KR101530848B1 - Apparatus and method for quality control using datamining in manufacturing process - Google Patents
Apparatus and method for quality control using datamining in manufacturing process Download PDFInfo
- Publication number
- KR101530848B1 KR101530848B1 KR1020120104587A KR20120104587A KR101530848B1 KR 101530848 B1 KR101530848 B1 KR 101530848B1 KR 1020120104587 A KR1020120104587 A KR 1020120104587A KR 20120104587 A KR20120104587 A KR 20120104587A KR 101530848 B1 KR101530848 B1 KR 101530848B1
- Authority
- KR
- South Korea
- Prior art keywords
- quality
- data
- kernel
- ontology
- svm
- Prior art date
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
An apparatus and method for managing quality in a production process using data mining are disclosed. According to an embodiment of the present invention, a method for managing quality in a production process includes inputting a quality parameter for a good and a defective product corresponding to a process factor corresponding to a factor affecting quality in a production process, The ontology representing the relationship between the factor and the quality is generated, the observation point to be observed is determined based on the generated ontology, the observation coordinates of the process parameter and the measurement data are input in real time through the production process, Linearly separates nonlinear values for the input observation coordinates and measurement data, performs SVM (Support Vector Machine) learning based on the separated linear values, and predicts the current process state from SVM learning results.
Description
The present invention relates to a technique for managing quality in a production process, and more particularly, to a device, a method, and a method for managing quality by predicting quality in a production process by using data mining techniques and controlling factors influencing predicted quality And the like.
When a single product is produced through a series of processes consisting of a plurality of steps, the integrity and reliability according to the organic connection is very important for each process. In order to achieve this integrity, it is necessary to develop an efficient quality control system that can identify the abnormalities of each process and the cause diagnosis in the production process.
In the past, we mainly focused on post-quality control by screening finished products and disposing or reprocessing low-quality products. However, it has been pointed out that such post-quality control is not an effective countermeasure in that the quality improvement of the product is not large compared to the cost of quality control. Therefore, the industry has begun to put a greater emphasis on pre-quality control, which can control the manufacturing process itself rather than the process of the finished product, thereby preventing the production of defective products or assuring the reliability of the quality.
Statistical Process Control (SPC) is a method of managing the process efficiently by repeating the Plan-Do-Check-Act (PDCA) cycle in order to achieve the quality or productivity targets required by the process. . In particular, the statistical process management, with the help of the statistical analysis technique, grasps the cause of fluctuation of the process quality and the ability status of the process, performs the PDCA cycle so that the given quality target can be achieved, and manages the continuous quality improvement . Non-patent literature cited below provides an overview of this statistical process control.
SUMMARY OF THE INVENTION It is an object of the present invention to solve the problems that the conventional multivariate statistical methods lose efficiency due to nonlinearity, To overcome the limitation that there is no technical means to directly and quickly grasp the reliability change even if the process is destabilized due to the deterioration of the machine performance.
According to an aspect of the present invention, there is provided a method for managing quality in a production process, the method comprising: Generating an ontology indicating a relationship between 'factor' and 'quality' by receiving quality variables for good and defective products; Determining an observation point to be observed on the basis of the generated ontology and inputting observation coordinates and measurement data of the process parameter for each observation point in real time through a production process; Linearly separating nonlinear values of the input observation coordinates and measurement data, and performing SVM (Support Vector Machine) learning based on the separated linear values; And predicting a current process state from the SVM learning result.
The method of managing quality in the production process according to an exemplary embodiment may further include storing an experimentally verified rule so as to predict a quality parameter based on the ontology and a process parameter.
In the method of managing quality in the production process according to an embodiment, the relation between the 'factor' and the 'quality' may be set through learning using a neural network or a decision tree . In addition, the relation between the 'factor' and the 'quality' is determined by considering the influence of the weight of the process parameters affecting the quality variable and the interactions among the process parameters, Can be calculated by reflecting the contribution of the user.
In the method of managing quality in the production process according to an exemplary embodiment, the step of linearly separating the non-linear values includes mapping non-linear values to a virtual space using a kernel function, The hyperplane of the data can be calculated.
In the method of managing quality in the production process according to an exemplary embodiment, the step of predicting the current process state may include calculating a stability or a predicted yield of a process that is calculated in consideration of a management degree criterion based on the ontology, .
In the method of managing quality in the production process according to an exemplary embodiment, the step of predicting the current process state is performed based on current measurement data from the ontology and history data using pattern recognition based on SVM And inducing a change in the process by providing the predicted process state to the user. In addition, when the change of the process is reflected in the production process, the step of updating the ontology by receiving a new process parameter and a quality parameter may be further included.
According to another aspect of the present invention, there is provided a method of managing quality in a PCB (Printed Circuit Board) production process using heat fusion, Generating an ontology that indicates a relationship between 'parameter' and 'quality' by receiving a variable; Determining an observation point to be observed on the basis of the generated ontology and receiving observation coordinates and measurement data of the process parameter for each observation point in real time using a sensor; Linearly separating nonlinear values of the input observation coordinates and measurement data using a kernel function and performing SVM learning using pattern matching based on the separated linear values; And predicting a process state occurring due to an influence of a current process parameter from the SVM learning result.
