TWI682294B - Soldering process parameters suggestion method - Google Patents

Abstract

A soldering process parameters suggestion method includes steps of: establishing a material component database; establishing a working parameter database; analyzing material and component characteristics required for a new soldering process, and comparing the characteristics with information in the material component database; selecting the operating parameters corresponding to the materials and component characteristics most similar to those required for the new soldering process; performing the soldering process using the operating parameters corresponding to the materials and component characteristics most similar to those required for the new soldering process; measuring and recording the soldering process information and the final product information; determining whether the final product of the solder process meets the quality control requirements; and using the machine learning method to fit the solder process information and the final product information of the solder process to get the operating parameters for the next soldering process when the final product does not meet the quality control requirements.

Description

Recommended method of soldering process parameters

The present invention relates to a soldering process parameter recommendation method, and in particular to a soldering process parameter recommendation method using a machine learning method.

Obtaining soldering parameters is extremely important in the soldering process, at the same time, product yield, production stability, production efficiency, production energy consumption, etc. must be considered. How to quickly find the operating parameters of new products, adjust the operating parameters of modified products, and how to continuously optimize the soldering parameters after the on-line production operation, has always relied on the guidance of expert experts with deep soldering experience, and spent a lot of time to repeat Test verification not only consumes manpower and costs, but also cannot determine the time required to obtain the best parameters, but also limits the development of production line automation and flexible production.

The solder joint operation is related to the complex parameter condition learning, which results in different optimal solder joint generation conditions. In the conventional process method, the trial and error method is often used to find its solder parameters, and the parameters are determined through accumulation of multiple tests, but it is often limited to the overall batch to determine the parameters, which causes a lot of time to test the parameters and cannot be achieved. The best quality is in production.

The invention proposes a soldering process parameter suggestion method to solve the problems of the prior art.

In an embodiment of the invention, a soldering process parameter suggestion method includes the following steps. Collect the material and component characteristics related to soldering process to create a material component database; collect the operation information of materials or components in the corresponding material component database to create a working parameter database; analyze the materials and components required for a new soldering process Characteristics, and compare the similarity with the information in the material component database; select the operation parameters corresponding to the most similar materials and component characteristics required in the new soldering process in the operation parameter database; use the corresponding most similar to the new soldering process The required material and component characteristics of the operating parameters to perform the soldering process; measure and record the soldering process information and final product information of the soldering process; detect whether the final product of the soldering process meets the quality control requirements; and when the final product does not meet the quality control requirements At the time, the machine learning method is used to fit the soldering process information of the soldering process and the final product information to obtain the operating parameters of the next soldering process.

In an embodiment of the present invention, when the final product meets the quality control requirements, the operating parameters included in the soldering process information of the soldering process are accepted as the optimal operating parameters.

In an embodiment of the present invention, when the final product does not meet the quality control requirements, the machine learning model is continuously used to fit the soldering process information of the soldering process and the final product information to obtain the operating parameters of the next soldering process until The final product meets the quality control requirements, and accepts the operation parameters contained in the soldering process information of the soldering process as the optimal operation parameters.

In an embodiment of the present invention, the material and component database at least includes thermal characteristics, optical characteristics, and structural dimension information of materials and components related to the soldering process.

In an embodiment of the invention, the working parameter database at least includes temperature parameters, laser spot parameters and heating time parameters.

In an embodiment of the invention, the soldering process information includes at least a soldering temperature curve, a dynamic process image of soldering, and a dynamic process image of soldering.

In an embodiment of the invention, the machine learning method further includes the use of Decision Tree Algorithm, Support Vector Machine, Support Neural Network, Artificial Neural Network, and Bayesian Algorithms ) Or any combination thereof for Dimensionality Reduction Algorithms.

In an embodiment of the invention, the machine learning method further includes using at least one of distance clustering (distance-based clustering) and density clustering (density-based clustering) or a combination thereof to perform structural clustering. The distance clustering includes, for example, K-means (H-means), Hierarchical clustering (Hierarchical clustering), and Gaussian mixture model (GMM). The density clustering includes, for example, a density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise), a level set tree clustering (Level Set Tree clustering) algorithm, and the like.

In an embodiment of the present invention, the machine learning method further includes using at least one of regression algorithm and classification algorithm or a combination thereof to establish the conditional probability calculation core. Regression algorithms include, for example, generalized linear regression, support vector machine, decision tree, random forest, gradient boosting machine, neural network Wait. Classification algorithms include, for example, generalized linear regression, support vector machine, decision tree, random forest, gradient boosting machine, neural network Wait.

