Parameter generating method and parameter generating device for printer

Abstract

The invention discloses a parameter generation method and a parameter generation device for a printer. And inputting the target solder paste amount data to a first machine learning model, wherein the first machine learning model outputs a plurality of predicted manufacturing process parameters for the printer. And inputting the plurality of predicted manufacturing process parameters to a second machine learning model, wherein the second machine learning model outputs predicted solder paste amount data for the printer. According to the minimum prediction error, the weight of the first machine learning model is updated, and the prediction error is the error between the target solder paste amount data and the predicted solder paste amount data. And inputting the target solder paste amount data to the updated first machine learning model, and outputting a plurality of new manufacturing process parameters by the updated first machine learning model, and controlling the operation of the printer. Therefore, the operation efficiency can be improved.

Description

Technical Field

[0001] This invention relates to a manufacturing process for surface mount technology (SMT), and particularly to a parameter generation method and apparatus for printers. Background Technology

[0002] Surface Mount Technology (SMT) is an indispensable key manufacturing process in the electronics industry. This technology mainly involves directly bonding electronic components to the surface of a printed circuit board (PCB). It has advantages such as high density, high efficiency, and low cost, and is widely used in consumer electronics, communications, automotive, medical and other fields.

[0003] However, the SMT manufacturing process involves a variety of complex parameters, and even slight changes in these parameters can have a significant impact on product quality and reliability. Traditionally, the setting of SMT manufacturing process parameters has relied mainly on engineers' experience and trial-and-error methods, which is not only time-consuming and labor-intensive but also difficult to optimize. Summary of the Invention

[0004] This invention provides a method and apparatus for generating parameters for a printer, and can provide suitable manufacturing process parameters.

[0005] The parameter generation method for a printer according to embodiments of the present invention includes (but is not limited to) the following steps: inputting target solder paste amount data into a first machine learning model, the first machine learning model outputting a plurality of predicted manufacturing process parameters for the printer; inputting the plurality of predicted manufacturing process parameters into a second machine learning model, the second machine learning model outputting predicted solder paste amount data for the printer, wherein the second machine learning model is trained to learn the correlation between a plurality of solder paste amount samples and a plurality of parameter samples; updating the weights of the first machine learning model based on minimizing the prediction error, wherein the prediction error is the error between the target solder paste amount data and the predicted solder paste amount data; and inputting the target solder paste amount data into the updated first machine learning model, the updated first machine learning model outputting a plurality of new manufacturing process parameters, wherein these new manufacturing process parameters are used to control the operation of the printer.

[0006] The parameter generation apparatus for a printer according to embodiments of the present 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 is configured to load the program code and execute: input target solder paste amount data to a first machine learning model, the first machine learning model outputting multiple predicted manufacturing process parameters for the printer; input the multiple predicted manufacturing process parameters to a second machine learning model, the second machine learning model outputting predicted solder paste amount data for the printer, wherein the second machine learning model is trained to learn the correlation between multiple solder paste amount samples and multiple parameter samples; update the weights of the first machine learning model based on minimizing the prediction error, wherein the prediction error is the error between the target solder paste amount data and the predicted solder paste amount data; and input the target solder paste amount data to the updated first machine learning model, the updated first machine learning model outputting multiple new manufacturing process parameters, wherein these new manufacturing process parameters are used to control the operation of the printer.

[0007] Based on the above, the parameter generation method and apparatus for the printer according to embodiments of the present invention can input the predicted manufacturing process parameters generated by the first machine learning model into the second machine learning model, and update the first machine learning model by minimizing the prediction error between the target solder paste amount data input to the first machine learning model and the predicted solder paste amount data generated by the second machine learning model. Then, the updated first machine learning model can be used to generate new manufacturing process parameters. Therefore, by using the new manufacturing process parameters, the printer can improve operational efficiency, reduce the chance of misjudgment, and reduce material waste.

[0008] 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

[0009] Figure 1 This is a block diagram of a parameter generation device according to an embodiment of the present invention;

[0010] Figure 2 This is a flowchart of a parameter generation method according to an embodiment of the present invention;

[0011] Figure 3 This is a schematic diagram of a first machine learning model and a second machine learning model according to an embodiment of the present invention;

[0012] Figure 4 This is a schematic diagram illustrating the output adjustment of a first machine learning model according to an embodiment of the present invention;

[0013] Figure 5 This is a schematic diagram illustrating the step function and the sigmoid function according to an embodiment of the present invention;

[0014] Figure 6 This is a schematic diagram illustrating the positioning position according to an embodiment of the present invention;

[0015] Figure 7 This is a schematic diagram illustrating the acceptance screening process according to an embodiment of the present invention;

[0016] Figure 8 This is a schematic diagram of a third machine learning model according to an embodiment of the present invention;

[0017] Figure 9 This is a schematic diagram illustrating the opening ratio of a hole according to an embodiment of the present invention;

[0018] Figure 10 This is a schematic diagram illustrating the prediction of solder paste volume by zone according to an embodiment of the present invention;

[0019] Figure 11 This is a schematic diagram comparing the actual and predicted solder paste amount according to an embodiment of the present invention;

[0020] Figure 12 This is a flowchart of model diagnosis and adjustment according to an embodiment of the present invention;

[0021] Figure 13 This is a schematic diagram illustrating clustering according to an embodiment of the present invention;

[0022] Figure 14 This is a schematic diagram illustrating a hierarchical clustering method according to an embodiment of the present invention;

[0023] Figure 15 This is an experimental diagram illustrating and verifying an embodiment of the present invention.