In a method of managing quality in a PCB production process using the thermal fusion according to another embodiment, the process parameter may include at least one of temperature, vibration, or noise.
In a method of managing quality in a PCB manufacturing process using the thermal fusion welding according to another embodiment, the input values include at least one of an equipment identification number, a date and time of a job, a process parameter (temperature), an equipment point number Value, set value, upper temperature limit, and lower temperature limit, and can be transferred for SVM learning.
In the method of managing quality in a PCB manufacturing process using the thermal fusion welding according to another embodiment, the kernel function may be selected from at least one of an RBF kernel, a Polynomial kernel, and a Linear kernel to designate a kernel initial variable, The non-linear values of the non-linear data can be calculated by mapping the non-linear values on the virtual space.
The present invention also provides a computer readable recording medium on which a program for executing quality control in a production process described above and a method for managing quality in a PCB production process using thermal welding is recorded.
According to an aspect of the present invention, there is provided an apparatus for managing quality in a production process, the apparatus comprising: a sensor for sensing a process factor corresponding to a factor affecting quality in a production process; An input unit for inputting quality parameters for good and defective products corresponding to the results of the comparison; A processing unit for generating an ontology representing a relationship between 'factor' and 'quality' from the input process parameter and quality variable, and determining an observation point to be observed based on the generated ontology; And a storage unit for storing experimentally-validated rules for predicting quality parameters based on the generated ontology and process parameters, wherein the input unit is capable of observing the process parameters in real- The processing unit linearly separates the non-linear values of the observation coordinates and the measurement data input through the input unit, performs SVM learning based on the separated linear values, and extracts, from the SVM learning result, Predict the current process state.
In an apparatus for managing quality in the production process according to an embodiment, the processing unit may perform a non-linear process by mapping the nonlinear values to a virtual space using a kernel function selected from at least one of an RBF kernel, a Polynomial kernel, The hyperplane of the data can be calculated.
In an apparatus for managing quality in the production process according to an exemplary embodiment, the processing unit predicts a current process state based on current measurement data from the ontology and history data using pattern recognition based on SVM, The process state can be induced to the user. In addition, when the change of the process is reflected on the production process, the processing unit can receive the new process parameter and the quality variable through the input unit and update the ontology.
Embodiments of the present invention accumulate various process variables extracted from a manufacturing process and utilize data mining techniques to derive a relationship between factors and results from the stored data and to create a new process set combination that can maintain optimal quality therefrom Thus, it is possible to predict a process state that changes in accordance with a situation occurring in real time while deviating from a process standard of a fixed value, and it is possible to quickly grasp the change in reliability in real time even if the performance of the machine deteriorates over time It is possible to provide the user with a dynamic setting value suitable for a given environmental factor.
1 is a diagram illustrating a quality management model using data mining adopted by embodiments of the present invention.
Figure 2 is a more detailed block diagram of the functional blocks of the quality management model of Figure 1 that the embodiments of the present invention employ.
3 is a flowchart illustrating a method of managing quality in a production process according to an embodiment of the present invention.
4 is a flowchart illustrating an additional process that may optionally be added in the quality management method of FIG. 3 according to an embodiment of the present invention.
FIG. 5 is a view for explaining process parameters that appear in a heat seal mass ram process in a PCB (Printed Circuit Board) production process of an automobile door trim shown as an example of a production process.
FIG. 6 is a diagram for explaining a kernel function utilized by embodiments of the present invention to pre-separate non-linear values. FIG.
FIG. 7 is a graph illustrating a result of performing operations using various kernel functions.
8 is a diagram illustrating pseudo code that implements the support vector machine (SVM) learning and verification method adopted by the embodiments of the present invention.
9 is a diagram illustrating data converted according to a method of managing quality in a PCB manufacturing process according to another embodiment of the present invention.
10 is a view for explaining a process of determining an optimal SVM variable according to a method of managing quality in a PCB production process according to another embodiment of the present invention.
11 is a diagram illustrating an SVM classification model in which learned data is expressed in three dimensions according to a method of managing quality in a production process according to embodiments of the present invention.
12 is a block diagram illustrating an apparatus for managing quality in a production process according to an embodiment of the present invention.
Prior to describing the embodiments of the present invention, the technical means adopted by the embodiments of the present invention will be outlined to solve the problems occurring in the conventional statistical process control.
The statistical process monitoring method is advantageous in that it can be relatively easily monitored by combining statistical processing if only good quality process data is given, and tools for analyzing the data of the process can be obtained. However, due to nonlinearity, multiple operation mode, and process state change, the existing multivariate statistical process monitoring techniques are inefficient, and the process monitoring performance degradation often leads to unreliable results.
In general, characteristics of various types such as mechanical abrasion, sorption, deformation and fatigue are changed after a certain period of time after the first development of the equipment used in the production process and a certain time after the operation of the equipment. And the performance of the machine is deteriorated, which leads to a destabilization of the process and a deterioration of the performance and reliability of the product. Therefore, the reliability of the equipment and the system may be deteriorated due to the degradation of the performance and reliability of the product.
