SMT printing machine technological parameter intelligent optimization method based on multi-algorithm fusion

Abstract

The invention belongs to the technical field of equipment process parameter optimization, and particularly relates to an SMT printer process parameter intelligent optimization method based on multi-algorithm fusion, and the method comprises the steps: firstly carrying out the preliminary parameter screening and data collection through the DOE, and building a high-quality data set; building a printing quality prediction model by using an artificial intelligence algorithm Gaussian process regression model; and finally, an operation planning optimization algorithm is introduced, multi-target production parameter decision making is carried out under multiple constraint conditions, a system integrating the algorithms is established, target parameters are received through a man-machine interaction interface, an optimization process can be automatically executed, and an optimal parameter combination is output to the SMT printing machine. The method has a very good effect when solving the problem of printer parameter optimization in the existing SMT industry, and compared with the existing printer parameter optimization method, the method has higher efficiency and precision and stronger robustness, and can better achieve the effect of parameter optimization through data driving and dynamic adaptation.

Description

Technical Field

[0001] This invention belongs to the field of equipment process parameter optimization technology, specifically involving an intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion. Background Technology

[0002] With the rapid development of intelligent manufacturing, surface mount technology (SMT) has been widely used in the electronics manufacturing industry. SMT production lines typically consist of multiple high-precision devices, such as printers, pick-and-place machines, reflow ovens, and AOI (automatic optical inspection). SMT printing is the first critical process in the SMT production line; the quality of solder paste printing (such as solder paste volume, height, and area) directly affects the yield of subsequent placement and reflow soldering, as well as the reliability of the final product. Because numerous and interdependent key parameters affect printing quality, including but not limited to: squeegee speed, pressure, angle, and material; stencil tension, cleaning frequency, and aperture design; substrate support height and ejector pin position; ambient temperature and humidity; and solder paste viscosity and particle size, optimizing SMT printer parameters is extremely difficult, often resulting in compromises. Therefore, how to achieve an efficient, accurate, and intelligent parameter optimization method that comprehensively considers multiple objectives and constraints, and is flexible enough to adapt to the different production needs of different companies, is a pressing issue that needs to be addressed.

[0003] Currently, the main methods for optimizing SMT printer parameters on the market include: 1. Trial and error method: This method relies on engineer experience, is inefficient, costly (wasting solder paste and time), difficult to find the global optimum, and has poor stability. 2. Traditional DOE method: This method involves full factorial experiments, which leads to an explosion in the number of experiments when there are many parameters, making it impractical; partial factorial or response surface methodology can reduce the number of experiments, but still requires many experiments, resulting in high costs, long cycles, and difficulty in handling highly nonlinear relationships; it is also sensitive to noise and fluctuations, and the optimization results may not be robust enough. 3. Prediction through neural networks: This method requires a large amount of high-quality training data, which is costly to acquire initially, has a "black box" problem, poor interpretability, and low engineer trust. 4. Operations research optimization method: This method usually requires an accurate mathematical model, but it is difficult to establish an accurate physical model of the SMT printing process, has limited ability to handle complex nonlinear relationships, and lacks effective integration with physical experiments and domain knowledge.

[0004] In general, existing technologies mainly suffer from the following problems: 1. High dependence on manual labor: Current parameter tuning methods rely heavily on engineers' experience, resulting in low efficiency, difficulty in finding the global optimum, and poor consistency. 2. High cost of parameter optimization: Current full factorial or response surface methodology methods require an exponential increase in the number of experiments as the number of parameters increases, leading to high costs, long cycles, and the ability to find the optimal value only within a small range. Summary of the Invention