[0024] Symbol Explanation

[0025] 10: Parameter generation device

[0026] 11: Memory

[0027] 13: Processor

[0028] S210~S240, S410~S430, S121~S126, S131~S135: Steps

[0029] ML1: The First Machine Learning Model

[0030] ML2: The Second Machine Learning Model

[0031] SP T Target solder paste quantity data

[0032] MP P1 ~MP Pi Predict manufacturing process parameters

[0033] SF, SF0: Step function

[0034] NOR: Normalization

[0035] SF1~SF8: S-shaped functions

[0036] SP P Predicted solder paste quantity data

[0037] AP: Auxiliary Parameters

[0038] P: Printing press

[0039] P Y P RX P FX Location

[0040] EF1, EF2, EF j Features to be evaluated

[0041] ML3: The Third Machine Learning Model

[0042] SP P2 Solder paste amount data for holes

[0043] S1: Metal plate

[0044] PD1: Hole

[0045] W: Width

[0046] T: Thickness

[0047] L: Long

[0048] G11, G12, G13, G21, G22, G23, G31, G32, G33: Blocks

[0049] G1, G2, G3, G4, G5: Samples Detailed Implementation

[0050] Figure 1 This is a block diagram of a parameter generating apparatus 10 according to an embodiment of the present invention. Please refer to... Figure 1 The parameter generating device 10 includes (but is not limited to) a memory 11 and a processor 13. The parameter generating device 10 may be a mobile phone, tablet computer, laptop computer, desktop computer, server, voice assistant device, smart home appliance, wearable device, manufacturing equipment or other electronic device.

[0051] The memory 11 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, the memory 11 is used to store program code, software modules, configuration settings, data (e.g., solder paste amount data, manufacturing process parameters, features, or physical information), or files, as will be detailed in subsequent embodiments.

[0052] Processor 13 is coupled to memory 11. Processor 13 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 elements or combinations thereof. In one embodiment, processor 13 is used to execute all or part of the operations of parameter generation device 10, and can load and execute various program codes, software modules, files, and data stored in memory 11.

[0053] The method described in this embodiment of the invention will be explained below in conjunction with the various devices, components, and modules in the parameter generation apparatus 10. The various processes of this method may be adjusted according to the implementation situation, and are not limited thereto.

[0054] Figure 2 This is a flowchart of a parameter generation method according to an embodiment of the present invention. Please refer to... Figure 2The processor 13 inputs the target solder paste amount data to the first machine learning model, and the first machine learning model outputs multiple predicted manufacturing process parameters for the printer (step S210). Specifically, the printing operation of the printer for surface mount technology (SMT) includes: placing a metal plate (e.g., a steel plate or other metal plate) on a printed circuit board (PCB); aligning the holes of the metal plate with the pads of the printed circuit board; and using a squeegee to apply / print solder paste through the holes in the metal plate onto the printed circuit board.

[0055] The target solder paste quantity data includes the numerical value of the amount of solder paste to be applied / printed onto the printed circuit board in the above printing operation. Solder paste quantity can be expressed as a percentage, volume, or area. A percentage is expressed as, for example, the proportion of solder paste quantity to a reference quantity. For example, 100 (%) indicates that the solder paste quantity is the same as the reference quantity. The reference quantity can be predefined or changed as needed.

[0056] Predicted manufacturing process parameters refer to the manufacturing process parameters predicted / output / generated by the first machine learning model. The types of manufacturing process parameters mentioned in this paper may include pressure, print speed, separate speed, separation distance, print gap, and clean rate. However, other types of manufacturing process parameters may also exist.

[0057] The first machine learning model is a model trained by a machine learning algorithm. This model can be trained to understand the relationship between solder paste quantity and manufacturing process parameters. Machine learning algorithms can be, for example, Artificial Neural Networks (ANNs), extreme gradient boosting (XGboost), Support Vector Regression (SVR), Convolutional Neural Networks (CNNs), or other algorithms. During training, the machine learning algorithm can adjust the parameters (e.g., weights) in the model by minimizing a loss function (e.g., a function based on the error between the real and output data). At the initial stage of training, the model can use preset or initial parameters.

[0058] Figure 3 This is a schematic diagram of a first machine learning model ML1 and a second machine learning model ML2 according to an embodiment of the present invention. Please refer to... Figure 3 The first machine learning model, ML1, is a single-input, multiple-output model. That is, the first machine learning model, ML1, only takes the target solder paste quantity data SP as input. T It also outputs multiple predicted manufacturing process parameters MP. P1 ~MP Pi (i is a positive integer greater than one). For example, if i is 6, then the predicted manufacturing process parameter MP... P1 The pressure value is used to predict the manufacturing process parameter MP. P2 The value of printing speed is used to predict the manufacturing process parameter MP. P3 The value of the separation rate is used to predict the manufacturing process parameter MP. P4 The value of the separation distance is used to predict the manufacturing process parameter MP. P5 This represents the printing pitch value and predicts the manufacturing process parameter MP. P6 This represents the cleaning rate; however, the predicted manufacturing process parameters are not limited to this.