However, there is no way to directly and quickly grasp the degradation of the reliability of such equipment and systems. The engineer who manages and operates the system has only a knowledge of the operation of the equipment and the parameter setting of the equipment, but it is difficult to know how the reliability of the equipment and the system changes. Therefore, in order to grasp the abnormality of such manufacturing process and to grasp the reliability, it is necessary to accumulate a large number of process variables arising from the manufacturing process in a database, and to dynamically operate and construct a quality control system using data mining techniques.
In an environment in which embodiments of the present invention are utilized, an abnormal signal observed during the production process of a product can affect the quality of the product, and particularly by analyzing patterns of signals that can be measured during a process such as temperature or vibration, It is possible to make a judgment as to what kind of tendency to have.
In the embodiments of the present invention, when the data measured by the sensor attached to the process facility is analyzed, if the data history of the good product and the defective product are different, analysis of the pattern of the data may be performed to determine the quality Can be predicted in advance. That is, embodiments of the present invention are based on the idea that preventive quality control is possible from data measured through a sensor.
1 is a diagram illustrating a quality management model using data mining adopted by embodiments of the present invention.
The management of a typical production process has the same model as the
It is not easy to know how the demand system and the result system are related to the manufacturing process, but it is very important to find out the correlation to improve the productivity such as stable quality and yield. In order to find out how the individual factors of the demand system and the individual results of the result system are linked to each other and how the interaction between the demand system and the demand system affects the result system, conventional conventional process management methods include an experiment planning method and regression analysis Statistical methods are mainly used.
The functions of the data mining-based quality management system proposed by the embodiments of the present invention described below are the processes that can produce optimum quality from a large number of process variables (demand factors) and quality variables (result systems) Lt; / RTI > It is possible to generate a process standard which is most suitable for a given situation in almost real time without interruption of production so that the process standard becomes a dynamically changing value which changes according to the situation rather than a fixed fixed value will be.
Referring to FIG. 1, a process of generating a new standard / rule from a process variable and a quality variable by a
The selected data sets the relationship between process variables and quality variables through learning using neural networks or decision trees. The weight of each factor on the result is calculated and the effect of the interaction of each factor on the result can be considered. At the same time, or through other processes, the level of the factor (process factor) that produces the best results is calculated for each factor. In order to verify the newly created standard / rule, it is necessary to check whether the predicted value using the process variable only matches with the result of the process, and then set the new standard. The newly set standard is stored in a rule database, and a new standard is applied to the process, and another new standard is created by the data read out from the process produced by the standard. It can continue.
Figure 2 is a more detailed block diagram of the functional blocks of the quality management model of Figure 1 that the embodiments of the present invention employ.
Figure 2 shows the detailed system and components of a data mining based quality management system. The proposed system can roughly predict the yield (230) by extracting (230) the process parameters which are the process parameters of the main process from the
The data mining related techniques and application schemes employed in the quality management system according to embodiments of the present invention will be described below.
The neural network is proposed in consideration of the physiological model of the neuron, which is the basic unit of the human cerebrum, and includes a node receiving input from the outside and a node performing an output to the outside, The weight of the connection between the node and the node is given and the value of one node is multiplied with the connection weight and transmitted to the other node. In the quality management system, neural networks can be used mainly for forecasting, and it is possible to determine important variables in each process and to understand how these variables affect the yield. Environment, or how the final yield will change if new variables are added.
A decision tree is used to determine which class a new data belongs to in a predefined class by a rule that classifies the class. In order to create a rule, it is necessary to have the data in which the class is defined, and it is necessary to consider the algorithm for generating the rule and the form in which the derived rule is to be displayed.
Pattern analysis using clustering can be utilized. Unlike classification, clustering does not have a predefined class in data, and groups data of similar characteristics according to a predetermined algorithm. Through such clustering, it is possible to analyze the characteristics of the process patterns, to know what statistical characteristics are possessed by the process patterns, and to identify the process variables that significantly affect the formation of a specific pattern. This provides a better understanding of the status of the manufacturing process by showing the control charts familiar to equipment technicians for important process variables.
Control charts and process capability monitoring systems can be utilized. General equipment technicians usually only set and manage process parameters for their equipment. It takes a lot of time and effort to understand the overall contents of the quality control system and utilize them in quality control. Therefore, by providing the existing management form, it is possible to manage each equipment easily, so that it is possible to manage at least the minimum process even if the quality control on the overall manufacturing process is impossible, It can play a role of assisting in immediate response.
Furthermore, embodiments of the present invention utilize an ontology in the course of performing data mining. The ontology can be thought of as a kind of dictionary composed of words and relations, in which words related to a specific domain are represented hierarchically, and in addition, an inference rule for extending the ontology is included, Knowledge sharing and reuse among application programs are possible. This ontology is the core concept of the semantic web application, and the language that defines the schema and the syntax structure to express it is the ontology language, and DSML + OIL, OWL, Ontolingun etc. are utilized now.
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the following description and the accompanying drawings, detailed description of well-known functions or constructions that may obscure the subject matter of the present invention will be omitted. It should be noted that the same constituent elements are denoted by the same reference numerals as possible throughout the drawings.
FIG. 3 is a flowchart illustrating a method of managing quality in a production process according to an embodiment of the present invention, and may be performed by a quality management system having at least one processor.
