Conference Agenda

Prompt engineering - improving the command given to a large language model (LLM) - is becoming increasingly useful for maximizing the performance of the model and, therefore, the quality of the output. However, in certain circumstances, LLM outputs need to be dynamically personalised personalized to specific users, for them to be effective. Prompting should, in this case, adapt dynamically according to the needs and preferences of the individual users. It is, therefore hence, useful to enrich prompt engineering with models that express user's preferences and implement them directly in the prompt. Logic-based machine learning has recently witnessed the development of human-interpretable, robust, and data-efficient algorithms and systems, called (ILASP), capable of learning preference models from data and background knowledge [1, 2]. These systems can play an important role in the development and advancement of digitalization strategies. They can be used, for instance, to learn personal user's preferences without sacrificing the human interpretability of the learned outcomes.

The Technical University of Denmark (DTU) offers a course on Good Manufacturing Practices (GMP). As part of the course, students are required to participate in an audit exercise, in which they interview a fictional company, represented by teachers, about its good manufacturing practices. In spring 2024, teachers have agreed to test the replacement of the physical teacher with ChatGMP, an AI-powered digital audit tool to represent the company. The chatbot, tested on a subset of volunteering groups, proved to be considered a viable alternative by both teachers and students, and it is therefore now regularly used in the course. Currently, the prompt given to the chatbot is the question formulated by the students. However, the prompt could be enriched with other features learned, such as the students' personal preferences.

In this work, we demonstrate, as proof of concept, how ILASP can be used to learn personal preferences of students to tailor and improve the prompts and generate targeted responses. More specifically, three different cases are investigated: (i) detecting missing questions in the audit, i.e., automatically assessing whether the groups have prepared questions regarding two central topics; (ii) checking for students' question repetition, which would suggest that ChatGMP's answer was inadequate, being either incomplete or not understood by the students; and, (iii) learn students' preferences, intended as both general rules to achieve a good performance (e.g., questions about a specific topics are to be preferred to maximize students' performance), as well as personal preferences of the groups, such as length and complexity of the questions or order of the topics.

References

[1] Mark Law, Alessandra Russo, and Krysia Broda. Logic-Based Learning of Answer Set

Programs. Lecture Notes in Computer Science (including subseries Lecture Notes in

Artificial Intelligence and Lecture Notes in Bioinformatics), 11810 LNCS:196–231, 2019.

[2] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael

Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learn-

ing with a unified text-to-text transformer. Journal of Machine Learning Research,

21(140):1–67, 2020

In this paper, we present our recent advances and achievements in automatic control course in the engineering study of cybernetics at the Faculty of Chemical and Food Technology STU in Bratislava. We describe the course elements and procedures used to improve teaching and learning experience. We discuss on-line learning management system, various teaching aids like e-books with/without solutions to practice examples, computer generated questions, video lectures, choice of computation and simulation tools.
The course is provided in the presence form of study for about 20 students but it relies heavily on on-line tools and methods. Starting from this academic year, flipped design of the course was designed. We describe our experience in the preparation of such a change and some initial feedback from the students.
The course concentrates on input/output linear approximation of processes in chemical and food technology and discusses poles/zeros, process dynamics, frequency and time methods, closed-loop system, design and tuning of PID controllers, root-locus analysis.
Although majority of laboratory excercises is performed using simulation tools, we also provide easy-to-use laboratory device that can be given to all students to work either at school or at home.

Extended Abstract

In the 2023-24 academic year, the lead author of this paper was on sabbatical at London’s Imperial College Department of Chemical Engineering, where he taught three undergraduate courses, all taught using variations of the “flipped class” paradigm:

1. The “good” course, on numerical methods, used flipping in each of the ten weeks of the course involving an online lesson, which students needed to complete in advance of a weekly class meeting of three hours, mostly focusing on problem-solving exercises by the students themselves with assistance from the course staff. This was the implementation that most closely followed the flipped class paradigm implemented at the Technion (Lewin and Barzilai, 2022, 2023).
2. The “bad” course consisted of six weeks covering process dynamics, taught by conventional lecturing with no hands-on exercises by the students, followed by five week covering process control, taught with a flipped approach, in which each week, students completed an online lesson in advance of weekly class meetings of three hours in total, mostly focusing on problem-solving exercises by the students themselves assisted by the course staff.
3. The “surprising,” one-week workshop was an integral part of the process design course. It involved three online lessons completed by the students in advance of class activity, with each online lesson associated with a hands-on problem-solving class meeting, to a total of seven contact hours held during a single week of the course.