[0059] Figure 4 This is a schematic diagram illustrating the output adjustment of the first machine learning model ML1 according to an embodiment of the present invention. Please refer to... Figure 4 The processor 13 can use the first machine learning model ML1 to output initial parameters (step S410). Specifically, the first machine learning model ML1 may include N layers, where N is a positive integer greater than one, and the number of layers N can be adjusted according to actual needs. These M layers may be fully connected layers, convolutional layers, pooling layers, etc. The initial parameters are the manufacturing process parameters generated by the first machine learning model ML1 at the beginning. However, the values ​​of the initial parameters may exceed the value range of the training data, thus affecting the reliability.

[0060] Table (1) shows the manufacturing process parameters and their value ranges:

[0061] Table (1)

[0062] Manufacturing process parameters Numerical range Printing speed (mm / sec) 0~200 (Minimum unit: 1) Pressure (kg) 0~20 (smallest unit: 0.2) Printing spacing (mm) -1, -1 (smallest unit: 0.1) Cleaning rate (cycle) Integers greater than or equal to 1 (smallest unit: 1) Separation speed (mm / s) 0.1~5 (smallest unit: 0.1) Separation distance (mm) 0.1~5 (smallest unit: 0.1)

[0063] The smallest unit can also be called the basic unit or the fundamental unit. Taking printing speed as an example, the numbers from smallest to largest are 20, 25, 30, 35, 40, 45 and 50, with the smallest unit (i.e., 5) separating any two consecutive values.

[0064] Processor 13 can convert the initial parameters into controllable values ​​using the transformation function SF (step S420). Specifically, the transformation function is a step function. For example, Figure 5 This is a schematic diagram illustrating the step function and the sigmoid function according to an embodiment of the present invention. Please refer to... Figure 13The horizontal axis corresponds to the initial parameters, and the vertical axis corresponds to the controllable value after transformation by the step function SF0. ​​It should be noted that the units corresponding to the horizontal axis and the vertical axis depend on the type of the corresponding manufacturing process parameters, as shown in Table (1), but are not limited to this.

[0065] Processor 13 can define a ladder function SF0 that exhibits a ladder-like shape and has fixed function values ​​(corresponding to the function's range x, i.e., controllable values) within different numerical intervals (corresponding to the function's domain f(x), i.e., initial parameters). The mathematical expression f(x) for the ladder function SF0 is:

[0066]

[0067] by Figure 5 For example, when the initial parameter value is between 0 and 2, the controllable value after transformation by the step function SF0 is 0 × 5 = 0 (e.g., 0.7, rounded down in this example, hence 0); when the initial parameter value is between 2 and 3 (e.g., 2.3, rounded down in this example, hence 2), the controllable value after transformation by the step function SF0 is 2 × 5 = 10; when the initial parameter value is between 3 and 4 (e.g., 3.8, rounded down in this example, hence 3), the controllable value after transformation by the step function SF0 is 4 × 5 = 20; and so on. When the initial parameter value is not greater than 9, the controllable value after transformation by the step function SF0 is 40. It should be noted that... Figure 5 The values ​​of the domain and range shown are for illustrative purposes only and may be changed according to actual needs.

[0068] In one embodiment, the step function is formed by adding multiple sigmoid functions. The step function may be non-differentiable or differentiate to zero, leading to vanishing gradients. To avoid this, [the following is used]. Figure 5 For example, the step function SF0 is formed by adding multiple sigmoid functions SF1, SF2, SF3, SF4, SF5, SF6, SF7, and SF8. That is, the processor 13 can approximate the step function SF0 by adding multiple different sigmoid functions SF1, SF2, SF3, SF4, SF5, SF6, SF7, and SF8. The sigmoid function SF1 maps any real number input to the range of 0 to 5; the sigmoid function SF2 maps any real number input to the range of 5 to 10; and so on, with the sigmoid function SF8 mapping any real number input to the range of 35 to 40.

[0069] Please refer to Figure 4The processor 13 can normalize the controllable values ​​(NOR) into predicted manufacturing process parameters (step S430). Specifically, normalization refers to the process of scaling data to a specific range. For example, the minMaxScaler uses the following equation (2) to scale data to the range of 0 to 1:

[0070] z_scaled=(z-z_min) / (z_max-z_min)…(2)

[0071] Where z_scaled is the scaled value (i.e., the predicted manufacturing process parameter value), z is the original value (i.e., the controllable value), z_min is the minimum value of the original data, and z_max is the maximum value of the original data. It should be noted that the minimum value z_min and the maximum value z_max in equation (2) can be adjusted according to actual needs.