In
The relationship between these 'factors' and 'quality' can be established through learning using a neural network or a decision tree. In addition, the relation between the 'factor' and the 'quality' is determined by considering the influence of the weight of the process parameters affecting the quality variable and the interactions among the process parameters, Is calculated by reflecting the contribution of the user.
In
In
The step of linearly separating nonlinear values may calculate a hyperplane of nonlinear data by mapping the nonlinear values to a virtual space using a kernel function. That is, the inputted observation coordinates and measurement data are linearly projected to search the margins. At this time, the prediction error can be minimized by calculating the optimal separation boundary where the margin becomes the maximum. Here, the margin refers to the distance from the data closest to the classification boundary to the classification boundary, and the data located closest to the classification boundary is called a support vector.
For example, suppose that a first group of data and a second group of data are given for learning by a general linear discriminant analysis method. You can learn how to divide the first group and the second group by measuring the distance between each pair of data in two groups to obtain the two centers and then finding the optimal hyperplane among them.
In SVM learning, however, the focus is on the data at the boundary between two groups, not the center of each group. First, H1 and H2 lines are drawn at the boundary between the data of the first group and the data of the second group to obtain a pipe, and then a new line is drawn in the middle of the pipe to determine an optimal hyperplane. In this case, the method of obtaining H1 and H2 can be infinitely existed. However, since there is no data between two segments of H1 and H2, and the constraint that the margins, which are the distances between the two segments, is maximized, .
Comparing the above two learning methods, when data having the attribute of the first group is newly appeared under a given environment, the first learning method (general linear discrimination) described above can be mistakenly classified as the second group, The second learning method (SVM) has an advantage in that it can accurately determine the attributes and can be correctly predicted as the first group.
The kernel function may be at least one of a Radial Basis Function (RBF) kernel, a Polynomial kernel, or a Linear kernel.
In
Furthermore, the embodiment of the present invention may further include storing an experimentally-verified rule so as to predict a quality parameter based on the ontology and the process parameter.
FIG. 4 is a flowchart illustrating an additional process that can be selectively added in the quality management method of FIG. 3 according to an exemplary embodiment of the present invention. In the
In
If the change in the process is reflected in the production process in
Hereinafter, an example in which the present invention is applied to a direct production process using an embodiment of the present invention will be described.
For example, a car consists of about 20,000 parts. In order to reduce the production and assembling time and cost of these parts, the carmaker manufactures these parts through hierarchical partners. Especially, when the concept of modularization is generalized as it is today, a single component with a defect leads to a failure of a modular assembly, which causes a failure of the finished vehicle. Therefore, suppliers who produce finished cars and module parts pay great attention to the quality inspection of each part.
Currently, the quality inspection carried out by domestic H and K car makers is managed through the supplier-quality mark certification system. Second and third-tier suppliers can only supply components to the primary supplier or subcontractor. However, most small and medium-sized companies have been supplying parts from non-certified suppliers, that is, unauthorized suppliers, in order to meet the volume that is difficult to handle on their own, even if they are certified. Therefore, it can be said that the complete management of product quality can not be achieved. This can be said to be an obstacle to the delivery industry as well as to suppliers who are burdened with delivery deadlines. The time and cost to cope with defective parts in the early part can be immediate and economical, but if such defects are detected in the assemblies of the module or the finished vehicle in the future, the economic loss to them will increase exponentially do. Therefore, an effective real-time / open-ended process monitoring system for quality control in the manufacturing supply chain is urgent.
FIG. 5 is a view for explaining the process parameters that appear in the process of manufacturing a printed circuit board (PCB) of a door trim of a vehicle, which is an example of a production process. The following will be outlined.
Registration is the top priority when discussing imaging in PCB production. Registration refers to the degree to which a pattern is placed at a specified location, and in the registration process it is required to combine all images, features and processes appropriately. Each should form an appropriate relationship in terms of space from different images, features, and processes on the PCB. This combination is applied from the design stage through the design of the data set synchronized to specific Cartesian (X, Y, Z) coordinates.
Among the registration methods, the mass lamination (mass ram) technique uses a heat fusion mass ram method among the mass ram systems in a way that does not require high accuracy of the inter register registration system. The principle of such a heat-welding mass ram is highly dependent on the prepreg layer bonded to the inner layer due to the resin polymerization effect. To polymerize the prepreg resin, a hot melt bonding system draws heat from the outside to the center of the stack using a hot electric heater similar to a soldering iron. In this case, the heater is raised to about 300 ° C (572 ° F) and typically 3 to 8 engagement points (points) are used for each long edge of the panel.
In this process, when the central portion of the thickness is heated to achieve the proper polymerization temperature, the temperature of the outer layer contacting the bonding head may be heated to cause poor bonding. If the central portion of the thickness is too cold, It may cause polymerization failure. Referring to FIG. 5, (a) a bass fusion process through hot air injection and (b) a cooling process after forming a bass part by descending a piston tip are introduced.
In the situation shown in FIG. 5, in order to predict the quality, three temperature data for each product may be expressed on the three-dimensional coordinates for the data history of the good product / defective product, so that the pattern of the data can be analyzed using the SVM learning.