This paper describes these three implementations in detail and presents and analyzes the responses from student surveys intended to ascertain students’ perceptions about the level of their satisfaction with the flipped class approach and the degree to which they achieved mastery of the taught materials. The actual learning outcomes are also analyzed, that is, the exam results in the case of the first two courses, and the performance of the students on their design project, for the last course. The paper ends with conclusions and recommendations concerning how the flipped class method should be implemented for success, depending for which classroom situation it is intended.

Keywords: Chemical engineering education, numerical methods, process control, process design, heat exchanger network synthesis, flipped classroom, active learning.

References

Lewin, D. R. and A. Barzilai (2022). “The Flip Side of Teaching Process Design and Process Control to Chemical Engineering Undergraduates – and Completely Online to Boot,” Education for Chemical Engineers, 39, 44-57.

Lewin, D. R. and A. Barzilai (2023). “A Hybrid-Flipped Course in Numerical Methods for Chemical Engineers,” Comput. Chem. Eng., 172, 108167.

This Chemical Thermodynamics course is designed for Bachelor of Engineering Students as well as first-year master’s students in Chemical Engineering and Materials Science.

The approach is not about introducing innovative activities but rather re-phasing existing activities based on the three-phase competency evaluation model described by Carette (2007) and cited by Dierendonck (2014):

1. Phase 1: Students are given a project, which is a complex task requiring the selection and combination of several elementary tasks they have performed during the tutorials (TDs).
2. Phase 2: The project remains, but the complex task is broken down into elementary tasks. Each week, at the end of the TD, students receive explicit instructions that allow them to progress in their project using the knowledge they have just acquired. These instructions will enable students to complete the overall complex task. However, for each elementary task, it is up to the student to determine the procedure to implement from those they are supposed to know.
3. Phase 3: During the TDs, students are presented with a series of simple, decontextualized tasks, which are the exercises proposed. These tasks correspond to the elementary procedures that will need to be mobilized to accomplish the complex task from Phase 1.

The TDs will systematically follow the same structure:

• Theoretical reminders related to the exercises performed, constructed with the help of students via interactive tools like Wooclap (or others).
• A specific exercise solved on the board, with each step detailed aloud to allow students to reproduce the reasoning. Pauses are planned during the resolution to enable students to develop their own reasoning, followed by a sharing of ideas.
• Additional exercises to be solved in pairs, with the teacher available to answer questions and provide hints.
• Definition of the part of the project that can already be solved thanks to the exercises performed. Students will be invited to solve it for the following week.

The course is divided into two parts:

1. The first part is dedicated to methods for evaluating the physical and thermodynamic properties of pure substances. It also reviews the main methods for predicting thermodynamic properties and the major families of equations of state.
2. The second part of the course focuses on these methods for mixtures of different substances. It also covers the description of multiphase equilibria of mixtures and the thermochemical quantities characterizing chemical reactions.

Projects are presented at the beginning of each part and will be subject to formative evaluation at mid-term. The goal is to lead students to a better understanding of various theoretical concepts through the implementation of an authentic project. Additionally, the aim is to support their motivation by specifying the different tasks to be accomplished each week.

References

Carette, V. (2007). L’évaluation au service de la gestion des paradoxes liés à la notion de compétence. Mesure et évaluation en éducation, 30(2), 49-71.

Dierendonck C. et Fagnant A. (2014) Approche par compétences et évaluation à large échelle : deux logiques incompatibles ?

Exergy is a very useful process systems engineering analysis tool, particularly as an aid for process design, synthesis, modelling, and performance benchmarking. At its core, exergy is a form of thermodynamic value that takes into account both the quantity and quality of energy. Although many engineers want to learn about exergy, they often find the topic impenetrable and soon discover that the effort required to compute exergy values can be too time consuming to be worth the effort. However, my colleagues and I have developed some new educational materials that make it drastically easier to not only learn the concept but also actually apply it for something useful, gaining insights into your process that you could not easily see without it.

