused by the specific process parameters is that the amount of solder paste for one or more apertures on the stencil is normal.

[0056] Similarly, the individual parameter model uses one or more process parameters related to solder paste printing operations on the stencil collected in the past or present as training samples. These include, for example, squeegee pressure and speed, stencil demolding speed and distance, stencil cleaning frequency, stencil aperture distribution and density, solder paste particle size and viscosity, ambient temperature and humidity, pad and aperture volume ratio, aperture length and width, aperture density, and corresponding process results. These process parameters and aperture dimensions are all variables affecting the process results. Therefore, a trained individual parameter model can be used to evaluate the process results obtained using process parameters and aperture dimensions (e.g., solder quantity). That is, the individual parameter model is a multivariate classifier. In one embodiment, the processor 120 can filter process parameters and aperture dimensions as training samples.

[0057] For example, the input parameters and process results of individual parameter models are shown in Table (5):

[0058]

[0059] In other words, processor 120 infers that the amount of solder paste obtained using these process parameters is normal.

[0060] In one embodiment, the processor 120 can correct the process result corresponding to the candidate parameter to the normal solder amount by using a reinforcement learning algorithm based on the prediction result output by the individual parameter model (i.e., the predicted process result). Figure 4 This is a schematic diagram illustrating reinforcement learning according to an embodiment of the present invention. Please refer to... Figure 4 In reinforcement learning, the processor 120 formulates a policy (e.g., setting the action range to a scraper pressure between ±1 and a scraper speed between ±5) (step S410), predicts the process results corresponding to this policy and the state provided by the agent (e.g., solder paste machine parameters) in the environment through an individual parameter model (step S420), and gives a reward accordingly (e.g., the lower the error rate, the better) (step S430). Thus, it can be seen that the amount of solder paste on each aperture of the stencil corresponding to the candidate parameters corrected by reinforcement learning is normal.

[0061] In one embodiment, the process results also include operational efficiency, which is related to cycle time and / or failure rate. Cycle time is the duration of a printing operation and is related to the stencil release speed and stencil cleaning frequency. For example, cycle time is the sum of the square root of the reciprocal of the stencil release speed and the square of the stencil cleaning frequency. Failure rate is the proportion of holes with abnormal apertures on each solder pad in the stencil.

[0062] Processor 120 further refines the candidate parameters based on the prediction results output by the individual parameter model, using reinforcement learning algorithms to adjust the corresponding work efficiency and thus become the optimal one. Figure 4 For example, the prediction results (i.e., process results) obtained from individual parameter models are failure rate and / or cycle time. The rewards are based on the rule that "the lower the failure rate, the better, and the longer the cycle time, the better."

[0063] In one embodiment, the processor 120 iteratively processes the correction results of the candidate parameters. Figure 4 For example, the processor 120 can re-set the movement range based on the updated machine parameters corresponding to the previous machine parameters and simulate the process results in the environment. This process is repeated to achieve iterative processing.

[0064] Take Tables (6) and (7) as examples:

[0065] Table (6)

[0066] Serial Number scraper pressure scraper speed Steel plate demolding speed Steel plate demolding distance Steel plate cleaning frequency 1 6 90 0.1 2 7 2 6.5 85 0.12 2 8 3 6.2 85 0.14 1.8 8

[0067] Table (7)

[0068]

[0069] like Figure 4 As shown in Table (7), the processor 120 can adjust settings such as scraper pressure and speed, and steel plate removal speed and distance to set the movement range in the reinforcement learning algorithm, and sets the failure rate and operation cycle as the reward for optimization. That is, the lower the failure rate, the better, and the longer the operation cycle, the better. In addition, the processor 120 uses the solder paste machine parameters as all states in the environment. Accordingly, the processor 120 performs iterative processing on candidate parameters based on the above settings.

[0070] Table (7) shows the parameters obtained by performing 10 iterations, taking the first parameter in Table (6) as an example. The processor 120 inputs the candidate parameters into the individual parameter model, which then adjusts the values ​​of scraper pressure, scraper speed, steel plate demolding speed, steel plate demolding distance, and steel plate cleaning frequency (i.e., the movement in reinforcement learning). Next, the processor 120 inputs each parameter into the environment of the individual parameter model to obtain the corresponding job cycle and failure rate, and then gives a reward accordingly. After several iterations, the processor 120 can obtain the optimal solution with the lowest failure rate and the longest job cycle in the 9th iteration.

[0071] To illustrate the application scenario further, the generating device 100 receives input from the user regarding input parameters via an input interface (e.g., keyboard, mouse, or touchscreen). The processor 120, based on process parameters such as ambient temperature and humidity and solder paste type, uses a parameter model to estimate recommended process parameters: squeegee pressure 6.5, squeegee speed 88, stencil demolding speed 0.2, stencil demolding distance 2, and stencil cleaning frequency 9. Furthermore, using these process parameters, a failure rate of 0.0135 and a cycle time of 20 can be obtained in the printing operation.

[0072] In summary, in the optimization parameter generation apparatus and method of this invention, it is determined whether the steel plate under test is a non-outlier value to ensure that the steel plate under test belongs to the category of steel plates that can be inferred and predicted from existing steel plate data. This invention dynamically corrects the target parameters in the circuit board stage by combining heuristic algorithms and parameter models to obtain the corresponding machine setting parameters and environmental parameters (i.e., candidate parameters) for the steel plate under test. Furthermore, this invention combines reinforcement learning and parameter models in the solder pad stage to make the parameters approach the set process results. Therefore, the provided optimization parameters can achieve normal solder paste amounts for each aperture in the steel plate, resulting in high cycle time and low failure rate.