[0072] Please refer to Figure 2 The processor 130 inputs multiple predicted manufacturing process parameters to a second machine learning model, which outputs predicted solder paste amount data for the printer (step S220). Specifically, the second machine learning model is a model trained using a machine learning algorithm. The machine learning algorithm may be, for example, an artificial neural network (ANN), extreme gradient boosting (XGboost) algorithm, support vector regression (SVR), convolutional neural network (CNN), or other algorithms. The second machine learning model can be trained to understand the correlation between multiple solder paste amount samples and multiple parameter samples. The dataset used for training and / or validation includes solder paste amount samples and multiple parameter samples.

[0073] Solder paste quantity samples include historical solder paste quantities applied / printed onto printed circuit boards during printing operations. Solder paste quantity can be expressed as a percentage, volume, or area. A percentage is expressed as the proportion of solder paste quantity to a reference quantity. For example, 100 (%) indicates that the solder paste quantity is the same as the reference quantity. The reference quantity can be predefined or changed as needed. Furthermore, each solder paste quantity sample corresponds to a set of parameter samples. Parameter samples are training samples for manufacturing process parameters. The set of parameter samples (including multiple parameter samples) used by the printer in a printing operation serves as a solder paste quantity sample corresponding to this set of parameter samples. Since the solder paste quantity samples and their corresponding set of parameter samples are known, machine learning algorithms can analyze labeled solder paste quantity samples (e.g., solder paste quantities with determined manufacturing process parameters) and establish a correlation between the solder paste quantity samples (as output samples) and the parameter samples (as input samples). The second machine learning model is a model constructed after training and learning, and can infer from the data to be evaluated (e.g., the (expected) manufacturing process parameters to be evaluated, i.e. the model input) to generate corresponding predicted solder paste amount data (i.e., the model output).

[0074] Please refer to Figure 3 The second machine learning model, ML2, is a multi-input, single-output model. That is, the second machine learning model ML2 takes multiple predicted manufacturing process parameters (MP) as input. P1 ~MP Pi (i is a positive integer greater than one), and output the predicted solder paste amount data SP. P .

[0075] In one embodiment, the processor 13 can input multiple predicted manufacturing process parameters MP. P1 ~MP Pi And the auxiliary parameter AP is fed into the second machine learning model ML2. That is, in addition to predicting the manufacturing process parameter MP... P1 ~MP Pi The second machine learning model, ML2, also receives inputs of auxiliary parameters, AP. These auxiliary parameters include temperature, humidity, and the position of the metal plate on the printer. Temperature and humidity are environmental parameters sensed by the printer during the printing process.

[0076] Figure 6 This is a schematic diagram illustrating the positioning position according to an embodiment of the present invention. Please refer to... Figure 6 The printer P includes a horizontal axis actuator, a vertical axis rear actuator, and a vertical axis front actuator. The horizontal axis actuator is used to position the metal plate at its horizontal axis position P. Y Furthermore, the rear actuator and the front actuator of the vertical axis are used to position the metal plate at position P in the direction of the vertical axis. RX P FXThese location points P Y P RX P FX It is used to align holes in metal plates and pads on printed circuit boards during printing operations. Position P Y P RX P FX It can be represented by coordinates, relative distance, or relative position. Location P Y P RX P FX This may affect the amount of solder paste.

[0077] Furthermore, the dataset also includes auxiliary samples during the training of the second machine learning model. These auxiliary samples include measured temperatures, humidity, and actuator positioning during the printing process. Since the solder paste quantity samples, their corresponding set of parameter samples, and the auxiliary samples are known, the machine learning algorithm can analyze the labeled solder paste quantity samples (e.g., solder paste quantities with determined manufacturing process parameters and auxiliary parameters) and establish relationships between the solder paste quantity samples (as output samples), parameter samples, and auxiliary samples (as input samples). The second machine learning model is the model constructed after training and learning, and it can infer from the data to be evaluated (e.g., the (expected) manufacturing process parameters and auxiliary parameters to be evaluated, i.e., the model input) to generate corresponding predicted solder paste quantity data (i.e., the model output).

[0078] For example, the model's inputs are: "Front squeegee printing pressure": 8.4 kg, "Rear squeegee printing pressure": 9.0 kg, "Printing gap": -0.6 mm, "Front squeegee printing speed": 60 mm / s, "Rear squeegee printing speed": 60 mm / s, "Demolding distance": 0.4 mm, "Demolding speed": 0.2 mm / s, and "Cleaning frequency": 1.0 board count. The model's outputs are: Area percentage (area detection value / area target value): 100.3%, Height percentage (height detection value / height target value): 101.3%, and Volume percentage (volume detection value / volume target value): 103.9%.

[0079] Figure 7 This is a schematic diagram illustrating the acceptance screening process according to an embodiment of the present invention. Please refer to... Figure 7Solder Paste Inspection (SPI) machines can detect the printing quality on circuit boards after the printing process is complete. For example, as shown in the 3D diagram of the circuit board, problematic pads can be clearly marked with color. Furthermore, the SPI machine can measure the actual amount of solder paste on the circuit board. The actual amount of solder paste is the height, area, and / or volume of the solder paste. The acceptance criteria for the second machine learning model can be that the difference between the area, height, and / or volume and the predicted solder paste amount data is less than a corresponding threshold value (e.g., 10%, 5%, or 3%). That is, only the second machine learning model whose difference from the predicted solder paste amount data is less than the corresponding threshold value is used.