[0005] This invention aims to provide an intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion, which improves the efficiency, accuracy, and practicality of SMT printing machine parameter setting, and enhances the first-pass yield and quality stability of solder paste printing process.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for intelligent optimization of SMT printing machine process parameters based on multi-algorithm fusion is provided, including: Step (1): Conduct preliminary experiments and data collection using the DOE method: Step (1.1) Determine the output response Y: Determine the index that is affected by the input parameters of the SMT printer and can reflect the printing quality as the output response Y. The output response Y includes solder paste volume, solder paste height, solder paste area, bridging defect rate, and offset. Step (1.2): Obtain the experimental dataset: Based on the working conditions of the SMT printer, design an input parameter data matrix for the experiment. Set the input parameters of the SMT printer according to the input parameter data matrix and conduct the experiment. After each experiment, detect and record the data about the output response Y to obtain an experimental dataset containing input parameter data and output response data. Step (1.3): Determine the input parameters X and obtain the initial dataset: Perform variance analysis on the experimental dataset to determine the significance and optimal level combination of each input parameter. The input parameters X corresponding to the optimal level combination include squeegee pressure, squeegee speed, demolding speed, and printing gap. Obtain the initial dataset containing [input parameters X, output response Y]. Step (1.4): Obtain the optimized dataset: Filter the initial dataset to obtain the optimized dataset; Step (2), Establishment of the Gaussian process regression model: Step (2.1): Train the initial prediction model: Construct a training dataset and a candidate dataset using the optimized dataset. Use a Gaussian process regression model to train an initial prediction model for predicting the printing quality of SMT printing machines using the training dataset. Then, use the initial prediction model to predict the mean and variance of all candidate data in the candidate dataset. Step (2.2): Training to obtain a high-precision prediction model: Select the data point with the largest prediction variance in the candidate dataset as the next experimental point, obtain new training data through experiments, and retrain the initial prediction model with the new training data. After multiple rounds of adding new training data and rebuilding the initial prediction model, the initial prediction model reaches a satisfactory accuracy, and a high-precision prediction model is obtained. Step (3), Global optimization based on operations research / intelligent algorithms: Step (3.1): Use the trained high-precision prediction model as a printing quality evaluator for the SMT printer. Step (3.2) Set optimization goals: Set the target values ​​of solder paste height, solder paste area and solder paste volume as expected values, and minimize bridging defect rate and offset; Step (3.3), Set constraints: Set the range of values ​​for squeegee pressure, squeegee speed, printing gap and demolding speed; Step (3.4): The high-precision prediction model is optimized by running a multi-objective particle swarm optimization algorithm with constraints on the model. Step (3.5) Finally, the optimal solution is output through the improved particle swarm algorithm, which yields the best combination of printing parameters.

[0007] Preferred options also include: Step (4), Robustness verification and closed-loop feedback: Step (4.1): Perform a physical verification experiment on the optimized parameter combination on the SMT printer, repeat the printing multiple times, and the results match the prediction of the high-precision prediction model, indicating that the model is good. Step (4.2): Perform Monte Carlo simulation on the optimal parameter combination on the high-precision prediction model to verify the robustness of the optimization scheme; Step (4.3): Feed the verification results back to the database and the high-precision prediction model to form a closed-loop learning process.

[0008] Preferred options also include: Step (5), Dynamic Adaptability and Continuous Learning: Step (5.1) Continuously collect the actual printing input parameters of the SMT printer on the production line, the output response results of the SPI detection, and environmental data; (5.2) Periodically fine-tune and update the high-precision prediction model using new data to adapt to changes in solder paste, equipment status drift, and environmental fluctuations; (5.3) Construct a knowledge base to store successful optimization cases, failure experiences, and parameter-quality mapping relationships to support case reasoning and rapid initialization.

[0009] Preferably, in step (1.4), the process of filtering the initial dataset to obtain the optimized dataset includes: firstly, filtering the initial dataset using the Taguchi method to obtain the key dataset, then using the response surface methodology to accurately establish the nonlinear relationship model between the input parameter X and the output response Y, and finding the optimal region in the key dataset to obtain the final optimized dataset.

[0010] Preferably, in step (3.3), the constraints are set as follows: squeegee pressure 30-150N, squeegee speed between 20-100mm / s, printing gap between -1-0.07mm, and demolding speed between 0.1-3mm / s.

