Post by account_disabled on Sept 14, 2023 10:26:28 GMT
Director Choi Byeong-wook shared a hypothetical example of using SAS' solution in the semiconductor blank mask manufacturing process to identify optimal manufacturing conditions and increase productivity. First of all, the task that must precede ML model development is problem definition. In this case, due to the nature of the manufacturing process, quality inspection values tended to gradually increase as production volume increased.
After defining the problem, we develop an Phone Number List ML model that can predict quality inspection values based on control parameters.
In this case, the differences between past and recent actual UV spectrum values and equipment, all data from the manufacturing production lot, and environmental information such as temperature and pressure were set as basic input parameters, and gas and power information were set as control parameters.
For example, in this case, if the allowable standard value is set to 0.3nm and the value obtained by entering the control parameters is 0.25nm, the gas and power values must be changed to optimize by 0.05nm. In this case, you can use the Optimization Parameter Search Algorithm to get an answer on how to configure the control parameters.
Specific predictive optimization work is centered around two types of users: analysts and operators. First, an analyst designs a quality prediction model, including data preparation and processing steps. During this process, 5 to 20 models are tested to identify the ‘champion model’ with the best prediction ability and use this to develop optimal parameter search logic. and automates the optimized model.
The operator specifies target quality test values and derives optimal parameter values that satisfy these values. In this case, the operator inputs the optimal gas and power information values considered by the operator and derives predicted quality test values to minimize the gap with the goal.
Director Choi Byeong-wook said, “If you use a predefined prediction model in the form of a template or are unsure how to use the algorithm of a model, you can create an automated pipeline by utilizing the automation function of the SAS Byya platform. In the latter case, when data is entered, an automated pipeline is created and the software derives the target prediction model. “Overall, the overall life cycle includes model creation and use, and direct operation by users.”
The ML model development and operations lifecycle included in the SAS Byya platform can also be applied to computer vision. For example, a manufacturing company that produces gypsum board may experience delays in product shipment time if multiple boards overlap or movement delays occur as the boards move along rails. In these cases, process problems can be quickly responded to by collecting monitoring image information captured by factory cameras and detecting problems in real time.
Generally, the image analysis step proceeds as follows. First, extract frames from the video, resize them, and then select the necessary parts using the object detection model YOLO. After identifying the location of a specific object in the image using a bounding box, the image size is adjusted accordingly, and feature points are detected in the image using KeyPoint Detection. The distance between products is tracked in real time through Kalman Filtering.
ⓒSAS
SAS seamlessly integrates a series of processes through solutions such as Event Stream Processing (ESP) and Visual Data Mining and Machine Learning, and even provides real-time notifications in connection with existing systems. Director Choi Byeong-wook said, “We predict the area, distance, and movement of a specific object from real-time frame data collected from the existing controller, and provide a message to Kepware or PLC owned by the customer or a third party by saying, ‘There is a problem, so the operator should check.’ “It will be notified,” he said.
A system ‘for everyone’, from analysis experts to citizen developers
SAS' analysis life cycle from data, exploration, and deployment follows DataOps and ModelOps. Director Choi Byeong-wook said, “It largely consists of a part that collects and manages data, a part that creates models, and a part that distributes models and manages them in PLCs or third parties. In between, as seen in the case, data is visualized or analyzed to make decisions. There is an application process. “The lifecycle includes not only ML and computer vision, but also various text analytics such as natural language processing.”
In particular, Director Choi Byeong-wook emphasized that SAS’s focus is “analysis for everyone.” Creating ML or deep learning models is a specialized area only for data engineers and data scientists, but SAS supports analysis experts and citizen data scientists to create, apply, and manage them as they wish through easy automation functions.
In the era of digital transformation, it is important for manufacturing companies to build solutions that optimize processes and performance. To this end, SAS supports the SAS analysis life cycle in real time through various open sources and solutions. Director Choi Byeong-wook said, “AI-based enterprise analysis platform SAS Viya manages data, visualizes managed data, generates expressed data as a model, and extracts and utilizes the actual created model as desired. Supports ‘one shot’ support. “This is the life cycle that SAS provides for manufacturing today and in the future,” he emphasized.