[0080] Please refer to Figure 2 The processor 13 updates the weights of the first machine learning model based on minimizing the prediction error (step S230). Specifically, the prediction error is the error between the target solder paste quantity data and the predicted solder paste quantity data. In machine learning, the goal of model training is to minimize the prediction error to improve the model's prediction accuracy. This process involves several key elements: loss function, optimization algorithm, gradient, iterative update, and convergence.

[0081] In one embodiment, processor 13 may use a loss function to determine the prediction error. The loss function is half the square of the numerical difference between the target solder paste amount data and the predicted solder paste amount data. The mathematical expression of the loss function LOSS is as follows:

[0082]

[0083] The predicted solder paste amount data is generated / output by the second machine learning model, which is generated by inputting the predicted manufacturing process parameters (and auxiliary parameters).

[0084] Optimization algorithms are used to adjust the parameters of a model to minimize the loss function. Examples of optimization algorithms include gradient descent, Newton's method, or Adam. The gradient is the derivative of the loss function at a given point, indicating the direction and magnitude of its change at that point. Optimization algorithms use gradient information to guide parameter updates. Furthermore, training the first machine learning model is an iterative process. In each iteration, the optimization algorithm updates the model's parameters (e.g., node weights) based on gradient information, causing the loss function to gradually decrease. Training stops when the loss function no longer decreases significantly or when a preset stopping condition is met. At this point, the parameters of the first machine learning model have reached a relatively optimized state, and an updated first machine learning model is generated accordingly.

[0085] Please refer to Figure 2 The processor 13 inputs the target solder paste amount data into the updated first machine learning model, which outputs multiple new manufacturing process parameters (step S240). Specifically, these new manufacturing process parameters are used to control the operation of the printer (e.g., the printing operation described above). The processor 13 inputs the target solder paste amount data into the updated first machine learning model. The new manufacturing process parameters are those inferred or predicted by the updated first machine learning model that conform to the target solder paste amount data.

[0086] For example, the model inputs are: area percentage (area detection value / area target value): 100%, height percentage (height detection value / height target value): 100%, and volume percentage (volume detection value / volume target value): 100%. The model outputs are: "front squeegee printing pressure": 8.8 kg, "rear squeegee printing pressure": 9.6 kg, "printing gap": -0.4 mm, "front squeegee printing speed": 45 mm / s, "rear squeegee printing speed": 45 mm / s, "demolding distance": 0.4 mm, "demolding speed": 0.2 mm / s, and "cleaning frequency": 2.0 boards.

[0087] In one embodiment, the processor 13 may generate operation instructions or operation settings based on new manufacturing process parameters, and the printer may use these operation instructions or operation settings to perform printing operations and operate under the new manufacturing process parameters.

[0088] The solder paste quantity data described above can be applied to the printing operation of an entire metal board. For example, it can be referred to as the total solder paste quantity or the amount of solder paste used for the entire board. The following describes providing appropriate solder paste quantities for one or more pads corresponding to the metal board. Figure 8 This is a schematic diagram of a third machine learning model according to an embodiment of the present invention. Please refer to... Figure 8 The processor 13 takes as input multiple features to be evaluated EF1, EF2, ..., EF1. j (j is a positive integer greater than one) to the third machine learning model ML3, generating SP data on the amount of solder paste for the printer's holes. P2 Specifically, the third machine learning model, ML3, is a model trained using a machine learning algorithm. This algorithm can be, for example, an artificial neural network (ANN), extreme gradient boosting (XGboost), support vector regression (SVR), convolutional neural network (CNN), or other algorithms. The third machine learning model, ML3, can be trained to learn the relationship between multiple hole features and solder paste amounts.

[0089] Hole characteristics include the proportion of multiple holes on the metal plate and the distribution of these holes on the metal plate. For example, Figure 9This is a schematic diagram illustrating the opening ratio according to an embodiment of the present invention. Please refer to... Figure 9 The aperture ratio can be either the width-to-thickness ratio or the aperture area ratio. Taking a hole PD1 in metal plate S1 as an example, the width-to-thickness ratio of hole PD1 is: width W of hole PD1 / thickness T of metal plate S1; while the aperture area ratio of hole PD1 is: area of ​​pad (i.e., width W of hole PD1 × length L of hole PD1) / area of ​​hole wall (i.e., 2 × (width W of hole PD1 + length L of hole PD1) × thickness T of metal plate S1). The distribution position of hole PD1 on metal plate S1 is the position of hole PD1 on metal plate S1, which can be represented by coordinates, relative distance, or relative position.

[0090] The solder paste quantity data for vias includes the amount of solder paste used for multiple vias. After the printing process, a solder paste inspection machine can measure the amount of solder paste for each of the multiple vias and store its area, volume, or percentage relative to a reference amount. The solder paste quantity data for a specific via or pad can be the solder paste quantity for the entire board multiplied by the percentage of the total solder paste quantity for each pad.

