Pick-and-Place Machine Rejection Analysis and Abnormal Notification System and Method

Abstract

The present invention discloses a component placement machine reject analysis and anomaly notification system and a component placement machine reject analysis and anomaly notification method. Among them, the component placement machine reject analysis and anomaly notification system links a communication interface to the component placement machine to collect reject data, and converts it into serialized data by a data processing module and stores it in a database. Moreover, it can autonomously check whether there is updated data in the database through a data learning and analysis module, and then classify the anomalies of rejects according to the results of learning and analyzing the serialized data, including anomalies of the suction nozzles, feeders, vacuum pressure values, or incoming materials of the component placement machine. Then, through the anomaly feedback and notification learning module, it can autonomously learn and automatically feedback and adjust the threshold of the reject rate according to the reject data stored in the database. When it is checked that the reject rate of the component placement machine is not less than the feedback threshold, the results of the anomaly classification by the data learning and analysis module can be notified to the engineer in real time for processing to improve its reject rate.

Description

Technical Field

[0001] The present invention provides a component rejection analysis and anomaly notification system and method, in particular, a component rejection analysis and anomaly notification system and method for a pick-and-place machine. Background Art

[0002] In the current process of manufacturing electronic products, various electronic components must be assembled on a circuit board. These electronic components can generally adopt the surface mount technology (SMT) or the dual in-line package (DIP) process. In the SMT production line, a pick-and-place machine is mostly used to place leadless or short-leaded surface mount components (hereinafter referred to as SMC / SMD) on the pads on the surface of the circuit board or other substrates at high speed and high precision, and then welded and assembled by methods such as reflow soldering or dip soldering, so as to improve the production efficiency of surface mounting.

[0003] In the surface mount production line, the most commonly concerned issue is how to effectively reduce production costs and improve production efficiency, which involves the problem of the component rejection rate during the production of the pick-and-place machine. The level of the component rejection rate will not only affect production costs, but also seriously affect the production efficiency and quality of surface mounting. Currently, the technology for reducing the component rejection rate of the pick-and-place machine generally defines a threshold value of the component rejection rate according to the experience of engineers. When the component rejection rate of the pick-and-place machine is greater than this threshold value, it means that the equipment of the pick-and-place machine has an anomaly, and engineers need to conduct on-site inspections. After actual testing and hardware replacement verification, it can be confirmed whether the component rejection rate of the pick-and-place machine can be reduced. However, such a complicated testing and verification process cannot accurately adjust the threshold value of its component rejection rate. In addition, the production conditions, characteristics, etc. of different surface mount components or components are mostly different. Simply relying on human experience and thinking analysis, not only can the problem of too high a component rejection rate not be quickly solved, but also a large amount of man-hours are required for observation and analysis. In addition to being difficult to define the threshold value or standard deviation range of the component rejection rate, it is even more difficult to bring good improvement measures.

[0004] In addition, in the actual production process, there are many key factors that will affect the component rejection rate of the pick-and-place machine. They can be mainly divided into problems such as nozzle deformation, feeder malfunction, abnormal vacuum pressure value, and defective incoming materials. These keys will each generate different anomaly details. For example, the feeder may accumulate dust, the gears may be old and worn, damaged, etc., resulting in poor feeding or poor component picking and rejection, thus affecting its component rejection rate. Therefore, in traditional experience, in addition to consuming man-hours and labor costs, plus these various detail factor variations, it is even more difficult to accurately classify the problems of anomaly occurrence based on human experience, so as to achieve effective anomaly analysis, which is the direction that those engaged in this industry are eager to research and improve. Summary of the Invention

[0005] In order to improve and solve the problem of component ejection of the above-mentioned mounter, the inventor of the present invention designed a component ejection analysis and anomaly notification system and method for the mounter, which can collect the data of the component ejection of the mounter, perform self-learning and analysis, automatically adjust the threshold of the component ejection rate, and classify and notify anomalies of component ejection in real time.

