Open questions

Every now and then, an open problem comes up that I believe is worth investigating. I think about the problem for an extended amount of time, and reach a road block (which is often based on time, resources and/or theory). This is the place where I share some of those ideas on which I am not actively working on, but that bug my mind every once in a while.

This a way to invite people to collaborate with each other to solve problems that I believe are important. Please do not hesitate to contact me if you have an idea or would like to discuss/work on any of these problems. I would be more than happy to collaborate or discuss them in more detail (over coffee if you are around Cambridge!).

Supercritical lubrication

Supercritical fluids, and more specifically supercritical CO2-based fluids, appear to be the holy grail of the solar thermal energy industry. However, while they present a number of advantages over conventional fluids (such as non-degradability at high temperatures, high coefficient of thermal expansion, etc), they also come with several issues that need to be addressed before they can become a commercially feasible solution. A prominent one is the lack of appropriate turbomachinery that work well with dense, highly compressible fluids at high temperatures. For example, supercritical CO2 has a coefficient of thermal expansion that is similar to that of air, but with a density that is 2 orders of magnitude higher! In addition to this challenge, the transport properties of these fluids can change quite a lot depending on the thermodynamic state on which they operate, which makes it more difficult to design a turbomachinery capable of operating well in off-design conditions.

My interest on this topic is to learn more about the behavior of supercritical fluids flowing along very narrow slits and channels, similar to what we would expect to find in a compact turbomachine. Under these conditions, significant load, pressure and temperature gradients can be expected, which in turn may reveal interesting heat transfer and fluid flow phenomena. More specifically, are there any kind of fluid mechanical instabilities that can be expected to arise at pressures and temperatures past the critical point?

While many computational and experimental studies on the thermo-hydraulic behavior of supercritical fluids have been published, instabilities are difficult to find if you are not actively looking for them. They are easy to be disregarded as computational errors or can be within instrumentation errors, at least at first glance. A prominent example is the only recent discovery of the piston effect. My progress on this has been mostly taking a step back and looking into simple analytical models, such as the Reynolds lubrication theory, and seeing what interesting phenomena they can reveal. I expect that a lot can be discovered using existing theoretical fluid mechanical machinery.

How can complex, high speed fluid phenomena benefit from machine learning?

Phase-change research is one of the most (if not the most) data-intensive and data-rich fields in heat transfer. Video imaging is often acquired at very high frequencies, and rarely used for quantitative analysis of the phenomenon. Even today, with the large amount of data that we have available, we still know very little about the mechanisms behind phase-change heat transfer. This was expected, because algorithms to handle such a large amount of data–and to make sense out of them–were mostly inexistent. Today, however, statistical machine learning can do wonders in revealing interesting patterns hidden in data. Do start demonstrating that, I (and other collaborators) have published a few papers showing that computer algorithms can learn how to classify and quantify heat transfer data from complicated boiling phenomena.

I have two main complementary ideas of how machine learning could grow into a fruitful subfield in the transport phenomena community:

  1. We should start building an open-access library of phase-change heat transfer data and images

    The ability for machine learning to succeed in revealing interesting phenomena is strongly correlated with the amount of data that we have available and the amount of variation that we have in the data. Therefore, it is extremely important for anyone working in this field to have access to as much different data as possible. The machine learning community has come with interesting ways to deal with this problem, one of which is the creation of standardized databases such as the ImageNET and MNIST collections. However, differently from the objectives of algorithm developers, our objective is not to compare the performance of specific models against each other, but rather to use these algorithms to extract quantitative and qualitative information about the combined phase-change heat transfer and multiphase flow phenomenon. Hence, I believe that it would be far more useful to construct an open-access library with curated data acquired for different fluids, on different experimental setups and for different operating conditions, from which more general conclusions could be drawn.

  2. We should look into semi-supervised or unsupervised machine learning methods to reveal interesting patterns in data

    While supervised learning is interesting and may be promising for many, many applications, we would hopefully like to move towards a direction along which we are learning from the data, and not teaching an algorithm how to interpret it. Using existing data and possibly data from an open-access database as described in the first item, my idea is to be able to use machine learning to extract significantly more information from high speed data than we currently can using humans and current algorithms to do it.

While most of my research in this topic has been on boiling heat transfer, I believe this technique can be extended to any complex, fluid-related phenomenon that has a significant amount of data available, such as turbulent transport and instability theory.

What is the best power cycle?