[0091] Furthermore, each set of via solder paste amount data corresponds to multiple via features. The printer fills each via with solder paste using the same manufacturing process parameters during the printing operation. Since the via solder paste amount data and its corresponding set of via features are known, machine learning algorithms can analyze the labeled via solder paste amount data (e.g., solder paste amount for vias or pads with defined via features) and establish a correlation between solder paste amount (as output sample) and via features (as input sample). The third machine learning model is the model constructed after training and learning, and can be used to evaluate the data (e.g., features to be evaluated EF1, EF2, ..., EF...). j (For example, the proportion and distribution of the holes to be evaluated on the metal board, i.e., the model input) inferences are made to generate corresponding solder paste amount data for the holes (i.e., the model output). The proportion of each pad to the total solder paste amount of the board is the target solder paste amount of that pad / hole divided by the target solder paste amount of all pads / holes on the same metal board.

[0092] For example, Table (2) shows the experimental data for the model's input and output:

[0093] Table (2)

[0094]

[0095] Figure 10 This is a schematic diagram illustrating solder paste volume zoning prediction according to an embodiment of the present invention. Please refer to... Figure 10The metal plate can be divided into multiple blocks, as shown in the figure: nine blocks G11, G12, G13, G21, G22, G23, G31, G32, and G33. The prediction result is, for example:

[0096] Table (3)

[0097] 1.10 / 1.48 / 6.10 1.09 / 1.51 / 6.11 1.08 / 1.41 / 5.69 1.02 / 1.14 / 4.57 1.05 / 1.24 / 4.95 1.01 / 1.14 / 4.57 1.08 / 1.41 / 5.69 1.08 / 1.51 / 6.11 1.09 / 1.48 / 6.10

[0098] Each cell in the table represents a percentage of volume / area / width-to-thickness ratio.

[0099] Figure 11 This is a schematic diagram comparing the actual and predicted solder paste amounts according to an embodiment of the present invention. Please refer to... Figure 11 The error between the actual result and the prediction can be less than 10%, and the correlation can be greater than 0.87.

[0100] Figure 12 This is a flowchart of model diagnosis and adjustment according to an embodiment of the present invention. Please refer to... Figure 12 The solder paste inspection machine can measure the amount of solder paste (e.g., height solder paste amount, area solder paste amount, and / or volume solder paste amount) on a slab (i.e., metal plate) or printed circuit board, and use this as a measurement value. The processor 13 can determine whether there is a corresponding artificial intelligence model (AI model) (e.g., the first, second, or third machine learning model mentioned above) for the slab or printed circuit board (step S121). If there is no corresponding model, the processor 13 can perform model training (step S122). For example, data of the slab over the past 30 days (e.g., hole features) is collected, missing or outliers are removed, and retraining is only performed whenever the amount of collected data is greater than or equal to thirty and the parameter settings are greater than or equal to two. Finally, it is verified whether the error between the predicted area, height, and / or volume and the measured value is less than the corresponding threshold value (e.g., 10%, 5%, or 3%). If the verification result is that the error is less than the corresponding threshold value, the third machine learning model can be launched online.

[0101] If a corresponding model already exists, the processor 13 determines whether the work order number has changed (step S123). If it has not changed, the processor 13 confirms whether the prediction error of this model is greater than the error threshold value (e.g., 10%, 5%, or 3%) (step S124). If the prediction error is not greater than the error threshold value, the processor 13 waits for the measurement value of the next steel plate.

[0102] If the prediction error exceeds the error threshold or the work order number has changed, the model is taken offline (step S125), and further trained using measured values ​​(step S126). For example, using data from the most recent work order (e.g., hole features), missing or outliers are removed, and retraining is only performed whenever the collected data volume is greater than or equal to five. Finally, it is verified whether the error between the predicted area, height, and / or volume and the measured value is less than the corresponding threshold (e.g., 10%, 5%, or 3%). If the verification result shows that the error is less than the corresponding threshold, the AI ​​model can be deployed online.

[0103] In one embodiment, the processor 13 may determine that the physical characteristics of the metal plate belong to a first group among a plurality of metal plate groups. The physical characteristics of the metal plate may include, for example, the type of metal plate, its size, the number of components, the number of holes, the upper and lower limits of the hole width-to-thickness ratio, and / or the upper and lower limits of the hole area ratio. For unknown or new metal plates, the processor 13 may obtain measurements of the physical characteristics of that metal plate.

[0104] For a group of metal plates, processor 13 can provide multiple second machine learning models. These second machine learning models have been described previously and will not be repeated here. Each second machine learning model corresponds to one of multiple historical metal plate physical information sets, and the historical metal plate physical information corresponding to any one second machine learning model is different from the historical metal plate physical information corresponding to another second machine learning model. A piece of historical metal plate physical information is a measurement of the physical characteristics of a known metal plate. For example, the measurement of the metal plate corresponding to the training sample mentioned above as a second machine learning model. Different metal plates may have different measurements of their physical characteristics. Therefore, processor 13 can train the training samples corresponding to different metal plates separately, thereby generating multiple second machine learning models. Alternatively, model training can be performed by other devices, and processor 13 can obtain multiple trained second machine learning models.

[0105] Next, processor 13 can group multiple historical metal plate physical information records into multiple metal plate groups. The grouping method (also known as clustering) 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. Grouping can classify historical metal plate physical information and group similar historical metal plate physical information into the same metal plate group.

