(6il) Online Process Optimization of Complex Cyber-Physical Systems

Online process optimization deals with driving a process towards optimal operation in the presence of varying disturbances and uncertainties. In the presence of increasing competition and increasing focus on sustainable operation, real-time optimization (RTO) is widely recognized as one of the most crucial technologies in a wide range of application areas and as an important factor in enabling new application areas such as renewable energy systems, medical and bio processes, water management, food production etc. In spite of this, traditional real-time optimization is not widely used in practice. From my four years of industrial research experience, together with three years of academic research experience in the field of process optimization, I have identified three main challenges that impede the practical application of real-time optimization in many processes, which are:

1. Lack of good models that can be used in optimization algorithms (offline model development)
2. Wrong value of model parameters used in the optimization problem (online model update)
3. Numerical robustness including computational issues

My primary research interest is to address this gap in online process optimization and develop novel decision-making methods and algorithms to tackle these challenges.

Based on the current challenges and knowledge needs, important research question that I aim to address are:

• How can process data be used more efficiently to address the challenges with the traditional model-based optimization tools for large-scale systems?
• How can we handle different types of uncertainty in the real time optimization (RTO) problem?

Building on the previous work on process optimization and control structure design, I aim to address the challenges with the traditional RTO schemes and take a new look at the possibilities and advantages of exploiting data more efficiently to address these challenges. I propose to achieve this by means of different approaches ranging from classical feedback control to more advanced machine learning and artificial intelligence (AI) tools. To this end, I make the following hypotheses that forms the basis for my research goals in the next three to five years.

Hypothesis 1: Combining machine learning algorithms that uses online process data with traditional model-based RTO methods, can address the challenge of developing and updating the model (Challenge 1)

Hypothesis 2: Modified RTO algorithms that use also transient data more efficiently, can alleviate the steady-state wait time issues (Challenge 2)

Hypothesis 3: Developing systematic approaches to convert the optimization problem into a feedback control problem can alleviate computational issues associated with solving numerical optimization problems online (Challenge 3).

Hypothesis 4: A hierarchical combination of the different RTO approaches that exploits the different time scales, can handle a wider class of uncertainty in an efficient manner and provide a unified RTO paradigm that exploits the synergy between the two existing paradigms.

By addressing the main challenges with traditional optimization approaches, the potential of process data can be better exploited in many industrial applications leading widespread application of autonomous decision-making tools in the process industries and at the same time enable new application areas.

Teaching Interests:

There is no greater joy than imparting knowledge to a generation of young minds to provide them with the tools to succeed in life and tackle the future challenges facing our society. Not only my research interests, but also my teaching interests are motivated by the need for technology to address the societal challenges. I believe that process systems engineering is an important subject area in achieving this and teaching these concepts at both undergraduate and graduate level becomes increasingly important.

My primary interest lies in teaching topics from process systems engineering (PSE), including process modelling, control, design and optimization. Having a background in mechatronics engineering with specialization in control theory, along with industrial experience in process control and PhD in chemical engineering, I have found my multidisciplinary background to be advantageous, since PSE as a field is becoming increasingly multidisciplinary. Therefore, to reflect this transformation in the PSE community, I am also interested in teaching multidisciplinary courses from a PSE context such as system identification, neural networks and optimal process control.

Teaching process systems engineering as a subject area involves a significant amount of scientific and mathematical concepts. When teaching mathematical concepts involving theorems, and formulae, it is very important that my students clearly understand the underlying concepts and how to derive these concepts. I tend to use the blackboard often when I teach. This is mainly because, I find that using a blackboard and writing the concepts is much easier for the students to follow along than when they just appear as slides in front of them. Due to varying learning rates among students, I also noticed that this gives enough time for the majority of the classroom to follow along. Since different learning style uses different parts of the brain, by involving more of the brain during learning, we remember more of what we learn. Therefore, in addition to visual and auditory teaching aides, I also advocate my students to take notes to engage in kinesthetic learning, thereby covering a broad range of learning styles in the classroom. The clarity and aesthetics of my handwriting and teaching style has also been well received by students.

I find that I myself learn a lot about the subject area from teaching, which further strengthens my research activities. I strongly believe that teaching and research are complementary.