[0011] Preferably, in step (3.4), the optimization of the high-precision prediction model using the multi-objective particle swarm optimization algorithm with constraints includes: Constructing a multi-objective particle swarm optimization algorithm with constraints: (3.4.1) Coding Design: Suppose there are a total of N particles, and the position of the i-th particle at time t is... The velocity at time t is They are all 4-dimensional vectors, with each dimension representing the squeegee pressure, squeegee speed, demolding speed, and printing gap, respectively. (3.4.2) The formula for updating the velocity is: ; In the formula: ω is the inertia weight; , The learning factor is used to adjust the step size of the particle's flight towards the individual optimal solution and the global optimal solution, respectively. , A random number within the interval [0,1]; The optimal position of particle i up to time t; The global optimal position found for the entire particle swarm up to time t; (3.4.3) The position update formula is: ; In the formula: , These are the upper and lower boundaries of the position, used to limit the particle's position within a feasible range; It is the position of particle i at time t in dimension d; is the velocity of particle i at time t in dimension d; r is the random perturbation term introduced when the velocity is 0; Where r has a one-third probability of being 1, a one-third probability of being 0, and a one-third probability of being -1, and a random number p between 0 and 1 is generated, the following formula is obtained: ; (3.4.4) Calculation of the objective function: The multi-objective optimization problem is transformed into a single-objective optimization problem, namely, minimizing the following objective: ; in These represent the height, area, and volume of the solder paste, respectively. These represent the expected values ​​for solder paste height, area, and volume, respectively. Represents the bridging defect rate; Represents the offset; (3.4.5) Linear decreasing strategy for inertia weight: ; In the formula, t is the current iteration round number. To be the maximum number of iteration rounds, It is the initial inertia weight; It is the termination inertia weight; The operation and optimization process of the above-mentioned multi-objective particle swarm optimization algorithm with constraints is as follows: (1) Initialize the position and velocity of the particle swarm according to the range of parameters and uniform distribution; (2) Calculate the objective function value of the particle at its current position; (3) Update the individual optimal value based on the objective function. and global optimal (4) Adjust the velocity and position of the particles according to the velocity and position update rules; repeat steps (2)-(4) until convergence or the highest number of iterations is reached.

[0012] Preferably, in step (4.2), the Monte Carlo simulation of the optimal parameter combination on the high-precision prediction model includes: adding ±3% random perturbation to each parameter to simulate production fluctuations, simulating multiple times; calculating the average solder paste volume, solder paste area, solder paste height, bridging defect rate, offset and their distribution, confirming that the fluctuation range of the printing results is within an acceptable range, indicating that the robustness of the optimization scheme meets the requirements.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion first uses Design of Experiments (DOE) for preliminary parameter screening and data collection to establish a high-quality dataset; then, it uses the Gaussian process regression model, an artificial intelligence algorithm, to construct a printing quality prediction model; finally, it introduces operations research optimization algorithms to make multi-objective production parameter decisions under multiple constraints. By establishing a system integrating the above algorithms, the system receives target parameters through a human-computer interaction interface, automatically executes the optimization process, and outputs the optimal parameter combination to the SMT printing machine. This method has a very good effect in solving the printing machine parameter optimization problem in the existing SMT industry. Compared with existing printing machine parameter optimization methods, this method has higher efficiency and accuracy, stronger robustness, and can achieve better parameter optimization results through data-driven and dynamic adaptation. This method greatly improves the efficiency of SMT printing parameter setting and significantly improves the accuracy of the model, making parameter setting more intelligent and ensuring printing stability. Through continuous learning and model fine-tuning, the model's anti-interference ability is significantly improved, effectively enhancing the model's accuracy and stability. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an embodiment of the intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion according to the present invention.

[0015] Figure 2 This is a flowchart illustrating the parameter adjustment process of an SMT printer using an embodiment of the intelligent optimization method for SMT printer process parameters based on multi-algorithm fusion according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] In one embodiment, a method for intelligent optimization of SMT printing machine process parameters based on multi-algorithm fusion is provided, combining... Figure 1 As shown, the intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion includes the following steps: 1. Data acquisition based on improved DOE.