After defining the problem, we develop an Phone Number List ML model that can predict quality inspection values based on control parameters.
In this case, the differences between past and recent actual UV spectrum values and equipment, all data from the manufacturing production lot, and environmental information such as temperature and pressure were set as basic input parameters, and gas and power information were set as control parameters.
For example, in this case, if the allowable standard value is set to 0.3nm and the value obtained by entering the control parameters is 0.25nm, the gas and power values must be changed to optimize by 0.05nm. In this case, you can use the Optimization Parameter Search Algorithm to get an answer on how to configure the control parameters.
Specific predictive optimization work is centered around two types of users: analysts and operators. First, an analyst designs a quality prediction model, including data preparation and processing steps. During this process, 5 to 20 models are tested to identify the ‘champion model’ with the best prediction ability and use this to develop optimal parameter search logic. and automates the optimized model.
The operator specifies target quality test values and derives optimal parameter values that satisfy these values. In this case, the operator inputs the optimal gas and power information values considered by the operator and derives predicted quality test values to minimize the gap with the goal.
Director Choi Byeong-wook said, “If you use a predefined prediction model in the form of a template or are unsure how to use the algorithm of a model, you can create an automated pipeline by utilizing the automation function of the SAS Byya platform. In the latter case, when data is entered, an automated pipeline is created and the software derives the target prediction model. “Overall, the overall life cycle includes model creation and use, and direct operation by users.”
The ML model development and operations lifecycle included in the SAS Byya platform can also be applied to computer vision. For example, a manufacturing company that produces gypsum board may experience delays in product shipment time if multiple boards overlap or movement delays occur as the boards move along rails. In these cases, process problems can be quickly responded to by collecting monitoring image information captured by factory cameras and detecting problems in real time.
Generally, the image analysis step proceeds as follows. First, extract frames from the video, resize them, and then select the necessary parts using the object detection model YOLO. After identifying the location of a specific object in the image using a bounding box, the image size is adjusted accordingly, and feature points are detected in the image using KeyPoint Detection. The distance between products is tracked in real time through Kalman Filtering.
ⓒSAS
SAS seamlessly integrates a series of processes through solutions such as Event Stream Processing (ESP) and Visual Data Mining and Machine Learning, and even provides real-time notifications in connection with existing systems. Director Choi Byeong-wook said, “We predict the area, distance, and movement of a specific object from real-time frame data collected from the existing controller, and provide a message to Kepware or PLC owned by the customer or a third party by saying, ‘There is a problem, so the operator should check.’ “It will be notified,” he said.
A system ‘for everyone’, from analysis experts to citizen developers
SAS' analysis life cycle from data, exploration, and deployment follows DataOps and ModelOps. Director Choi Byeong-wook said, “It largely consists of a part that collects and manages data, a part that creates models, and a part that distributes models and manages them in PLCs or third parties. In between, as seen in the case, data is visualized or analyzed to make decisions. There is an application process. “The lifecycle includes not only ML and computer vision, but also various text analytics such as natural language processing.”
In particular, Director Choi Byeong-wook emphasized that SAS’s focus is “analysis for everyone.” Creating ML or deep learning models is a specialized area only for data engineers and data scientists, but SAS supports analysis experts and citizen data scientists to create, apply, and manage them as they wish through easy automation functions.
In the era of digital transformation, it is important for manufacturing companies to build solutions that optimize processes and performance. To this end, SAS supports the SAS analysis life cycle in real time through various open sources and solutions. Director Choi Byeong-wook said, “AI-based enterprise analysis platform SAS Viya manages data, visualizes managed data, generates expressed data as a model, and extracts and utilizes the actual created model as desired. Supports ‘one shot’ support. “This is the life cycle that SAS provides for manufacturing today and in the future,” he emphasized.