In this talk, I will explain the concept of exergy in plain language, focusing on the thermomechanical (heat/pressure/phase) and chemical (chemical bond/phase) forms of exergy that are most relevant to chemical engineers. I will demonstrate how engineers can use the new book Exergy Tables which contains a compendium of pre-computed exergy values for thousands of chemicals at a wide variety of temperatures and pressures using a rigorous and well-defined referencing system. Moreover, all exergy computation results using Exergy Tables can be directly compared between different applications, researchers, and studies because they all use the same set of reference standards, which is particularly important for chemical exergies relevant for chemical engineering applications. In other words, the book does all the hard parts of the computation up front so that engineers can very quickly apply the results to real problems and immediately see the benefits.

In the talk, I will show how to use the book to compute exergy quickly and easily for substances and systems at high and low temperatures, high and low pressures, and for various chemicals. I will discuss how this material can be integrated into chemical engineering education (especially unit operation and process design) by providing many in-class examples for applications such as fuel combustion, steam and power generation, CO₂ capture from power plants, CO₂ sequestration, direct air capture, heat exchanger network design, water treatment systems, work-heat integration, and others. I will discuss how to use exergy numbers as a part of systems analyses through methods such as visualizing exergy flows, computing exergy-based efficiencies or other key performance indicators, utility and capital cost estimation, value proposition, and thermodynamic benchmarking, all in an educational context. The result is that with a relatively small amount of effort, you can give your students a powerful tool to use throughout their chemical engineering education.

Current progress in digitalization has led to a wide interest in learning more from available data. Advanced data analytics can be achieved through commercially available software; however, learning to program allows for more flexibility and ultimately more freedom in the potentially tailor-suited research. Among other programming languages, Python is one of the most requested. However, the integration of programming into the curricula might imply fundamental restructuring [1].

To train engineers willing to take on the challenge, we previously implemented sPyCE [2], an open-source series of Python courses tailored to chemical engineers. These courses cover topics such as design of chemical reactors, stoichiometry relationships, data pre-processing, data analysis, and data science. We also implemented FermentAI [3], a chatbot trained to answer questions about fermentation processes. To intensify the previously carried out efforts and create both a pedagogical framework to teach programming to (bio)chemical engineers, as well as to provide students with the opportunity to ask questions in a semi-private and tailored environment, we explore the integration of sPyCE and FermentAI into the BioVL platform [4], a virtual laboratory for teaching (bio)processes.

The main goal of this work is to enable students to (i) learn about bioprocesses and, simultaneously, (ii) learn how to model them, and (iii) ask questions to a chatbot phrasing their doubts about the teaching content. We believe that the Python tutorials, given the stimulant material used along with gamification, will be an engaging and insightful way to learn to model bioprocesses. Moreover, the chatbot will enable students to ask for clarification right away without any time delay – facilitating their learning progress.

In a small-scale test, we evaluated how the tools are integrated into the educational platform and conducted quantitative interviews with the volunteering students. The results show positive feedback regarding the usefulness of implementing the Python tutorials in BioVL. Moreover, students appreciate the possibility of having a chatbot available to ask questions and clarify doubts. All the collected feedback is used to further improve the platform, with the goal to provide as seamless an experience as possible. Finally, it was suggested that further development of other tailored chatbots could facilitate the students' learning and progress.

References

[1] Hermann J. Feise and Eric Schaer. “Mastering digitized chemical engineering”. In: Education for Chemical Engineers 34 (2021), pp. 78–86. ISSN: 1749-7728. DOI: https://doi.org/10.1016/j.ece.2020.11.011. URL: https://www.sciencedirect.com/science/article/pii/S1749772820300622.

[2] Caccavale, F., Gargalo, C. L., Gernaey, K. V., & Krühne, U. (2023). SPyCE: A structured and tailored series of Python courses for (bio) chemical engineers. Education for Chemical Engineers, 45, 90-103.

[3] Caccavale, F., Gargalo, C. L., Gernaey, K. V., & Krühne, U. (2024). FermentAI: Large Language Models in Chemical Engineering Education for Learning Fermentation Processes. In Computer Aided Chemical Engineering (Vol. 53, pp. 3493-3498). Elsevier.

[4] Caño de las Heras, S., Gargalo, C. L., Caccavale, F., Kensington-Miller, B., Gernaey, K. V., Baroutian, S., & Krühne, U. (2022). From Paper to web: Students as partners for virtual laboratories in (Bio) chemical engineering education. Frontiers in Chemical Engineering, 4, 959188.