In summary, the soldering process parameter suggestion method of the present invention can be applied to production systems with complete or incomplete data systems and a large number of samples or a small variety of production lines, so that various production lines can move towards automated production with improved efficiency.

The above description will be described in detail in the following embodiments, and the technical solutions of the present invention will be further explained.

In order to make the description of the present invention more detailed and complete, reference may be made to the accompanying drawings and various embodiments described below. The same numbers in the drawings represent the same or similar elements. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessary restrictions to the present invention.

The present invention proposes an intelligent soldering process parameter recommendation method, which uses machine learning methods to target solder joint background parameters, including printed circuit board material characteristics, circuit design, component characteristics, tin material characteristics, tin supply conditions, solder joint dynamic generation temperature, plus Temperature mode and parameters, machine vision images, etc. are analyzed to obtain the best recommended operating parameters, so as to effectively reduce the number of tests for obtaining operating parameters, and then realize solder automation.

Please refer to FIG. 1, which illustrates a soldering process parameter recommendation method according to an embodiment of the present invention. Step 102 of this method is to collect information about the characteristics of the soldering process that can be used as machine learning.

In step 104, the characteristic information of the soldering process collected in step 102 is divided into appropriate categories to facilitate the establishment of a database and execution of machine learning. In the embodiment of this case, the characteristic information of the soldering process collected in step 102 can be divided into four categories. The first category is the characteristics of materials and components related to soldering processes, such as the thermal characteristics (specific heat, thermal conductivity, etc.), optical characteristics (reflectivity, etc.) of printed circuit boards, pads, and tin materials, and the three-dimensional spatial distribution of the pins of electronic components. To create a material component database. The second type is soldering operation information, such as reference temperature settings, laser spot size and power, heating time, etc., to create a database of operation parameters. The operation parameters in the operation parameter database should correspond to the materials and component characteristics in the material component database. The third type is the soldering process execution information, such as the soldering temperature curve, the dynamic process image of soldering or the dynamic process image of soldering, etc., to establish a soldering process database. The fourth category is the solder quality results, that is, the final product information of the soldering process, such as the solder structure test results of the final product, including ICT test (In-Circuit-Test, referred to as ICT test), X-ray (X-RAY) test, The results of AOI inspection (Automated Optical Inspection, referred to as AOI inspection) or solder slice inspection are used to establish a final product database.

In steps 106 and 108, the machine learning is used to establish the correspondence between the solder process information and the solder quality results for the information in the third and fourth types of databases, that is, a quality prediction model is generated. There are many machine learning methods that can be used to build quality prediction models.

In some embodiments, multiple machine learning models can be combined, such as deep learning, random forest, support vector machine, etc., and the temperature curve and solder image information are fused to establish the correspondence between the solder process information and the solder quality results.

In some embodiments, multiple machine learning models can be combined to establish a corresponding relationship between the soldering process temperature curve and the soldering quality result, for example, the soldering process time series is divided into several intervals, and the characteristic values (such as average Value, standard deviation, area, etc.). For example, the time series of the soldering process is divided into the temperature curve of the soldering stage, the temperature curve of the (solder) infiltration stage, and the temperature curve of the (solder) curing conversion stage, etc., and machine learning is used to find the temperature curve and solder quality of each stage Correspondence of results.

In some embodiments, the solder quality result may be that the final product of the solder structure meets the requirements for quality control or does not meet the requirements for quality control, such as non-conforming products such as tin explosion, empty soldering, cold soldering, or no soldering of the solder structure Manage the required status, but not limited to this. Using machine learning to find the correspondence between the temperature curve at each stage and the solder structure that does not meet the quality control requirements is still quite helpful for the automation of the soldering process.

In some embodiments, the machine learning method may include the use of Decision Tree Algorithm, Support Vector Machine, Artificial Neural Network, Bayesian Algorithms or Bayesian Algorithms. Any combination of the above information of the third and fourth types of databases can be used to perform dimensionality reduction algorithms (Dimensionality Reduction Algorithms) to help find the correspondence between the soldering process information and the solder quality results.

In some embodiments, the machine learning method may include structural grouping using at least one of distance grouping and density grouping or a combination thereof. The distance clustering includes, for example, K-means (H-means), Hierarchical clustering (Hierarchical clustering), and Gaussian mixture model (GMM). Density clustering includes, for example, density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm, level set tree clustering (Level Set Tree clustering), etc., in order to find cases with similar characteristics.