[0006] The main purpose of the present invention is that the communication interface of the component ejection analysis and anomaly notification system for the mounter is connected to the mounter to collect the data of component ejection, and the data is converted into serialized data by the data processing module and stored in the database for data sharing. The data learning and analysis module can automatically check whether there is updated data in the database, and then classify the anomalies of component ejection according to the results of learning and analysis of the serialized data, including anomalies of the nozzle of the mounter, anomalies of the feeder, anomalies of the vacuum pressure value, or anomalies of the incoming material. The anomaly feedback and notification learning module can perform self-learning and analysis based on the component ejection data in the database and automatically feedback and adjust the threshold of the component ejection rate of the mounter for different components. When it is checked that the component ejection rate of the mounter is not less than (i.e., greater than or equal to) the feedback threshold, the results of the corresponding anomaly classification by the data learning and analysis module can be notified to the engineer in real time, so as to help the engineer reduce the process of analyzing different component ejection anomalies and reduce the man-hours consumed by manual analysis, more accurately define the threshold of the component ejection rate, classify different component ejection anomalies, thereby improving the component ejection rate of the mounter, effectively improving the quality and qualification rate of production, and reducing costs. Description of the Drawings

[0007] Figure 1 It is a block diagram of a preferred embodiment of the present invention.

[0008] Figure 2 It is a step flow chart of a preferred embodiment of the present invention.

[0009] Description of the Reference Numerals: 100 - Component Ejection Analysis and Anomaly Notification System for Mounter; 110 - Communication Interface; 120 - Data Processing Module; 130 - Database; 140 - Data Learning and Analysis Module; 150 - Anomaly Feedback and Notification Learning Module; 160 - Monitoring Platform; 200 - Mounter. Detailed Description of the Preferred Embodiment

[0010] To achieve the above-mentioned purposes and effects, the technical means and its structure adopted by the present invention are hereby described in detail with reference to the drawings of the preferred embodiments of the present invention for its structure and function as follows.

[0011] Please refer to Figures 1 to 2As shown, they are respectively the block diagram and the step flow chart of a preferred embodiment of the present invention. It can be clearly seen from the figure that the component placement machine reject analysis and anomaly notification system 100 of the present invention includes a communication interface 110 connected to a component placement machine 200 and extracting its relevant original reject data; a data processing module 120 connected to the communication interface 110 for converting the reject data into a serialized data; a database 130 connected to the communication interface 110 and the data processing module 120 for collecting and storing the reject data and the serialized data; a data learning and analysis module 140 connected to the database 130 for autonomously checking whether the database 130 has updated data, obtaining a learning and analysis result based on the serialized data, and then classifying the anomalies of the component placement machine 200's rejects according to the learning and analysis result, including but not limited to four groups: a nozzle anomaly, a feeder anomaly, a vacuum pressure value anomaly, or a material incoming anomaly on the placement head of the component placement machine 200; and an anomaly feedback notification learning module 150 connected to the data learning and analysis module 140 and the database 130, autonomously learning and analyzing based on the reject data stored in the database 130 and automatically feedback-adjusting the threshold of the reject rate of the component placement machine 200. When it is checked that the reject rate of the component placement machine 200 is not less than (i.e., greater than or equal to) the feedback threshold, a prompt message is output according to the anomaly classification result corresponding to the data learning and analysis module 140 to actively notify the engineer, and the reject state in the production process of the component placement machine 200 is improved in real time.

[0012] In this embodiment, when the data learning and analysis module 140 analyzes the nozzle anomaly in the classification of the component placement machine 200's reject anomalies, the reject data it extracts includes the component number (abbreviated as part number) picked up by the nozzle, the nozzle identifier, the part number of the rejected component (abbreviated as reject part number), the number of rejects, and the total number of picked-up components. When it is analyzed that a certain specific nozzle identifier is particularly prone to rejection behavior, it can be determined that the nozzle may have anomalies such as breakage or blockage, resulting in its inability to pick up components normally.

[0013] If it is analyzing the feeder anomaly, the reject data it extracts includes the feeder identifier, the feeder position, the part number thrown out from the feeder, the number of rejects thrown out from the feeder, and the total number of feeds. When it is analyzed that a feeder at a certain specific position is particularly prone to rejection behavior, and after accumulating a certain amount of data according to different combinations of the part numbers thrown out, it can be determined that an anomaly event has occurred in the feeder. The possible causes of the anomaly include dust accumulation in the feeder, old and worn gears, breakage, etc., resulting in poor feeding and thus affecting its reject rate.

[0014] When analyzing abnormal vacuum pressure values, the extracted rejection data includes the part number picked up by the nozzle, the nozzle identifier, the vacuum pressure value, etc. For example, when picking up a part, a certain pressure threshold will be generated. When it is analyzed that the picking pressure is greater than the threshold, it can be determined that abnormal events such as the nozzle may be damaged, leaking air, or there is a problem with the vacuum pressure machine have occurred. Similarly, when the picking pressure is less than the threshold, it can be determined that the quality of the incoming material is abnormal (such as differences in the flatness of the part surface, different qualities, etc. will all affect the picking pressure).