[0073] Although the present invention has been disclosed above with reference to embodiments, it is not intended to limit the present invention. Those skilled in the art can make some modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A method for generating optimized parameters for parameters related to solder paste printing in surface mount technology, characterized in that, The generation method includes: A group parameter model is used to determine whether at least one input parameter deviates from at least one parameter group, wherein each input parameter is related to at least one aperture distribution of a steel plate, solder paste information, and at least one of a process environment. The printing result of multiple target parameters in the at least one input parameter is determined by an overall parameter model, which is trained by a first machine learning algorithm, and each target parameter is related to at least one process parameter of a solder paste printing operation on the slab. At least one candidate parameter is selected from a plurality of said target parameters using a heuristic algorithm, wherein each of the at least one process parameter in the target parameters serves as a variable in an objective function of the heuristic algorithm; and The at least one candidate parameter is corrected by an individual parameter model, wherein the individual parameter model is trained by a second machine learning algorithm based on a process result of the solder paste printing operation and the size of the at least one aperture, and the process result includes the result of the solder pad corresponding to the at least one aperture in the solder paste printing operation using the at least one process parameter.

2. The method of claim 1, wherein the step of generating the optimized parameters is performed by a computer. The heuristic algorithm is a particle swarm optimization algorithm, and the step of selecting at least one candidate parameter from a plurality of said target parameters includes: The particle swarm optimization algorithm determines at least one optimal solution to the objective function, wherein the at least one optimal solution is used as the at least one candidate parameter.

3. The method of claim 1, wherein the step of generating the optimized parameters comprises the steps of: determining a plurality of parameters; and selecting a parameter from the plurality of parameters based on a predetermined criterion. The steps for modifying the at least one candidate parameter using the individual parameter model include: Based on a prediction result output by the individual parameter model, a reinforcement learning algorithm is used to correct the process result corresponding to at least one candidate parameter to achieve the normal tin content.

4. The method for generating the optimization parameters as described in claim 1, characterized in that, The process results also include an operational efficiency, which is related to at least one of operational cycle time and failure rate, and the step of correcting the at least one candidate parameter through the individual parameter model includes: Based on the prediction result output by the individual parameter model, the job efficiency corresponding to at least one candidate parameter is adjusted through a reinforcement learning algorithm to achieve the optimal result.

5. The method of claim 1, wherein the step of generating the optimized parameters further comprises: determining a plurality of parameters; and selecting a subset of the plurality of parameters based on a predetermined criterion. The clustering parameter model was trained using a third machine learning algorithm related to the clustering method.

6. The method of claim 1, wherein the step of generating the optimized parameters further comprises the step of: The distribution of at least one aperture includes the area ratio of the pads to the apertures of one of the plurality of steel plates and the density of the apertures therein. ​ 7. The method of claim 3 or 4, wherein the parameter is a parameter of a filter. The steps for correcting the at least one candidate parameter using the reinforcement learning algorithm include: The modification of at least one candidate parameter is performed in an iterative process.

8. An apparatus for generating optimization parameters, characterized in that, include: One memory, storing one piece of program code; as well as A processor, coupled to the memory, loads and executes the code configured to: A group parameter model is used to determine whether at least one input parameter deviates from at least one parameter group, wherein each input parameter is related to at least one aperture distribution of a steel plate, solder paste information, and at least one process environment. The printing result of multiple target parameters in the at least one input parameter is determined by an overall parameter model, which is trained by a first machine learning algorithm, and each target parameter is related to at least one process parameter of a solder paste printing operation on the slab. At least one candidate parameter is selected from a plurality of said target parameters by a heuristic algorithm, wherein the at least one process parameter in each of the target parameters is a variable in a target function in the heuristic algorithm; as well as The at least one candidate parameter is corrected by an individual parameter model, wherein the individual parameter model is trained by a second machine learning algorithm based on a process result of the solder paste printing operation and the size of the at least one aperture, and the process result includes the result of the solder pad corresponding to the at least one aperture in the solder paste printing operation using the at least one process parameter.

9. The apparatus for generating an optimized parameter of claim 8, wherein, The heuristic algorithm is a particle swarm optimization algorithm, and the processor is further configured to: The particle swarm optimization algorithm determines at least one optimal solution to the objective function, wherein the at least one optimal solution is used as the at least one candidate parameter.

10. The apparatus for generating an optimized parameter of claim 8, wherein, The processor is further configured to: Based on a prediction result output by the individual parameter model, a reinforcement learning algorithm is used to correct the process result corresponding to at least one candidate parameter to achieve the normal tin content.

11. The apparatus for generating an optimized parameter of claim 8, wherein, The process results also include an operating efficiency related to at least one of cycle time and failure rate, and the processor is further configured to: Based on a prediction result output by the individual parameter model, a reinforcement learning algorithm is used to correct the job efficiency corresponding to at least one candidate parameter to become the best.

12. The apparatus for generating an optimized parameter of claim 8, wherein, The clustering parameter model was trained using a third machine learning algorithm related to the clustering method.

13. The apparatus for generating optimization parameters as described in claim 8, characterized in that, The distribution of at least one aperture includes the area ratio of the pads to the apertures of one of the plurality of steel plates and the density of the apertures therein.

14. The apparatus for generating an optimized parameter according to claim 10 or 11, wherein The processor is further configured to: The modification of at least one candidate parameter is performed in an iterative process.

Images