NMR or plasmon | AIChE

NMR or plasmon

Research Interests:

I am interested in using mathematical modeling and optimization to solve practical chemical engineering problems. Mathematical Optimization is the selection of the optimal elements from many, possibly infinite number of, alternatives, based on a proposed model for the process to be studied, which will provide timely and intelligent aids in decision-making. Over the past few decades of study, optimization has been proved to improve many process performances, including but not limited to: product quality, process robustness, environment sustainability, capital and operational cost, etc. To use optimization properly for solving chemical engineering problems, one should (i) model the process effectively and accurately enough, and (ii) solve the models in a timely fashion.

Previously, I studied the modeling and optimization in three process system engineering problems, including (i) inventory routing, (ii) production scheduling, and (iii) process characterization for the pharmaceutical industry. In the future, I plan to study how to solve the online scheduling and inventory routing problem through the perspectives of finding the “better” solution from the pool of symmetric solutions. I also plan to combine the knowledge of optimization and design of experiments together, which will not only provide the optimal solution, but also offer statistical insights. Additionally, I want to study how to use the modeling method called DRSM, which I have been working on, together with model predictive control. These works will be very useful because of their direct industrial relevance. These problems are also very interesting academically because they are closely related to different areas of process system engineering community, with potential for follow-up and additional work by many other researchers.

Teaching Interests:

I am very interested in teaching courses on (i) process synthesis, design and optimization, (ii) process dynamics and control, and (iii) design of experiments and statistics. First, I have a solid background on process synthesis, mathematical optimization, design of experiments, and artificial intelligence. Therefore, I will love to teach how to analyze and synthesize the chemical processes, as well as how to design and optimize them using the modern optimization or artificial intelligence tools (e.g., integer programming, neural networks, etc.); additionally, design of experiments will also be a very tool for many chemical engineering students to include in such a course. Second, I have studied process dynamics and control, including the advanced model predictive control (MPC), in both undergraduate and graduate years; I have also assisted the teaching of such a course for many semesters (including guest-lecturing). Thus, I would love to teach students how to study the dynamics of process, and how to control the dynamic processes, using the important ideas such as feedback and close-loop; moreover, topics such as MPC, the comparison of knowledge-driven and data-driven models will be interesting to introduce in this course. Finally, design of experiments are not offered in many schools, but it presents a very useful statistics tool for researchers in almost any area of the broad chemical and biological engineering discipline. I am open to teaching other courses too.

Overall, I wish I can lead and guide the students to learn the knowledge, as well as to obtain the tools for self-education, because learning should be a lifelong process, especially considering the rapid growth of technology.

Previous Research Topics and Results:

First, inventory routing problem (IRP) is a supply chain optimization problem that arises in a wide range of manufacturing sectors, including industrial gases and petrochemicals. To improve the efficiency of their supply chains, vendors in a wide range of manufacturing sectors switch to vendor-managed inventory (VMI) practices. VMI refers to an agreement between a vendor and a customer in which the latter allows the vendor to choose the timing and volume of deliveries, while the former agrees to ensure that the customer does not run out of product. A switch to such a practice has a distinct advantage because the vendor can combine deliveries to make more efficient use of the resources and can therefore reduce the distribution cost. However, executing VMI is non-trivial, which requires the solution of IRP, a difficult combinatorial problem. We have conducted studies for IRP in the following three aspects: propose mixed-integer programming models that handle different features and constraints that appear in IRP, especially, driver constraints (1); develop solution methods that can solve IRP efficiently (2); and study how reoptimization should be carried out for IRP (3).

Second, production scheduling has been widely studied, because it is a complex and practical problem that needs to be addressed for many chemical manufacturers. In the face of new information and uncertainties, online scheduling is required, and terminal constraints should be added in the model. However, how to write the terminal constraints is not apparent. The traditional approach would require that the terminal inventory level of each material is required to be greater than a lead-time-based threshold, but this method does not exploit the relationships of inventory levels among materials. Considering this issue, we propose new types of terminal constraints on inventory levels for different network structures (4). The proposed terminal constraints lead to better closed-loop solutions in two aspects, compared to the traditional approach: they prevent stockout and save inventory holding cost. Additionally, we proposed different formulations, and compared their theoretical tightness and computational performances for modeling the sequence-dependent changeovers (5).

Third, Dynamic Response Surface Modeling (DRSM) is a recently proposed data-driven method that models time-resolved outputs, and it can be used for the classical Design of Experiments (DoE) and the generalized Design of Dynamic Experiments (DoDE). Compared to the classical approaches, DoDE and DRSM methodologies allow time-varying inputs, accommodate time-resolved output measurements, and model their interrelationship. Because DRSM quantifies the effects of experimental conditions on the process outputs over time, it can enable more convenient and comprehensive process optimization. We apply the DRSM model in a pharmaceutical problem to identify stoichiometries and obtain the optimal operating condition. The optimization is aimed to maximize the length of an operational window, which is defined based on the species concentrations. Because DRSM provides the models of the concentrations under different conditions and at different times, it is very convenient and useful to apply it for this optimization problem. Moreover, we also incorporate two types of robust constraints in the optimization to explicitly handle the uncertainty in the operating condition as well as in the DRSM model.

Reference

  • Dong, Y., Pinto, J.M., Sundaramoorthy, A., and Maravelias, C.T. MIP Model for Inventory Routing in Industrial Gases Supply Chain. Industrial & Engineering Chemistry Research, 2014, 53, 17214-17225.
  • Dong, Y., Maravelias, C.T., Pinto, J.M., and Sundaramoorthy, A. Solution Methods for Vehicle-based Inventory Routing Problems. Computers & Chemical Engineering, 2017, 101, 259-278.
  • Dong, Y., Maravelias, C.T., and Jerome, N.F. Reoptimization and Policy Analysis for Maritime Inventory Routing under uncertainty. Optimization & Engineering, 2018, doi: 10.1007/s11081-018-9383-8.
  • Dong, Y. and Maravelias, C.T. Terminal Constraints for Single-stage Multi-product Scheduling Problems. In Foundations of Computer Aided Process Operations (FOCAPO), Tucson, AZ, January 8-12 2017.
  • Velez, S., Dong, Y., and Maravelias, C.T. Changeover Formulations for Discrete-Time Mixed-integer Programming Scheduling Models. European Journal of Operational Research. 2017, doi: 10.1016/j.ejor.2017.01.004.
  • Dong, Y., Georgakis, C., Mustakis, J., Hawkins, J. M., Lu, H., Wang, K., McMullen, J. P., Grosser, S. T., Stone, K., Constrained Version of the Dynamic Response Surface Methodology for Modeling Pharmaceutical Reactions. Industrial & Engineering Chemistry Research. 2019, doi: 10.1021/acs.iecr.9b00731.