More specifically, first, a process parameter and a quality parameter for the process parameter are input in a PCB production process to generate an ontology indicating a relationship between 'factor' and 'quality', and the observation to be observed based on the generated ontology And receives observation coordinates and measurement data of the process parameters for each observation point in real time using a sensor. These process parameters may include at least one of temperature, vibration or noise, and it is assumed here that the temperature is measured using a temperature sensor. At this time, the collection of the input data may be performed by collecting data from all the fusion points of the hot wind fusion apparatus, collecting data once from the fusion point for each product, or collecting data from the pressure point. In particular, since the reference temperature is different for each fusing point, the data collected through the sensor will be the reference temperature and the measured temperature.
Then, the kernel function is used to linearly separate nonlinear values of the input observation coordinates and measurement data (which can be measured temperature), SVM learning is performed based on the separated linear values, and SVM From the learning results, it is possible to predict the process state that occurs due to the influence of current process parameters (which can be temperature). As described above, it is desirable to solve the problem of nonlinear data separation by using a kernel function since most data in various application fields can not be linearly separated.
FIG. 6 is a diagram for explaining a kernel function utilized by embodiments of the present invention to pre-separate non-linear values. FIG.
Choosing a kernel function is one of the most important problems in applying SVM, but there is no standardized method, so it is necessary to select an appropriate kernel function depending on the situation. The hyperplane of the nonlinear data can be calculated by mapping the data on the virtual space using the kernel function. When the kernel function having the vector (x i , x j ) as a parameter is K (x i , x j ) , And can be expressed by the following equation (1).
Then, we can graph the results of using the same data as that of RBF (Radial Basis Function), Polynomial, and Linear kernel used in SVM.
FIG. 7 is a graph illustrating the results of performing operations using various kernel functions, and shows results according to (a) a Radial Basis Function (RBF) kernel, (b) a Polynomial kernel, and (c) a Linear kernel . In addition, the expression of the kernel corresponding to each kernel function is expressed by the following equations (2) to (4).
Now, an embodiment of the present invention performs SVM learning and reflects temperature data among process parameters through SVM modeling to quality monitoring. For this purpose, this quality management system performs SVM using temperature data that can be measured in process parameters defined in ontology modeling. At this time, the measured temperature data may be collected using the PLC from the temperature sensor installed in the production equipment.
In addition, the collected data may be transmitted to the quality management system (for example, from the production equipment) in the form of an array having the equipment identification number, the date and time of the operation, the process parameter, the equipment point number, the measured value, the set value, It may be a physically separated server). Then, the quality management system converts the received data into a data structure for SVM modeling, and performs SVM training. In this case, the kernel variable is compared with the training data using the K-packed cross validation method. By selecting the variable with the smallest error value, it can be verified that the model is the best approximate model and the kernel variable can be determined. Furthermore, the quality management system can test real-time data with SVM to indicate graphs and error probabilities.
FIG. 8 is a diagram illustrating a pseudo code implementing a support vector machine (SVM) learning and verification method adopted by embodiments of the present invention. The input data is first read through SVM Training, After calculating the difference between the measured temperature value and the set temperature value, the process of setting the initial kernel option and selecting the optimal kernel option are described. Meanwhile, FIG. 8 also shows a process of testing SVM set through SVM Testing.
Data collection through a temperature sensor is collected by detecting data from a temperature data measurement sensor on the production process. At this time, the data can be collected in real time without interfering with the operation of the existing facility by installing the data communicating with the temperature sensor installed in the facility to enable PLC communication. The collected data is transmitted to the quality control system, where the data transmitted from the PLC are [identification number of equipment, date and time of operation, process parameters (for example, temperature), point / , Set value, temperature lower limit value, upper temperature limit value].
FIG. 9 is a diagram illustrating data converted according to a method of managing quality in a PCB manufacturing process according to another embodiment of the present invention, and includes data collected by a series of items.
For the SVM classification model, it is necessary to convert the data to the format and set the necessary parameters. The data format conversion process is as follows.
(STEP 1) [Equipment identification number, date and time of operation, process factor, point / sensor number, measured value, set value] of equipment are read out from the equipment data.
(STEP 2) Determine the number of measured values and store them in variable N (N is a natural number).
(STEP 3) Calculate the deviation [measured value - set value] of all measurement data.
(STEP 4) Convert to data with the following arrangement.
A = number of equipment with an array of [2 × number of products]
B = set value with an array of [number of sensors per product x number of products]
X = temperature deviation value with an array of [number of sensors per product x number of products]
Y = [1 × number of products] Pass / fail judgment value
After performing the above data conversion process, the quality management system determines the applied kernel for the SVM classification model and designates the kernel initial variable.
Now, the quality management system determines the optimal SVM variable. To do this, reconstruct an approximate model with K-fold cross validation (a method of evaluating the accuracy of an approximate model) by sequentially removing the variables of the size specified to find the kernel variable with the smallest Errors . This selected variable can be determined by verifying the smallest variable as an approximate model that best represents the real model.
10 is a diagram for explaining a process of determining an optimal SVM variable according to a method of managing quality in a PCB production process according to another embodiment of the present invention, wherein the x-axis represents a
The SVM classification model represents the data learning result using the SVM algorithm in three dimensions. 11 is a diagram illustrating an SVM classification model in which the learned data is expressed in three dimensions according to a method of managing quality in a production process according to embodiments of the present invention. The results are shown. Through the SVM learning, it is possible to calculate the probability of error occurrence by analyzing the modeling result using the data for learning model and SVM test, and to express the quality problem of the product caused by the influence of temperature.
