Page 118 - Kỷ yếu hội thảo quốc tế: Ứng dụng công nghệ mới trong công trình xanh - lần thứ 9 (ATiGB 2024)

th
HỘI THẢO QUỐC TẾ ATiGB LẦN THỨ CHÍN - The 9 ATiGB 2024 109

is comprehensively analyzed to measure the accuracy embedded device mentioned above. Of course, even if
of causality analysis between raw materials, equipment, this is not the case, if you devise and apply a more
and process defects. Mean Absolute Percentage Error efficient and productive method, it can be helpful in
(MAPE) represents error expressed as a percentage and operating a smart factory by applying it to the
is the average of the percentage error between the process. Although this paper only discussed
predicted value and the actual value. appearance inspection and quality monitoring, it
 actual value  appears that more efficient and effective production is
possible if AI is applied to facility operation and
MAPE = (l / n)*   - predicted value   *100% manufacturing processes.
 
 / actual value  ACKNOWLEDGMENT
Fig 9. MAPE calculation method This work was supported by the 2023 Korea
When evaluating a model, use cross-validation to Industrial Complex Corporation Industrial Cluster
evaluate the model's generalization performance and Competitiveness Reinforcement Project R&D Support
visually analyze the prediction results. Scatterplots of Project (Project No. VCKB2301).
actual and predicted values, distribution of prediction REFERENCES
errors, distribution of predicted values, etc. are
displayed graphically to visually check the [1]. Jun-Tae Park, Ar-Chim Ryu, Kyu-Phil Han, “A Deep-Learning
Object Recognition Algorithm Using Real-time Object
performance of the model. Data acquired from the Detection and Data-set Structuring”. Korean Society of
quality management integrated control solution is Electronic Engineers conference, 2018 .
stored and becomes big data. If a real-time monitoring [2]. Tae-Seob Shim, Sang-In Lee, Sang-Goo Yoon, & Jae-Chul
system is further developed based on the created big Kim, “A Study on Image Preprocessing for Object Detection”
data, a vast amount of data can be used for real-time KIIT Conference, Korea, 2022
analysis and efficient use in process management [4]. [4]. Seong-won Hong, Hye-rin Park, Beom-sik Shin, Seon-hwa Oh,
In addition, equipment data is collected and analyzed Seok-hyeon Cho. “Construction of a quality prediction and
in real time to identify abnormal signs in advance and management system using real-time process monitoring”.
Korean Society of Industrial Engineers, 2014
provide alarms to take action. In order to maintain [5]. Kim Mi-jin, Yoo Yun-sik, “Big data-based real-time
optimal manufacturing conditions, changes in time monitoring system development and application case”, Korean
series data are monitored to maintain the tolerance Information Science Society Academic Conference Papers,
range of set and status values. In case of deviation, 2015
automatic or semi-automatic equipment control is
possible. Optimization of manufacturing conditions is
based on Condition based maintenance (CBM), which
diagnoses and predicts the condition of equipment to
determine whether to repair or reset it. It utilizes IIoT
(Industrial Internet of Things), one of the key
technologies of smart factories, to provide various by
collecting data through sensors, etc., you can monitor
trends, determine whether there are any abnormalities,
and decide whether to preserve them.
5. CONCLUSION
In this paper, we discussed a method of detecting
defects in appearance using deep learning based on AI
and a method of quality monitoring through GUI
development for efficient operation of the factory. In
the case of external defects, the shape of the defect is
not always consistent. Although they may have a
similar shape, unexpected defects may occur and the
shape may also appear in an unspecified shape. In
order not to miss this, when a new type is discovered,
it must be learned through deep learning and tested
repeatedly to ensure the reliability of the equipment.
The image preprocessing function must also improve
image quality to increase the detection rate, so the
recognition rate must be increased by using the

ISBN: 978-604-80-9779-0