In some embodiments, the machine learning method may include using at least one of a regression model and a classification algorithm or a combination thereof to establish a conditional probability calculation core to facilitate the establishment of a correspondence between solder process information and solder quality results . Regression algorithms include, for example, generalized linear regression, support vector machine, decision tree, random forest, random boosting machine, gradient boosting machine, neural network Wait. Classification algorithms include, for example, generalized linear regression, support vector machine, decision tree, random forest, gradient boosting machine, neural network Wait.

In step 110, the predictive model created by the machine learning described above is used to generate operating parameters of the soldering process, and the soldering process is executed using the operating parameters.

In step 112, it is verified whether the operation parameters of the soldering process can obtain the final product that meets the quality control requirements. The verification method includes testing and recording the final product of the solder structure. The method for detecting the final product of the solder structure may be the aforementioned ICT test, X-RAY test, AOI test, solder slice test, or a combination of two or more of the above tests. The test results of X-RAY test, AOI test and solder slice test are images, but the test results of ICT test are electrical tests to determine whether the solder structure has defects such as short circuit, open circuit or open circuit.

In step 114, when the final product of the soldering process meets or passes the quality control requirements, the operating parameters included in the soldering process information of the soldering process are accepted as the optimal operating parameters.

In most cases, it is difficult to obtain the best operating parameters in the first prediction model using machine learning methods. Therefore, in step 116, it is necessary to add characteristic information of the soldering process, and perform steps 106, 108, 110, and 112 again until the final product of the soldering process meets or passes the quality control requirements. In some embodiments, the information related to the soldering process added in step 116 is the process information of the soldering process performed in step 110 and the solder quality result information. In some embodiments, the feature information of the soldering process added in step 116 may be an image difference comparison between the final product after the latest soldering process and the database of the final product.

Please refer to FIGS. 2 and 3, which illustrate a soldering process parameter recommendation method according to another embodiment of the present invention. In steps 104a and 104b, materials and component characteristics related to the soldering process are sequentially collected to create a material component database, and operation information of materials or components in the corresponding material component database is collected to create an operation parameter database.

In step 106a, when a new soldering process requirement is proposed, the materials and device characteristics required by the new soldering process are collated and analyzed, and compared with the information in the material and component database created in step 104a, for example Use correlation analysis and density clustering algorithms to select cases with similar characteristics.

In step 108a, select the operation parameter corresponding to the most similar material and component characteristics required by the new soldering process from the operation parameter database created in step 104b.

In step 110a, the soldering process is performed using the operating parameters obtained in step 108a that correspond to the most similar materials and device characteristics required by the new soldering process.

In step 112a, the soldering process information and final product information of the soldering process of step 110a are measured and recorded.

In step 112b, it is checked whether the final product of the soldering process recorded in step 112a meets the quality control requirements. The inspection method includes inspecting and recording the final product of the solder structure. The method for detecting the final product of the solder structure may be the aforementioned ICT test, X-RAY test, AOI test, solder slice test, or a combination of two or more of the above tests. The test results of X-RAY test, AOI test and solder slice test are images, but the test results of ICT test are electrical tests to determine whether the solder structure has defects such as short circuit, open circuit or open circuit.

In step 118, when the final product does not meet the quality control requirements, the machine learning method is used to fit the soldering process information recorded in step 112a to the final product information to obtain the operation parameters of the next soldering process. In some embodiments, the soldering process information of the soldering process and the final product information are input into a Bayesian Optimization model to be fitted to obtain the operating parameters of the next soldering process, that is, repeating the steps using the new operating parameters 110a, 112a, 112b until the final product of the soldering process meets or passes quality control requirements to obtain the best operating parameters.

When the above soldering process parameter recommendation method is applied to a production line with a complete data system, the production line material information system can be connected to an intelligent soldering process parameter recommendation system. After machine learning and intelligent calculation, the system will provide users with recommended operating parameters. And in the test process, the actual soldering situation is recorded in real time and fed back to the system adjustment calculation model to automatically converge the optimal operating parameters, which not only effectively shortens the parameter test, but also completely records the soldering production process. When the above soldering process parameter recommendation method is applied to a production line where the data system is not yet sound, a limited range of recommended operating parameters can be obtained for the limited production material information in the existing database, and the system test loop can be entered to effectively converge the best operation Parameters, and shorten the timeline for a production line that lacks a sound information system to move towards automated production, and establish an information system simultaneously. When the above soldering process parameter recommendation method is applied to a large number of production lines with a small number of high-capacity production lines, data can be quickly accumulated to establish the optimal parameter model to improve the efficiency of the production line. When the above soldering process parameter recommendation method is applied to a small number of diverse production lines, it can effectively shorten the time to achieve stable production and reduce test costs, and maintain continuous production of the production line.