[0015] Furthermore, when analyzing abnormal incoming materials, the extracted original rejection data includes the part number picked up by the nozzle, the number of rejected parts, the total number of rejected parts, the total number of picked-up parts, and the data extracted from all the above groups. When a certain part number that is thrown out often occurs at different nozzle or feeder positions, but there is no concentrated rejection phenomenon, it can be determined that the incoming material is abnormal.

[0016] In addition, the pick-and-place machine rejection analysis and exception notification system 100 further includes a monitoring platform 160. The preferred implementation of the monitoring platform 160 can be a display device, but it is not limited thereto. It can also further include an alarm device. Taking the display device as an example, it has an operating system or a user interface and is connected to the database 130 and the data learning and analysis module 140 to display the results of learning and analysis stored in the database 130 by the data learning and analysis module 140. It can also be further connected to the exception feedback and notification learning module 150 to receive the prompt information output by the exception feedback and notification learning module 150, and actively notify the engineer to handle it by displaying or issuing an alarm, so as to achieve the function of improving the rejection rate in real time.

[0017] As Figure 2 shown, the present invention further provides a method for analyzing and notifying exceptions in pick-and-place machine rejections, which is applicable to the pick-and-place machine rejection analysis and exception notification system 100 to analyze the reasons for abnormal classification of pick-and-place machine 200 rejections. The pick-and-place machine rejection analysis and exception notification system 100 includes the above-mentioned communication interface 110, data processing module 120, database 130, data learning and analysis module 140, and exception feedback and notification learning module 150. The method for analyzing and notifying exceptions in pick-and-place machine rejections includes the following implementation steps:

[0018] (S101) Start.

[0019] (S102) Connect the communication interface 110 to the pick-and-place machine 200.

[0020] (S103) Collect the data of the rejection data.

[0021] (S104) The data processing module 120 performs data serialization conversion.

[0022] (S105) Store in database 130.

[0023] (S106) Does the data learning and analysis module 140 check for updated data? If yes, execute step (S107); if no, repeat step (S103).

[0024] (S107) Data learning and analysis, and then synchronously execute step (S108) and step (S109).

[0025] (S108) Abnormality classification, and then execute step (S110).

[0026] (S109) The abnormality feedback notification learning module 150 feeds back and adjusts the threshold of the component throwing rate of the mounter 200, and then executes step (S110).

[0027] (S110) Check whether the throwing rate is greater than the threshold? If yes, execute step (S111); if no, repeat step (S107).

[0028] (S111) Actively issue a notification of the result of the abnormality classification.

[0029] (S112) End.

[0030] It can be clearly seen from the figure and the above implementation steps that the component throwing analysis and abnormality notification system 100 of the present invention is an automatic control system connected to the mounter 200 through the communication interface 110, and extracts the relevant values when detecting the operation of the nozzle of the component mounting head, the feeder, the vacuum press, etc., to collect the data as the original throwing data. And the data processing module 120 can perform data serialization conversion on the original throwing data, and then store it in the database 130. Then, when the data learning and analysis module 140 independently checks that there is updated data in the database 130, it will perform the process of data learning and analysis, and classify the abnormalities of the component throwing of the learning and analysis results. At the same time, the abnormality feedback notification learning module 150 will independently learn and analyze according to the throwing data stored in the database 130 and automatically feedback and adjust the threshold of the component throwing rate of the mounter 200 for different components. When it is checked that the throwing rate of the mounter 200 is not less than the feedback threshold, the result of the corresponding analysis and abnormality classification of the data learning and analysis module 140 will be actively notified to the engineer for subsequent related processing procedures to improve the component throwing rate of the mounter 200 in real time.

[0031] Specifically, the above data learning and analysis module 140 processes the serialized data of the component rejection data through data learning and analysis. It mainly uses the regression model algorithm of machine learning to process the data serialized and converted by the data processing module 120, including linear regression models and logistic regression models. Among them, the linear regression model can be further divided into simple linear regression, polynomial regression, and multivariable regression. Preferably, multivariable regression is used to perform curve fitting on the characteristic data of multiple continuous variables, and similar characteristic data are grouped / clustered through unsupervised learning, including centroid clustering (or K-means clustering), hierarchical clustering, affinity propagation clustering, or density clustering, to construct a regression model for classification. However, it is not limited thereto, and supervised learning can also be used for the classification of characteristic data, including support vector machines, Bayesian classifiers, or random forest models (models including multiple decision trees), etc., so as to be able to independently perform classification inference or analyze and predict the results of abnormal classification.

