(3gs) Data-Driven Uncertainty Aware Optimal Design

My research, including the safe model predictive control through stability analysis, machine learning-based dynamic optimisation for PDEs, CFDs and the reinforcement learning-based dynamic optimisation under uncertainty, matches the department's requirements.

Optimisation of industries is driven by the need for reducing costs and enhancing production to remain competitive in the global marketplace. A domain that is currently struggling in this regard is the design, operation, and scale-up of sustainable production systems, in specific, bio-production processes.

Partial differential equations combined with data-driven approaches are major tools for the modelling of bioprocesses. These pose a particular challenge, as they are governed by highly complex dynamics, have large process uncertainty, and present low yields, which endangers their economic viability. Major players in addressing this matter include the optimisation of plant-wide superstructures, real-time optimisation, and process control, all of which require solving numerically intricate nonlinear problems. Additionally, hybrid modeling approaches are required to compensate their inherent stochastic nature, using combined physics based laws and data-driven models. Furthermore, in the era of big data and automation, one of the key features of modern manufacturing systems is the integration of information and decision-making to take optimal actions, which has led to the recent progress of data science and predictive analytics in many fields.
Traditionally, process intensification strategies in chemical and biochemical processes apply physics-based models along with optimisation and control frameworks to improve performance of the system. Although chemical engineers make use of many fields, few works adopted recent advances in the fast-growing areas of Artificial Intelligence (AI) and Machine Learning (ML). However, incorporating ML and AI tools into the chemical engineering repertoire is paramount to transform the chemical industries into the era of the digital economy, biotechonomy and industry 4.0. As a result, my research within the next 5 years will focus on the adoption and development of data-driven (i.e. AI and ML) based methodologies to address the major challenges in the design, optimisation, and control of bioprocesses towards a cognitive manufacturing approach.

Teaching Experience

As I fulfil my research career, I have volunteered for as many teaching and mentoring activities as I can. I have initiated 3 mini courses so far. In 2018 in the University of Manchester I initiated and organized the Model Predictive Control reading group between the schools of Electrical and Electronic Engineering (EEE) and Chemical Engineering and Analytical Science (CEAS), in which I gave a series of introductory lectures and podcasts regarding convex optimisation, dynamic optimisation, MPC and stochastic MPC. When I moved to London, I decided to continue this exciting project in the centre for process system engineering (CPSE). Hence, I initiated the MPC reading group between Imperial College of London and University College London. For computational exercises of this mini-course, Python and Collaboratory notebooks were developed, which I consider great tools for learning. This semester I started a mini course with hands-on examples for machine learning in CPSE. More classic techniques are discussed that are unknown for the most chemical engineers regarding classification technologies.

I think that my most significant ability as a lecturer is the ability to explain complex and abstract concepts in a way that is intuitive and easy to understand. I am confident in my ability to teach a variety of undergraduate and graduate courses, probably most notably, Process Analysis, Mathematics Fundamentals, Engineering Mathematics, Process Dynamics and Control, Process Optimisation , Process Design and the process systems engineering advanced courses. Whenever starting a new topic in a lecture/tutorial/supervision, I first go through a less-formal and more intuition-based explanation behind the technical apparatus of the topic. Such an explanation gives students a `big picture' and a rough outline of what to expect. Then, I like to present the material as if we, as a class, were re-discovering it together, and explaining what benefits this new method or topic has against previous existing ones. Finally, I show how to apply this new approach and draw a logical conclusion about it. My excitement in teaching comes from trying to make a topic not only easy to understand, but also enjoyable and even exhilarating, one that students will be eager to revise or even pursue.

Furthermore, during my PhD I mentored 2 MEng students and 1 MSc students in the University of Manchester and Currently, I have already supervised 1 Msc student at Imperial College London (ICL), and I am currently supervising 3 Msc students and 1 PhD student at Imperial College London, 2 MEng students and 1 PhD student at University College London (UCL).
Currently, I have already supervised 1 Msc student at Imperial College London (ICL), and I am currently supervising 3 Msc students and 1 PhD student at Imperial College London, 2 MEng students and 1 PhD student at University College London (UCL). The topics include: Modular flowsheet optimisation using surrogate models embedding Gaussian processes, Evolutionary optimisation for black-box models, Policy gradient methods in reinforcement learning for control of continuous bioprocesses, Reinforcement learning for process optimisation of batch bioprocesses and Model-based data mining approach for identification of kinetic models and model discrimination. Additionally, I supervise 2 PhD students at Imperial College London and University College London with the following projects. Project (ICL): Flowsheet optimisation using modular Gaussian processes and uncertainty propagation. Project (UCL): Dynamic optimisation for model-based design of experiment using mixed-integer nonlinear programming. In parallel, I supervised and helped on the development and organization of the relevant course-work projects of the computer aided process design (graduate level) course, the chemical reactor design (graduate level), the process control course, and the computing courses where students learned to use Matlab, GAMS, and Aspen for process simulation. In these courses, I supervised (roughly) 30 students per year, per course.

I am confident in my ability to teach a variety of undergraduate and graduate courses, probably most notably, Process Analysis, Mathematics Fundamentals, Engineering Mathematics, Process Dynamics and Control, Process Optimisation, Process Design and the process systems engineering advanced courses in the fourth year. Whenever starting a new topic in a lecture/tutorial/supervision, I first go through a less-formal and more intuition-based explanation behind the technical apparatus of the topic. Such explanation gives students a “big picture” and a rough outline of what to expect. Then, I like to present the material as if we, as a class, were re-discovering it together, and explaining what benefits this new method or topic has against previous existing ones. Finally, I show how to apply this new approach and draw logical conclusion about it. Complementary to a well-explained topic, I think that experience in problem-solving is crucial to build conceptual understanding. Therefore, I present and work through different examples that enhance the mental picture that students have for an abstract concept. With each example I stress a particular issue, concept or advantage of the topic. For computational exercises, I like using Python or Collaboratory notebooks, which I consider great tools for learning. Engaging students is probably one of the most difficult parts of teaching, and my favourite tool to address this is Mentimeter. With this tool you can ask questions to your students interactively, and award (participation) points to students that answer questions correctly and even quickly. This is a superb way of engaging students and tracking participation in class. In the end, my excitement in teaching comes from trying to make a topic not only easy to understand, but also enjoyable and even exhilarating, one that students will be eager to revise or even pursue.