Jun-Geol Baek’s research while affiliated with Korea University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (105)


Construct training exemplar set
Switch ON/OFF learning of 1D CNN and 1D GAN
Example of an LAM dataset: true alarm (red) and false alarm (black)
Optimal 1D GAN structure for LAM dataset
Optimal 1D GAN structure for SECOM dataset

+4

Switch ON/OFF learning of one-dimensional convolutional neural network and one-dimensional generative adversarial network for fault detection
  • Article
  • Publisher preview available

January 2025

·

5 Reads

Journal of Intelligent Manufacturing

Seunghwan Song

·

·

Cheolsoon Park

·

Jun-Geol Baek

Fault detection and classification (FDC) has been adopted to minimize equipment errors in semiconductor manufacturing. The FDC monitors production facilities in real-time to detect abnormalities and ensure compliance with specifications. If the FDC detects that it is out of specification, an interlock is triggered. This interlock forces a safe action. Recent advancements in FDC require a more thorough understanding of interlocked data. This analysis process is called the problem of true/false alarm detection. However, true/false alarm detection is challenging not only because it requires more sophisticated work but also a smaller percentage of true alarms. In this study, we propose a novel approach for fault detection in semiconductor manufacturing using Switch ON/OFF learning that combines one-dimensional convolutional neural network (1D CNN) and one-dimensional generative adversarial network (1D GAN). The noise-resistant 1D CNN is responsible for feature extraction and distinguishing between true/false alarms, whereas the 1D GAN generates additional true alarm samples to address the imbalance in the dataset. In Switch OFF, both the models are trained in parallel, and in Switch ON, the true alarms generated by the trained 1D GAN are used to update the 1D CNN. The results of experiments conducted on two real-world semiconductor datasets demonstrated the superior performance of the proposed model over other sampling techniques. Our study provides a foundation for advancing neural networks in monitoring process facilities, specifically for detecting equipment errors in semiconductor manufacturing.

View access options

Deep Clustering and Regression Ensemble Network for Lot Cycle Time Prediction in Semiconductor Wafer Fabrication

January 2025

·

2 Reads

IEEE Transactions on Semiconductor Manufacturing

Accurate cycle time (CT) prediction is crucial in semiconductor manufacturing. Hybrid models integrating classification and prediction models can enhance CT prediction accuracy. However, existing methods have limitations, including challenges in capturing the dynamic conditions of the production line and optimizing job classification to ensure high CT prediction performance. In this study, we propose a novel hybrid framework for predicting process step-level CT in semiconductor wafer fabrication, thereby addressing the limitations of previous methods. Moreover, the paper formalizes and introduces dynamically changing manufacturing environment attributes as variables that contribute to CT. The proposed method combines deep embedded clustering (DEC) with a regression ensemble network. First, the DEC extracts cluster-friendly representative features from high-dimensional CT datasets and classifies jobs accordingly. Then, a weighted ensemble approach merges regression networks based on cluster membership probabilities. Unlike existing methods that separately handle feature extraction, job classification, and CT prediction, the proposed unified network synchronizes these processes. Experimental results using real-world operational data from a semiconductor manufacturing system indicate that the proposed prediction method considerably outperforms previous approaches in terms of prediction accuracy. To the best of our knowledge, this is the first study to integrate deep clustering with a regression ensemble network for predicting cycle time at the process step level in semiconductor manufacturing. By synchronizing feature extraction, clustering, and prediction tasks, the proposed framework achieves enhanced accuracy and robustness in dynamically changing manufacturing environments.


