(6ig) Accelerating Chemical Discovery and Process Development with Theoretical Models and Machine Learning | AIChE

(6ig) Accelerating Chemical Discovery and Process Development with Theoretical Models and Machine Learning


Gao, H. - Presenter, Massachusetts Institute of Technology
Research Interests:

Computation and modeling have played an important role in chemical engineering. Theoretical models can be used to determine the thermodynamic limits and kinetic evolution of chemical systems, thus shedding light upon controlling and optimizing chemical processes. Machine learning models have more recently demonstrated their power in predicting complicated system behaviors and have gained fast-growing interest, particularly in molecular property prediction and optimization, materials design and synthetic path planning. With the complexity of chemical systems quickly increasing, it is desirable to further expand the scope and capability of different models and explore the synergies between them. I am excited about combining these modeling techniques to study chemical reactions and process systems, in order to make the best use of historical and new experimental data, thus enabling more effective experimental efforts and speeding up process development and optimization. Initially, I would like to develop research projects on 1) applying transfer learning to expediate new polymer design, 2) using reinforcement learning to identify mechanisms for complicated reaction systems, and 3) automating holistic chemical process design with the help of artificial intelligence (AI).

Research Experience:

My PhD dissertation focuses on simulation and optimization of radical polymerization processes, combining mechanistic kinetic modeling and optimization algorithms. I developed schemes to accelerate kinetic Monte Carlo (KMC) method, which is used to simulate the explicit sequence of polymer chains during polymerization process. These methods significantly reduce the computational cost of KMC simulations so that they were successfully incorporated into derivative-free optimization algorithms to design and optimize polymer sequence distributions.

During my post-doc, I worked on AI-assisted synthesis route design and evaluation. I developed a deep learning model trained on a large reaction database to recommend conditions (catalyst, reagent and solvent) for given organic reactions with a high accuracy. Without explicitly given the properties of the chemicals, the model learned the similarity of solvents/reagents/catalysts, which was used to assess compatibility of reactions and identify retrosynthetic suggestions that are amenable to one-pot synthesis development. I also used optimization techniques to minimize the chemical inventory for multiple synthetic targets.

Teaching Interests:

As a chemical engineer by training, I am capable of and passionate about teaching core chemical engineering courses in general. During the 4th year of my PhD, I participated in the Teaching Apprenticeship Program at Northwestern University and taught half of the graduate kinetics class under the supervision of Prof. Justin Notestein. I have also recently participated in the Kaufman Teaching Certificate Program at MIT to fine tune my teaching methodologies and philosophy.

Selected Publications:

  1. Gao, H., Struble, T. J., Coley, C. W., Green, W. H., Jensen, K. F. (2018). Using Machine Learning to Predict Suitable Conditions for Organic Reactions. ACS central science, 11, 1465-1476. Highlighted by Science - DOI: 10.1126/science.362.6420.1260-b
  2. Gao, H., Waechter, A., Konstantinov, I. A., Arturo, S. G., Broadbelt, L. J. (2018). Application and Comparison of Derivative-free Optimization Algorithms to Control and Optimize Free Radical Polymerization Simulated Using the Kinetic Monte Carlo Method. Computers & Chemical Engineering, 108, 268-275.
  3. Gao, H., Konstantinov, I. A., Arturo, S. G., Broadbelt, L. J. (2017). On the Modeling of Number and Weight Average Molecular Weight of Polymers. Chemical Engineering Journal, 327, 906-913
  4. Gao, H., Konstantinov, I. A., Arturo, S. G., Broadbelt, L. J. (2016). Acceleration of Kinetic Monte Carlo Simulations for Free Radical Copolymerization: a Hybrid Approach with Scaling of Kinetic Parameters. AIChE Journal, 63(9), 4013-4021. Selected as Editor's Choice: Reaction Engineering, Kinetics and Catalysis
  5. Gao, H., Oakley, L. H., Konstantinov, I. A., Arturo, S. G., Broadbelt, L. J. (2015). Acceleration of Kinetic Monte Carlo Method for the Simulations of Free Radical Copolymerization through Scaling. Industrial & Engineering Chemistry Research, vol. 54(48), 11975-11985. Selected as the ACS Editor’s Choice for October 26th, 2015