Research Highlights

(1) Interpretable machine learning for tree ensembles

Research detailsThis work was part of the Advanced Training Program for Smart Manufacturing Intelligence which aimed to cultivate experts for smart factory. In this project I developed a rule extraction algorithm to understand the decision mechanism of complex tree ensembles that can be used to explain predictive models in different domains.
Research achievements
  1. The size of the models is reduced 96.5% in average while performance is maintained in most of the experiments using 33 datasets (average F-score= 91.82)
  2. The project produced two research papers in top-ranked journals, ESWA (2019) and Information Fusion (2023), and a public implementation available in GitHub.
TasksMy involvement in this project was into conceptualization, methodology, software implementation, formal analysis, investigation, writing and visualization.
PublicationsRuleCOSI: Combination and Simplification of Production Rules from Boosted Decision Trees for Imbalanced Classification, Expert Systems with Applications (2019)

RuleCOSI+: Rule extraction for interpreting classification tree ensembles, Information Fusion (2023)

(2) Fault detection in the injection molding process

Research detailsThis work was developed in collaboration with the Singapore Institute of Manufacturing Technology (SIMTech), A*STAR, Singapore. It consisted in generating explainable visualizations of predictive models for fault detection in the injection molding manufacturing process. A rule-based explanations framework was developed that combines decision rules, feature importance and dependence visualization plots, to aid quality engineers in understanding the main manufacturing process conditions affecting the quality of the products
Research achievements
  1. The framework can reduce the size of ensemble predictive models from 186 decision rules to only 5 decision rules, while maintaining the classification performance (average F-score=95.2).
  2. The project produced a research paper in the Journal of Manufacturing Systems (2021)
TasksMy involvement in this project was into conceptualization, methodology, software implementation, formal analysis, investigation, writing and visualization.
PublicationsRule-based Explanations Based on Ensemble Machine Learning for Detecting Sink Mark Defects in the Injection Moulding Process, Journal of Manufacturing Systems (2021)

(3) Visualization method for quality prediction in the die-casting process

Research detailsThis work was also part of the Advanced Training Program for Smart Manufacturing Intelligence, and it also consisted in the generation of explanations of predictive models for fault detection in the die-casting manufacturing process. The framework is based in a mathematical formal language called Formal Concept Analysis (FCA).
Research achievements
  1. A novel explainable model for visualizing the hierarchies of manufacturing process conditions with respect to the quality of the final products was developed.
  2. The visualization model was used to identify the main process conditions interactions related to different type of faults in the die-casting process
  3. The project produced one research paper in the Journal of Intelligent Manufacturing.
TasksMy involvement in this project was into conceptualization, methodology, software implementation, formal analysis, investigation, writing and visualization.
PublicationsRule-based visualization of faulty process conditions in the die-casting manufacturing, Journal of Intelligent Manufacturing Systems (2022)

(4) State-of-Health prediction for lithium-ion batteries

Research detailsThis work is part of the Next-Generation Engineers for Smart Energy program, and it consisted in the application of Convolutional Autoencoders (CAE) and deep neural networks (DNN) to SOH prediction for lithium-ion batteries using electrochemical impedance spectroscopy. This work was developed in collaboration with researchers of the Chemical Engineering department or Kyung Hee University.
Research achievements
  1. The performance (R2) of the previous state-of-the-art model (Gaussian Process Regressor) was improved from 0.89 to 0.98 in average on the test cells.
  2. The project produced one research paper in the Journal of Energy Storage.
TasksMy involvement in this project was into conceptualization, methodology, software implementation, formal analysis, investigation, writing and visualization.
PublicationsConvolutional autoencoder-based SOH estimation of lithium-ion batteries using electrochemical impedance spectroscopy“, Journal of Energy Storage (2023)

(5) Photovoltaic power generation forecasting

Research detailsThis work was part of the Next-Generation Engineers for Smart Energy program, and it consisted in the application of Transformer networks for multistep ahead photovoltaic power generation forecasting for Ulsan and Dangjin solar power stations.
Research achievements
  1. Compared to previous state-of-the-art LSTM forecasting models, the performance (R2) of the transformer model was increased from 0.75 to 0.81 for Dangjin station and from 0.59 to 0.83 for Ulsan station
  2. Best Paper Award in the International Conference on Innovation Convergence Technology (ICICT2021)
TasksMy involvement in this project was into validation, conceptualization, and writing.
PublicationsThe project produced one research paper that is on the final stage of writing, and it will be submitted to Applied Energy journal.