[0018] This step involves preliminary experiments and data collection using the Design of Experiments (DOE) method. DOE is a statistical method for optimizing processes through systematic experimental design, aiming to identify key factors and their interactions with the fewest possible experiments, and to establish a mathematical model to determine the optimal parameter combination. This step specifically includes: 1.1 Determine the key output responses: The output response (Y) includes: solder paste volume (% of target value), solder paste height, solder paste area, bridging defect rate, and offset. These parameters in the output response Y—solder paste volume, solder paste height, solder paste area, bridging defect rate, and offset—are influenced by the input parameters of the SMT printer and reflect print quality. The relevant parameters of the output response Y are measured using the SMT printer's SPI (Solder Paste Inspection) function. SPI inspection is a key process in surface mount technology (SMT) used to inspect solder paste printing quality, automating the measurement of solder paste parameters on PCB pads using optical or laser technology. The inspection includes measuring the geometric parameters of solder paste, such as height, volume, area, and shape, to ensure that the printing amount meets the design requirements; detecting solder paste offset to avoid short circuits or cold solder joints caused by misalignment, and ensuring the accuracy of the soldering position; and identifying printing defects such as solder spikes, bridging, and collapse. Bridging refers to the accidental connection of adjacent pads / pins caused by excessive solder or printing offset. The bridging defect rate is the occurrence rate of bridging defects and is a key indicator for measuring the soldering quality of SMT (Surface Mount Technology).

[0019] 1.2 Conduct experiments and collect data to obtain experimental datasets.

[0020] Based on the working conditions of the SMT printer, an input parameter data matrix was designed for experiments. The printer parameters were set according to the designed experimental matrix. After printing, SPI was used to detect and record the quality data (output response Y) corresponding to each experiment, resulting in an experimental dataset containing input parameter data and output response data.

[0021] 1.3 Data analysis: Determine the input parameter X and obtain the initial dataset.

[0022] Analysis of variance (ANOVA) was performed on the previous experimental dataset to determine the significance and optimal level combination of each input parameter. The optimal level combination corresponds to input parameters X, including squeegee pressure, squeegee speed, demolding speed, and printing gap. An initial dataset containing [input parameters X, output response Y] was obtained. Here, ANOVA is used to determine the significance and optimal level combination of each parameter, which is to identify the input parameter combination that has a significant impact on the output response Y.

[0023] 1.4. Select the DOE method to filter the initial dataset and obtain the optimized dataset.

[0024] In step 1.4, the initial dataset is filtered to obtain the optimized dataset. This includes filtering the initial dataset to obtain the key dataset using the Taguchi method. In the Taguchi method, an L9 (3^4) orthogonal array is selected, and four parameters (scraper pressure, scraper speed, demolding speed, and printing gap) are arranged. Each parameter has three levels, and a total of nine experiments are conducted.

[0025] Then, the Response Surface Method (RSM) and Central Composite Design (CCD) are used to accurately establish the nonlinear relationship model between the input parameter X and the output response Y, and to find the optimal region in the key dataset to obtain the final optimized dataset.

[0026] 2. Establish a Gaussian process regression model to obtain a refined AI prediction model.

[0027] 2.1 Using the DOE experimental data from step 1 (X = squeegee pressure, squeegee speed, demolding speed, printing gap; Y = solder paste volume, solder paste height, solder paste area, offset, bridging defect rate), train an initial AI prediction model using a Gaussian process regression model. Predict the mean and variance of all candidate parameter points. Specifically, use the optimized dataset obtained in step 1 to construct a training dataset and a candidate dataset. Use a Gaussian process regression model to train an initial prediction model for predicting the printing quality of SMT printers using the training dataset. Then, use this initial prediction model to predict the mean and variance of all candidate data in the candidate dataset.

[0028] In step 2.1, the training principle of the Gaussian process regression model is as follows: If random variable Follows a mean variance is The probability distribution of , and its probability density function is: ; This random variable is called a normally distributed random variable, and the distribution it follows is called the normal distribution (also known as the Gaussian distribution), denoted as . .

[0029] If a multidimensional random vector exhibits a probability pattern similar to a univariate Gaussian distribution, it is said that this random vector follows a multivariate Gaussian distribution. The multivariate Gaussian distribution can be derived from the univariate Gaussian distribution for a D-dimensional random variable. The multivariate Gaussian distribution is as follows: ; In the formula It is a D-dimensional mean vector. It is a D × D covariance matrix. yes The determinant of .