FIG. 12 is a block diagram illustrating an apparatus for managing quality in a production process according to an embodiment of the present invention, and includes a configuration corresponding to each process of FIG. 3 described above. Therefore, in order to avoid duplication of explanation, the function is outlined mainly in the hardware device. 12, the
The
The
The
In addition, the
In addition, the
Furthermore, the
According to the embodiments of the present invention described above, various process variables extracted from a manufacturing process are accumulated, data mining techniques are utilized to derive a relationship between factors and results from the accumulated data, and a new By creating a combination of process setpoints, it is possible to predict a process state that changes in response to a situation occurring in real time, deviating from a process specification of a fixed value, and even if the performance of the machine deteriorates over time, , It is possible to provide the user with a dynamic setting value suitable for a given environmental factor.
Further, embodiments of the present invention are able to apply global quality management standards in the supply chain through quality management techniques based on data mining techniques. In particular, it is possible to monitor and manage integrated real-time standards in the supply chain through knowledge modeling of qualitative / quantitative standard indicators, which has the advantage of facilitating ISO / GS standardization work in the future.
Meanwhile, the embodiments of the present invention can be embodied as computer readable codes on a computer readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored.
Examples of the computer-readable recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device and the like, and also a carrier wave (for example, transmission via the Internet) . In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner. In addition, functional programs, codes, and code segments for implementing the present invention can be easily deduced by programmers skilled in the art to which the present invention belongs.
The present invention has been described above with reference to various embodiments. It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is defined by the appended claims rather than by the foregoing description, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.
110: Process control system
130: Quality management system based on data mining
210: Process database
230: Main Process Parameter Extraction System
250: rule base 270: monitoring system
10: Production process / production machine 15: Sensor
20: quality control device 21: input unit
23: processing section 25:
Claims (20)
The input unit inputs the process parameters corresponding to the factors affecting the quality in the production process and the quality parameters for the good and defective products corresponding to the results of the process parameters and determines the relationship between the 'factor' and the 'quality' Generating an ontology representing the object;
Wherein the processing unit determines an observation point to be observed based on the generated ontology and receives observation coordinates and measurement data of the process parameter by the observation point in real time through a production process through the input unit;
Wherein the processing unit linearly separates the nonlinear values by mapping hyperplanes of the nonlinear data by mapping the nonlinear values of the input observation coordinates and measurement data on a virtual space, Performing Support Vector Machine (SVM) learning based on the SVM; And
And the processing unit predicting a current process state from the SVM learning result.
Storing the experimentally verified rule in the storage so that the quality variable can be predicted based on the ontology and the process parameter.
The relation between 'factor' and 'quality'
Wherein the learning is set through learning using a neural network or a decision tree.
The relation between 'factor' and 'quality'
Wherein the weighting factor is calculated by taking into consideration the weight of the process parameters influencing the quality variable and the effect of the interaction between the process parameters, and reflecting the contribution of the process parameters causing the quality variable representing good products.
Linearly separating the non-linear values comprises:
Linear values are linearly projected on a virtual space by using a kernel function, and the separation boundary surface which maximizes the margin indicating the distance from the data closest to the classification boundary to the classification boundary, ≪ / RTI >
The kernel function may include:
A Radial Basis Function (RBF) kernel, a Polynomial kernel, or a Linear kernel.
Wherein the step of predicting the current process state comprises:
Calculating a stability or a predicted yield of the process that is calculated in consideration of the management metric based on the ontology, and providing the calculated stability or predicted yield to the user.
Wherein the predicting the current process state is performed based on current measurement data from the ontology and historical data using pattern recognition based on SVM,
And inducing a change in the process by providing the predicted process state to a user.
When the change of the process is reflected in the production process,
Receiving the new process parameters and quality parameters and updating the ontology.
Inputting a process parameter and a quality parameter for the process parameter in a PCB production process, and generating an ontology indicating a relation between a 'factor' and a 'quality';
Wherein the processing unit determines an observation point to be observed based on the generated ontology and receives observation coordinates and measurement data of the process parameter for each observation point in real time using the input unit sensor;
Linearly separating the nonlinear values by mapping a nonlinear value of the input observation coordinates and measurement data to a virtual space using a kernel function by calculating a hyperplane of the nonlinear data, Performing SVM learning using pattern matching based on the separated linear values; And
And predicting a process state in which the processing unit originates from an influence of a current process parameter from the SVM learning result.
The process parameters include,
Temperature, vibration, or noise.
The input values are data converted as an array including the equipment identification number, the operation date and time, the process parameter (temperature), the equipment point number (temperature sensor number), the measured value, the set value, the upper temperature limit and the lower temperature limit, ≪ / RTI >
The data conversion may include:
Equipment identification number, job date and time, process parameters, equipment point number, measured value and set value,
The number of measured values is counted and stored in a variable,
The deviation of the measured value and the set value is calculated for all measurement data,
An apparatus identification number having an arrangement of twice the number of products, an operation date and time, a setting value having an array of the number of sensors multiplied by the number of products per product, and an array of the number of products per product multiplied by the product number And a pass / fail determination value of a product having an arrangement as many as the number of products.