[0106] Figure 13 This is a schematic diagram illustrating clustering according to an embodiment of the present invention, and Figure 14 This is a schematic diagram illustrating a hierarchical clustering method according to an embodiment of the present invention. Please refer to... Figure 13 and Figure 14 Taking the hierarchical clustering method as an example, processor 13 calculates the distance between samples (e.g., historical metal plate physical information) (step S131). For example, Euclidean distance in the feature coordinate system. Processor 13 groups the closest samples into a group, which becomes a new combined sample. Sample G1 and sample G2 are grouped together (step S132). Next, the distance between samples and / or combined samples is continuously calculated, and the closest samples and / or combined samples are grouped together. Sample G4 and sample G5 are grouped together (step S133). The combined sample of sample G1 and sample G2 is grouped together with sample G3 (step S134). Finally, the combined samples of samples G1 to G3 and the combined samples of samples G4 and G5 are grouped together (step S135), so that all samples G1 to G5 become a combined sample. Figure 14 As shown, the processor 13 cuts the samples according to the distance between samples G1 and G5, and determines the number of groups to be 5 accordingly.

[0107] When a group of metal plates has only one piece of physical information (i.e., corresponding to one metal plate), the second machine learning model corresponding to this physical information can be directly used as the representative of this group of metal plates.

[0108] For a first group among multiple groups of metal plates, the processor 13 can select the second machine learning model with the smallest prediction error among multiple second machine learning models corresponding to the first group as the second machine learning model corresponding to the first group. That is, when a certain group of metal plates (e.g., the first group) has multiple pieces of metal plate physical information (i.e., corresponding to multiple metal plates), the processor 13 can compare the prediction errors of the second machine learning models corresponding to these metal plate physical information and select the second machine learning model with the smallest prediction error as the representative of this metal plate group (i.e., the second machine learning model corresponding to the first group).

[0109] When the physical characteristics of the metal board to be evaluated belong to a first group among multiple metal board groups, the processor 13 may select a second machine learning model corresponding to this first group to generate predicted solder paste amount data. For example, the processor 13 inputs multiple manufacturing process parameters to be evaluated into the second machine learning model, and the second machine learning model outputs predicted solder paste amount data for the printer.

[0110] Figure 15 These are experimental diagrams illustrating verification based on an embodiment of the present invention. Please refer to... Figure 15The study validated the volume percentage, height percentage, and area percentage of solder paste volume. The left half of the graph shows the process parameters used, while the right half shows the parameters generated by the updated first machine learning model. The volume percentage of solder paste volume in the left half exhibits greater variation (between 1.01 and 1.43), while the variation in the right half is smaller (between 1.21 and 1.31). The height percentage of solder paste volume in the left half shows greater variation (between 1.15 and 1.42), even detecting defects, while the variation in the height percentage of solder paste volume in the right half is smaller (between 1.21 and 1.29). The area percentage of solder paste volume in the left half shows greater variation (between 0.83 and 1.04), while the variation in the area percentage of solder paste volume in the right half is smaller (between 0.99 and 1.05). This demonstrates that printing operations using the new process parameters not only exhibit less variation in solder paste volume but also better avoid defects. In addition, printing operations using the new manufacturing process parameters can improve efficiency (e.g., by 8%).

[0111] In summary, in the parameter generation method for printers according to embodiments of the present invention, two machine learning models are used to update the model weights, and the updated machine learning models are used to generate new manufacturing process parameters. Therefore, the amount of solder paste used in printing operations using the new manufacturing process parameters can be close to or the same as the target amount of solder paste. Furthermore, the variability in solder paste amount can be reduced, defects can be reduced or avoided, and the efficiency of printing operations can be improved.

[0112] Although the present invention has been disclosed above by way of embodiments, it is not intended to limit the present invention. Any person 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 defined by the appended claims.

Claims

1. A method for generating parameters for a printer, comprising: inputting target solder paste amount data to a first machine learning model, the first machine learning model outputting predicted manufacturing process parameters for a printer; inputting the predicted manufacturing process parameters to a second machine learning model, the second machine learning model outputting predicted solder paste amount data for the printer, wherein the second machine learning model is trained to learn a correlation between a plurality of solder paste amount samples and a plurality of parameter samples; updating weights of the first machine learning model according to minimizing a prediction error, wherein the prediction error is an error between the target solder paste amount data and the predicted solder paste amount data; and inputting the target solder paste amount data to the updated first machine learning model, the updated first machine learning model outputting new manufacturing process parameters, wherein the new manufacturing process parameters are used to control operation of the printer. 2.The method of claim 1, wherein the step of updating the weights of the first machine learning model according to minimizing the prediction error comprises: determining the prediction error using a loss function, wherein the loss function is half of a square of a numerical difference between the target solder paste amount data and the predicted solder paste amount data. 3.The method of claim 1, wherein the new manufacturing process parameters comprise printing parameters used for operation of the printer, and the target solder paste amount data comprises a solder paste amount for a metal plate. 4.The method of claim 3, wherein types of the printing parameters comprise pressure, print speed, separate speed, separation distance, print gap, and clean rate. 5.The method of claim 1, further comprising: inputting a plurality of features to be evaluated to a third machine learning model, the third machine learning model outputting hole solder paste amount data for the printer, wherein the third machine learning model is trained to learn a correlation between a plurality of hole features and solder paste amount, the hole features comprising a ratio of a plurality of holes to be opened on a metal plate and distribution positions of the holes on the metal plate, and the hole solder paste amount data comprising solder paste amounts for the holes. 6.The method of claim 1, wherein the step of outputting the predicted manufacturing process parameters for the printer by the first machine learning model comprises: outputting initial parameters using the first machine learning model; converting the initial parameters into controllable values using a conversion function, wherein the conversion function is a step function; and normalizing the controllable values into the predicted manufacturing process parameters. 7.The method of claim 6, wherein the step function is formed by adding a plurality of sigmoid functions.