[0030] In a Gaussian process, any linear combination of random variables follows a multivariate Gaussian distribution, and each finite-dimensional distribution is a joint Gaussian distribution. For an arbitrary set... A definition in A Gaussian process (GP) is a set of random variables. , such that for any and ,satisfy It is a multivariate Gaussian distribution. Since a Gaussian distribution is specified by a mean vector and a covariance matrix, a Gaussian process is also entirely determined by the mean function. Sum of covariance functions (kernel functions) A common and unique decision is expressed in the following form:

[0031] The mean function reflects the mean of the function at the input point x:

[0032] Kernel function Characterizes the function value at and Relationship between places:

[0033] Consider a general regression model with noise:

[0034] in It is a Gaussian process with undetermined parameters, and the noise satisfies ; Assuming dataset When the observed points satisfy the regression model, according to the definition of a Gaussian process, the joint distribution of these points needs to satisfy a multidimensional Gaussian distribution, that is:

[0035] here It is a mean vector. yes The matrix, where the first... The element is .

[0036] To predict the function value at the new point ,in According to the properties of the Gaussian distribution, the joint distribution of the training points and the prediction points is still a Gaussian distribution, that is:

[0037] in: , , ; It is The matrix, where the first... element ; It is The matrix, where the first... element ; Finally, using the conditional distribution properties of the Gaussian distribution, we can obtain the conditional probability distribution of the predicted value:

[0038] in:

[0039]

[0040] If we take noise into account in the predicted values, the conditional probability distribution is as follows:

[0041] 2.2 Training to Obtain a High-Precision Prediction Model: An active learning strategy is employed. The data points with the largest prediction variance in the candidate dataset are selected as the next experimental points. Three experimental points are chosen, and physical experiments are performed to obtain new training data. This new data is added to the training set, and the initial prediction model is retrained using this new data. This process is repeated multiple times with the addition of new training data and the reconstruction of the initial prediction model until the initial prediction model reaches satisfactory accuracy (evaluated through cross-validation), resulting in a high-precision prediction model. This high-precision prediction model is an AI prediction model capable of capturing complex nonlinear relationships. Here, to improve the accuracy of the prediction model during training, data points with large prediction variance in the candidate dataset are selected as experimental points, thus achieving the effect of building a high-precision prediction model with fewer physical experiments.

[0042] 3. Global Optimization Based on Operations Research / Intelligent Algorithms. Operations research optimization here is based on the previously constructed high-precision prediction model. Under constraints, it achieves optimal resource allocation to realize global optimization. Global optimization involves systematically verifying and finding the globally optimal parameter values ​​among all possible parameter points. 3.1 Use the trained high-precision prediction model as a print quality evaluator.

[0043] 3.2 Setting optimization goals: Set the target values ​​for printing height, area, and volume as expected values, and minimize bridging defect rate and offset.

[0044] 3.3 Setting constraints: Squeegee pressure is between 30-150N, speed is between 20-100mm / s, printing gap is between -1-0.07mm, and demolding speed is between 0.1-3mm / s.

[0045] 3.4. A multi-objective particle swarm optimization algorithm with constraints is used to optimize the AI ​​model.

[0046] The particle swarm optimization algorithm is designed as follows: 3.4.1 Coding Design: Suppose there are a total of N particles, and the position of the i-th particle at time t is... The velocity at time t is They are all 4-dimensional vectors, with each dimension representing the squeegee pressure, squeegee speed, demolding speed, and printing gap, respectively.

[0047] 3.4.2 The formula for updating the speed is:

[0048] In the formula: ω is the inertia weight; , The learning factor is used to adjust the step size of the particle's flight towards the individual optimal solution and the global optimal solution, respectively. , A random number within the interval [0,1]; The optimal position of particle i up to time t; The global optimal position found for the entire particle swarm up to time t; 3.4.3 The formula for updating the position is:

[0049] In the formula: , These are the upper and lower boundaries of the position, used to limit the particle's position within a feasible range; It is the position of particle i at time t in dimension d; is the velocity of particle i at time t in dimension d; r is the random perturbation term introduced when the velocity is 0; Where r has a one-third probability of being 1, a one-third probability of being 0, and a one-third probability of being -1, and a random number p between 0 and 1 is generated, then:

[0050] 3.4.4 Calculation of the objective function: The multi-objective optimization problem is transformed into a single-objective optimization problem, namely, minimizing the following objective:

[0051] in These represent the height, area, and volume of the solder paste, respectively. These represent the expected values ​​for solder paste height, area, and volume, respectively. Represents the bridging defect rate; This represents the offset.