[0032] In addition, the process of the abnormal feedback notification learning module 150 defining and feedback-adjusting the threshold of the component rejection rate of the mounter 200 also uses the component rejection data shared by the database 130 and performs autonomous learning and analysis using the regression model algorithm of machine learning for prediction, so as to define the threshold of the component rejection rate of the mounter 200 that conforms to the actual production process. It can also automatically feedback and adjust the threshold of the component rejection rate of the mounter 200 according to different component production conditions, characteristics, etc., or obtain a standard deviation of the component rejection rate of the mounter 200, and use the upper and lower limit ranges of the standard deviation as the determination of whether there is a sudden increase in the component rejection rate and whether it is necessary to notify the engineer of component rejection abnormalities. When the abnormal feedback notification learning module 150 checks that the component rejection rate of the mounter 200 is greater than the feedback threshold or the standard deviation range, it will actively notify the engineer of the corresponding abnormal classification result of the data learning and analysis module 140 for subsequent processing procedures to improve the component rejection rate of the mounter 200 in real time.

[0033] Therefore, the component placement machine reject analysis and anomaly notification system 100 of the present invention can collect reject data from the component placement machine 200 through the communication interface 110. The data processing module 120 serializes and converts the data and stores it in the database 130 for data sharing. After the data learning and analysis module 140 and the anomaly feedback and notification learning module 150 independently analyze and learn using the regression model algorithm of machine learning, in addition to automatically feedback-adjusting the threshold of the reject rate of the component placement machine 200 according to different components, it can also classify (but not limited to the above four anomaly classification groups) and notify the reject anomaly phenomenon in real time, so as to help engineers reduce the process of analyzing different reject anomalies, while reducing the man-hours consumed by manual analysis, more accurately define the threshold of the reject rate of the component placement machine 200, appropriately classify different reject anomalies, thereby improving the reject rate, effectively improving the production quality and qualification rate, and reducing costs.

[0034] The above detailed description is only for a preferred feasible embodiment of the present invention, but this embodiment is not used to limit the scope of the patent application of the present invention. All other equivalent changes and modifications made without departing from the technical spirit disclosed by the present invention should be included in the patent scope covered by the present invention.

Claims

1. A component placement machine material ejection analysis and anomaly notification system, characterized in that Comprising: A communication interface is connected to a placement machine and captures its original component rejection data; A data processing module is connected to the communication interface for converting the component rejection data into serialized data; A database is connected to the communication interface and the data processing module for storing the component rejection data and the serialized data; A data learning and analysis module is connected to the database for autonomously checking whether the database has updated data, obtaining a learning and analysis result based on the serialized data, and then classifying component rejection anomalies according to the learning and analysis result, including a nozzle anomaly, a feeder anomaly, a vacuum pressure value anomaly, or a material supply anomaly of the placement machine; And An anomaly feedback notification and learning module is connected to the data learning and analysis module and the database, and autonomously learns and analyzes based on the component rejection data stored in the database and automatically feedback-adjusts the threshold of the component rejection rate of the placement machine. When the anomaly feedback notification and learning module checks that the component rejection rate of the placement machine is not less than the feedback threshold, it actively issues a notification of the anomaly classification result of the data learning and analysis module; Wherein, the process of the anomaly feedback notification and learning module defining and feedback-adjusting the threshold of the component rejection rate of the placement machine includes: Using the component rejection data shared by the database and autonomously learning and analyzing using the regression model algorithm of machine learning for prediction to define the threshold of the component rejection rate of the placement machine that conforms to the actual production process; Automatically feedback-adjusting the threshold of the component rejection rate of the placement machine according to different component production conditions and characteristics; or Obtaining a standard deviation of the component rejection rate of the placement machine, and using the upper and lower limit ranges of the standard deviation as the determination of whether there is a sudden increase in the component rejection rate and whether to notify component rejection anomalies. When the anomaly feedback notification and learning module checks that the component rejection rate of the placement machine is greater than the standard deviation range, it actively issues a notification of the anomaly classification result of the data learning and analysis module.

2. The component placement machine reject analysis and anomaly notification system according to claim 1, wherein The data learning and analysis module processes the serialized data using the regression model algorithm of machine learning, and classifies similar feature data through supervised or unsupervised learning, and can thus autonomously predict the anomaly classification result.