Wherein the kernel function is selected from at least one of an RBF kernel, a Polynomial kernel, and a Linear kernel to designate a kernel initial variable,
Wherein the nonlinear values are projected linearly using the kernel function and a separation boundary surface having a maximum margin from the data located closest to the classification boundary to the classification boundary is calculated .
The kernel initial variable,
K-fold cross validation is used to compare kernel configuration variables as learning data and to select the smallest variance variable to verify the model as an approximate model that best represents the actual model. Lt; / RTI >
An input unit for inputting a process parameter corresponding to a factor affecting quality in a production process using a sensor and a quality parameter for a good and defective product corresponding to the result of the process parameter;
A processing unit for generating an ontology representing a relationship between 'factor' and 'quality' from the input process parameter and quality variable, and determining an observation point to be observed based on the generated ontology; And
And a storage unit for storing experimentally-validated rules for predicting quality parameters based on the generated ontology and process parameters,
Wherein the input unit receives observation coordinates and measurement data of the process parameters for each observation point in real time through a production process,
Wherein the processing unit linearly separates the nonlinear values by mapping hyperplanes of nonlinear data by mapping the nonlinear values of the observation coordinates and measurement data input through the input unit on a virtual space, Performing SVM learning based on the linear values, and predicting a current process state from the SVM learning result.
Wherein,
Linear values are linearly projected using a kernel function selected from at least one of an RBF kernel, a polynomial kernel, and a linear kernel, and a margin indicating a distance from a data closest to the classification boundary to a classification boundary is linearly projected And the separation interface is calculated as the separation interface.
Wherein,
Predicting a current process state based on current measurement data from the ontology and historical data using pattern recognition based on SVM,
And provides the predicted process state to a user to induce a change in the process.
When the change of the process is reflected in the production process,
Wherein the processing unit receives new process parameters and quality parameters through the input unit and updates the ontology.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR1020120104587A KR101530848B1 (en) | 2012-09-20 | 2012-09-20 | Apparatus and method for quality control using datamining in manufacturing process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR1020120104587A KR101530848B1 (en) | 2012-09-20 | 2012-09-20 | Apparatus and method for quality control using datamining in manufacturing process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| KR20140039380A KR20140039380A (en) | 2014-04-02 |
| KR101530848B1 true KR101530848B1 (en) | 2015-06-24 |
Family
ID=50650005
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| KR1020120104587A KR101530848B1 (en) | 2012-09-20 | 2012-09-20 | Apparatus and method for quality control using datamining in manufacturing process |
Country Status (1)
| Country | Link |
|---|---|
| KR (1) | KR101530848B1 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170024262A (en) | 2015-08-25 | 2017-03-07 | 위즈코어 주식회사 | statistical process management system |
| KR101888637B1 (en) | 2017-03-20 | 2018-08-14 | 한국생산기술연구원 | Analysis methodology and platform architecture system for big data based on manufacturing specialized algorithm template |
| KR20190060547A (en) | 2017-11-24 | 2019-06-03 | 한국생산기술연구원 | Method of Deriving and Visualizing the Causes of Process Malfunctions Through Machine Learning Model In Data Imbalance Environment |
| KR20190060548A (en) | 2017-11-24 | 2019-06-03 | 한국생산기술연구원 | Method of analyzing and visualizing the cause of process failure by deriving the defect occurrence index by variable sections |
| KR102100499B1 (en) * | 2019-10-11 | 2020-04-13 | 강현광 | Quality management system using big data |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101466798B1 (en) * | 2014-05-20 | 2014-12-01 | 삼성전자주식회사 | Method and apparatus for discovering the equipment causing product faults in manufacturing process |
| KR101661818B1 (en) * | 2014-12-23 | 2016-09-30 | (주)해인씨앤에스 | Method for managing a process and quality improvement of manufacturing process |
| KR101632073B1 (en) * | 2015-06-04 | 2016-06-20 | 장원중 | Method, device, system and non-transitory computer-readable recording medium for providing data profiling based on statistical analysis |
| CN105427016A (en) * | 2015-10-28 | 2016-03-23 | 南车株洲电力机车研究所有限公司 | Locomotive vehicle data processing method and system |
| CN107578104B (en) * | 2017-08-31 | 2018-11-06 | 江苏康缘药业股份有限公司 | A kind of Chinese Traditional Medicine knowledge system |
| US11067964B2 (en) | 2018-01-17 | 2021-07-20 | Kymeta Corporation | Method to improve performance, manufacturing, and design of a satellite antenna |
| KR102265235B1 (en) * | 2019-01-31 | 2021-06-14 | 상명대학교 산학협력단 | Machine Learning System using Big Data |
| KR102348006B1 (en) * | 2020-09-28 | 2022-01-07 | 주식회사 휴먼시스템 | Factory manufacturing facility system monitoring apparatus for monitoring the defective rate for the manufacturing process |