[0052] 3.4.5 Inertia Weight: Linear Decreasing Strategy

[0053] In the formula, t is the current iteration round number. To be the maximum number of iteration rounds, It is the initial inertia weight; It is the termination inertia weight.

[0054] The operation and optimization process of the above-mentioned multi-objective particle swarm optimization algorithm with constraints is as follows: (1) Initialize the position and velocity of the particle swarm according to the range of parameters and uniform distribution; (2) Calculate the objective function value of the particle at its current position; (3) Update the individual optimal value based on the objective function. and global optimal (4) Adjust the velocity and position of the particles according to the velocity and position update rules; repeat steps (2)-(4) until convergence or the highest number of iterations is reached.

[0055] 3.5 Finally, the optimal solution, i.e. the best printing parameters, is output through the improved particle swarm algorithm described above.

[0056] 4. Robustness Validation and Closed-Loop Feedback. Robustness validation assesses the high-precision prediction model's ability to maintain stability and performance under uncertainty. Closed-loop feedback feeds the validation results back into the database, allowing the high-precision prediction model to make targeted improvements in subsequent learning.

[0057] The methods for robustness verification and closed-loop feedback are as follows: 4.1 A physical verification experiment was performed on the optimized parameter combination on the SMT printer (repeated printing multiple times). The results matched the high-precision prediction model, indicating that the model is good.

[0058] 4.2. Conduct Monte Carlo simulations on the optimal parameter points using a high-precision prediction model: Add ±3% random perturbation to each parameter (simulating production fluctuations), and perform multiple simulations. Calculate the average volume, area, height, bridging defect rate, offset, and their distribution to confirm that the fluctuation range of the printing results is within an acceptable range, indicating that the robustness of the optimization scheme meets the requirements.

[0059] 4.3. Feed the verification results back to the database and high-precision model to form a closed-loop learning process.

[0060] 5. Dynamic Adaptability and Continuous Learning. Through dynamic adaptation and continuous learning, the high-precision prediction model can adapt to dynamic changes in solder paste, equipment status, environment, etc.

[0061] 5.1 Continuously collect actual printing parameters, SPI test results, and environmental data from the SMT production line.

[0062] 5.2 Regularly use new data to fine-tune / update the high-precision prediction model to adapt to changes in solder paste, equipment status drift, and environmental fluctuations.

[0063] 5.3 Construct a knowledge base to store successful optimization cases, failure experiences, and parameter-quality mapping relationships (the mapping relationship between input parameters and output responses) to support case-based reasoning and rapid initialization. Case-based reasoning enables the high-precision prediction model to make rapid decisions using the historical case database. Rapid initialization uses data from the knowledge base to assign initial values ​​to the parameters of the high-precision model, ensuring that the high-precision prediction model can quickly achieve effective predictive performance. Combination Figure 2 As shown, when the SMT printer is adjusted using the intelligent optimization method for SMT printer process parameters based on multi-algorithm fusion in the above embodiment, the adjustment is performed through a system containing a data layer, an AI layer, an operations research optimization layer, and a decision layer. First, preliminary experiments and data collection are conducted using the DOE method in the data layer. In the AI ​​layer, a Gaussian process regression model is established to obtain a high-precision prediction model. In the operations research optimization layer, the optimal printing parameters are optimized by improving the particle swarm optimization algorithm. In the decision layer, the AI ​​optimization results (high-precision prediction model) and operations research constraints are fused to output the final parameter set. Finally, the result is output to the SMT printer.

[0064] As can be seen from the above embodiments, this intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion first uses Design of Experiments (DOE) for preliminary parameter screening and data collection to establish a high-quality dataset; then, it uses the Gaussian process regression model, an artificial intelligence algorithm, to construct a printing quality prediction model; finally, it introduces operations research optimization algorithms to make multi-objective production parameter decisions under multiple constraints. By establishing a system integrating the above algorithms, the system receives target parameters through a human-computer interaction interface, automatically executes the optimization process, and outputs the optimal parameter combination to the SMT printing machine.