3. The component placement machine material throwing analysis and anomaly notification system according to claim 2, wherein, The machine learning uses the multiple regression algorithm of the linear regression model to perform curve fitting on the feature data of the serialized data with multiple continuous variables, and clusters / group similar feature data through unsupervised learning, and the clustering includes centroid clustering, hierarchical clustering, affinity propagation clustering, or density clustering to construct a regression model for classification.

4. The component placement machine material throwing analysis and exception notification system according to claim 2, characterized in that, The machine learning uses the multiple regression algorithm of the linear regression model to perform curve fitting on the feature data of the serialized data with multiple continuous variables, and uses the Bayesian classifier of supervised learning to classify the feature data.

5. The component placement machine material throwing analysis and anomaly notification system according to claim 1, wherein, It further includes a monitoring platform, which is connected to the database and the data learning and analysis module for displaying the learning and analysis results of the data learning and analysis module stored in the database.

6. The component placement machine reject analysis and anomaly notification system according to claim 5, wherein, The monitoring platform is connected to the anomaly feedback notification and learning module for receiving a prompt message of the anomaly classification result output by the anomaly feedback notification and learning module and actively notifying by means of display or alarm.

7. A method for analyzing component rejection and reporting anomalies in a pick-and-place machine, applicable to a component rejection analysis and anomaly reporting system of a pick-and-place machine, characterized in that, Including the following steps: Provide a communication interface to connect to a placement machine and collect the original component rejection data of the placement machine; Provide a data processing module to convert the component rejection data of the placement machine into serialized data and store it in a database; Provide a data learning and analysis module to check whether there is updated data in the database, obtain a learning and analysis result based on the serialized data, and then classify the component rejection anomalies according to the learning and analysis result, including a nozzle anomaly, a feeder anomaly, a vacuum pressure value anomaly, or a material supply anomaly of the placement machine; Provide an anomaly feedback notification and learning module to autonomously learn, analyze, and automatically feedback and adjust the threshold of the component rejection rate of the placement machine based on the component rejection data stored in the database. When it is checked that the component rejection rate of the placement machine is not less than the feedback threshold, actively issue a notification of the anomaly classification result corresponding to the data learning and analysis module; Among them, the process of the anomaly feedback notification and learning module defining and feedback-adjusting the threshold of the component rejection rate of the placement machine includes: Use the component rejection data shared by the database and use the regression model algorithm of machine learning for autonomous learning and analysis for prediction to define the threshold of the component rejection rate of the placement machine that conforms to the actual production process; Automatically feedback and adjust the threshold of the component rejection rate of the placement machine according to the production conditions and characteristics of different components; or Obtain a standard deviation of the component rejection rate of the placement machine, and use the upper and lower limit ranges of the standard deviation as the determination of whether there is a sudden increase in the component rejection rate and whether to notify component rejection anomalies. When the anomaly feedback notification and learning module checks that the component rejection rate of the placement machine is greater than the standard deviation range, actively issue a notification of the anomaly classification result corresponding to the data learning and analysis module.

8. The method for analyzing component rejection and abnormal notification of a placement machine according to claim 7, characterized in that, The data learning and analysis module processes the serialized data using the regression model algorithm of machine learning, and classifies similar feature data through supervised or unsupervised learning, and then can autonomously predict the anomaly classification result.

9. The method for analyzing component ejection and abnormal notification of a mounter according to claim 8, wherein, The machine learning uses the multiple regression algorithm of the linear regression model to perform curve fitting on the feature data of the serialized data with multiple continuous variables, and clusters similar feature data through unsupervised learning. The clustering includes center clustering, hierarchical clustering, affinity propagation clustering, or density clustering to construct a regression model for classification.

10. The method for analyzing component ejection and reporting anomalies of a component mounter according to claim 8, wherein, The machine learning uses the multiple regression algorithm of the linear regression model to perform curve fitting on the feature data of the serialized data with multiple continuous variables, and uses the Bayesian classifier of supervised learning to classify the feature data.

11. The method for analyzing component ejection and abnormal notification of a pick-and-place machine according to claim 7, wherein There is also provided a monitoring platform, and the monitoring platform displays the learning and analysis results of the data learning and analysis module stored in the database.

12. The method for analyzing component rejection and abnormal notification of a component mounter according to claim 11, characterized in that The monitoring platform actively notifies the anomaly classification result of the anomaly feedback notification and learning module by means of display or alarm.

Images