Unsupervised fault detection using frequency-wise angular filtering in contaminated vibration signals

November 2024

·

4 Reads


GlocalCLIP: Object-agnostic Global-Local Prompt Learning for Zero-shot Anomaly Detection

November 2024

·

9 Reads

Zero-shot anomaly detection (ZSAD) is crucial for detecting abnormal patterns in target datasets without using training samples, specifically in scenarios where there are distributional differences between the target domain and training data or where data scarcity arises because of restricted access. Although recently pretrained vision-language models demonstrate strong zero-shot performance across various visual tasks, they focus on learning class semantics, which makes their direct application to ZSAD challenging. To address this scenario, we propose GlocalCLIP, which uniquely separates global and local prompts and jointly optimizes them. This approach enables the object-agnostic glocal semantic prompt design to effectively capture general normal and anomalous patterns without dependency on specific objects in the image. We refine the text prompts for more precise adjustments by utilizing deep-text prompt tuning in the text encoder. In the vision encoder, we apply V-V attention layers to capture detailed local image features. Finally, we introduce glocal contrastive learning to improve the complementary learning of global and local prompts, effectively detecting abnormal patterns across various domains. The generalization performance of GlocalCLIP in ZSAD was demonstrated on 15 real-world datasets from both the industrial and medical domains, achieving superior performance compared to existing methods.



Enhancing Long-Term Cloud Workload Forecasting Framework: Anomaly Handling and Ensemble Learning in Multivariate Time Series

April 2024

·

3 Reads

·

2 Citations

IEEE Transactions on Cloud Computing

Yeong-Min Kim

·

Seunghwan Song

·

Byoung-Mo Koo

·

[...]

·

Jun-Geol Baek

Forecasting workloads and responding promptly with resource scaling and migration is critical to optimizing operations and enhancing resource management in cloud environments. However, the diverse and dynamic nature of devices within cloud environments complicates workload forecasting. These challenges often lead to service level agreement violations or inefficient resource usage. Hence, this paper proposes an Enhanced Long-Term Cloud Workload Forecasting (E-LCWF) framework designed specifically for efficient resource management in these heterogeneous and dynamic environments. The E-LCWF framework processes individual resource workloads as multivariate time series and enhances model performance through anomaly detection and handling. Additionally, the E-LCWF framework employs an error-based ensemble approach, using transformer-based models and Long-Term Time Series Forecasting (LTSF) linear models, each of which has demonstrated exceptional performance in LTSF. Experimental results obtained using virtual machine data from real-world management information systems and manufacturing execution systems show that the E-LCWF framework outperforms state-of-the-art models in forecasting accuracy.





Anomaly Detection Framework With Contrastive Learning and Multiview Augmentation for Time-Series Domain Generalization

January 2024

·

4 Reads

IEEE Transactions on Instrumentation and Measurement

Time series data are generated and collected in manufacturing processes. Time series data have different distribution characteristics, with the underlying assumption being that all data originate from the same distribution. Time series anomaly detection methods are trained on stationary data derived from the source domain. However, the performance of the trained model diminishes when confronted with a non-stationary distribution in the target domain. Consequently, there is a need for a model that can adapt to any distribution using only the source domain. In this study, we propose an anomaly detection framework with contrastive learning and multi-view augmentation for time series domain generalization. The proposed method uses multi-view augmentation to learn three representations for time series domain generalization. Moreover, the augmented samples use a temporal convolutional network (TCN) to extract these representations. Subsequently, the extracted representations are trained with a contrastive learning method inspired by simple triple representation learning (SimTriplet). Therefore, the proposed method can attain the diversity and generalization performance of the source domain. The performance superiority of our proposed method is experimentally validated across various domain shift scenarios. Additionally, the effectiveness of the modules used in the proposed method is confirmed through qualitative analysis.


Citations (59)


... This increases profit and lowers total energy consumption, CO2 emissions, and maintenance costs. With an emphasis on probabilistic and uncertain elements, machine learning (ML) and deep learning (DL) techniques have been widely employed to estimate future demand in the cloud environment [11,17,24]. Usually, cloud providers oversee several clusters, each with distinct properties, resource requirements, and workload patterns. ...