[0065] This method, through a systematic parameter optimization approach integrating physical experimental design, data-driven modeling, and intelligent optimization algorithms, effectively addresses the shortcomings of traditional single optimization methods in terms of efficiency, accuracy, and practicality. It significantly improves the first-pass yield and quality stability of solder paste printing processes, enabling the rapid acquisition of optimal SMT printer parameter settings with minimal experiments and low cost. This achieves efficient, accurate, robust, and dynamic SMT printer parameter optimization. The high efficiency of this multi-algorithm-based intelligent optimization method for SMT printer process parameters is reflected in: significantly reducing the number of physical experiments compared to traditional full-factor DOE methods, thus significantly reducing time and material costs. High accuracy is reflected in: utilizing high-precision prediction models to capture complex nonlinear relationships and finding global or quasi-global optimal solutions that are difficult to discover using traditional methods. Strong robustness is reflected in: through robustness analysis, ensuring that optimized parameters maintain good performance under real-world production fluctuations, improving production stability. Dynamic adaptation is reflected in: continuously learning to adapt to changes in the production environment, ensuring the accuracy and stability of parameter optimization. In addition, this intelligent optimization method for SMT printing machine process parameters based on multi-algorithm fusion can integrate data-driven and knowledge-based approaches. By combining the power of data with the experience of engineers, it can effectively improve the scientific nature and reliability of the optimization process.

[0066] It should be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-algorithm fusion-based intelligent optimization method for process parameters of an SMT printer, characterized in that, Comprise: Step (1), preliminary experiment and data collection by DOE method: Step (1.1), determine output response Y: determine the index affected by the input parameters of SMT printer and capable of reflecting the printing quality as the output response Y, the output response Y includes tin paste volume, tin paste height, tin paste area, bridging defect rate, offset; Step (1.2), obtain experimental data set: based on the working conditions of SMT printer, design input parameter data matrix for experiment, set the input parameters of SMT printer according to the input parameter data matrix, after each experiment, detect and record the data about the output response Y, obtain the experimental data set containing input parameter data and output response data; Step (1.3), determine input parameter X and obtain initial data set: perform variance analysis on the experimental data set to determine the significance and optimal level combination of each input parameter, the optimal level combination of the input parameter X includes squeegee pressure, squeegee speed, demolding speed, printing gap, and obtain the initial data set containing [input parameter X, output response Y]; Step (1.4), obtain optimized data set: screen the initial data set to obtain the optimized data set; Step (2), establishment of Gaussian process regression model: Step (2.1), training to obtain initial prediction model: use the optimized data set to construct training data set and candidate data set, use Gaussian process regression model, train the initial prediction model for predicting the printing quality of SMT printer through the training data set, and predict the mean and variance of all candidate data in the candidate data set through the initial prediction model; Step (2.2), training to obtain high-precision prediction model: select the data point with the maximum prediction variance in the candidate data set as the next experimental point, obtain new training data through experiment, retrain the initial prediction model through new training data, and thus through multiple rounds of addition of new training data and re-construction of the initial prediction model, until the initial prediction model reaches satisfactory precision, to obtain high-precision prediction model; Step (3), global optimization based on operation / intelligent algorithm: Step (3.1), use the trained high-precision prediction model as the printing quality evaluator of SMT printer; Step (3.2), set optimization target: set the target values of tin paste height, tin paste area and tin paste volume as expected values, and minimize bridging defect rate and offset; Step (3.3), set constraints: set the value range of squeegee pressure, squeegee speed, printing gap and demolding speed; Step (3.4), run optimization on the high-precision prediction model by using multi-objective particle swarm optimization algorithm with constraint processing; Step (3.5), finally output the optimal solution by the improved particle swarm algorithm, that is, obtain the best printing parameter combination.

2. The method of claim 1, wherein the method is characterized by, Also include: Step (4), robustness verification and closed-loop feedback: Step (4.1), perform physical verification experiment on the optimized parameter combination on SMT printer, repeat printing multiple times, and the results consistent with the prediction of the high-precision prediction model indicate that the model is good; Step (4.2), Monte Carlo simulation on the optimal parameter combination on the high-precision prediction model to verify the robustness of the optimization scheme; Step (4.3), feedback the verification results back to the database and the high-precision prediction model to form a closed-loop learning.