8. The method of claim 1, wherein inputting the predicted manufacturing process parameters to the second machine learning model comprises: inputting the predicted manufacturing process parameters and auxiliary parameters to the second machine learning model, wherein the auxiliary parameters include temperature, humidity, and a positioning location of a metal plate in the printer.

9. The method of claim 1, further comprising: determining that a physical feature of a metal plate belongs to a first group of a plurality of metal plate groups; and selecting the second machine learning model corresponding to the first group for generating the predicted solder paste amount data.

10. The method of claim 9, further comprising: providing a plurality of the second machine learning models, wherein each of the second machine learning models corresponds to one of a plurality of historical metal plate physical information, and any of the second machine learning models corresponds to a different historical metal plate physical information than another of the second machine learning models; grouping the historical metal plate physical information into the metal plate groups; and for the first group of the metal plate groups, selecting one of the second machine learning models corresponding to the first group with a smallest prediction error as the second machine learning model corresponding to the first group.

11. An apparatus for generating parameters for a printer, comprising: a memory storing program codes; and a processor coupled to the memory, loading the program codes and performing: inputting target solder paste amount data to a first machine learning model, the first machine learning model generating a plurality of predicted manufacturing process parameters for a printer; inputting the predicted manufacturing process parameters to a second machine learning model, the second machine learning model generating predicted solder paste amount data for the printer, wherein the second machine learning model is trained to learn a correlation between a plurality of solder paste amount samples and a plurality of parameter samples; updating weights of the first machine learning model according to minimizing a prediction error, wherein the prediction error is an error between the target solder paste amount data and the predicted solder paste amount data; and inputting the target solder paste amount data to the updated first machine learning model, the updated first machine learning model outputting a plurality of new manufacturing process parameters, wherein the new manufacturing process parameters are used to control an operation of the printer.

12. The apparatus of claim 11, wherein the processor further performs: determining the prediction error using a loss function, wherein the loss function is a half of a square of a numerical difference between the target solder paste amount data and the predicted solder paste amount data.

13. The apparatus of claim 11, wherein the new manufacturing process parameters include a plurality of printing parameters used for the operation of the printer, and the target solder paste amount data includes a solder paste amount for a metal plate. ​ ​ ​ 14. The parameter generating device for a printer of claim 13, wherein the types of the printing parameters include pressure, print speed, separate speed, separation distance, print gap, and clean rate.

15. The parameter generating device for a printer of claim 11, wherein the processor further performs: inputting the plurality of evaluated features to a third machine learning model to generate via-hole solder paste amount data for the printer, wherein the third machine learning model is trained to learn the association between a plurality of via features and solder paste amount, the via features include the proportion of openings of the plurality of vias on the metal plate and the distribution positions of the plurality of vias on the metal plate, and the via solder paste amount data includes the solder paste amount for the plurality of vias.

16. The parameter generating device for a printer of claim 11, wherein the processor further performs: outputting initial parameters using the first machine learning model; converting the initial parameters into controllable values using a conversion function, wherein the conversion function is a step function; and normalizing the controllable values into the predicted manufacturing process parameters.

17. The parameter generating device for a printer of claim 16, wherein the step function is formed by adding a plurality of sigmoid functions.

18. The parameter generating device for a printer of claim 11, wherein the processor further performs: inputting the predicted manufacturing process parameters and auxiliary parameters to the second machine learning model, wherein the types of the auxiliary parameters include temperature, humidity, and the positioning position of the metal plate in the printer.

19. The parameter generating device for a printer of claim 11, wherein the processor further performs: determining that the metal plate physical features belong to a first group of a plurality of metal plate groups, wherein the types of the metal plate physical features are related to the size of the metal plate, and the area and thickness of the plurality of vias on the metal plate; and selecting the second machine learning model corresponding to the first group for generating the predicted solder paste amount data.

20. The parameter generating device for a printer of claim 19, wherein the processor further performs: providing a plurality of the second machine learning models, wherein each of the second machine learning models corresponds to one of a plurality of historical metal plate physical information, and any of the second machine learning models corresponds to the historical metal plate physical information different from the historical metal plate physical information corresponding to another of the second machine learning models; grouping the plurality of historical metal plate physical information into the plurality of metal plate groups; and for the first group of the plurality of metal plate groups, selecting one of the second machine learning models corresponding to the first group having the smallest prediction error among the second machine learning models corresponding to the first group as the second machine learning model corresponding to the first group.

Images