| KR102548849B1 (en) * | 2022-04-07 | 2023-06-28 | 김재성 | Decision Optimization System |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20050062576A (en) * | 2002-09-30 | 2005-06-23 | 어드밴스드 마이크로 디바이시즈, 인코포레이티드 | Comprehensive integrated lithographic process control system based on product design and yield feedback system |
| KR20080092881A (en) * | 2007-04-12 | 2008-10-16 | 도쿄엘렉트론가부시키가이샤 | Controlling a fabrication tool using support vector machine |
| KR20090061856A (en) * | 2007-12-12 | 2009-06-17 | (주) 이아이비코리아 | A system and method for statistical process control enabling process quality rules to be changed |
-
2012
- 2012-09-20 KR KR1020120104587A patent/KR101530848B1/en active IP Right Grant
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20050062576A (en) * | 2002-09-30 | 2005-06-23 | 어드밴스드 마이크로 디바이시즈, 인코포레이티드 | Comprehensive integrated lithographic process control system based on product design and yield feedback system |
| KR20080092881A (en) * | 2007-04-12 | 2008-10-16 | 도쿄엘렉트론가부시키가이샤 | Controlling a fabrication tool using support vector machine |
| KR20090061856A (en) * | 2007-12-12 | 2009-06-17 | (주) 이아이비코리아 | A system and method for statistical process control enabling process quality rules to be changed |
Non-Patent Citations (1)
| Title |
|---|
| 박용민. 원격 자동 생산시스템의 실시간 EQM 구현을 위한 DATA MINING 방법론에 대한 연구 : 아주대학교(석사학위 논문), 2011.02. * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20170024262A (en) | 2015-08-25 | 2017-03-07 | 위즈코어 주식회사 | statistical process management system |
| KR101888637B1 (en) | 2017-03-20 | 2018-08-14 | 한국생산기술연구원 | Analysis methodology and platform architecture system for big data based on manufacturing specialized algorithm template |
| KR20190060547A (en) | 2017-11-24 | 2019-06-03 | 한국생산기술연구원 | Method of Deriving and Visualizing the Causes of Process Malfunctions Through Machine Learning Model In Data Imbalance Environment |
| KR20190060548A (en) | 2017-11-24 | 2019-06-03 | 한국생산기술연구원 | Method of analyzing and visualizing the cause of process failure by deriving the defect occurrence index by variable sections |
| KR102100499B1 (en) * | 2019-10-11 | 2020-04-13 | 강현광 | Quality management system using big data |
Also Published As
| Publication number | Publication date |
|---|---|
| KR20140039380A (en) | 2014-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR101530848B1 (en) | Apparatus and method for quality control using datamining in manufacturing process | |
| KR101508641B1 (en) | Apparatus and method for predicting product condition by using datamining in manufacturing process | |
| Schmidt et al. | Cloud-enhanced predictive maintenance | |
| Vrignat et al. | Sustainable manufacturing, maintenance policies, prognostics and health management: A literature review | |
| Qamsane et al. | A methodology to develop and implement digital twin solutions for manufacturing systems | |
| Bengtsson et al. | Design of condition based maintenance system—A case study using sound analysis and case-based reasoning | |
| WO2013024613A1 (en) | Malfunction detection/diagnostic method, and malfunction detection/diagnostic system | |
| Vogl et al. | Standards for prognostics and health management (PHM) techniques within manufacturing operations | |
| Brundage et al. | Smart manufacturing through a framework for a knowledge-based diagnosis system | |
| Yacout | Fault detection and diagnosis for condition based maintenance using the logical analysis of data | |
| Adhikari et al. | Machine learning based data driven diagnostics & prognostics framework for aircraft predictive maintenance | |
| Groba et al. | Architecture of a predictive maintenance framework | |
| Navinchandran et al. | Discovering critical KPI factors from natural language in maintenance work orders | |
| CN109522193A (en) | A kind of processing method of operation/maintenance data, system and device | |
| CN116882978B (en) | Deep sea submersible operation and maintenance support system based on product information frame | |
| Yen et al. | Application of a neural network integrated with the internet of things sensing technology for 3D printer fault diagnosis | |
| Hagedorn et al. | Understanding unforeseen production downtimes in manufacturing processes using log data-driven causal reasoning | |
| Kang et al. | Predicting just‐in‐time software defects to reduce post‐release quality costs in the maritime industry | |
| Shukla et al. | Digital twin-based prediction for CNC machines inspection using blockchain for industry 4.0 | |
| US11544580B2 (en) | Methods and systems for improving asset operation based on identification of significant changes in sensor combinations in related events | |
| CN116894211A (en) | System for generating human perceptible interpretive output, method and computer program for monitoring anomaly identification | |
| KR102437917B1 (en) | Equipment operation system | |
| Xia | A systematic graph-based methodology for cognitive predictive maintenance of complex engineering equipment | |
| Giannoulidis et al. | A context-aware unsupervised predictive maintenance solution for fleet management | |
| Keivanpour | A Conceptual Framework for Estimating the Remaining Operational Lifetime of the Recovered Components from End of Life Aircraft Using Fuzzy Simulation and Digital Twin |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A201 | Request for examination | ||
| E902 | Notification of reason for refusal | ||
| E701 | Decision to grant or registration of patent right | ||
| GRNT | Written decision to grant | ||
| FPAY | Annual fee payment |
Payment date: 20180409 Year of fee payment: 4 |
|
| FPAY | Annual fee payment |
Payment date: 20190401 Year of fee payment: 5 |