3. The multi-algorithm fusion based intelligent optimization method of SMT printer process parameters according to claim 2, characterized in that, Also includes: Step (5), dynamic adaptability and continuous learning: Step (5.1) continuously collect the actual printing input parameters of the SMT printer on the production line, the output response results of SPI detection, and environmental data; (5.2) periodically fine-tune and update the high-precision prediction model using new data to adapt to changes in solder paste, equipment state drift, and environmental fluctuations; (5.3) build a knowledge base to store successful optimization cases, failure experiences, and parameter-quality mapping relationships to support case reasoning and rapid initialization.

4. The multi-algorithm fusion based intelligent optimization method of SMT printer process parameters according to claim 1, characterized in that, In step (1.4), the filtering of the initial data set to obtain the optimization data set includes: first filtering the initial data set by Taguchi method to obtain a key data set, then using response surface method to accurately establish the nonlinear relationship model between the input parameter X and the output response Y, and finding the optimal region in the key data set to obtain the final optimization data set.

5. The multi-algorithm fusion based intelligent optimization method of SMT printer process parameters according to claim 1, characterized in that, In step (3.3), the constraints are set as follows: doctor blade pressure 30-150N, doctor blade speed between 20-100mm / s, printing gap between -1-0.07mm, and demolding speed between 0.1-3mm / s.

6. The multi-algorithm fusion based intelligent optimization method of SMT printer process parameters according to claim 1, characterized in that, In step (3.4), the multi-objective particle swarm optimization algorithm with constraint processing runs optimization on the high-precision prediction model, including: Constructing a multi-objective particle swarm optimization algorithm with constraint processing: (3.4.1) Coding design: Let there be N particles in total, the position of the i-th particle at time t is , and the velocity at time t is , which are all 4-dimensional vectors, each dimension representing the squeegee pressure, squeegee velocity, release velocity, and printing gap, respectively; (3.4.2) the update formula of the velocity is: ； where ω is the inertia weight; , is the learning factor, respectively adjusting the step length of the particle flying towards the individual optimal solution and the global optimal solution; , is a random number in the interval [0, 1]; is the individual optimal position found by the particle i up to time t; is the global optimal position found by the entire particle swarm up to time t; (3.4.3) the update formula of the position is: ； wherein: , is the upper and lower boundary of the position to limit the particle position within a feasible range; is the position of particle i in dimension d at time t; is the velocity of particle i in dimension d at time t; r is a random perturbation term introduced when the velocity is 0; Where r has a probability of 1 for one-third, a probability of 0 for one-third, and a probability of -1 for one-third, and a random number p between 0 and 1 is randomly generated, then the following formula is obtained: ； (3.4.4) calculation of the objective function: Convert the multi-objective optimization problem into a single-objective optimization, i.e. minimize the following objective: ； wherein respectively represent height, area, volume of the solder paste; respectively represent expected values of height, area, volume of the solder paste; represents a bridging defect rate; represents an offset amount; (3.4.5) linear decreasing strategy of inertia weight: ； where t is the current iteration round, is the maximum iteration round, is the initial inertia weight; is the terminal inertia weight; The operation optimization process of the multi-objective particle swarm optimization algorithm with the above-mentioned constraint processing is as follows: (1) initializing the position and speed of the particle swarm according to the uniform distribution according to the range of parameters; (2) calculating the objective function value of the current position of the particle; (3) judging the update of the individual optimum and the global optimum according to the objective function; (4) adjusting the speed and position of the particle according to the speed and position update rule; repeating steps (2)-(4) until convergence or the highest iteration number is reached.

7. The multi-algorithm fusion based intelligent optimization method of SMT printer process parameters according to claim 2, characterized in that, In step (4.2), the Monte Carlo simulation on the optimal parameter combination on the high-precision prediction model includes: adding ±3% random disturbance to simulate production fluctuations for multiple times; calculating the average solder paste volume, solder paste area, solder paste height, bridging defect rate, offset, and their distribution to confirm that the fluctuation range of the printing result is within the acceptable range, indicating that the robustness of the optimization scheme meets the requirements.

Images