In summary, the soldering process parameter suggestion method of the present invention can be applied to production systems with complete or incomplete data systems and a large number of samples or a small variety of production lines, so that various production lines can move towards automated production with improved efficiency.

Although the present invention has been disclosed as above in an embodiment, it is not intended to limit the present invention. Anyone who is familiar with this skill can make various modifications and retouching without departing from the spirit and scope of the present invention, so the protection of the present invention The scope shall be determined by the scope of the attached patent application.

In order to make the above and other objects, features, advantages and embodiments of the present invention more obvious and understandable, the attached symbols are described as follows:

102‧‧‧Step

104‧‧‧Step

104a‧‧‧Step

104b‧‧‧Step

106‧‧‧Step

106a‧‧‧Step

108‧‧‧Step

108a‧‧‧Step

110‧‧‧Step

110a‧‧‧Step

112‧‧‧Step

112a‧‧‧Step

112b‧‧‧Step

114‧‧‧Step

116‧‧‧Step

118‧‧‧Step

In order to make the above and other objects, features, advantages and embodiments of the present invention more obvious and understandable, the drawings are described as follows: FIG. 1 illustrates a method for recommending soldering process parameters according to an embodiment of the present invention; And FIGS. 2 and 3 illustrate a method for recommending soldering process parameters according to another embodiment of the present invention.

102~116‧‧‧Step

Claims (14)

A method for recommending soldering process parameters, including: collecting material and component characteristics related to the soldering process to create a material component database; collecting the material or operation information of the component in the material component database to create an operation parameter database ; Analyze the characteristics of materials and components required by a new soldering process, and compare the similarity with the information in the material component database; select the materials in the operation parameter database that are most similar to the new soldering process Operating parameters corresponding to component characteristics; using the operating parameters corresponding to the most similar materials and component characteristics required by the new soldering process to execute the soldering process; measuring and recording the soldering process information and final product information of the soldering process; detecting the soldering process Whether the final product meets the quality control requirements; and when the final product does not meet the quality control requirements, the machine learning method is used to fit the soldering process information of the soldering process and the final product information to obtain the operating parameters of the next soldering process. The method as described in claim 1, further comprising: when the final product meets the quality control requirements, accepting the operation parameters included in the soldering process information of the soldering process as the optimal operation parameters. The method as described in claim 1 further includes: when the final product does not meet the quality control requirements, continuously using the machine learning model to fit the soldering process information of the soldering process and the final product information to obtain the next soldering process Operating parameters until the final product meets quality control requirements. The method as described in claim 3 further includes: when the final product meets the quality control requirements, accepting the operation parameters included in the soldering process information of the soldering process as the optimal operation parameters. The method according to claim 1, wherein the material component database at least contains thermal characteristics, optical characteristics, and structural dimension information of materials and components related to the soldering process. The method according to claim 1, wherein the operation parameter database includes at least a temperature parameter, a laser spot parameter and a heating time parameter. The method according to claim 1, wherein the soldering process information includes at least a soldering temperature curve, a dynamic process image of soldering, and a dynamic process image of soldering. The method according to claim 1, wherein the machine learning method further includes: using a decision tree algorithm (Decision Tree Algorithm), support vector machine (Support Vector Machine), artificial neural network (Artificial Neural Network), Bayesian classification Dimensionality Reduction Algorithms (Bayesian Algorithms) or any combination thereof. The method according to claim 8, wherein the machine learning method further comprises: using at least one of distance clustering (distance-based clustering) and density clustering (density-based clustering) or a combination thereof for structural clustering. The method according to claim 9, wherein the distance clustering is at least one of K-means, Hierarchical clustering, and Gaussian mixture model (GMM) methods or a combination thereof. The method according to claim 9, wherein the density clustering is at least one of a density-based clustering algorithm (Density-Based Spatial Clustering of Applications with Noise) and a level set tree clustering algorithm (Level Set Tree clustering) One or a combination of them. The method according to claim 9, wherein the machine learning method further comprises: using at least one of a regression algorithm and a classification algorithm (classification) or a combination thereof to establish a conditional probability calculation core. The method according to claim 12, wherein the regression algorithm is generalized linear regression, support vector machine (support vector machine), decision tree (decision tree), random forest (random forest), gradient boosting machine (gradient boosting machine) At least one of neural network (neural network) methods or a combination thereof. The method according to claim 12, wherein the classification algorithm is generalized linear regression, support vector machine (support vector machine), decision tree (decision tree), random forest (random forest), gradient boosting machine (gradient boosting machine) At least one of neural network (neural network) methods or a combination thereof.

Images