Development of high-speed semiconductor dispatching and explainable AI technology using machine learning

Research detailsThis work is being developed in conjunction with a private manufacturing company. It has two main areas. The first area is dedicated to the development of technology for speeding up the semiconductor dispatching process using deep learning and ensemble learning ranking models. The second area consists of extracting interpretable dispatching rules from accurate ensemble models.
Research achievementsThe machine learning ranking models developed had 100% dispatching accuracy while being 1.22 times faster compared to other models. The execution of rulesets extracted from ensemble models took 0.05 seconds for dispatching 1000 jobs compared to 0.26 seconds of LightGBM. The interpretable rules generated from accurate ensemble models obtained a 99.99% dispatching accuracy
Period2021.9~present (1 year and a half)

Smart Factory Applications for Big Data in Manufacturing

Research detailsDevelopment of applications for Big Data processing and analysis in manufacturing with the purpose of enabling smart factories in small and medium-size enterprises
Research achievementsWe developed a Smart Factory framework containing several libraries for analysis and processing of Big Data in manufacturing. The main functions carried out by the libraries were: production line sensor data pre-processing, machine learning modeling, analysis of process variables, equipment life prediction, anomaly or outlier detection, process novelty detection, and energy consumption forecasting.
Period2015.7~2018.6 (3 years)
PublicationPRANAS: A Process Analytics System Using Process Warehouse and Cube“, Applied Sciences (2020)

Development of Reverse Engineering Techniques Using CAD Models and 3D Scan Data

Research detailsDeveloped technology to generate CAD drawings from 3D point cloud to be used for quality inspection of parts.
Research achievementsWe developed procedures to extract and measure key quality inspection areas from 3D point cloud by applying machine learning clustering of algorithms. My work consisted of developing a deep learning model that can generate surfaces based on sample points on cross-sections.
Period2018.7~2019.2 (8 months)

Deep Learning Architecture for Molecular Dynamics

Research detailsPredicting the movement of atoms using deep learning
Research achievementsApplying LSTM technique to predict atomic movement with good performance (R2=0.9964)
Period2017.4~2018.6 (1 year, 3 months)

Decision Mining for Business Processes

Research detailsUsing state-of-art data and process mining techniques, discover business decisions that are often hidden in process flows and process activities. This work was done in collaboration with KU Leuven
Research achievementsWe surveyed the current landscape and future directions for decision mining in the business process area and presented our work in the International Conference on Business Process Management 2016. Additionally, we earned the grand award in the Process mining Case competition (AP-BPM 2015) with a study case that combine Process Mining with Decision Trees.
Period2013.12~2016.1 (2 years, 2 months)
Publication“Decision Mining in a Broader Context: A Survey on the Current Landscape and Future Directions”, In Proc. of the 4th Int. Workshop on Decision Mining & Modeling for Business Processes (DeMiMoP’ 2016) in conjunction with BPM 2016, Sep 2016.

Reality Mining for Regional Big Data

Research detailsDevelopment of Reality Mining Models and Systems from Regional Big Data for Comprehension of Individual and Social Behaviors.
Research achievementsCreation of a new Process Mining algorithm for the generation of information diffusion models in Social Network Services data. My work was published in the IEEE Access journal (2019).
Period2013.6~2016.5 (3 years)
PublicationsInfoFlow: Mining Information Flow Based on User Community in Social Networking Services“, IEEE Access (2019)

Analyzing Information Flow and Context for Facebook Fan Pages“, IEICE Transactions on Information and Systems (2014)