Reference:

ADAPT: Attention-Driven Domain Adaptation for Inter-cluster Workload Forecasting in Cloud Data Centers
Enhancing Long-Term Cloud Workload Forecasting Framework: Anomaly Handling and Ensemble Learning in Multivariate Time Series
  • Citing Article
  • April 2024

IEEE Transactions on Cloud Computing

... Hao et al. [77] adopted the WGAN to generate new defect samples and integrated attention mechanisms into a CNN model to recognize defects on a hot-rolled strip steel surface. Song et al. [78] constructed a mapping network for the latent space, which made it possible to compose defect images of various sizes and optimized the proposed method in terms of convergence and computational cost. Jin et al. [79] replaced the activation functions DCGAN with ELUs (EDCGAN) for sample generation and adopted the MobileNet for steel surface defect classification. ...

Defect Synthesis Using Latent Mapping Adversarial Network for Automated Visual Inspection

Electronics

... Rights reserved. where H j and G J 0 are both T s -dimensional matrices (Kim & Baek, 2022). Finally, the time series X can recover from MODWT through the following equation (Gupta, 2018;Wen, 2022 suggests the overall trend of X. ...

Bagging ensemble-based novel data generation method for univariate time series forecasting
  • Citing Article
  • May 2022

Expert Systems with Applications

... Teymourifar et al. [23] constructed a due date allocation model of a dynamic FJSP, propose a new SDR based on the combination characteristics of job shops, and validate the model through simulation. Đurasević and Jakobović [24] proposed a DR selection procedure to select the appropriate DR from the set of evolved DRs based on the features of the problem instances to be solved, which achieves better results than those obtained when only a single DR is selected. Now, some literatures use deep reinforcement learning (DRL) to solve such problems [25] . ...

AGV dispatching algorithm based on deep Q-network in CNC machines environment
  • Citing Article
  • October 2021

... Particularly, the SVM-based classification trained on ResNet50-extracted features surpasses its rivals, obtaining 98% accuracy. Throughout these cases, the use of transfer learning consistently improves performance while decreasing train time and computing system necessities [6]. This paper describes a new deep multimodal cascades technique that combines single-frame or multi-frame image processing to recognize and identify foreign particles. ...

Pattern Classification for Small-Sized Defects Using Multi-Head CNN in Semiconductor Manufacturing
  • Citing Article
  • August 2021

International Journal of Precision Engineering and Manufacturing

... Wafers are processed sequentially in several facilities, and failure to detect facility errors in advance can lead to losses due to defective wafers. To proactively prevent such losses, several companies are attempting to reduce equipment errors by introducing fault detection and classification (FDC) (Lee et al., 2021). Efficient FDC can contribute to productivity and bring improvements by enabling process engineers to detect and act on equipment errors in advance (Hsu & Liu, 2020). ...

Incremental Learning Using Generative-Rehearsal Strategy for Fault Detection and Classification
  • Citing Article
  • June 2021

Expert Systems with Applications

... It measures the local deviation of a data unit with respect to its neighbors. The basic concept behind the LOF is that outliers will have fewer neighbors in their local neighborhood than nonoutliers [46], [47]. The mathematical formula for the LOF can be expressed as [46] LOF (p) = 1 k ...

Anomaly Detection Using Signal Segmentation and One-Class Classification in Diffusion Process of Semiconductor Manufacturing

Sensors

... The RM can help companies increase revenue growth and survive in a fiercely competitive environment. It effectively maximizes expected profits [5,15]. Therefore, this is the first papers that adopts a delayed delivery strategy [16,17] in the RM theory to study the CSAECRP, making new contributions from an applied research perspective. ...

A Reinforcement Learning Approach to Distribution-free Capacity Allocation for Sea Cargo Revenue Management
  • Citing Article
  • May 2021

Information Sciences

... Random forest is a ML method that could be used in classification and regression applications [20][21][22], and it is able to deal with high dimensional data with high performance [23,24], and recently it had been successfully applied to evaluate parameters related to petroleum engineering [25]. RF aims at building an ensemble model composed of a set of decision trees. ...

Random-forest-based real-time contrasts control chart using adaptive breakpoints with symbolic aggregate approximation
  • Citing Article
  • May 2020

Expert Systems with Applications